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/netdevice.h>
53 #ifdef CONFIG_NET_CLS_ACT
54 #include <net/pkt_sched.h>
56 #include <linux/string.h>
57 #include <linux/skbuff.h>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
66 #include <net/protocol.h>
69 #include <net/checksum.h>
70 #include <net/ip6_checksum.h>
73 #include <asm/uaccess.h>
74 #include <trace/events/skb.h>
75 #include <linux/highmem.h>
77 struct kmem_cache
*skbuff_head_cache __read_mostly
;
78 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
81 * skb_panic - private function for out-of-line support
85 * @msg: skb_over_panic or skb_under_panic
87 * Out-of-line support for skb_put() and skb_push().
88 * Called via the wrapper skb_over_panic() or skb_under_panic().
89 * Keep out of line to prevent kernel bloat.
90 * __builtin_return_address is not used because it is not always reliable.
92 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
95 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
96 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
97 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
98 skb
->dev
? skb
->dev
->name
: "<NULL>");
102 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
104 skb_panic(skb
, sz
, addr
, __func__
);
107 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
109 skb_panic(skb
, sz
, addr
, __func__
);
113 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
114 * the caller if emergency pfmemalloc reserves are being used. If it is and
115 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
116 * may be used. Otherwise, the packet data may be discarded until enough
119 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
120 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
122 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
123 unsigned long ip
, bool *pfmemalloc
)
126 bool ret_pfmemalloc
= false;
129 * Try a regular allocation, when that fails and we're not entitled
130 * to the reserves, fail.
132 obj
= kmalloc_node_track_caller(size
,
133 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
135 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
138 /* Try again but now we are using pfmemalloc reserves */
139 ret_pfmemalloc
= true;
140 obj
= kmalloc_node_track_caller(size
, flags
, node
);
144 *pfmemalloc
= ret_pfmemalloc
;
149 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
155 struct sk_buff
*__alloc_skb_head(gfp_t gfp_mask
, int node
)
160 skb
= kmem_cache_alloc_node(skbuff_head_cache
,
161 gfp_mask
& ~__GFP_DMA
, node
);
166 * Only clear those fields we need to clear, not those that we will
167 * actually initialise below. Hence, don't put any more fields after
168 * the tail pointer in struct sk_buff!
170 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
172 skb
->truesize
= sizeof(struct sk_buff
);
173 atomic_set(&skb
->users
, 1);
175 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
181 * __alloc_skb - allocate a network buffer
182 * @size: size to allocate
183 * @gfp_mask: allocation mask
184 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
185 * instead of head cache and allocate a cloned (child) skb.
186 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
187 * allocations in case the data is required for writeback
188 * @node: numa node to allocate memory on
190 * Allocate a new &sk_buff. The returned buffer has no headroom and a
191 * tail room of at least size bytes. The object has a reference count
192 * of one. The return is the buffer. On a failure the return is %NULL.
194 * Buffers may only be allocated from interrupts using a @gfp_mask of
197 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
200 struct kmem_cache
*cache
;
201 struct skb_shared_info
*shinfo
;
206 cache
= (flags
& SKB_ALLOC_FCLONE
)
207 ? skbuff_fclone_cache
: skbuff_head_cache
;
209 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
210 gfp_mask
|= __GFP_MEMALLOC
;
213 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
218 /* We do our best to align skb_shared_info on a separate cache
219 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
220 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
221 * Both skb->head and skb_shared_info are cache line aligned.
223 size
= SKB_DATA_ALIGN(size
);
224 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
225 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
228 /* kmalloc(size) might give us more room than requested.
229 * Put skb_shared_info exactly at the end of allocated zone,
230 * to allow max possible filling before reallocation.
232 size
= SKB_WITH_OVERHEAD(ksize(data
));
233 prefetchw(data
+ size
);
236 * Only clear those fields we need to clear, not those that we will
237 * actually initialise below. Hence, don't put any more fields after
238 * the tail pointer in struct sk_buff!
