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 <asm/uaccess.h>
76 #include <trace/events/skb.h>
77 #include <linux/highmem.h>
78 #include <linux/capability.h>
79 #include <linux/user_namespace.h>
81 struct kmem_cache
*skbuff_head_cache __read_mostly
;
82 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
83 int sysctl_max_skb_frags __read_mostly
= MAX_SKB_FRAGS
;
84 EXPORT_SYMBOL(sysctl_max_skb_frags
);
87 * skb_panic - private function for out-of-line support
91 * @msg: skb_over_panic or skb_under_panic
93 * Out-of-line support for skb_put() and skb_push().
94 * Called via the wrapper skb_over_panic() or skb_under_panic().
95 * Keep out of line to prevent kernel bloat.
96 * __builtin_return_address is not used because it is not always reliable.
98 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
101 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
103 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
104 skb
->dev
? skb
->dev
->name
: "<NULL>");
108 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
110 skb_panic(skb
, sz
, addr
, __func__
);
113 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
115 skb_panic(skb
, sz
, addr
, __func__
);
119 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120 * the caller if emergency pfmemalloc reserves are being used. If it is and
121 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122 * may be used. Otherwise, the packet data may be discarded until enough
125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
128 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
129 unsigned long ip
, bool *pfmemalloc
)
132 bool ret_pfmemalloc
= false;
135 * Try a regular allocation, when that fails and we're not entitled
136 * to the reserves, fail.
138 obj
= kmalloc_node_track_caller(size
,
139 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
141 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
144 /* Try again but now we are using pfmemalloc reserves */
145 ret_pfmemalloc
= true;
146 obj
= kmalloc_node_track_caller(size
, flags
, node
);
150 *pfmemalloc
= ret_pfmemalloc
;
155 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
156 * 'private' fields and also do memory statistics to find all the
161 struct sk_buff
*__alloc_skb_head(gfp_t gfp_mask
, int node
)
166 skb
= kmem_cache_alloc_node(skbuff_head_cache
,
167 gfp_mask
& ~__GFP_DMA
, node
);
172 * Only clear those fields we need to clear, not those that we will
173 * actually initialise below. Hence, don't put any more fields after
174 * the tail pointer in struct sk_buff!
176 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
178 skb
->truesize
= sizeof(struct sk_buff
);
179 atomic_set(&skb
->users
, 1);
181 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
187 * __alloc_skb - allocate a network buffer
188 * @size: size to allocate
189 * @gfp_mask: allocation mask
190 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191 * instead of head cache and allocate a cloned (child) skb.
192 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193 * allocations in case the data is required for writeback
194 * @node: numa node to allocate memory on
196 * Allocate a new &sk_buff. The returned buffer has no headroom and a
197 * tail room of at least size bytes. The object has a reference count
198 * of one. The return is the buffer. On a failure the return is %NULL.
200 * Buffers may only be allocated from interrupts using a @gfp_mask of
203 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
206 struct kmem_cache
*cache
;
207 struct skb_shared_info
*shinfo
;
212 cache
= (flags
& SKB_ALLOC_FCLONE
)
213 ? skbuff_fclone_cache
: skbuff_head_cache
;
215 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
216 gfp_mask
|= __GFP_MEMALLOC
;
219 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
224 /* We do our best to align skb_shared_info on a separate cache
225 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227 * Both skb->head and skb_shared_info are cache line aligned.
229 size
= SKB_DATA_ALIGN(size
);
230 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
231 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
234 /* kmalloc(size) might give us more room than requested.
235 * Put skb_shared_info exactly at the end of allocated zone,
236 * to allow max possible filling before reallocation.
238 size
= SKB_WITH_OVERHEAD(ksize(data
));
239 prefetchw(data
+ size
);
242 * Only clear those fields we need to clear, not those that we will
243 * actually initialise below. Hence, don't put any more fields after
244 * the tail pointer in struct sk_buff!
246 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
247 /* Account for allocated memory : skb + skb->head */
248 skb
->truesize
= SKB_TRUESIZE(size
);
249 skb
->pfmemalloc
= pfmemalloc
;
250 atomic_set(&skb
->users
, 1);
253 skb_reset_tail_pointer(skb
);
254 skb
->end
= skb
->tail
+ size
;
255 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
256 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
258 /* make sure we initialize shinfo sequentially */
259 shinfo
= skb_shinfo(skb
);
260 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
261 atomic_set(&shinfo
->dataref
, 1);
262 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
264 if (flags
& SKB_ALLOC_FCLONE
) {
265 struct sk_buff_fclones
*fclones
;
267 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
269 kmemcheck_annotate_bitfield(&fclones
->skb2
, flags1
);
270 skb
->fclone
= SKB_FCLONE_ORIG
;
271 atomic_set(&fclones
->fclone_ref
, 1);
273 fclones
->skb2
.fclone
= SKB_FCLONE_CLONE
;
274 fclones
->skb2
.pfmemalloc
= pfmemalloc
;
279 kmem_cache_free(cache
, skb
);
283 EXPORT_SYMBOL(__alloc_skb
);
286 * __build_skb - build a network buffer
287 * @data: data buffer provided by caller
288 * @frag_size: size of data, or 0 if head was kmalloced
290 * Allocate a new &sk_buff. Caller provides space holding head and
291 * skb_shared_info. @data must have been allocated by kmalloc() only if
292 * @frag_size is 0, otherwise data should come from the page allocator
294 * The return is the new skb buffer.
295 * On a failure the return is %NULL, and @data is not freed.
297 * Before IO, driver allocates only data buffer where NIC put incoming frame
298 * Driver should add room at head (NET_SKB_PAD) and
299 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
300 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
301 * before giving packet to stack.
302 * RX rings only contains data buffers, not full skbs.
304 struct sk_buff
*__build_skb(void *data
, unsigned int frag_size
)
306 struct skb_shared_info
*shinfo
;
308 unsigned int size
= frag_size
? : ksize(data
);
310 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
314 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
316 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
317 skb
->truesize
= SKB_TRUESIZE(size
);
318 atomic_set(&skb
->users
, 1);
321 skb_reset_tail_pointer(skb
);
322 skb
->end
= skb
->tail
+ size
;
323 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
324 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
326 /* make sure we initialize shinfo sequentially */
327 shinfo
= skb_shinfo(skb
);
328 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
329 atomic_set(&shinfo
->dataref
, 1);
330 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
335 /* build_skb() is wrapper over __build_skb(), that specifically
336 * takes care of skb->head and skb->pfmemalloc
337 * This means that if @frag_size is not zero, then @data must be backed
338 * by a page fragment, not kmalloc() or vmalloc()
340 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
342 struct sk_buff
*skb
= __build_skb(data
, frag_size
);
344 if (skb
&& frag_size
) {
346 if (page_is_pfmemalloc(virt_to_head_page(data
)))
351 EXPORT_SYMBOL(build_skb
);
353 #define NAPI_SKB_CACHE_SIZE 64
355 struct napi_alloc_cache
{
356 struct page_frag_cache page
;
358 void *skb_cache
[NAPI_SKB_CACHE_SIZE
];
361 static DEFINE_PER_CPU(struct page_frag_cache
, netdev_alloc_cache
);
362 static DEFINE_PER_CPU(struct napi_alloc_cache
, napi_alloc_cache
);
364 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
366 struct page_frag_cache
*nc
;
370 local_irq_save(flags
);
371 nc
= this_cpu_ptr(&netdev_alloc_cache
);
372 data
= __alloc_page_frag(nc
, fragsz
, gfp_mask
);
373 local_irq_restore(flags
);
378 * netdev_alloc_frag - allocate a page fragment
379 * @fragsz: fragment size
381 * Allocates a frag from a page for receive buffer.
382 * Uses GFP_ATOMIC allocations.
384 void *netdev_alloc_frag(unsigned int fragsz
)
386 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
388 EXPORT_SYMBOL(netdev_alloc_frag
);
390 static void *__napi_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
392 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
394 return __alloc_page_frag(&nc
->page
, fragsz
, gfp_mask
);
397 void *napi_alloc_frag(unsigned int fragsz
)
399 return __napi_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
401 EXPORT_SYMBOL(napi_alloc_frag
);
404 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
405 * @dev: network device to receive on
406 * @len: length to allocate
407 * @gfp_mask: get_free_pages mask, passed to alloc_skb
409 * Allocate a new &sk_buff and assign it a usage count of one. The
410 * buffer has NET_SKB_PAD headroom built in. Users should allocate
411 * the headroom they think they need without accounting for the
412 * built in space. The built in space is used for optimisations.
414 * %NULL is returned if there is no free memory.
416 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
, unsigned int len
,
419 struct page_frag_cache
*nc
;
427 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
428 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
429 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
435 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
436 len
= SKB_DATA_ALIGN(len
);
438 if (sk_memalloc_socks())
439 gfp_mask
|= __GFP_MEMALLOC
;
441 local_irq_save(flags
);
443 nc
= this_cpu_ptr(&netdev_alloc_cache
);
444 data
= __alloc_page_frag(nc
, len
, gfp_mask
);
445 pfmemalloc
= nc
->pfmemalloc
;
447 local_irq_restore(flags
);
452 skb
= __build_skb(data
, len
);
453 if (unlikely(!skb
)) {
458 /* use OR instead of assignment to avoid clearing of bits in mask */
464 skb_reserve(skb
, NET_SKB_PAD
);
470 EXPORT_SYMBOL(__netdev_alloc_skb
);
473 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
474 * @napi: napi instance this buffer was allocated for
475 * @len: length to allocate
476 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
478 * Allocate a new sk_buff for use in NAPI receive. This buffer will
479 * attempt to allocate the head from a special reserved region used
480 * only for NAPI Rx allocation. By doing this we can save several
481 * CPU cycles by avoiding having to disable and re-enable IRQs.
483 * %NULL is returned if there is no free memory.
485 struct sk_buff
*__napi_alloc_skb(struct napi_struct
*napi
, unsigned int len
,
488 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
492 len
+= NET_SKB_PAD
+ NET_IP_ALIGN
;
494 if ((len
> SKB_WITH_OVERHEAD(PAGE_SIZE
)) ||
495 (gfp_mask
& (__GFP_DIRECT_RECLAIM
| GFP_DMA
))) {
496 skb
= __alloc_skb(len
, gfp_mask
, SKB_ALLOC_RX
, NUMA_NO_NODE
);
502 len
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
503 len
= SKB_DATA_ALIGN(len
);
505 if (sk_memalloc_socks())
506 gfp_mask
|= __GFP_MEMALLOC
;
508 data
= __alloc_page_frag(&nc
->page
, len
, gfp_mask
);
512 skb
= __build_skb(data
, len
);
513 if (unlikely(!skb
)) {
518 /* use OR instead of assignment to avoid clearing of bits in mask */
519 if (nc
->page
.pfmemalloc
)
524 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
);
525 skb
->dev
= napi
->dev
;
530 EXPORT_SYMBOL(__napi_alloc_skb
);
532 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
533 int size
, unsigned int truesize
)
535 skb_fill_page_desc(skb
, i
, page
, off
, size
);
537 skb
->data_len
+= size
;
538 skb
->truesize
+= truesize
;
540 EXPORT_SYMBOL(skb_add_rx_frag
);
542 void skb_coalesce_rx_frag(struct sk_buff
*skb
, int i
, int size
,
543 unsigned int truesize
)
545 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
547 skb_frag_size_add(frag
, size
);
549 skb
->data_len
+= size
;
550 skb
->truesize
+= truesize
;
552 EXPORT_SYMBOL(skb_coalesce_rx_frag
);
554 static void skb_drop_list(struct sk_buff
**listp
)
556 kfree_skb_list(*listp
);
560 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
562 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
565 static void skb_clone_fraglist(struct sk_buff
*skb
)
567 struct sk_buff
*list
;
569 skb_walk_frags(skb
, list
)
573 static void skb_free_head(struct sk_buff
*skb
)
575 unsigned char *head
= skb
->head
;
583 static void skb_release_data(struct sk_buff
*skb
)
585 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
589 atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
593 for (i
= 0; i
< shinfo
->nr_frags
; i
++)
594 __skb_frag_unref(&shinfo
->frags
[i
]);
597 * If skb buf is from userspace, we need to notify the caller
598 * the lower device DMA has done;
600 if (shinfo
->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
601 struct ubuf_info
*uarg
;
603 uarg
= shinfo
->destructor_arg
;
605 uarg
->callback(uarg
, true);
608 if (shinfo
->frag_list
)
609 kfree_skb_list(shinfo
->frag_list
);
615 * Free an skbuff by memory without cleaning the state.
617 static void kfree_skbmem(struct sk_buff
*skb
)
619 struct sk_buff_fclones
*fclones
;
621 switch (skb
->fclone
) {
622 case SKB_FCLONE_UNAVAILABLE
:
623 kmem_cache_free(skbuff_head_cache
, skb
);
626 case SKB_FCLONE_ORIG
:
627 fclones
= container_of(skb
, struct sk_buff_fclones
, skb1
);
629 /* We usually free the clone (TX completion) before original skb
630 * This test would have no chance to be true for the clone,
631 * while here, branch prediction will be good.
633 if (atomic_read(&fclones
->fclone_ref
) == 1)
637 default: /* SKB_FCLONE_CLONE */
638 fclones
= container_of(skb
, struct sk_buff_fclones
, skb2
);
641 if (!atomic_dec_and_test(&fclones
->fclone_ref
))
644 kmem_cache_free(skbuff_fclone_cache
, fclones
);
647 static void skb_release_head_state(struct sk_buff
*skb
)
651 secpath_put(skb
->sp
);
653 if (skb
->destructor
) {
655 skb
->destructor(skb
);
657 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
658 nf_conntrack_put(skb
->nfct
);
660 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
661 nf_bridge_put(skb
->nf_bridge
);
665 /* Free everything but the sk_buff shell. */
666 static void skb_release_all(struct sk_buff
*skb
)
668 skb_release_head_state(skb
);
669 if (likely(skb
->head
))
670 skb_release_data(skb
);
674 * __kfree_skb - private function
677 * Free an sk_buff. Release anything attached to the buffer.