240 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
241 /* Account for allocated memory : skb + skb->head */
242 skb
->truesize
= SKB_TRUESIZE(size
);
243 skb
->pfmemalloc
= pfmemalloc
;
244 atomic_set(&skb
->users
, 1);
247 skb_reset_tail_pointer(skb
);
248 skb
->end
= skb
->tail
+ size
;
249 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
250 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
252 /* make sure we initialize shinfo sequentially */
253 shinfo
= skb_shinfo(skb
);
254 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
255 atomic_set(&shinfo
->dataref
, 1);
256 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
258 if (flags
& SKB_ALLOC_FCLONE
) {
259 struct sk_buff
*child
= skb
+ 1;
260 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
262 kmemcheck_annotate_bitfield(child
, flags1
);
263 kmemcheck_annotate_bitfield(child
, flags2
);
264 skb
->fclone
= SKB_FCLONE_ORIG
;
265 atomic_set(fclone_ref
, 1);
267 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
268 child
->pfmemalloc
= pfmemalloc
;
273 kmem_cache_free(cache
, skb
);
277 EXPORT_SYMBOL(__alloc_skb
);
280 * build_skb - build a network buffer
281 * @data: data buffer provided by caller
282 * @frag_size: size of fragment, or 0 if head was kmalloced
284 * Allocate a new &sk_buff. Caller provides space holding head and
285 * skb_shared_info. @data must have been allocated by kmalloc() only if
286 * @frag_size is 0, otherwise data should come from the page allocator.
287 * The return is the new skb buffer.
288 * On a failure the return is %NULL, and @data is not freed.
290 * Before IO, driver allocates only data buffer where NIC put incoming frame
291 * Driver should add room at head (NET_SKB_PAD) and
292 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
293 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
294 * before giving packet to stack.
295 * RX rings only contains data buffers, not full skbs.
297 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
299 struct skb_shared_info
*shinfo
;
301 unsigned int size
= frag_size
? : ksize(data
);
303 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
307 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
309 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
310 skb
->truesize
= SKB_TRUESIZE(size
);
311 skb
->head_frag
= frag_size
!= 0;
312 atomic_set(&skb
->users
, 1);
315 skb_reset_tail_pointer(skb
);
316 skb
->end
= skb
->tail
+ size
;
317 skb
->mac_header
= (typeof(skb
->mac_header
))~0U;
318 skb
->transport_header
= (typeof(skb
->transport_header
))~0U;
320 /* make sure we initialize shinfo sequentially */
321 shinfo
= skb_shinfo(skb
);
322 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
323 atomic_set(&shinfo
->dataref
, 1);
324 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
328 EXPORT_SYMBOL(build_skb
);
330 struct netdev_alloc_cache
{
331 struct page_frag frag
;
332 /* we maintain a pagecount bias, so that we dont dirty cache line
333 * containing page->_count every time we allocate a fragment.
335 unsigned int pagecnt_bias
;
337 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
339 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
341 struct netdev_alloc_cache
*nc
;
346 local_irq_save(flags
);
347 nc
= &__get_cpu_var(netdev_alloc_cache
);
348 if (unlikely(!nc
->frag
.page
)) {
350 for (order
= NETDEV_FRAG_PAGE_MAX_ORDER
; ;) {
351 gfp_t gfp
= gfp_mask
;
354 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
355 nc
->frag
.page
= alloc_pages(gfp
, order
);
356 if (likely(nc
->frag
.page
))
361 nc
->frag
.size
= PAGE_SIZE
<< order
;
363 atomic_set(&nc
->frag
.page
->_count
, NETDEV_PAGECNT_MAX_BIAS
);
364 nc
->pagecnt_bias
= NETDEV_PAGECNT_MAX_BIAS
;
368 if (nc
->frag
.offset
+ fragsz
> nc
->frag
.size
) {
369 /* avoid unnecessary locked operations if possible */
370 if ((atomic_read(&nc
->frag
.page
->_count
) == nc
->pagecnt_bias
) ||
371 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->frag
.page
->_count
))
376 data
= page_address(nc
->frag
.page
) + nc
->frag
.offset
;
377 nc
->frag
.offset
+= fragsz
;
380 local_irq_restore(flags
);
385 * netdev_alloc_frag - allocate a page fragment
386 * @fragsz: fragment size
388 * Allocates a frag from a page for receive buffer.
389 * Uses GFP_ATOMIC allocations.
391 void *netdev_alloc_frag(unsigned int fragsz
)
393 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
395 EXPORT_SYMBOL(netdev_alloc_frag
);
398 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
399 * @dev: network device to receive on
400 * @length: length to allocate
401 * @gfp_mask: get_free_pages mask, passed to alloc_skb
403 * Allocate a new &sk_buff and assign it a usage count of one. The
404 * buffer has unspecified headroom built in. Users should allocate
405 * the headroom they think they need without accounting for the
406 * built in space. The built in space is used for optimisations.
408 * %NULL is returned if there is no free memory.
410 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
411 unsigned int length
, gfp_t gfp_mask
)
413 struct sk_buff
*skb
= NULL
;
414 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
415 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
417 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
420 if (sk_memalloc_socks())
421 gfp_mask
|= __GFP_MEMALLOC
;
423 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
426 skb
= build_skb(data
, fragsz
);
428 put_page(virt_to_head_page(data
));
431 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
432 SKB_ALLOC_RX
, NUMA_NO_NODE
);
435 skb_reserve(skb
, NET_SKB_PAD
);
440 EXPORT_SYMBOL(__netdev_alloc_skb
);
442 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
443 int size
, unsigned int truesize
)
445 skb_fill_page_desc(skb
, i
, page
, off
, size
);
447 skb
->data_len
+= size
;
448 skb
->truesize
+= truesize
;
450 EXPORT_SYMBOL(skb_add_rx_frag
);
452 void skb_coalesce_rx_frag(struct sk_buff
*skb
, int i
, int size
,
453 unsigned int truesize
)
455 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
457 skb_frag_size_add(frag
, size
);
459 skb
->data_len
+= size
;
460 skb
->truesize
+= truesize
;
462 EXPORT_SYMBOL(skb_coalesce_rx_frag
);
464 static void skb_drop_list(struct sk_buff
**listp
)
466 kfree_skb_list(*listp
);
470 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
472 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
475 static void skb_clone_fraglist(struct sk_buff
*skb
)
477 struct sk_buff
*list
;
479 skb_walk_frags(skb
, list
)
483 static void skb_free_head(struct sk_buff
*skb
)
486 put_page(virt_to_head_page(skb
->head
));
491 static void skb_release_data(struct sk_buff
*skb
)
494 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
495 &skb_shinfo(skb
)->dataref
)) {
496 if (skb_shinfo(skb
)->nr_frags
) {
498 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
499 skb_frag_unref(skb
, i
);
503 * If skb buf is from userspace, we need to notify the caller
504 * the lower device DMA has done;
506 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
507 struct ubuf_info
*uarg
;
509 uarg
= skb_shinfo(skb
)->destructor_arg
;
511 uarg
->callback(uarg
, true);
514 if (skb_has_frag_list(skb
))
515 skb_drop_fraglist(skb
);
522 * Free an skbuff by memory without cleaning the state.
524 static void kfree_skbmem(struct sk_buff
*skb
)
526 struct sk_buff
*other
;
527 atomic_t
*fclone_ref
;
529 switch (skb
->fclone
) {
530 case SKB_FCLONE_UNAVAILABLE
:
531 kmem_cache_free(skbuff_head_cache
, skb
);
534 case SKB_FCLONE_ORIG
:
535 fclone_ref
= (atomic_t
*) (skb
+ 2);
536 if (atomic_dec_and_test(fclone_ref
))
537 kmem_cache_free(skbuff_fclone_cache
, skb
);
540 case SKB_FCLONE_CLONE
:
541 fclone_ref
= (atomic_t
*) (skb
+ 1);
544 /* The clone portion is available for
545 * fast-cloning again.