678 * Clean the state. This is an internal helper function. Users should
679 * always call kfree_skb
682 void __kfree_skb(struct sk_buff
*skb
)
684 skb_release_all(skb
);
687 EXPORT_SYMBOL(__kfree_skb
);
690 * kfree_skb - free an sk_buff
691 * @skb: buffer to free
693 * Drop a reference to the buffer and free it if the usage count has
696 void kfree_skb(struct sk_buff
*skb
)
700 if (likely(atomic_read(&skb
->users
) == 1))
702 else if (likely(!atomic_dec_and_test(&skb
->users
)))
704 trace_kfree_skb(skb
, __builtin_return_address(0));
707 EXPORT_SYMBOL(kfree_skb
);
709 void kfree_skb_list(struct sk_buff
*segs
)
712 struct sk_buff
*next
= segs
->next
;
718 EXPORT_SYMBOL(kfree_skb_list
);
721 * skb_tx_error - report an sk_buff xmit error
722 * @skb: buffer that triggered an error
724 * Report xmit error if a device callback is tracking this skb.
725 * skb must be freed afterwards.
727 void skb_tx_error(struct sk_buff
*skb
)
729 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
730 struct ubuf_info
*uarg
;
732 uarg
= skb_shinfo(skb
)->destructor_arg
;
734 uarg
->callback(uarg
, false);
735 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
738 EXPORT_SYMBOL(skb_tx_error
);
741 * consume_skb - free an skbuff
742 * @skb: buffer to free
744 * Drop a ref to the buffer and free it if the usage count has hit zero
745 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
746 * is being dropped after a failure and notes that
748 void consume_skb(struct sk_buff
*skb
)
752 if (likely(atomic_read(&skb
->users
) == 1))
754 else if (likely(!atomic_dec_and_test(&skb
->users
)))
756 trace_consume_skb(skb
);
759 EXPORT_SYMBOL(consume_skb
);
761 void __kfree_skb_flush(void)
763 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
765 /* flush skb_cache if containing objects */
767 kmem_cache_free_bulk(skbuff_head_cache
, nc
->skb_count
,
773 static inline void _kfree_skb_defer(struct sk_buff
*skb
)
775 struct napi_alloc_cache
*nc
= this_cpu_ptr(&napi_alloc_cache
);
777 /* drop skb->head and call any destructors for packet */
778 skb_release_all(skb
);
780 /* record skb to CPU local list */
781 nc
->skb_cache
[nc
->skb_count
++] = skb
;
784 /* SLUB writes into objects when freeing */
788 /* flush skb_cache if it is filled */
789 if (unlikely(nc
->skb_count
== NAPI_SKB_CACHE_SIZE
)) {
790 kmem_cache_free_bulk(skbuff_head_cache
, NAPI_SKB_CACHE_SIZE
,
795 void __kfree_skb_defer(struct sk_buff
*skb
)
797 _kfree_skb_defer(skb
);
800 void napi_consume_skb(struct sk_buff
*skb
, int budget
)
805 /* Zero budget indicate non-NAPI context called us, like netpoll */
806 if (unlikely(!budget
)) {
807 dev_consume_skb_any(skb
);
811 if (likely(atomic_read(&skb
->users
) == 1))
813 else if (likely(!atomic_dec_and_test(&skb
->users
)))
815 /* if reaching here SKB is ready to free */
816 trace_consume_skb(skb
);
818 /* if SKB is a clone, don't handle this case */
819 if (skb
->fclone
!= SKB_FCLONE_UNAVAILABLE
) {
824 _kfree_skb_defer(skb
);
826 EXPORT_SYMBOL(napi_consume_skb
);
828 /* Make sure a field is enclosed inside headers_start/headers_end section */
829 #define CHECK_SKB_FIELD(field) \
830 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
831 offsetof(struct sk_buff, headers_start)); \
832 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
833 offsetof(struct sk_buff, headers_end)); \
835 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
837 new->tstamp
= old
->tstamp
;
838 /* We do not copy old->sk */
840 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
841 skb_dst_copy(new, old
);
843 new->sp
= secpath_get(old
->sp
);
845 __nf_copy(new, old
, false);
847 /* Note : this field could be in headers_start/headers_end section
848 * It is not yet because we do not want to have a 16 bit hole
850 new->queue_mapping
= old
->queue_mapping
;
852 memcpy(&new->headers_start
, &old
->headers_start
,
853 offsetof(struct sk_buff
, headers_end
) -
854 offsetof(struct sk_buff
, headers_start
));
855 CHECK_SKB_FIELD(protocol
);
856 CHECK_SKB_FIELD(csum
);
857 CHECK_SKB_FIELD(hash
);
858 CHECK_SKB_FIELD(priority
);
859 CHECK_SKB_FIELD(skb_iif
);
860 CHECK_SKB_FIELD(vlan_proto
);
861 CHECK_SKB_FIELD(vlan_tci
);
862 CHECK_SKB_FIELD(transport_header
);
863 CHECK_SKB_FIELD(network_header
);
864 CHECK_SKB_FIELD(mac_header
);
865 CHECK_SKB_FIELD(inner_protocol
);
866 CHECK_SKB_FIELD(inner_transport_header
);
867 CHECK_SKB_FIELD(inner_network_header
);
868 CHECK_SKB_FIELD(inner_mac_header
);
869 CHECK_SKB_FIELD(mark
);
870 #ifdef CONFIG_NETWORK_SECMARK
871 CHECK_SKB_FIELD(secmark
);
873 #ifdef CONFIG_NET_RX_BUSY_POLL
874 CHECK_SKB_FIELD(napi_id
);
877 CHECK_SKB_FIELD(sender_cpu
);
879 #ifdef CONFIG_NET_SCHED
880 CHECK_SKB_FIELD(tc_index
);
881 #ifdef CONFIG_NET_CLS_ACT
882 CHECK_SKB_FIELD(tc_verd
);
889 * You should not add any new code to this function. Add it to
890 * __copy_skb_header above instead.
892 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
894 #define C(x) n->x = skb->x
896 n
->next
= n
->prev
= NULL
;
898 __copy_skb_header(n
, skb
);
903 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
906 n
->destructor
= NULL
;
913 atomic_set(&n
->users
, 1);
915 atomic_inc(&(skb_shinfo(skb
)->dataref
));
923 * skb_morph - morph one skb into another
924 * @dst: the skb to receive the contents
925 * @src: the skb to supply the contents
927 * This is identical to skb_clone except that the target skb is
928 * supplied by the user.
930 * The target skb is returned upon exit.
932 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
934 skb_release_all(dst
);
935 return __skb_clone(dst
, src
);
937 EXPORT_SYMBOL_GPL(skb_morph
);
940 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
941 * @skb: the skb to modify
942 * @gfp_mask: allocation priority
944 * This must be called on SKBTX_DEV_ZEROCOPY skb.
945 * It will copy all frags into kernel and drop the reference
946 * to userspace pages.
948 * If this function is called from an interrupt gfp_mask() must be
951 * Returns 0 on success or a negative error code on failure
952 * to allocate kernel memory to copy to.
954 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
957 int num_frags
= skb_shinfo(skb
)->nr_frags
;
958 struct page
*page
, *head
= NULL
;
959 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
961 for (i
= 0; i
< num_frags
; i
++) {
963 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
965 page
= alloc_page(gfp_mask
);
968 struct page
*next
= (struct page
*)page_private(head
);
974 vaddr
= kmap_atomic(skb_frag_page(f
));
975 memcpy(page_address(page
),
976 vaddr
+ f
->page_offset
, skb_frag_size(f
));
977 kunmap_atomic(vaddr
);
978 set_page_private(page
, (unsigned long)head
);
982 /* skb frags release userspace buffers */
983 for (i
= 0; i
< num_frags
; i
++)
984 skb_frag_unref(skb
, i
);
986 uarg
->callback(uarg
, false);
988 /* skb frags point to kernel buffers */
989 for (i
= num_frags
- 1; i
>= 0; i
--) {
990 __skb_fill_page_desc(skb
, i
, head
, 0,
991 skb_shinfo(skb
)->frags
[i
].size
);
992 head
= (struct page
*)page_private(head
);
995 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
998 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
1001 * skb_clone - duplicate an sk_buff
1002 * @skb: buffer to clone
1003 * @gfp_mask: allocation priority
1005 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1006 * copies share the same packet data but not structure. The new
1007 * buffer has a reference count of 1. If the allocation fails the
1008 * function returns %NULL otherwise the new buffer is returned.
1010 * If this function is called from an interrupt gfp_mask() must be
1014 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
1016 struct sk_buff_fclones
*fclones
= container_of(skb
,
1017 struct sk_buff_fclones
,
1021 if (skb_orphan_frags(skb
, gfp_mask
))
1024 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
1025 atomic_read(&fclones
->fclone_ref
) == 1) {
1027 atomic_set(&fclones
->fclone_ref
, 2);
1029 if (skb_pfmemalloc(skb
))
1030 gfp_mask
|= __GFP_MEMALLOC
;
1032 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
1036 kmemcheck_annotate_bitfield(n
, flags1
);
1037 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
1040 return __skb_clone(n
, skb
);
1042 EXPORT_SYMBOL(skb_clone
);
1044 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
1046 /* Only adjust this if it actually is csum_start rather than csum */
1047 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1048 skb
->csum_start
+= off
;
1049 /* {transport,network,mac}_header and tail are relative to skb->head */
1050 skb
->transport_header
+= off
;
1051 skb
->network_header
+= off
;
1052 if (skb_mac_header_was_set(skb
))
1053 skb
->mac_header
+= off
;
1054 skb
->inner_transport_header
+= off
;
1055 skb
->inner_network_header
+= off
;
1056 skb
->inner_mac_header
+= off
;
1059 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
1061 __copy_skb_header(new, old
);
1063 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
1064 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
1065 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
1068 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
1070 if (skb_pfmemalloc(skb
))
1071 return SKB_ALLOC_RX
;
1076 * skb_copy - create private copy of an sk_buff
1077 * @skb: buffer to copy
1078 * @gfp_mask: allocation priority
1080 * Make a copy of both an &sk_buff and its data. This is used when the
1081 * caller wishes to modify the data and needs a private copy of the
1082 * data to alter. Returns %NULL on failure or the pointer to the buffer
1083 * on success. The returned buffer has a reference count of 1.
1085 * As by-product this function converts non-linear &sk_buff to linear
1086 * one, so that &sk_buff becomes completely private and caller is allowed
1087 * to modify all the data of returned buffer. This means that this
1088 * function is not recommended for use in circumstances when only
1089 * header is going to be modified. Use pskb_copy() instead.
1092 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
1094 int headerlen
= skb_headroom(skb
);
1095 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
1096 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
1097 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
1102 /* Set the data pointer */
1103 skb_reserve(n
, headerlen
);
1104 /* Set the tail pointer and length */
1105 skb_put(n
, skb
->len
);
1107 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
1110 copy_skb_header(n
, skb
);
1113 EXPORT_SYMBOL(skb_copy
);
1116 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1117 * @skb: buffer to copy
1118 * @headroom: headroom of new skb
1119 * @gfp_mask: allocation priority
1120 * @fclone: if true allocate the copy of the skb from the fclone
1121 * cache instead of the head cache; it is recommended to set this
1122 * to true for the cases where the copy will likely be cloned
1124 * Make a copy of both an &sk_buff and part of its data, located
1125 * in header. Fragmented data remain shared. This is used when
1126 * the caller wishes to modify only header of &sk_buff and needs
1127 * private copy of the header to alter. Returns %NULL on failure
1128 * or the pointer to the buffer on success.
1129 * The returned buffer has a reference count of 1.
1132 struct sk_buff
*__pskb_copy_fclone(struct sk_buff
*skb
, int headroom
,
1133 gfp_t gfp_mask
, bool fclone
)
1135 unsigned int size
= skb_headlen(skb
) + headroom
;
1136 int flags
= skb_alloc_rx_flag(skb
) | (fclone
? SKB_ALLOC_FCLONE
: 0);
1137 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
, flags
, NUMA_NO_NODE
);
1142 /* Set the data pointer */
1143 skb_reserve(n
, headroom
);
1144 /* Set the tail pointer and length */
1145 skb_put(n
, skb_headlen(skb
));
1146 /* Copy the bytes */
1147 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
1149 n
->truesize
+= skb
->data_len
;
1150 n
->data_len
= skb
->data_len
;
1153 if (skb_shinfo(skb
)->nr_frags
) {
1156 if (skb_orphan_frags(skb
, gfp_mask
)) {
1161 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1162 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1163 skb_frag_ref(skb
, i
);
1165 skb_shinfo(n
)->nr_frags
= i
;
1168 if (skb_has_frag_list(skb
)) {
1169 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1170 skb_clone_fraglist(n
);
1173 copy_skb_header(n
, skb
);
1177 EXPORT_SYMBOL(__pskb_copy_fclone
);
1180 * pskb_expand_head - reallocate header of &sk_buff
1181 * @skb: buffer to reallocate
1182 * @nhead: room to add at head
1183 * @ntail: room to add at tail
1184 * @gfp_mask: allocation priority
1186 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1187 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1188 * reference count of 1. Returns zero in the case of success or error,
1189 * if expansion failed. In the last case, &sk_buff is not changed.
1191 * All the pointers pointing into skb header may change and must be
1192 * reloaded after call to this function.
1195 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1200 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1205 if (skb_shared(skb
))
1208 size
= SKB_DATA_ALIGN(size
);
1210 if (skb_pfmemalloc(skb
))
1211 gfp_mask
|= __GFP_MEMALLOC
;
1212 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1213 gfp_mask
, NUMA_NO_NODE
, NULL
);
1216 size
= SKB_WITH_OVERHEAD(ksize(data
));
1218 /* Copy only real data... and, alas, header. This should be
1219 * optimized for the cases when header is void.