547 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
549 if (atomic_dec_and_test(fclone_ref
))
550 kmem_cache_free(skbuff_fclone_cache
, other
);
555 static void skb_release_head_state(struct sk_buff
*skb
)
559 secpath_put(skb
->sp
);
561 if (skb
->destructor
) {
563 skb
->destructor(skb
);
565 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
566 nf_conntrack_put(skb
->nfct
);
568 #ifdef CONFIG_BRIDGE_NETFILTER
569 nf_bridge_put(skb
->nf_bridge
);
571 /* XXX: IS this still necessary? - JHS */
572 #ifdef CONFIG_NET_SCHED
574 #ifdef CONFIG_NET_CLS_ACT
580 /* Free everything but the sk_buff shell. */
581 static void skb_release_all(struct sk_buff
*skb
)
583 skb_release_head_state(skb
);
584 if (likely(skb
->head
))
585 skb_release_data(skb
);
589 * __kfree_skb - private function
592 * Free an sk_buff. Release anything attached to the buffer.
593 * Clean the state. This is an internal helper function. Users should
594 * always call kfree_skb
597 void __kfree_skb(struct sk_buff
*skb
)
599 skb_release_all(skb
);
602 EXPORT_SYMBOL(__kfree_skb
);
605 * kfree_skb - free an sk_buff
606 * @skb: buffer to free
608 * Drop a reference to the buffer and free it if the usage count has
611 void kfree_skb(struct sk_buff
*skb
)
615 if (likely(atomic_read(&skb
->users
) == 1))
617 else if (likely(!atomic_dec_and_test(&skb
->users
)))
619 trace_kfree_skb(skb
, __builtin_return_address(0));
622 EXPORT_SYMBOL(kfree_skb
);
624 void kfree_skb_list(struct sk_buff
*segs
)
627 struct sk_buff
*next
= segs
->next
;
633 EXPORT_SYMBOL(kfree_skb_list
);
636 * skb_tx_error - report an sk_buff xmit error
637 * @skb: buffer that triggered an error
639 * Report xmit error if a device callback is tracking this skb.
640 * skb must be freed afterwards.
642 void skb_tx_error(struct sk_buff
*skb
)
644 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
645 struct ubuf_info
*uarg
;
647 uarg
= skb_shinfo(skb
)->destructor_arg
;
649 uarg
->callback(uarg
, false);
650 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
653 EXPORT_SYMBOL(skb_tx_error
);
656 * consume_skb - free an skbuff
657 * @skb: buffer to free
659 * Drop a ref to the buffer and free it if the usage count has hit zero
660 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
661 * is being dropped after a failure and notes that
663 void consume_skb(struct sk_buff
*skb
)
667 if (likely(atomic_read(&skb
->users
) == 1))
669 else if (likely(!atomic_dec_and_test(&skb
->users
)))
671 trace_consume_skb(skb
);
674 EXPORT_SYMBOL(consume_skb
);
676 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
678 new->tstamp
= old
->tstamp
;
680 new->transport_header
= old
->transport_header
;
681 new->network_header
= old
->network_header
;
682 new->mac_header
= old
->mac_header
;
683 new->inner_protocol
= old
->inner_protocol
;
684 new->inner_transport_header
= old
->inner_transport_header
;
685 new->inner_network_header
= old
->inner_network_header
;
686 new->inner_mac_header
= old
->inner_mac_header
;
687 skb_dst_copy(new, old
);
688 skb_copy_hash(new, old
);
689 new->ooo_okay
= old
->ooo_okay
;
690 new->no_fcs
= old
->no_fcs
;
691 new->encapsulation
= old
->encapsulation
;
693 new->sp
= secpath_get(old
->sp
);
695 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
696 new->csum
= old
->csum
;
697 new->ignore_df
= old
->ignore_df
;
698 new->pkt_type
= old
->pkt_type
;
699 new->ip_summed
= old
->ip_summed
;
700 skb_copy_queue_mapping(new, old
);
701 new->priority
= old
->priority
;
702 #if IS_ENABLED(CONFIG_IP_VS)
703 new->ipvs_property
= old
->ipvs_property
;
705 new->pfmemalloc
= old
->pfmemalloc
;
706 new->protocol
= old
->protocol
;
707 new->mark
= old
->mark
;
708 new->skb_iif
= old
->skb_iif
;
710 #ifdef CONFIG_NET_SCHED
711 new->tc_index
= old
->tc_index
;
712 #ifdef CONFIG_NET_CLS_ACT
713 new->tc_verd
= old
->tc_verd
;
716 new->vlan_proto
= old
->vlan_proto
;
717 new->vlan_tci
= old
->vlan_tci
;
719 skb_copy_secmark(new, old
);
721 #ifdef CONFIG_NET_RX_BUSY_POLL
722 new->napi_id
= old
->napi_id
;
727 * You should not add any new code to this function. Add it to
728 * __copy_skb_header above instead.
730 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
732 #define C(x) n->x = skb->x
734 n
->next
= n
->prev
= NULL
;
736 __copy_skb_header(n
, skb
);
741 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
744 n
->destructor
= NULL
;
751 atomic_set(&n
->users
, 1);
753 atomic_inc(&(skb_shinfo(skb
)->dataref
));
761 * skb_morph - morph one skb into another
762 * @dst: the skb to receive the contents
763 * @src: the skb to supply the contents
765 * This is identical to skb_clone except that the target skb is
766 * supplied by the user.
768 * The target skb is returned upon exit.
770 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
772 skb_release_all(dst
);
773 return __skb_clone(dst
, src
);
775 EXPORT_SYMBOL_GPL(skb_morph
);
778 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
779 * @skb: the skb to modify
780 * @gfp_mask: allocation priority
782 * This must be called on SKBTX_DEV_ZEROCOPY skb.
783 * It will copy all frags into kernel and drop the reference
784 * to userspace pages.
786 * If this function is called from an interrupt gfp_mask() must be
789 * Returns 0 on success or a negative error code on failure
790 * to allocate kernel memory to copy to.
792 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
795 int num_frags
= skb_shinfo(skb
)->nr_frags
;
796 struct page
*page
, *head
= NULL
;
797 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
799 for (i
= 0; i
< num_frags
; i
++) {
801 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
803 page
= alloc_page(gfp_mask
);
806 struct page
*next
= (struct page
*)page_private(head
);
812 vaddr
= kmap_atomic(skb_frag_page(f
));
813 memcpy(page_address(page
),
814 vaddr
+ f
->page_offset
, skb_frag_size(f
));
815 kunmap_atomic(vaddr
);
816 set_page_private(page
, (unsigned long)head
);
820 /* skb frags release userspace buffers */
821 for (i
= 0; i
< num_frags
; i
++)
822 skb_frag_unref(skb
, i
);
824 uarg
->callback(uarg
, false);
826 /* skb frags point to kernel buffers */
827 for (i
= num_frags
- 1; i
>= 0; i
--) {
828 __skb_fill_page_desc(skb
, i
, head
, 0,
829 skb_shinfo(skb
)->frags
[i
].size
);
830 head
= (struct page
*)page_private(head
);
833 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
836 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
839 * skb_clone - duplicate an sk_buff
840 * @skb: buffer to clone
841 * @gfp_mask: allocation priority
843 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
844 * copies share the same packet data but not structure. The new
845 * buffer has a reference count of 1. If the allocation fails the
846 * function returns %NULL otherwise the new buffer is returned.
848 * If this function is called from an interrupt gfp_mask() must be
852 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
856 if (skb_orphan_frags(skb
, gfp_mask
))
860 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
861 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
862 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
863 n
->fclone
= SKB_FCLONE_CLONE
;
864 atomic_inc(fclone_ref
);
866 if (skb_pfmemalloc(skb
))
867 gfp_mask
|= __GFP_MEMALLOC
;
869 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
873 kmemcheck_annotate_bitfield(n
, flags1
);
874 kmemcheck_annotate_bitfield(n
, flags2
);
875 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
878 return __skb_clone(n
, skb
);
880 EXPORT_SYMBOL(skb_clone
);
882 static void skb_headers_offset_update(struct sk_buff
*skb
, int off
)
884 /* Only adjust this if it actually is csum_start rather than csum */
885 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
886 skb
->csum_start
+= off
;
887 /* {transport,network,mac}_header and tail are relative to skb->head */
888 skb
->transport_header
+= off
;
889 skb
->network_header
+= off
;
890 if (skb_mac_header_was_set(skb
))
891 skb
->mac_header
+= off
;
892 skb
->inner_transport_header
+= off
;
893 skb
->inner_network_header
+= off
;
894 skb
->inner_mac_header
+= off
;
897 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
899 __copy_skb_header(new, old
);
901 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
902 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
903 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
906 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
908 if (skb_pfmemalloc(skb
))
914 * skb_copy - create private copy of an sk_buff
915 * @skb: buffer to copy
916 * @gfp_mask: allocation priority
918 * Make a copy of both an &sk_buff and its data. This is used when the
919 * caller wishes to modify the data and needs a private copy of the
920 * data to alter. Returns %NULL on failure or the pointer to the buffer
921 * on success. The returned buffer has a reference count of 1.