1221 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1223 memcpy((struct skb_shared_info
*)(data
+ size
),
1225 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1228 * if shinfo is shared we must drop the old head gracefully, but if it
1229 * is not we can just drop the old head and let the existing refcount
1230 * be since all we did is relocate the values
1232 if (skb_cloned(skb
)) {
1233 /* copy this zero copy skb frags */
1234 if (skb_orphan_frags(skb
, gfp_mask
))
1236 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1237 skb_frag_ref(skb
, i
);
1239 if (skb_has_frag_list(skb
))
1240 skb_clone_fraglist(skb
);
1242 skb_release_data(skb
);
1246 off
= (data
+ nhead
) - skb
->head
;
1251 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1255 skb
->end
= skb
->head
+ size
;
1258 skb_headers_offset_update(skb
, nhead
);
1262 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1270 EXPORT_SYMBOL(pskb_expand_head
);
1272 /* Make private copy of skb with writable head and some headroom */
1274 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1276 struct sk_buff
*skb2
;
1277 int delta
= headroom
- skb_headroom(skb
);
1280 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1282 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1283 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1291 EXPORT_SYMBOL(skb_realloc_headroom
);
1294 * skb_copy_expand - copy and expand sk_buff
1295 * @skb: buffer to copy
1296 * @newheadroom: new free bytes at head
1297 * @newtailroom: new free bytes at tail
1298 * @gfp_mask: allocation priority
1300 * Make a copy of both an &sk_buff and its data and while doing so
1301 * allocate additional space.
1303 * This is used when the caller wishes to modify the data and needs a
1304 * private copy of the data to alter as well as more space for new fields.
1305 * Returns %NULL on failure or the pointer to the buffer
1306 * on success. The returned buffer has a reference count of 1.
1308 * You must pass %GFP_ATOMIC as the allocation priority if this function
1309 * is called from an interrupt.
1311 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1312 int newheadroom
, int newtailroom
,
1316 * Allocate the copy buffer
1318 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1319 gfp_mask
, skb_alloc_rx_flag(skb
),
1321 int oldheadroom
= skb_headroom(skb
);
1322 int head_copy_len
, head_copy_off
;
1327 skb_reserve(n
, newheadroom
);
1329 /* Set the tail pointer and length */
1330 skb_put(n
, skb
->len
);
1332 head_copy_len
= oldheadroom
;
1334 if (newheadroom
<= head_copy_len
)
1335 head_copy_len
= newheadroom
;
1337 head_copy_off
= newheadroom
- head_copy_len
;
1339 /* Copy the linear header and data. */
1340 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1341 skb
->len
+ head_copy_len
))
1344 copy_skb_header(n
, skb
);
1346 skb_headers_offset_update(n
, newheadroom
- oldheadroom
);
1350 EXPORT_SYMBOL(skb_copy_expand
);
1353 * skb_pad - zero pad the tail of an skb
1354 * @skb: buffer to pad
1355 * @pad: space to pad
1357 * Ensure that a buffer is followed by a padding area that is zero
1358 * filled. Used by network drivers which may DMA or transfer data
1359 * beyond the buffer end onto the wire.
1361 * May return error in out of memory cases. The skb is freed on error.
1364 int skb_pad(struct sk_buff
*skb
, int pad
)
1369 /* If the skbuff is non linear tailroom is always zero.. */
1370 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1371 memset(skb
->data
+skb
->len
, 0, pad
);
1375 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1376 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1377 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1382 /* FIXME: The use of this function with non-linear skb's really needs
1385 err
= skb_linearize(skb
);
1389 memset(skb
->data
+ skb
->len
, 0, pad
);
1396 EXPORT_SYMBOL(skb_pad
);
1399 * pskb_put - add data to the tail of a potentially fragmented buffer
1400 * @skb: start of the buffer to use
1401 * @tail: tail fragment of the buffer to use
1402 * @len: amount of data to add
1404 * This function extends the used data area of the potentially
1405 * fragmented buffer. @tail must be the last fragment of @skb -- or
1406 * @skb itself. If this would exceed the total buffer size the kernel
1407 * will panic. A pointer to the first byte of the extra data is
1411 unsigned char *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
)
1414 skb
->data_len
+= len
;
1417 return skb_put(tail
, len
);
1419 EXPORT_SYMBOL_GPL(pskb_put
);
1422 * skb_put - add data to a buffer
1423 * @skb: buffer to use
1424 * @len: amount of data to add
1426 * This function extends the used data area of the buffer. If this would
1427 * exceed the total buffer size the kernel will panic. A pointer to the
1428 * first byte of the extra data is returned.
1430 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1432 unsigned char *tmp
= skb_tail_pointer(skb
);
1433 SKB_LINEAR_ASSERT(skb
);
1436 if (unlikely(skb
->tail
> skb
->end
))
1437 skb_over_panic(skb
, len
, __builtin_return_address(0));
1440 EXPORT_SYMBOL(skb_put
);
1443 * skb_push - add data to the start of a buffer
1444 * @skb: buffer to use
1445 * @len: amount of data to add
1447 * This function extends the used data area of the buffer at the buffer
1448 * start. If this would exceed the total buffer headroom the kernel will
1449 * panic. A pointer to the first byte of the extra data is returned.
1451 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1455 if (unlikely(skb
->data
<skb
->head
))
1456 skb_under_panic(skb
, len
, __builtin_return_address(0));
1459 EXPORT_SYMBOL(skb_push
);
1462 * skb_pull - remove data from the start of a buffer
1463 * @skb: buffer to use
1464 * @len: amount of data to remove
1466 * This function removes data from the start of a buffer, returning
1467 * the memory to the headroom. A pointer to the next data in the buffer
1468 * is returned. Once the data has been pulled future pushes will overwrite
1471 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1473 return skb_pull_inline(skb
, len
);
1475 EXPORT_SYMBOL(skb_pull
);
1478 * skb_trim - remove end from a buffer
1479 * @skb: buffer to alter
1482 * Cut the length of a buffer down by removing data from the tail. If
1483 * the buffer is already under the length specified it is not modified.
1484 * The skb must be linear.
1486 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1489 __skb_trim(skb
, len
);
1491 EXPORT_SYMBOL(skb_trim
);
1493 /* Trims skb to length len. It can change skb pointers.
1496 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1498 struct sk_buff
**fragp
;
1499 struct sk_buff
*frag
;
1500 int offset
= skb_headlen(skb
);
1501 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1505 if (skb_cloned(skb
) &&
1506 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1513 for (; i
< nfrags
; i
++) {
1514 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1521 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1524 skb_shinfo(skb
)->nr_frags
= i
;
1526 for (; i
< nfrags
; i
++)
1527 skb_frag_unref(skb
, i
);
1529 if (skb_has_frag_list(skb
))
1530 skb_drop_fraglist(skb
);
1534 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1535 fragp
= &frag
->next
) {
1536 int end
= offset
+ frag
->len
;
1538 if (skb_shared(frag
)) {
1539 struct sk_buff
*nfrag
;
1541 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1542 if (unlikely(!nfrag
))
1545 nfrag
->next
= frag
->next
;
1557 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1561 skb_drop_list(&frag
->next
);
1566 if (len
> skb_headlen(skb
)) {
1567 skb
->data_len
-= skb
->len
- len
;
1572 skb_set_tail_pointer(skb
, len
);
1577 EXPORT_SYMBOL(___pskb_trim
);
1580 * __pskb_pull_tail - advance tail of skb header
1581 * @skb: buffer to reallocate
1582 * @delta: number of bytes to advance tail
1584 * The function makes a sense only on a fragmented &sk_buff,
1585 * it expands header moving its tail forward and copying necessary
1586 * data from fragmented part.
1588 * &sk_buff MUST have reference count of 1.
1590 * Returns %NULL (and &sk_buff does not change) if pull failed
1591 * or value of new tail of skb in the case of success.
1593 * All the pointers pointing into skb header may change and must be
1594 * reloaded after call to this function.
1597 /* Moves tail of skb head forward, copying data from fragmented part,
1598 * when it is necessary.
1599 * 1. It may fail due to malloc failure.
1600 * 2. It may change skb pointers.
1602 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1604 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1606 /* If skb has not enough free space at tail, get new one
1607 * plus 128 bytes for future expansions. If we have enough
1608 * room at tail, reallocate without expansion only if skb is cloned.
1610 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1612 if (eat
> 0 || skb_cloned(skb
)) {
1613 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1618 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1621 /* Optimization: no fragments, no reasons to preestimate
1622 * size of pulled pages. Superb.
1624 if (!skb_has_frag_list(skb
))
1627 /* Estimate size of pulled pages. */
1629 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1630 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1637 /* If we need update frag list, we are in troubles.
1638 * Certainly, it possible to add an offset to skb data,
1639 * but taking into account that pulling is expected to
1640 * be very rare operation, it is worth to fight against
1641 * further bloating skb head and crucify ourselves here instead.
1642 * Pure masohism, indeed. 8)8)
1645 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1646 struct sk_buff
*clone
= NULL
;
1647 struct sk_buff
*insp
= NULL
;
1652 if (list
->len
<= eat
) {
1653 /* Eaten as whole. */
1658 /* Eaten partially. */
1660 if (skb_shared(list
)) {
1661 /* Sucks! We need to fork list. :-( */
1662 clone
= skb_clone(list
, GFP_ATOMIC
);
1668 /* This may be pulled without
1672 if (!pskb_pull(list
, eat
)) {
1680 /* Free pulled out fragments. */
1681 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1682 skb_shinfo(skb
)->frag_list
= list
->next
;
1685 /* And insert new clone at head. */
1688 skb_shinfo(skb
)->frag_list
= clone
;
1691 /* Success! Now we may commit changes to skb data. */
1696 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1697 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1700 skb_frag_unref(skb
, i
);
1703 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1705 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1706 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1712 skb_shinfo(skb
)->nr_frags
= k
;
1715 skb
->data_len
-= delta
;
1717 return skb_tail_pointer(skb
);
1719 EXPORT_SYMBOL(__pskb_pull_tail
);
1722 * skb_copy_bits - copy bits from skb to kernel buffer
1724 * @offset: offset in source
1725 * @to: destination buffer
1726 * @len: number of bytes to copy
1728 * Copy the specified number of bytes from the source skb to the
1729 * destination buffer.
1732 * If its prototype is ever changed,
1733 * check arch/{*}/net/{*}.S files,
1734 * since it is called from BPF assembly code.
1736 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1738 int start
= skb_headlen(skb
);
1739 struct sk_buff
*frag_iter
;
1742 if (offset
> (int)skb
->len
- len
)
1746 if ((copy
= start
- offset
) > 0) {
1749 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1750 if ((len
-= copy
) == 0)
1756 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1758 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1760 WARN_ON(start
> offset
+ len
);
1762 end
= start
+ skb_frag_size(f
);
1763 if ((copy
= end
- offset
) > 0) {
1769 vaddr
= kmap_atomic(skb_frag_page(f
));
1771 vaddr
+ f
->page_offset
+ offset
- start
,
1773 kunmap_atomic(vaddr
);
1775 if ((len
-= copy
) == 0)
1783 skb_walk_frags(skb
, frag_iter
) {
1786 WARN_ON(start
> offset
+ len
);
1788 end
= start
+ frag_iter
->len
;
1789 if ((copy
= end
- offset
) > 0) {
1792 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1794 if ((len
-= copy
) == 0)
1808 EXPORT_SYMBOL(skb_copy_bits
);
1811 * Callback from splice_to_pipe(), if we need to release some pages
1812 * at the end of the spd in case we error'ed out in filling the pipe.
1814 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1816 put_page(spd
->pages
[i
]);
1819 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1820 unsigned int *offset
,
1823 struct page_frag
*pfrag
= sk_page_frag(sk
);
1825 if (!sk_page_frag_refill(sk
, pfrag
))
1828 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1830 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1831 page_address(page
) + *offset
, *len
);
1832 *offset
= pfrag
->offset
;
1833 pfrag
->offset
+= *len
;
1838 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1840 unsigned int offset
)
1842 return spd
->nr_pages
&&
1843 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1844 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1845 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1849 * Fill page/offset/length into spd, if it can hold more pages.
1851 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1852 struct pipe_inode_info
*pipe
, struct page
*page
,
1853 unsigned int *len
, unsigned int offset
,
1857 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1861 page
= linear_to_page(page
, len
, &offset
, sk
);
1865 if (spd_can_coalesce(spd
, page
, offset
)) {
1866 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1870 spd
->pages
[spd
->nr_pages
] = page
;
1871 spd
->partial
[spd
->nr_pages
].len
= *len
;
1872 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1878 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1879 unsigned int plen
, unsigned int *off
,
1881 struct splice_pipe_desc
*spd
, bool linear
,
1883 struct pipe_inode_info
*pipe
)
1888 /* skip this segment if already processed */
1894 /* ignore any bits we already processed */
1900 unsigned int flen
= min(*len
, plen
);
1902 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1908 } while (*len
&& plen
);
1914 * Map linear and fragment data from the skb to spd. It reports true if the
1915 * pipe is full or if we already spliced the requested length.
1917 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1918 unsigned int *offset
, unsigned int *len
,
1919 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1922 struct sk_buff
*iter
;
1924 /* map the linear part :
1925 * If skb->head_frag is set, this 'linear' part is backed by a
1926 * fragment, and if the head is not shared with any clones then
1927 * we can avoid a copy since we own the head portion of this page.
1929 if (__splice_segment(virt_to_page(skb
->data
),
1930 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1933 skb_head_is_locked(skb
),
1938 * then map the fragments
1940 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1941 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1943 if (__splice_segment(skb_frag_page(f
),
1944 f
->page_offset
, skb_frag_size(f
),
1945 offset
, len
, spd
, false, sk
, pipe
))
1949 skb_walk_frags(skb
, iter
) {
1950 if (*offset
>= iter
->len
) {
1951 *offset
-= iter
->len
;
1954 /* __skb_splice_bits() only fails if the output has no room
1955 * left, so no point in going over the frag_list for the error
1958 if (__skb_splice_bits(iter
, pipe
, offset
, len
, spd
, sk
))
1965 ssize_t
skb_socket_splice(struct sock
*sk
,
1966 struct pipe_inode_info
*pipe
,
1967 struct splice_pipe_desc
*spd
)
1971 /* Drop the socket lock, otherwise we have reverse
1972 * locking dependencies between sk_lock and i_mutex
1973 * here as compared to sendfile(). We enter here
1974 * with the socket lock held, and splice_to_pipe() will
1975 * grab the pipe inode lock. For sendfile() emulation,
1976 * we call into ->sendpage() with the i_mutex lock held
1977 * and networking will grab the socket lock.
1980 ret
= splice_to_pipe(pipe
, spd
);
1987 * Map data from the skb to a pipe. Should handle both the linear part,
1988 * the fragments, and the frag list.
1990 int skb_splice_bits(struct sk_buff
*skb
, struct sock
*sk
, unsigned int offset
,
1991 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1993 ssize_t (*splice_cb
)(struct sock
*,
1994 struct pipe_inode_info
*,
1995 struct splice_pipe_desc
*))
1997 struct partial_page partial
[MAX_SKB_FRAGS
];
1998 struct page
*pages
[MAX_SKB_FRAGS
];
1999 struct splice_pipe_desc spd
= {
2002 .nr_pages_max
= MAX_SKB_FRAGS
,
2004 .ops
= &nosteal_pipe_buf_ops
,
2005 .spd_release
= sock_spd_release
,
2009 __skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
);
2012 ret
= splice_cb(sk
, pipe
, &spd
);
2016 EXPORT_SYMBOL_GPL(skb_splice_bits
);
2019 * skb_store_bits - store bits from kernel buffer to skb
2020 * @skb: destination buffer
2021 * @offset: offset in destination
2022 * @from: source buffer
2023 * @len: number of bytes to copy
2025 * Copy the specified number of bytes from the source buffer to the
2026 * destination skb. This function handles all the messy bits of
2027 * traversing fragment lists and such.
2030 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
2032 int start
= skb_headlen(skb
);
2033 struct sk_buff
*frag_iter
;
2036 if (offset
> (int)skb
->len
- len
)
2039 if ((copy
= start
- offset
) > 0) {
2042 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
2043 if ((len
-= copy
) == 0)
2049 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2050 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2053 WARN_ON(start
> offset
+ len
);
2055 end
= start
+ skb_frag_size(frag
);
2056 if ((copy
= end
- offset
) > 0) {
2062 vaddr
= kmap_atomic(skb_frag_page(frag
));
2063 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
2065 kunmap_atomic(vaddr
);
2067 if ((len
-= copy
) == 0)
2075 skb_walk_frags(skb
, frag_iter
) {
2078 WARN_ON(start
> offset
+ len
);
2080 end
= start
+ frag_iter
->len
;
2081 if ((copy
= end
- offset
) > 0) {
2084 if (skb_store_bits(frag_iter
, offset
- start
,
2087 if ((len
-= copy
) == 0)
2100 EXPORT_SYMBOL(skb_store_bits
);
2102 /* Checksum skb data. */
2103 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2104 __wsum csum
, const struct skb_checksum_ops
*ops
)
2106 int start
= skb_headlen(skb
);
2107 int i
, copy
= start
- offset
;
2108 struct sk_buff
*frag_iter
;
2111 /* Checksum header. */
2115 csum
= ops
->update(skb
->data
+ offset
, copy
, csum
);
2116 if ((len
-= copy
) == 0)
2122 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2124 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2126 WARN_ON(start
> offset
+ len
);
2128 end
= start
+ skb_frag_size(frag
);
2129 if ((copy
= end
- offset
) > 0) {
2135 vaddr
= kmap_atomic(skb_frag_page(frag
));
2136 csum2
= ops
->update(vaddr
+ frag
->page_offset
+
2137 offset
- start
, copy
, 0);
2138 kunmap_atomic(vaddr
);
2139 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2148 skb_walk_frags(skb
, frag_iter
) {
2151 WARN_ON(start
> offset
+ len
);
2153 end
= start
+ frag_iter
->len
;
2154 if ((copy
= end
- offset
) > 0) {
2158 csum2
= __skb_checksum(frag_iter
, offset
- start
,
2160 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
2161 if ((len
-= copy
) == 0)
2172 EXPORT_SYMBOL(__skb_checksum
);
2174 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
2175 int len
, __wsum csum
)
2177 const struct skb_checksum_ops ops
= {
2178 .update
= csum_partial_ext
,
2179 .combine
= csum_block_add_ext
,
2182 return __skb_checksum(skb
, offset
, len
, csum
, &ops
);
2184 EXPORT_SYMBOL(skb_checksum
);
2186 /* Both of above in one bottle. */
2188 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2189 u8
*to
, int len
, __wsum csum
)
2191 int start
= skb_headlen(skb
);
2192 int i
, copy
= start
- offset
;
2193 struct sk_buff
*frag_iter
;
2200 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2202 if ((len
-= copy
) == 0)
2209 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2212 WARN_ON(start
> offset
+ len
);
2214 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2215 if ((copy
= end
- offset
) > 0) {
2218 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2222 vaddr
= kmap_atomic(skb_frag_page(frag
));
2223 csum2
= csum_partial_copy_nocheck(vaddr
+
2227 kunmap_atomic(vaddr
);
2228 csum
= csum_block_add(csum
, csum2
, pos
);
2238 skb_walk_frags(skb
, frag_iter
) {
2242 WARN_ON(start
> offset
+ len
);
2244 end
= start
+ frag_iter
->len
;
2245 if ((copy
= end
- offset
) > 0) {
2248 csum2
= skb_copy_and_csum_bits(frag_iter
,
2251 csum
= csum_block_add(csum
, csum2
, pos
);
2252 if ((len
-= copy
) == 0)
2263 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2266 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2267 * @from: source buffer
2269 * Calculates the amount of linear headroom needed in the 'to' skb passed
2270 * into skb_zerocopy().
2273 skb_zerocopy_headlen(const struct sk_buff
*from
)
2275 unsigned int hlen
= 0;
2277 if (!from
->head_frag
||
2278 skb_headlen(from
) < L1_CACHE_BYTES
||
2279 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
2280 hlen
= skb_headlen(from
);
2282 if (skb_has_frag_list(from
))
2287 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen
);
2290 * skb_zerocopy - Zero copy skb to skb
2291 * @to: destination buffer
2292 * @from: source buffer
2293 * @len: number of bytes to copy from source buffer
2294 * @hlen: size of linear headroom in destination buffer
2296 * Copies up to `len` bytes from `from` to `to` by creating references
2297 * to the frags in the source buffer.
2299 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2300 * headroom in the `to` buffer.
2303 * 0: everything is OK
2304 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2305 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2308 skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
, int len
, int hlen
)
2311 int plen
= 0; /* length of skb->head fragment */
2314 unsigned int offset
;
2316 BUG_ON(!from
->head_frag
&& !hlen
);
2318 /* dont bother with small payloads */
2319 if (len
<= skb_tailroom(to
))
2320 return skb_copy_bits(from
, 0, skb_put(to
, len
), len
);
2323 ret
= skb_copy_bits(from
, 0, skb_put(to
, hlen
), hlen
);
2328 plen
= min_t(int, skb_headlen(from
), len
);
2330 page
= virt_to_head_page(from
->head
);
2331 offset
= from
->data
- (unsigned char *)page_address(page
);
2332 __skb_fill_page_desc(to
, 0, page
, offset
, plen
);
2339 to
->truesize
+= len
+ plen
;
2340 to
->len
+= len
+ plen
;
2341 to
->data_len
+= len
+ plen
;
2343 if (unlikely(skb_orphan_frags(from
, GFP_ATOMIC
))) {
2348 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++) {
2351 skb_shinfo(to
)->frags
[j
] = skb_shinfo(from
)->frags
[i
];
2352 skb_shinfo(to
)->frags
[j
].size
= min_t(int, skb_shinfo(to
)->frags
[j
].size
, len
);
2353 len
-= skb_shinfo(to
)->frags
[j
].size
;
2354 skb_frag_ref(to
, j
);
2357 skb_shinfo(to
)->nr_frags
= j
;
2361 EXPORT_SYMBOL_GPL(skb_zerocopy
);
2363 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2368 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2369 csstart
= skb_checksum_start_offset(skb
);
2371 csstart
= skb_headlen(skb
);
2373 BUG_ON(csstart
> skb_headlen(skb
));
2375 skb_copy_from_linear_data(skb
, to
, csstart
);
2378 if (csstart
!= skb
->len
)
2379 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2380 skb
->len
- csstart
, 0);
2382 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2383 long csstuff
= csstart
+ skb
->csum_offset
;
2385 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2388 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2391 * skb_dequeue - remove from the head of the queue
2392 * @list: list to dequeue from
2394 * Remove the head of the list. The list lock is taken so the function
2395 * may be used safely with other locking list functions. The head item is
2396 * returned or %NULL if the list is empty.
2399 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2401 unsigned long flags
;
2402 struct sk_buff
*result
;
2404 spin_lock_irqsave(&list
->lock
, flags
);
2405 result
= __skb_dequeue(list
);
2406 spin_unlock_irqrestore(&list
->lock
, flags
);
2409 EXPORT_SYMBOL(skb_dequeue
);
2412 * skb_dequeue_tail - remove from the tail of the queue
2413 * @list: list to dequeue from
2415 * Remove the tail of the list. The list lock is taken so the function
2416 * may be used safely with other locking list functions. The tail item is
2417 * returned or %NULL if the list is empty.
2419 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2421 unsigned long flags
;
2422 struct sk_buff
*result
;
2424 spin_lock_irqsave(&list
->lock
, flags
);
2425 result
= __skb_dequeue_tail(list
);
2426 spin_unlock_irqrestore(&list
->lock
, flags
);
2429 EXPORT_SYMBOL(skb_dequeue_tail
);
2432 * skb_queue_purge - empty a list
2433 * @list: list to empty
2435 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2436 * the list and one reference dropped. This function takes the list
2437 * lock and is atomic with respect to other list locking functions.
2439 void skb_queue_purge(struct sk_buff_head
*list
)
2441 struct sk_buff
*skb
;
2442 while ((skb
= skb_dequeue(list
)) != NULL
)
2445 EXPORT_SYMBOL(skb_queue_purge
);
2448 * skb_rbtree_purge - empty a skb rbtree
2449 * @root: root of the rbtree to empty
2451 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2452 * the list and one reference dropped. This function does not take
2453 * any lock. Synchronization should be handled by the caller (e.g., TCP
2454 * out-of-order queue is protected by the socket lock).
2456 void skb_rbtree_purge(struct rb_root
*root
)
2458 struct sk_buff
*skb
, *next
;
2460 rbtree_postorder_for_each_entry_safe(skb
, next
, root
, rbnode
)
2467 * skb_queue_head - queue a buffer at the list head
2468 * @list: list to use
2469 * @newsk: buffer to queue
2471 * Queue a buffer at the start of the list. This function takes the
2472 * list lock and can be used safely with other locking &sk_buff functions
2475 * A buffer cannot be placed on two lists at the same time.
2477 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2479 unsigned long flags
;
2481 spin_lock_irqsave(&list
->lock
, flags
);
2482 __skb_queue_head(list
, newsk
);
2483 spin_unlock_irqrestore(&list
->lock
, flags
);
2485 EXPORT_SYMBOL(skb_queue_head
);
2488 * skb_queue_tail - queue a buffer at the list tail
2489 * @list: list to use
2490 * @newsk: buffer to queue
2492 * Queue a buffer at the tail of the list. This function takes the
2493 * list lock and can be used safely with other locking &sk_buff functions
2496 * A buffer cannot be placed on two lists at the same time.
2498 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2500 unsigned long flags
;
2502 spin_lock_irqsave(&list
->lock
, flags
);
2503 __skb_queue_tail(list
, newsk
);
2504 spin_unlock_irqrestore(&list
->lock
, flags
);
2506 EXPORT_SYMBOL(skb_queue_tail
);
2509 * skb_unlink - remove a buffer from a list
2510 * @skb: buffer to remove
2511 * @list: list to use
2513 * Remove a packet from a list. The list locks are taken and this
2514 * function is atomic with respect to other list locked calls
2516 * You must know what list the SKB is on.
2518 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2520 unsigned long flags
;
2522 spin_lock_irqsave(&list
->lock
, flags
);
2523 __skb_unlink(skb
, list
);
2524 spin_unlock_irqrestore(&list
->lock
, flags
);
2526 EXPORT_SYMBOL(skb_unlink
);
2529 * skb_append - append a buffer
2530 * @old: buffer to insert after
2531 * @newsk: buffer to insert
2532 * @list: list to use
2534 * Place a packet after a given packet in a list. The list locks are taken
2535 * and this function is atomic with respect to other list locked calls.
2536 * A buffer cannot be placed on two lists at the same time.
2538 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2540 unsigned long flags
;
2542 spin_lock_irqsave(&list
->lock
, flags
);
2543 __skb_queue_after(list
, old
, newsk
);
2544 spin_unlock_irqrestore(&list
->lock
, flags
);
2546 EXPORT_SYMBOL(skb_append
);
2549 * skb_insert - insert a buffer
2550 * @old: buffer to insert before
2551 * @newsk: buffer to insert
2552 * @list: list to use
2554 * Place a packet before a given packet in a list. The list locks are
2555 * taken and this function is atomic with respect to other list locked
2558 * A buffer cannot be placed on two lists at the same time.
2560 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2562 unsigned long flags
;
2564 spin_lock_irqsave(&list
->lock
, flags
);
2565 __skb_insert(newsk
, old
->prev
, old
, list
);
2566 spin_unlock_irqrestore(&list
->lock
, flags
);
2568 EXPORT_SYMBOL(skb_insert
);
2570 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2571 struct sk_buff
* skb1
,
2572 const u32 len
, const int pos
)
2576 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2578 /* And move data appendix as is. */
2579 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2580 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2582 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2583 skb_shinfo(skb
)->nr_frags
= 0;
2584 skb1
->data_len
= skb
->data_len
;
2585 skb1
->len
+= skb1
->data_len
;
2588 skb_set_tail_pointer(skb
, len
);
2591 static inline void skb_split_no_header(struct sk_buff
*skb
,
2592 struct sk_buff
* skb1
,
2593 const u32 len
, int pos
)
2596 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2598 skb_shinfo(skb
)->nr_frags
= 0;
2599 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2601 skb
->data_len
= len
- pos
;
2603 for (i
= 0; i
< nfrags
; i
++) {
2604 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2606 if (pos
+ size
> len
) {
2607 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2611 * We have two variants in this case:
2612 * 1. Move all the frag to the second
2613 * part, if it is possible. F.e.
2614 * this approach is mandatory for TUX,
2615 * where splitting is expensive.
2616 * 2. Split is accurately. We make this.
2618 skb_frag_ref(skb
, i
);
2619 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2620 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2621 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2622 skb_shinfo(skb
)->nr_frags
++;
2626 skb_shinfo(skb
)->nr_frags
++;
2629 skb_shinfo(skb1
)->nr_frags
= k
;
2633 * skb_split - Split fragmented skb to two parts at length len.
2634 * @skb: the buffer to split
2635 * @skb1: the buffer to receive the second part
2636 * @len: new length for skb
2638 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2640 int pos
= skb_headlen(skb
);
2642 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2643 if (len
< pos
) /* Split line is inside header. */
2644 skb_split_inside_header(skb
, skb1
, len
, pos
);
2645 else /* Second chunk has no header, nothing to copy. */
2646 skb_split_no_header(skb
, skb1
, len
, pos
);
2648 EXPORT_SYMBOL(skb_split
);
2650 /* Shifting from/to a cloned skb is a no-go.
2652 * Caller cannot keep skb_shinfo related pointers past calling here!
2654 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2656 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2660 * skb_shift - Shifts paged data partially from skb to another
2661 * @tgt: buffer into which tail data gets added
2662 * @skb: buffer from which the paged data comes from
2663 * @shiftlen: shift up to this many bytes
2665 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2666 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2667 * It's up to caller to free skb if everything was shifted.
2669 * If @tgt runs out of frags, the whole operation is aborted.
2671 * Skb cannot include anything else but paged data while tgt is allowed
2672 * to have non-paged data as well.
2674 * TODO: full sized shift could be optimized but that would need
2675 * specialized skb free'er to handle frags without up-to-date nr_frags.
2677 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2679 int from
, to
, merge
, todo
;
2680 struct skb_frag_struct
*fragfrom
, *fragto
;
2682 BUG_ON(shiftlen
> skb
->len
);
2683 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2687 to
= skb_shinfo(tgt
)->nr_frags
;
2688 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2690 /* Actual merge is delayed until the point when we know we can
2691 * commit all, so that we don't have to undo partial changes
2694 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2695 fragfrom
->page_offset
)) {
2700 todo
-= skb_frag_size(fragfrom
);
2702 if (skb_prepare_for_shift(skb
) ||
2703 skb_prepare_for_shift(tgt
))
2706 /* All previous frag pointers might be stale! */
2707 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2708 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2710 skb_frag_size_add(fragto
, shiftlen
);
2711 skb_frag_size_sub(fragfrom
, shiftlen
);
2712 fragfrom
->page_offset
+= shiftlen
;
2720 /* Skip full, not-fitting skb to avoid expensive operations */
2721 if ((shiftlen
== skb
->len
) &&
2722 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2725 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2728 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2729 if (to
== MAX_SKB_FRAGS
)
2732 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2733 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2735 if (todo
>= skb_frag_size(fragfrom
)) {
2736 *fragto
= *fragfrom
;
2737 todo
-= skb_frag_size(fragfrom
);
2742 __skb_frag_ref(fragfrom
);
2743 fragto
->page
= fragfrom
->page
;
2744 fragto
->page_offset
= fragfrom
->page_offset
;
2745 skb_frag_size_set(fragto
, todo
);
2747 fragfrom
->page_offset
+= todo
;
2748 skb_frag_size_sub(fragfrom
, todo
);
2756 /* Ready to "commit" this state change to tgt */
2757 skb_shinfo(tgt
)->nr_frags
= to
;
2760 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2761 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2763 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2764 __skb_frag_unref(fragfrom
);
2767 /* Reposition in the original skb */
2769 while (from
< skb_shinfo(skb
)->nr_frags
)
2770 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2771 skb_shinfo(skb
)->nr_frags
= to
;
2773 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2776 /* Most likely the tgt won't ever need its checksum anymore, skb on
2777 * the other hand might need it if it needs to be resent
2779 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2780 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2782 /* Yak, is it really working this way? Some helper please? */
2783 skb
->len
-= shiftlen
;
2784 skb
->data_len
-= shiftlen
;
2785 skb
->truesize
-= shiftlen
;
2786 tgt
->len
+= shiftlen
;
2787 tgt
->data_len
+= shiftlen
;
2788 tgt
->truesize
+= shiftlen
;
2794 * skb_prepare_seq_read - Prepare a sequential read of skb data
2795 * @skb: the buffer to read
2796 * @from: lower offset of data to be read
2797 * @to: upper offset of data to be read
2798 * @st: state variable
2800 * Initializes the specified state variable. Must be called before
2801 * invoking skb_seq_read() for the first time.
2803 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2804 unsigned int to
, struct skb_seq_state
*st
)
2806 st
->lower_offset
= from
;
2807 st
->upper_offset
= to
;
2808 st
->root_skb
= st
->cur_skb
= skb
;
2809 st
->frag_idx
= st
->stepped_offset
= 0;
2810 st
->frag_data
= NULL
;
2812 EXPORT_SYMBOL(skb_prepare_seq_read
);
2815 * skb_seq_read - Sequentially read skb data
2816 * @consumed: number of bytes consumed by the caller so far
2817 * @data: destination pointer for data to be returned
2818 * @st: state variable
2820 * Reads a block of skb data at @consumed relative to the
2821 * lower offset specified to skb_prepare_seq_read(). Assigns
2822 * the head of the data block to @data and returns the length
2823 * of the block or 0 if the end of the skb data or the upper
2824 * offset has been reached.
2826 * The caller is not required to consume all of the data
2827 * returned, i.e. @consumed is typically set to the number
2828 * of bytes already consumed and the next call to
2829 * skb_seq_read() will return the remaining part of the block.
2831 * Note 1: The size of each block of data returned can be arbitrary,
2832 * this limitation is the cost for zerocopy sequential
2833 * reads of potentially non linear data.
2835 * Note 2: Fragment lists within fragments are not implemented
2836 * at the moment, state->root_skb could be replaced with
2837 * a stack for this purpose.
2839 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2840 struct skb_seq_state
*st
)
2842 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2845 if (unlikely(abs_offset
>= st
->upper_offset
)) {
2846 if (st
->frag_data
) {
2847 kunmap_atomic(st
->frag_data
);
2848 st
->frag_data
= NULL
;
2854 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2856 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2857 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2858 return block_limit
- abs_offset
;
2861 if (st
->frag_idx
== 0 && !st
->frag_data
)
2862 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2864 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2865 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2866 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2868 if (abs_offset
< block_limit
) {
2870 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2872 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2873 (abs_offset
- st
->stepped_offset
);
2875 return block_limit
- abs_offset
;
2878 if (st
->frag_data
) {
2879 kunmap_atomic(st
->frag_data
);
2880 st
->frag_data
= NULL
;
2884 st
->stepped_offset
+= skb_frag_size(frag
);
2887 if (st
->frag_data
) {
2888 kunmap_atomic(st
->frag_data
);
2889 st
->frag_data
= NULL
;
2892 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2893 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2896 } else if (st
->cur_skb
->next
) {
2897 st
->cur_skb
= st
->cur_skb
->next
;
2904 EXPORT_SYMBOL(skb_seq_read
);
2907 * skb_abort_seq_read - Abort a sequential read of skb data
2908 * @st: state variable
2910 * Must be called if skb_seq_read() was not called until it
2913 void skb_abort_seq_read(struct skb_seq_state
*st
)
2916 kunmap_atomic(st
->frag_data
);
2918 EXPORT_SYMBOL(skb_abort_seq_read
);
2920 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2922 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2923 struct ts_config
*conf
,
2924 struct ts_state
*state
)
2926 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2929 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2931 skb_abort_seq_read(TS_SKB_CB(state
));
2935 * skb_find_text - Find a text pattern in skb data
2936 * @skb: the buffer to look in
2937 * @from: search offset
2939 * @config: textsearch configuration
2941 * Finds a pattern in the skb data according to the specified
2942 * textsearch configuration. Use textsearch_next() to retrieve
2943 * subsequent occurrences of the pattern. Returns the offset
2944 * to the first occurrence or UINT_MAX if no match was found.
2946 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2947 unsigned int to
, struct ts_config
*config
)
2949 struct ts_state state
;
2952 config
->get_next_block
= skb_ts_get_next_block
;
2953 config
->finish
= skb_ts_finish
;
2955 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(&state
));
2957 ret
= textsearch_find(config
, &state
);
2958 return (ret
<= to
- from
? ret
: UINT_MAX
);
2960 EXPORT_SYMBOL(skb_find_text
);
2963 * skb_append_datato_frags - append the user data to a skb
2964 * @sk: sock structure
2965 * @skb: skb structure to be appended with user data.
2966 * @getfrag: call back function to be used for getting the user data
2967 * @from: pointer to user message iov
2968 * @length: length of the iov message
2970 * Description: This procedure append the user data in the fragment part
2971 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2973 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2974 int (*getfrag
)(void *from
, char *to
, int offset
,
2975 int len
, int odd
, struct sk_buff
*skb
),
2976 void *from
, int length
)
2978 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2982 struct page_frag
*pfrag
= ¤t
->task_frag
;
2985 /* Return error if we don't have space for new frag */
2986 if (frg_cnt
>= MAX_SKB_FRAGS
)
2989 if (!sk_page_frag_refill(sk
, pfrag
))
2992 /* copy the user data to page */
2993 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2995 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2996 offset
, copy
, 0, skb
);
3000 /* copy was successful so update the size parameters */
3001 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
3004 pfrag
->offset
+= copy
;
3005 get_page(pfrag
->page
);
3007 skb
->truesize
+= copy
;
3008 atomic_add(copy
, &sk
->sk_wmem_alloc
);
3010 skb
->data_len
+= copy
;
3014 } while (length
> 0);
3018 EXPORT_SYMBOL(skb_append_datato_frags
);
3020 int skb_append_pagefrags(struct sk_buff
*skb
, struct page
*page
,
3021 int offset
, size_t size
)
3023 int i
= skb_shinfo(skb
)->nr_frags
;
3025 if (skb_can_coalesce(skb
, i
, page
, offset
)) {
3026 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], size
);
3027 } else if (i
< MAX_SKB_FRAGS
) {
3029 skb_fill_page_desc(skb
, i
, page
, offset
, size
);
3036 EXPORT_SYMBOL_GPL(skb_append_pagefrags
);
3039 * skb_pull_rcsum - pull skb and update receive checksum
3040 * @skb: buffer to update
3041 * @len: length of data pulled
3043 * This function performs an skb_pull on the packet and updates
3044 * the CHECKSUM_COMPLETE checksum. It should be used on
3045 * receive path processing instead of skb_pull unless you know
3046 * that the checksum difference is zero (e.g., a valid IP header)
3047 * or you are setting ip_summed to CHECKSUM_NONE.
3049 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
3051 unsigned char *data
= skb
->data
;
3053 BUG_ON(len
> skb
->len
);
3054 __skb_pull(skb
, len
);
3055 skb_postpull_rcsum(skb
, data
, len
);
3058 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
3061 * skb_segment - Perform protocol segmentation on skb.
3062 * @head_skb: buffer to segment
3063 * @features: features for the output path (see dev->features)
3065 * This function performs segmentation on the given skb. It returns
3066 * a pointer to the first in a list of new skbs for the segments.
3067 * In case of error it returns ERR_PTR(err).
3069 struct sk_buff
*skb_segment(struct sk_buff
*head_skb
,
3070 netdev_features_t features
)
3072 struct sk_buff
*segs
= NULL
;
3073 struct sk_buff
*tail
= NULL
;
3074 struct sk_buff
*list_skb
= skb_shinfo(head_skb
)->frag_list
;
3075 skb_frag_t
*frag
= skb_shinfo(head_skb
)->frags
;
3076 unsigned int mss
= skb_shinfo(head_skb
)->gso_size
;
3077 unsigned int doffset
= head_skb
->data
- skb_mac_header(head_skb
);
3078 struct sk_buff
*frag_skb
= head_skb
;
3079 unsigned int offset
= doffset
;
3080 unsigned int tnl_hlen
= skb_tnl_header_len(head_skb
);
3081 unsigned int partial_segs
= 0;
3082 unsigned int headroom
;
3083 unsigned int len
= head_skb
->len
;
3086 int nfrags
= skb_shinfo(head_skb
)->nr_frags
;
3092 __skb_push(head_skb
, doffset
);
3093 proto
= skb_network_protocol(head_skb
, &dummy
);
3094 if (unlikely(!proto
))
3095 return ERR_PTR(-EINVAL
);
3097 sg
= !!(features
& NETIF_F_SG
);
3098 csum
= !!can_checksum_protocol(features
, proto
);
3100 /* GSO partial only requires that we trim off any excess that
3101 * doesn't fit into an MSS sized block, so take care of that
3104 if (sg
&& csum
&& (features
& NETIF_F_GSO_PARTIAL
)) {
3105 partial_segs
= len
/ mss
;
3106 if (partial_segs
> 1)
3107 mss
*= partial_segs
;
3112 headroom
= skb_headroom(head_skb
);
3113 pos
= skb_headlen(head_skb
);
3116 struct sk_buff
*nskb
;
3117 skb_frag_t
*nskb_frag
;
3121 if (unlikely(mss
== GSO_BY_FRAGS
)) {
3122 len
= list_skb
->len
;
3124 len
= head_skb
->len
- offset
;
3129 hsize
= skb_headlen(head_skb
) - offset
;
3132 if (hsize
> len
|| !sg
)
3135 if (!hsize
&& i
>= nfrags
&& skb_headlen(list_skb
) &&
3136 (skb_headlen(list_skb
) == len
|| sg
)) {
3137 BUG_ON(skb_headlen(list_skb
) > len
);
3140 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3141 frag
= skb_shinfo(list_skb
)->frags
;
3142 frag_skb
= list_skb
;
3143 pos
+= skb_headlen(list_skb
);
3145 while (pos
< offset
+ len
) {
3146 BUG_ON(i
>= nfrags
);
3148 size
= skb_frag_size(frag
);
3149 if (pos
+ size
> offset
+ len
)
3157 nskb
= skb_clone(list_skb
, GFP_ATOMIC
);
3158 list_skb
= list_skb
->next
;
3160 if (unlikely(!nskb
))
3163 if (unlikely(pskb_trim(nskb
, len
))) {
3168 hsize
= skb_end_offset(nskb
);
3169 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
3174 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
3175 skb_release_head_state(nskb
);
3176 __skb_push(nskb
, doffset
);
3178 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
3179 GFP_ATOMIC
, skb_alloc_rx_flag(head_skb
),
3182 if (unlikely(!nskb
))
3185 skb_reserve(nskb
, headroom
);
3186 __skb_put(nskb
, doffset
);
3195 __copy_skb_header(nskb
, head_skb
);
3197 skb_headers_offset_update(nskb
, skb_headroom(nskb
) - headroom
);
3198 skb_reset_mac_len(nskb
);
3200 skb_copy_from_linear_data_offset(head_skb
, -tnl_hlen
,
3201 nskb
->data
- tnl_hlen
,
3202 doffset
+ tnl_hlen
);
3204 if (nskb
->len
== len
+ doffset
)
3205 goto perform_csum_check
;
3208 if (!nskb
->remcsum_offload
)
3209 nskb
->ip_summed
= CHECKSUM_NONE
;
3210 SKB_GSO_CB(nskb
)->csum
=
3211 skb_copy_and_csum_bits(head_skb
, offset
,
3214 SKB_GSO_CB(nskb
)->csum_start
=
3215 skb_headroom(nskb
) + doffset
;
3219 nskb_frag
= skb_shinfo(nskb
)->frags
;
3221 skb_copy_from_linear_data_offset(head_skb
, offset
,
3222 skb_put(nskb
, hsize
), hsize
);
3224 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(head_skb
)->tx_flags
&
3227 while (pos
< offset
+ len
) {
3229 BUG_ON(skb_headlen(list_skb
));
3232 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3233 frag
= skb_shinfo(list_skb
)->frags
;
3234 frag_skb
= list_skb
;
3238 list_skb
= list_skb
->next
;
3241 if (unlikely(skb_shinfo(nskb
)->nr_frags
>=
3243 net_warn_ratelimited(
3244 "skb_segment: too many frags: %u %u\n",
3249 if (unlikely(skb_orphan_frags(frag_skb
, GFP_ATOMIC
)))
3253 __skb_frag_ref(nskb_frag
);
3254 size
= skb_frag_size(nskb_frag
);
3257 nskb_frag
->page_offset
+= offset
- pos
;
3258 skb_frag_size_sub(nskb_frag
, offset
- pos
);
3261 skb_shinfo(nskb
)->nr_frags
++;
3263 if (pos
+ size
<= offset
+ len
) {
3268 skb_frag_size_sub(nskb_frag
, pos
+ size
- (offset
+ len
));
3276 nskb
->data_len
= len
- hsize
;
3277 nskb
->len
+= nskb
->data_len
;
3278 nskb
->truesize
+= nskb
->data_len
;
3282 if (skb_has_shared_frag(nskb
)) {
3283 err
= __skb_linearize(nskb
);
3287 if (!nskb
->remcsum_offload
)
3288 nskb
->ip_summed
= CHECKSUM_NONE
;
3289 SKB_GSO_CB(nskb
)->csum
=
3290 skb_checksum(nskb
, doffset
,
3291 nskb
->len
- doffset
, 0);
3292 SKB_GSO_CB(nskb
)->csum_start
=
3293 skb_headroom(nskb
) + doffset
;
3295 } while ((offset
+= len
) < head_skb
->len
);
3297 /* Some callers want to get the end of the list.
3298 * Put it in segs->prev to avoid walking the list.
3299 * (see validate_xmit_skb_list() for example)
3303 /* Update GSO info on first skb in partial sequence. */
3305 int type
= skb_shinfo(head_skb
)->gso_type
;
3307 /* Update type to add partial and then remove dodgy if set */
3308 type
|= SKB_GSO_PARTIAL
;
3309 type
&= ~SKB_GSO_DODGY
;
3311 /* Update GSO info and prepare to start updating headers on
3312 * our way back down the stack of protocols.
3314 skb_shinfo(segs
)->gso_size
= skb_shinfo(head_skb
)->gso_size
;
3315 skb_shinfo(segs
)->gso_segs
= partial_segs
;
3316 skb_shinfo(segs
)->gso_type
= type
;
3317 SKB_GSO_CB(segs
)->data_offset
= skb_headroom(segs
) + doffset
;
3320 /* Following permits correct backpressure, for protocols
3321 * using skb_set_owner_w().
3322 * Idea is to tranfert ownership from head_skb to last segment.
3324 if (head_skb
->destructor
== sock_wfree
) {
3325 swap(tail
->truesize
, head_skb
->truesize
);
3326 swap(tail
->destructor
, head_skb
->destructor
);
3327 swap(tail
->sk
, head_skb
->sk
);
3332 kfree_skb_list(segs
);
3333 return ERR_PTR(err
);
3335 EXPORT_SYMBOL_GPL(skb_segment
);
3337 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
3339 struct skb_shared_info
*pinfo
, *skbinfo
= skb_shinfo(skb
);
3340 unsigned int offset
= skb_gro_offset(skb
);
3341 unsigned int headlen
= skb_headlen(skb
);
3342 unsigned int len
= skb_gro_len(skb
);
3343 struct sk_buff
*lp
, *p
= *head
;
3344 unsigned int delta_truesize
;
3346 if (unlikely(p
->len
+ len
>= 65536))
3349 lp
= NAPI_GRO_CB(p
)->last
;
3350 pinfo
= skb_shinfo(lp
);
3352 if (headlen
<= offset
) {
3355 int i
= skbinfo
->nr_frags
;
3356 int nr_frags
= pinfo
->nr_frags
+ i
;
3358 if (nr_frags
> MAX_SKB_FRAGS
)
3362 pinfo
->nr_frags
= nr_frags
;
3363 skbinfo
->nr_frags
= 0;
3365 frag
= pinfo
->frags
+ nr_frags
;
3366 frag2
= skbinfo
->frags
+ i
;
3371 frag
->page_offset
+= offset
;
3372 skb_frag_size_sub(frag
, offset
);
3374 /* all fragments truesize : remove (head size + sk_buff) */
3375 delta_truesize
= skb
->truesize
-
3376 SKB_TRUESIZE(skb_end_offset(skb
));
3378 skb
->truesize
-= skb
->data_len
;
3379 skb
->len
-= skb
->data_len
;
3382 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3384 } else if (skb
->head_frag
) {
3385 int nr_frags
= pinfo
->nr_frags
;
3386 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3387 struct page
*page
= virt_to_head_page(skb
->head
);
3388 unsigned int first_size
= headlen
- offset
;
3389 unsigned int first_offset
;
3391 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3394 first_offset
= skb
->data
-
3395 (unsigned char *)page_address(page
) +
3398 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3400 frag
->page
.p
= page
;
3401 frag
->page_offset
= first_offset
;
3402 skb_frag_size_set(frag
, first_size
);
3404 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3405 /* We dont need to clear skbinfo->nr_frags here */
3407 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3408 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3413 delta_truesize
= skb
->truesize
;
3414 if (offset
> headlen
) {
3415 unsigned int eat
= offset
- headlen
;
3417 skbinfo
->frags
[0].page_offset
+= eat
;
3418 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3419 skb
->data_len
-= eat
;
3424 __skb_pull(skb
, offset
);
3426 if (NAPI_GRO_CB(p
)->last
== p
)
3427 skb_shinfo(p
)->frag_list
= skb
;
3429 NAPI_GRO_CB(p
)->last
->next
= skb
;
3430 NAPI_GRO_CB(p
)->last
= skb
;
3431 __skb_header_release(skb
);
3435 NAPI_GRO_CB(p
)->count
++;
3437 p
->truesize
+= delta_truesize
;
3440 lp
->data_len
+= len
;
3441 lp
->truesize
+= delta_truesize
;
3444 NAPI_GRO_CB(skb
)->same_flow
= 1;
3447 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3449 void __init
skb_init(void)
3451 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3452 sizeof(struct sk_buff
),
3454 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3456 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3457 sizeof(struct sk_buff_fclones
),
3459 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3464 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3465 * @skb: Socket buffer containing the buffers to be mapped
3466 * @sg: The scatter-gather list to map into
3467 * @offset: The offset into the buffer's contents to start mapping
3468 * @len: Length of buffer space to be mapped
3470 * Fill the specified scatter-gather list with mappings/pointers into a
3471 * region of the buffer space attached to a socket buffer.
3474 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3476 int start
= skb_headlen(skb
);
3477 int i
, copy
= start
- offset
;
3478 struct sk_buff
*frag_iter
;
3484 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3486 if ((len
-= copy
) == 0)
3491 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3494 WARN_ON(start
> offset
+ len
);
3496 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3497 if ((copy
= end
- offset
) > 0) {
3498 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3502 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3503 frag
->page_offset
+offset
-start
);
3512 skb_walk_frags(skb
, frag_iter
) {
3515 WARN_ON(start
> offset
+ len
);
3517 end
= start
+ frag_iter
->len
;
3518 if ((copy
= end
- offset
) > 0) {
3521 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3523 if ((len
-= copy
) == 0)
3533 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3534 * sglist without mark the sg which contain last skb data as the end.
3535 * So the caller can mannipulate sg list as will when padding new data after
3536 * the first call without calling sg_unmark_end to expend sg list.
3538 * Scenario to use skb_to_sgvec_nomark:
3540 * 2. skb_to_sgvec_nomark(payload1)
3541 * 3. skb_to_sgvec_nomark(payload2)
3543 * This is equivalent to:
3545 * 2. skb_to_sgvec(payload1)
3547 * 4. skb_to_sgvec(payload2)
3549 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3550 * is more preferable.
3552 int skb_to_sgvec_nomark(struct sk_buff
*skb
, struct scatterlist
*sg
,
3553 int offset
, int len
)
3555 return __skb_to_sgvec(skb
, sg
, offset
, len
);
3557 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark
);
3559 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3561 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3563 sg_mark_end(&sg
[nsg
- 1]);
3567 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3570 * skb_cow_data - Check that a socket buffer's data buffers are writable
3571 * @skb: The socket buffer to check.
3572 * @tailbits: Amount of trailing space to be added
3573 * @trailer: Returned pointer to the skb where the @tailbits space begins
3575 * Make sure that the data buffers attached to a socket buffer are
3576 * writable. If they are not, private copies are made of the data buffers
3577 * and the socket buffer is set to use these instead.
3579 * If @tailbits is given, make sure that there is space to write @tailbits
3580 * bytes of data beyond current end of socket buffer. @trailer will be
3581 * set to point to the skb in which this space begins.
3583 * The number of scatterlist elements required to completely map the
3584 * COW'd and extended socket buffer will be returned.
3586 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3590 struct sk_buff
*skb1
, **skb_p
;
3592 /* If skb is cloned or its head is paged, reallocate
3593 * head pulling out all the pages (pages are considered not writable
3594 * at the moment even if they are anonymous).
3596 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3597 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3600 /* Easy case. Most of packets will go this way. */
3601 if (!skb_has_frag_list(skb
)) {
3602 /* A little of trouble, not enough of space for trailer.
3603 * This should not happen, when stack is tuned to generate
3604 * good frames. OK, on miss we reallocate and reserve even more
3605 * space, 128 bytes is fair. */
3607 if (skb_tailroom(skb
) < tailbits
&&
3608 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3616 /* Misery. We are in troubles, going to mincer fragments... */
3619 skb_p
= &skb_shinfo(skb
)->frag_list
;
3622 while ((skb1
= *skb_p
) != NULL
) {
3625 /* The fragment is partially pulled by someone,
3626 * this can happen on input. Copy it and everything
3629 if (skb_shared(skb1
))
3632 /* If the skb is the last, worry about trailer. */
3634 if (skb1
->next
== NULL
&& tailbits
) {
3635 if (skb_shinfo(skb1
)->nr_frags
||
3636 skb_has_frag_list(skb1
) ||
3637 skb_tailroom(skb1
) < tailbits
)
3638 ntail
= tailbits
+ 128;
3644 skb_shinfo(skb1
)->nr_frags
||
3645 skb_has_frag_list(skb1
)) {
3646 struct sk_buff
*skb2
;
3648 /* Fuck, we are miserable poor guys... */
3650 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3652 skb2
= skb_copy_expand(skb1
,
3656 if (unlikely(skb2
== NULL
))
3660 skb_set_owner_w(skb2
, skb1
->sk
);
3662 /* Looking around. Are we still alive?
3663 * OK, link new skb, drop old one */
3665 skb2
->next
= skb1
->next
;
3672 skb_p
= &skb1
->next
;
3677 EXPORT_SYMBOL_GPL(skb_cow_data
);
3679 static void sock_rmem_free(struct sk_buff
*skb
)
3681 struct sock
*sk
= skb
->sk
;
3683 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3687 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3689 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3691 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3692 (unsigned int)sk
->sk_rcvbuf
)
3697 skb
->destructor
= sock_rmem_free
;
3698 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3700 /* before exiting rcu section, make sure dst is refcounted */
3703 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3704 if (!sock_flag(sk
, SOCK_DEAD
))
3705 sk
->sk_data_ready(sk
);
3708 EXPORT_SYMBOL(sock_queue_err_skb
);
3710 struct sk_buff
*sock_dequeue_err_skb(struct sock
*sk
)
3712 struct sk_buff_head
*q
= &sk
->sk_error_queue
;
3713 struct sk_buff
*skb
, *skb_next
;
3714 unsigned long flags
;
3717 spin_lock_irqsave(&q
->lock
, flags
);
3718 skb
= __skb_dequeue(q
);
3719 if (skb
&& (skb_next
= skb_peek(q
)))
3720 err
= SKB_EXT_ERR(skb_next
)->ee
.ee_errno
;
3721 spin_unlock_irqrestore(&q
->lock
, flags
);
3725 sk
->sk_error_report(sk
);
3729 EXPORT_SYMBOL(sock_dequeue_err_skb
);
3732 * skb_clone_sk - create clone of skb, and take reference to socket
3733 * @skb: the skb to clone
3735 * This function creates a clone of a buffer that holds a reference on
3736 * sk_refcnt. Buffers created via this function are meant to be
3737 * returned using sock_queue_err_skb, or free via kfree_skb.
3739 * When passing buffers allocated with this function to sock_queue_err_skb
3740 * it is necessary to wrap the call with sock_hold/sock_put in order to
3741 * prevent the socket from being released prior to being enqueued on
3742 * the sk_error_queue.
3744 struct sk_buff
*skb_clone_sk(struct sk_buff
*skb
)
3746 struct sock
*sk
= skb
->sk
;
3747 struct sk_buff
*clone
;
3749 if (!sk
|| !atomic_inc_not_zero(&sk
->sk_refcnt
))
3752 clone
= skb_clone(skb
, GFP_ATOMIC
);
3759 clone
->destructor
= sock_efree
;
3763 EXPORT_SYMBOL(skb_clone_sk
);
3765 static void __skb_complete_tx_timestamp(struct sk_buff
*skb
,
3769 struct sock_exterr_skb
*serr
;
3772 serr
= SKB_EXT_ERR(skb
);
3773 memset(serr
, 0, sizeof(*serr
));
3774 serr
->ee
.ee_errno
= ENOMSG
;
3775 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3776 serr
->ee
.ee_info
= tstype
;
3777 if (sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
) {
3778 serr
->ee
.ee_data
= skb_shinfo(skb
)->tskey
;
3779 if (sk
->sk_protocol
== IPPROTO_TCP
&&
3780 sk
->sk_type
== SOCK_STREAM
)
3781 serr
->ee
.ee_data
-= sk
->sk_tskey
;
3784 err
= sock_queue_err_skb(sk
, skb
);
3790 static bool skb_may_tx_timestamp(struct sock
*sk
, bool tsonly
)
3794 if (likely(sysctl_tstamp_allow_data
|| tsonly
))
3797 read_lock_bh(&sk
->sk_callback_lock
);
3798 ret
= sk
->sk_socket
&& sk
->sk_socket
->file
&&
3799 file_ns_capable(sk
->sk_socket
->file
, &init_user_ns
, CAP_NET_RAW
);
3800 read_unlock_bh(&sk
->sk_callback_lock
);
3804 void skb_complete_tx_timestamp(struct sk_buff
*skb
,
3805 struct skb_shared_hwtstamps
*hwtstamps
)
3807 struct sock
*sk
= skb
->sk
;
3809 if (!skb_may_tx_timestamp(sk
, false))
3812 /* take a reference to prevent skb_orphan() from freeing the socket */
3815 *skb_hwtstamps(skb
) = *hwtstamps
;
3816 __skb_complete_tx_timestamp(skb
, sk
, SCM_TSTAMP_SND
);
3820 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp
);
3822 void __skb_tstamp_tx(struct sk_buff
*orig_skb
,
3823 struct skb_shared_hwtstamps
*hwtstamps
,
3824 struct sock
*sk
, int tstype
)
3826 struct sk_buff
*skb
;
3832 tsonly
= sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_TSONLY
;
3833 if (!skb_may_tx_timestamp(sk
, tsonly
))
3837 skb
= alloc_skb(0, GFP_ATOMIC
);
3839 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3844 skb_shinfo(skb
)->tx_flags
= skb_shinfo(orig_skb
)->tx_flags
;
3845 skb_shinfo(skb
)->tskey
= skb_shinfo(orig_skb
)->tskey
;
3849 *skb_hwtstamps(skb
) = *hwtstamps
;
3851 skb
->tstamp
= ktime_get_real();
3853 __skb_complete_tx_timestamp(skb
, sk
, tstype
);
3855 EXPORT_SYMBOL_GPL(__skb_tstamp_tx
);
3857 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3858 struct skb_shared_hwtstamps
*hwtstamps
)
3860 return __skb_tstamp_tx(orig_skb
, hwtstamps
, orig_skb
->sk
,
3863 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3865 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3867 struct sock
*sk
= skb
->sk
;
3868 struct sock_exterr_skb
*serr
;
3871 skb
->wifi_acked_valid
= 1;
3872 skb
->wifi_acked
= acked
;
3874 serr
= SKB_EXT_ERR(skb
);
3875 memset(serr
, 0, sizeof(*serr
));
3876 serr
->ee
.ee_errno
= ENOMSG
;
3877 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3879 /* take a reference to prevent skb_orphan() from freeing the socket */
3882 err
= sock_queue_err_skb(sk
, skb
);
3888 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3891 * skb_partial_csum_set - set up and verify partial csum values for packet
3892 * @skb: the skb to set
3893 * @start: the number of bytes after skb->data to start checksumming.
3894 * @off: the offset from start to place the checksum.
3896 * For untrusted partially-checksummed packets, we need to make sure the values
3897 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3899 * This function checks and sets those values and skb->ip_summed: if this
3900 * returns false you should drop the packet.
3902 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3904 if (unlikely(start
> skb_headlen(skb
)) ||
3905 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3906 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3907 start
, off
, skb_headlen(skb
));
3910 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3911 skb
->csum_start
= skb_headroom(skb
) + start
;
3912 skb
->csum_offset
= off
;
3913 skb_set_transport_header(skb
, start
);
3916 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3918 static int skb_maybe_pull_tail(struct sk_buff
*skb
, unsigned int len
,
3921 if (skb_headlen(skb
) >= len
)
3924 /* If we need to pullup then pullup to the max, so we
3925 * won't need to do it again.
3930 if (__pskb_pull_tail(skb
, max
- skb_headlen(skb
)) == NULL
)
3933 if (skb_headlen(skb
) < len
)
3939 #define MAX_TCP_HDR_LEN (15 * 4)
3941 static __sum16
*skb_checksum_setup_ip(struct sk_buff
*skb
,
3942 typeof(IPPROTO_IP
) proto
,
3949 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct tcphdr
),
3950 off
+ MAX_TCP_HDR_LEN
);
3951 if (!err
&& !skb_partial_csum_set(skb
, off
,
3952 offsetof(struct tcphdr
,
3955 return err
? ERR_PTR(err
) : &tcp_hdr(skb
)->check
;
3958 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct udphdr
),
3959 off
+ sizeof(struct udphdr
));
3960 if (!err
&& !skb_partial_csum_set(skb
, off
,
3961 offsetof(struct udphdr
,
3964 return err
? ERR_PTR(err
) : &udp_hdr(skb
)->check
;
3967 return ERR_PTR(-EPROTO
);
3970 /* This value should be large enough to cover a tagged ethernet header plus
3971 * maximally sized IP and TCP or UDP headers.
3973 #define MAX_IP_HDR_LEN 128
3975 static int skb_checksum_setup_ipv4(struct sk_buff
*skb
, bool recalculate
)
3984 err
= skb_maybe_pull_tail(skb
,
3985 sizeof(struct iphdr
),
3990 if (ip_hdr(skb
)->frag_off
& htons(IP_OFFSET
| IP_MF
))
3993 off
= ip_hdrlen(skb
);
4000 csum
= skb_checksum_setup_ip(skb
, ip_hdr(skb
)->protocol
, off
);
4002 return PTR_ERR(csum
);
4005 *csum
= ~csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
4008 ip_hdr(skb
)->protocol
, 0);
4015 /* This value should be large enough to cover a tagged ethernet header plus
4016 * an IPv6 header, all options, and a maximal TCP or UDP header.
4018 #define MAX_IPV6_HDR_LEN 256
4020 #define OPT_HDR(type, skb, off) \
4021 (type *)(skb_network_header(skb) + (off))
4023 static int skb_checksum_setup_ipv6(struct sk_buff
*skb
, bool recalculate
)
4036 off
= sizeof(struct ipv6hdr
);
4038 err
= skb_maybe_pull_tail(skb
, off
, MAX_IPV6_HDR_LEN
);
4042 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
4044 len
= sizeof(struct ipv6hdr
) + ntohs(ipv6_hdr(skb
)->payload_len
);
4045 while (off
<= len
&& !done
) {
4047 case IPPROTO_DSTOPTS
:
4048 case IPPROTO_HOPOPTS
:
4049 case IPPROTO_ROUTING
: {
4050 struct ipv6_opt_hdr
*hp
;
4052 err
= skb_maybe_pull_tail(skb
,
4054 sizeof(struct ipv6_opt_hdr
),
4059 hp
= OPT_HDR(struct ipv6_opt_hdr
, skb
, off
);
4060 nexthdr
= hp
->nexthdr
;
4061 off
+= ipv6_optlen(hp
);
4065 struct ip_auth_hdr
*hp
;
4067 err
= skb_maybe_pull_tail(skb
,
4069 sizeof(struct ip_auth_hdr
),
4074 hp
= OPT_HDR(struct ip_auth_hdr
, skb
, off
);
4075 nexthdr
= hp
->nexthdr
;
4076 off
+= ipv6_authlen(hp
);
4079 case IPPROTO_FRAGMENT
: {
4080 struct frag_hdr
*hp
;
4082 err
= skb_maybe_pull_tail(skb
,
4084 sizeof(struct frag_hdr
),
4089 hp
= OPT_HDR(struct frag_hdr
, skb
, off
);
4091 if (hp
->frag_off
& htons(IP6_OFFSET
| IP6_MF
))
4094 nexthdr
= hp
->nexthdr
;
4095 off
+= sizeof(struct frag_hdr
);
4106 if (!done
|| fragment
)
4109 csum
= skb_checksum_setup_ip(skb
, nexthdr
, off
);
4111 return PTR_ERR(csum
);
4114 *csum
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4115 &ipv6_hdr(skb
)->daddr
,
4116 skb
->len
- off
, nexthdr
, 0);
4124 * skb_checksum_setup - set up partial checksum offset
4125 * @skb: the skb to set up
4126 * @recalculate: if true the pseudo-header checksum will be recalculated
4128 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
)
4132 switch (skb
->protocol
) {
4133 case htons(ETH_P_IP
):
4134 err
= skb_checksum_setup_ipv4(skb
, recalculate
);
4137 case htons(ETH_P_IPV6
):
4138 err
= skb_checksum_setup_ipv6(skb
, recalculate
);
4148 EXPORT_SYMBOL(skb_checksum_setup
);
4151 * skb_checksum_maybe_trim - maybe trims the given skb
4152 * @skb: the skb to check
4153 * @transport_len: the data length beyond the network header
4155 * Checks whether the given skb has data beyond the given transport length.
4156 * If so, returns a cloned skb trimmed to this transport length.
4157 * Otherwise returns the provided skb. Returns NULL in error cases
4158 * (e.g. transport_len exceeds skb length or out-of-memory).
4160 * Caller needs to set the skb transport header and free any returned skb if it
4161 * differs from the provided skb.
4163 static struct sk_buff
*skb_checksum_maybe_trim(struct sk_buff
*skb
,
4164 unsigned int transport_len
)
4166 struct sk_buff
*skb_chk
;
4167 unsigned int len
= skb_transport_offset(skb
) + transport_len
;
4172 else if (skb
->len
== len
)
4175 skb_chk
= skb_clone(skb
, GFP_ATOMIC
);
4179 ret
= pskb_trim_rcsum(skb_chk
, len
);
4189 * skb_checksum_trimmed - validate checksum of an skb
4190 * @skb: the skb to check
4191 * @transport_len: the data length beyond the network header
4192 * @skb_chkf: checksum function to use
4194 * Applies the given checksum function skb_chkf to the provided skb.
4195 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4197 * If the skb has data beyond the given transport length, then a
4198 * trimmed & cloned skb is checked and returned.
4200 * Caller needs to set the skb transport header and free any returned skb if it
4201 * differs from the provided skb.
4203 struct sk_buff
*skb_checksum_trimmed(struct sk_buff
*skb
,
4204 unsigned int transport_len
,
4205 __sum16(*skb_chkf
)(struct sk_buff
*skb
))
4207 struct sk_buff
*skb_chk
;
4208 unsigned int offset
= skb_transport_offset(skb
);
4211 skb_chk
= skb_checksum_maybe_trim(skb
, transport_len
);
4215 if (!pskb_may_pull(skb_chk
, offset
))
4218 skb_pull_rcsum(skb_chk
, offset
);
4219 ret
= skb_chkf(skb_chk
);
4220 skb_push_rcsum(skb_chk
, offset
);
4228 if (skb_chk
&& skb_chk
!= skb
)
4234 EXPORT_SYMBOL(skb_checksum_trimmed
);
4236 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
4238 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4241 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
4243 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
4246 skb_release_head_state(skb
);
4247 kmem_cache_free(skbuff_head_cache
, skb
);
4252 EXPORT_SYMBOL(kfree_skb_partial
);
4255 * skb_try_coalesce - try to merge skb to prior one
4257 * @from: buffer to add
4258 * @fragstolen: pointer to boolean
4259 * @delta_truesize: how much more was allocated than was requested
4261 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
4262 bool *fragstolen
, int *delta_truesize
)
4264 int i
, delta
, len
= from
->len
;
4266 *fragstolen
= false;
4271 if (len
<= skb_tailroom(to
)) {
4273 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
4274 *delta_truesize
= 0;
4278 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
4281 if (skb_headlen(from
) != 0) {
4283 unsigned int offset
;
4285 if (skb_shinfo(to
)->nr_frags
+
4286 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
4289 if (skb_head_is_locked(from
))
4292 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
4294 page
= virt_to_head_page(from
->head
);
4295 offset
= from
->data
- (unsigned char *)page_address(page
);
4297 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
4298 page
, offset
, skb_headlen(from
));
4301 if (skb_shinfo(to
)->nr_frags
+
4302 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
4305 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
4308 WARN_ON_ONCE(delta
< len
);
4310 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
4311 skb_shinfo(from
)->frags
,
4312 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
4313 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
4315 if (!skb_cloned(from
))
4316 skb_shinfo(from
)->nr_frags
= 0;
4318 /* if the skb is not cloned this does nothing
4319 * since we set nr_frags to 0.
4321 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
4322 skb_frag_ref(from
, i
);
4324 to
->truesize
+= delta
;
4326 to
->data_len
+= len
;
4328 *delta_truesize
= delta
;
4331 EXPORT_SYMBOL(skb_try_coalesce
);
4334 * skb_scrub_packet - scrub an skb
4336 * @skb: buffer to clean
4337 * @xnet: packet is crossing netns
4339 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4340 * into/from a tunnel. Some information have to be cleared during these
4342 * skb_scrub_packet can also be used to clean a skb before injecting it in
4343 * another namespace (@xnet == true). We have to clear all information in the
4344 * skb that could impact namespace isolation.
4346 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
)
4348 skb
->tstamp
.tv64
= 0;
4349 skb
->pkt_type
= PACKET_HOST
;
4355 nf_reset_trace(skb
);
4363 EXPORT_SYMBOL_GPL(skb_scrub_packet
);
4366 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4370 * skb_gso_transport_seglen is used to determine the real size of the
4371 * individual segments, including Layer4 headers (TCP/UDP).
4373 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4375 unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
)
4377 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4378 unsigned int thlen
= 0;
4380 if (skb
->encapsulation
) {
4381 thlen
= skb_inner_transport_header(skb
) -
4382 skb_transport_header(skb
);
4384 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
4385 thlen
+= inner_tcp_hdrlen(skb
);
4386 } else if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))) {
4387 thlen
= tcp_hdrlen(skb
);
4388 } else if (unlikely(shinfo
->gso_type
& SKB_GSO_SCTP
)) {
4389 thlen
= sizeof(struct sctphdr
);
4391 /* UFO sets gso_size to the size of the fragmentation
4392 * payload, i.e. the size of the L4 (UDP) header is already
4395 return thlen
+ shinfo
->gso_size
;
4397 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen
);
4400 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4403 * @mtu: MTU to validate against
4405 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4408 bool skb_gso_validate_mtu(const struct sk_buff
*skb
, unsigned int mtu
)
4410 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
4411 const struct sk_buff
*iter
;
4414 hlen
= skb_gso_network_seglen(skb
);
4416 if (shinfo
->gso_size
!= GSO_BY_FRAGS
)
4419 /* Undo this so we can re-use header sizes */
4420 hlen
-= GSO_BY_FRAGS
;
4422 skb_walk_frags(skb
, iter
) {
4423 if (hlen
+ skb_headlen(iter
) > mtu
)
4429 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu
);
4431 static struct sk_buff
*skb_reorder_vlan_header(struct sk_buff
*skb
)
4433 if (skb_cow(skb
, skb_headroom(skb
)) < 0) {
4438 memmove(skb
->data
- ETH_HLEN
, skb
->data
- skb
->mac_len
- VLAN_HLEN
,
4440 skb
->mac_header
+= VLAN_HLEN
;
4444 struct sk_buff
*skb_vlan_untag(struct sk_buff
*skb
)
4446 struct vlan_hdr
*vhdr
;
4449 if (unlikely(skb_vlan_tag_present(skb
))) {
4450 /* vlan_tci is already set-up so leave this for another time */
4454 skb
= skb_share_check(skb
, GFP_ATOMIC
);
4458 if (unlikely(!pskb_may_pull(skb
, VLAN_HLEN
)))
4461 vhdr
= (struct vlan_hdr
*)skb
->data
;
4462 vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4463 __vlan_hwaccel_put_tag(skb
, skb
->protocol
, vlan_tci
);
4465 skb_pull_rcsum(skb
, VLAN_HLEN
);
4466 vlan_set_encap_proto(skb
, vhdr
);
4468 skb
= skb_reorder_vlan_header(skb
);
4472 skb_reset_network_header(skb
);
4473 skb_reset_transport_header(skb
);
4474 skb_reset_mac_len(skb
);
4482 EXPORT_SYMBOL(skb_vlan_untag
);
4484 int skb_ensure_writable(struct sk_buff
*skb
, int write_len
)
4486 if (!pskb_may_pull(skb
, write_len
))
4489 if (!skb_cloned(skb
) || skb_clone_writable(skb
, write_len
))
4492 return pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4494 EXPORT_SYMBOL(skb_ensure_writable
);
4496 /* remove VLAN header from packet and update csum accordingly. */
4497 static int __skb_vlan_pop(struct sk_buff
*skb
, u16
*vlan_tci
)
4499 struct vlan_hdr
*vhdr
;
4500 unsigned int offset
= skb
->data
- skb_mac_header(skb
);
4503 __skb_push(skb
, offset
);
4504 err
= skb_ensure_writable(skb
, VLAN_ETH_HLEN
);
4508 skb_postpull_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4510 vhdr
= (struct vlan_hdr
*)(skb
->data
+ ETH_HLEN
);
4511 *vlan_tci
= ntohs(vhdr
->h_vlan_TCI
);
4513 memmove(skb
->data
+ VLAN_HLEN
, skb
->data
, 2 * ETH_ALEN
);
4514 __skb_pull(skb
, VLAN_HLEN
);
4516 vlan_set_encap_proto(skb
, vhdr
);
4517 skb
->mac_header
+= VLAN_HLEN
;
4519 if (skb_network_offset(skb
) < ETH_HLEN
)
4520 skb_set_network_header(skb
, ETH_HLEN
);
4522 skb_reset_mac_len(skb
);
4524 __skb_pull(skb
, offset
);
4529 int skb_vlan_pop(struct sk_buff
*skb
)
4535 if (likely(skb_vlan_tag_present(skb
))) {
4538 if (unlikely((skb
->protocol
!= htons(ETH_P_8021Q
) &&
4539 skb
->protocol
!= htons(ETH_P_8021AD
)) ||
4540 skb
->len
< VLAN_ETH_HLEN
))
4543 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4547 /* move next vlan tag to hw accel tag */
4548 if (likely((skb
->protocol
!= htons(ETH_P_8021Q
) &&
4549 skb
->protocol
!= htons(ETH_P_8021AD
)) ||
4550 skb
->len
< VLAN_ETH_HLEN
))
4553 vlan_proto
= skb
->protocol
;
4554 err
= __skb_vlan_pop(skb
, &vlan_tci
);
4558 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4561 EXPORT_SYMBOL(skb_vlan_pop
);
4563 int skb_vlan_push(struct sk_buff
*skb
, __be16 vlan_proto
, u16 vlan_tci
)
4565 if (skb_vlan_tag_present(skb
)) {
4566 unsigned int offset
= skb
->data
- skb_mac_header(skb
);
4569 /* __vlan_insert_tag expect skb->data pointing to mac header.
4570 * So change skb->data before calling it and change back to
4571 * original position later
4573 __skb_push(skb
, offset
);
4574 err
= __vlan_insert_tag(skb
, skb
->vlan_proto
,
4575 skb_vlan_tag_get(skb
));
4577 __skb_pull(skb
, offset
);
4581 skb
->protocol
= skb
->vlan_proto
;
4582 skb
->mac_len
+= VLAN_HLEN
;
4584 skb_postpush_rcsum(skb
, skb
->data
+ (2 * ETH_ALEN
), VLAN_HLEN
);
4585 __skb_pull(skb
, offset
);
4587 __vlan_hwaccel_put_tag(skb
, vlan_proto
, vlan_tci
);
4590 EXPORT_SYMBOL(skb_vlan_push
);
4593 * alloc_skb_with_frags - allocate skb with page frags
4595 * @header_len: size of linear part
4596 * @data_len: needed length in frags
4597 * @max_page_order: max page order desired.
4598 * @errcode: pointer to error code if any
4599 * @gfp_mask: allocation mask
4601 * This can be used to allocate a paged skb, given a maximal order for frags.
4603 struct sk_buff
*alloc_skb_with_frags(unsigned long header_len
,
4604 unsigned long data_len
,
4609 int npages
= (data_len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
4610 unsigned long chunk
;
4611 struct sk_buff
*skb
;
4616 *errcode
= -EMSGSIZE
;
4617 /* Note this test could be relaxed, if we succeed to allocate
4618 * high order pages...
4620 if (npages
> MAX_SKB_FRAGS
)
4623 gfp_head
= gfp_mask
;
4624 if (gfp_head
& __GFP_DIRECT_RECLAIM
)
4625 gfp_head
|= __GFP_REPEAT
;
4627 *errcode
= -ENOBUFS
;
4628 skb
= alloc_skb(header_len
, gfp_head
);
4632 skb
->truesize
+= npages
<< PAGE_SHIFT
;
4634 for (i
= 0; npages
> 0; i
++) {
4635 int order
= max_page_order
;
4638 if (npages
>= 1 << order
) {
4639 page
= alloc_pages((gfp_mask
& ~__GFP_DIRECT_RECLAIM
) |
4646 /* Do not retry other high order allocations */
4652 page
= alloc_page(gfp_mask
);
4656 chunk
= min_t(unsigned long, data_len
,
4657 PAGE_SIZE
<< order
);
4658 skb_fill_page_desc(skb
, i
, page
, 0, chunk
);
4660 npages
-= 1 << order
;
4668 EXPORT_SYMBOL(alloc_skb_with_frags
);
4670 /* carve out the first off bytes from skb when off < headlen */
4671 static int pskb_carve_inside_header(struct sk_buff
*skb
, const u32 off
,
4672 const int headlen
, gfp_t gfp_mask
)
4675 int size
= skb_end_offset(skb
);
4676 int new_hlen
= headlen
- off
;
4679 size
= SKB_DATA_ALIGN(size
);
4681 if (skb_pfmemalloc(skb
))
4682 gfp_mask
|= __GFP_MEMALLOC
;
4683 data
= kmalloc_reserve(size
+
4684 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4685 gfp_mask
, NUMA_NO_NODE
, NULL
);
4689 size
= SKB_WITH_OVERHEAD(ksize(data
));
4691 /* Copy real data, and all frags */
4692 skb_copy_from_linear_data_offset(skb
, off
, data
, new_hlen
);
4695 memcpy((struct skb_shared_info
*)(data
+ size
),
4697 offsetof(struct skb_shared_info
,
4698 frags
[skb_shinfo(skb
)->nr_frags
]));
4699 if (skb_cloned(skb
)) {
4700 /* drop the old head gracefully */
4701 if (skb_orphan_frags(skb
, gfp_mask
)) {
4705 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
4706 skb_frag_ref(skb
, i
);
4707 if (skb_has_frag_list(skb
))
4708 skb_clone_fraglist(skb
);
4709 skb_release_data(skb
);
4711 /* we can reuse existing recount- all we did was
4720 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4723 skb
->end
= skb
->head
+ size
;
4725 skb_set_tail_pointer(skb
, skb_headlen(skb
));
4726 skb_headers_offset_update(skb
, 0);
4730 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
4735 static int pskb_carve(struct sk_buff
*skb
, const u32 off
, gfp_t gfp
);
4737 /* carve out the first eat bytes from skb's frag_list. May recurse into
4740 static int pskb_carve_frag_list(struct sk_buff
*skb
,
4741 struct skb_shared_info
*shinfo
, int eat
,
4744 struct sk_buff
*list
= shinfo
->frag_list
;
4745 struct sk_buff
*clone
= NULL
;
4746 struct sk_buff
*insp
= NULL
;
4750 pr_err("Not enough bytes to eat. Want %d\n", eat
);
4753 if (list
->len
<= eat
) {
4754 /* Eaten as whole. */
4759 /* Eaten partially. */
4760 if (skb_shared(list
)) {
4761 clone
= skb_clone(list
, gfp_mask
);
4767 /* This may be pulled without problems. */
4770 if (pskb_carve(list
, eat
, gfp_mask
) < 0) {
4778 /* Free pulled out fragments. */
4779 while ((list
= shinfo
->frag_list
) != insp
) {
4780 shinfo
->frag_list
= list
->next
;
4783 /* And insert new clone at head. */
4786 shinfo
->frag_list
= clone
;
4791 /* carve off first len bytes from skb. Split line (off) is in the
4792 * non-linear part of skb
4794 static int pskb_carve_inside_nonlinear(struct sk_buff
*skb
, const u32 off
,
4795 int pos
, gfp_t gfp_mask
)
4798 int size
= skb_end_offset(skb
);
4800 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
4801 struct skb_shared_info
*shinfo
;
4803 size
= SKB_DATA_ALIGN(size
);
4805 if (skb_pfmemalloc(skb
))
4806 gfp_mask
|= __GFP_MEMALLOC
;
4807 data
= kmalloc_reserve(size
+
4808 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
4809 gfp_mask
, NUMA_NO_NODE
, NULL
);
4813 size
= SKB_WITH_OVERHEAD(ksize(data
));
4815 memcpy((struct skb_shared_info
*)(data
+ size
),
4816 skb_shinfo(skb
), offsetof(struct skb_shared_info
,
4817 frags
[skb_shinfo(skb
)->nr_frags
]));
4818 if (skb_orphan_frags(skb
, gfp_mask
)) {
4822 shinfo
= (struct skb_shared_info
*)(data
+ size
);
4823 for (i
= 0; i
< nfrags
; i
++) {
4824 int fsize
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
4826 if (pos
+ fsize
> off
) {
4827 shinfo
->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
4831 * We have two variants in this case:
4832 * 1. Move all the frag to the second
4833 * part, if it is possible. F.e.
4834 * this approach is mandatory for TUX,
4835 * where splitting is expensive.
4836 * 2. Split is accurately. We make this.
4838 shinfo
->frags
[0].page_offset
+= off
- pos
;
4839 skb_frag_size_sub(&shinfo
->frags
[0], off
- pos
);
4841 skb_frag_ref(skb
, i
);
4846 shinfo
->nr_frags
= k
;
4847 if (skb_has_frag_list(skb
))
4848 skb_clone_fraglist(skb
);
4851 /* split line is in frag list */
4852 pskb_carve_frag_list(skb
, shinfo
, off
- pos
, gfp_mask
);
4854 skb_release_data(skb
);
4859 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4862 skb
->end
= skb
->head
+ size
;
4864 skb_reset_tail_pointer(skb
);
4865 skb_headers_offset_update(skb
, 0);
4870 skb
->data_len
= skb
->len
;
4871 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
4875 /* remove len bytes from the beginning of the skb */
4876 static int pskb_carve(struct sk_buff
*skb
, const u32 len
, gfp_t gfp
)
4878 int headlen
= skb_headlen(skb
);
4881 return pskb_carve_inside_header(skb
, len
, headlen
, gfp
);
4883 return pskb_carve_inside_nonlinear(skb
, len
, headlen
, gfp
);
4886 /* Extract to_copy bytes starting at off from skb, and return this in
4889 struct sk_buff
*pskb_extract(struct sk_buff
*skb
, int off
,
4890 int to_copy
, gfp_t gfp
)
4892 struct sk_buff
*clone
= skb_clone(skb
, gfp
);
4897 if (pskb_carve(clone
, off
, gfp
) < 0 ||
4898 pskb_trim(clone
, to_copy
)) {
4904 EXPORT_SYMBOL(pskb_extract
);