923 * As by-product this function converts non-linear &sk_buff to linear
924 * one, so that &sk_buff becomes completely private and caller is allowed
925 * to modify all the data of returned buffer. This means that this
926 * function is not recommended for use in circumstances when only
927 * header is going to be modified. Use pskb_copy() instead.
930 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
932 int headerlen
= skb_headroom(skb
);
933 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
934 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
935 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
940 /* Set the data pointer */
941 skb_reserve(n
, headerlen
);
942 /* Set the tail pointer and length */
943 skb_put(n
, skb
->len
);
945 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
948 copy_skb_header(n
, skb
);
951 EXPORT_SYMBOL(skb_copy
);
954 * __pskb_copy_fclone - create copy of an sk_buff with private head.
955 * @skb: buffer to copy
956 * @headroom: headroom of new skb
957 * @gfp_mask: allocation priority
958 * @fclone: if true allocate the copy of the skb from the fclone
959 * cache instead of the head cache; it is recommended to set this
960 * to true for the cases where the copy will likely be cloned
962 * Make a copy of both an &sk_buff and part of its data, located
963 * in header. Fragmented data remain shared. This is used when
964 * the caller wishes to modify only header of &sk_buff and needs
965 * private copy of the header to alter. Returns %NULL on failure
966 * or the pointer to the buffer on success.
967 * The returned buffer has a reference count of 1.
970 struct sk_buff
*__pskb_copy_fclone(struct sk_buff
*skb
, int headroom
,
971 gfp_t gfp_mask
, bool fclone
)
973 unsigned int size
= skb_headlen(skb
) + headroom
;
974 int flags
= skb_alloc_rx_flag(skb
) | (fclone
? SKB_ALLOC_FCLONE
: 0);
975 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
, flags
, NUMA_NO_NODE
);
980 /* Set the data pointer */
981 skb_reserve(n
, headroom
);
982 /* Set the tail pointer and length */
983 skb_put(n
, skb_headlen(skb
));
985 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
987 n
->truesize
+= skb
->data_len
;
988 n
->data_len
= skb
->data_len
;
991 if (skb_shinfo(skb
)->nr_frags
) {
994 if (skb_orphan_frags(skb
, gfp_mask
)) {
999 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1000 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1001 skb_frag_ref(skb
, i
);
1003 skb_shinfo(n
)->nr_frags
= i
;
1006 if (skb_has_frag_list(skb
)) {
1007 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1008 skb_clone_fraglist(n
);
1011 copy_skb_header(n
, skb
);
1015 EXPORT_SYMBOL(__pskb_copy_fclone
);
1018 * pskb_expand_head - reallocate header of &sk_buff
1019 * @skb: buffer to reallocate
1020 * @nhead: room to add at head
1021 * @ntail: room to add at tail
1022 * @gfp_mask: allocation priority
1024 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1025 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1026 * reference count of 1. Returns zero in the case of success or error,
1027 * if expansion failed. In the last case, &sk_buff is not changed.
1029 * All the pointers pointing into skb header may change and must be
1030 * reloaded after call to this function.
1033 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1038 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1043 if (skb_shared(skb
))
1046 size
= SKB_DATA_ALIGN(size
);
1048 if (skb_pfmemalloc(skb
))
1049 gfp_mask
|= __GFP_MEMALLOC
;
1050 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1051 gfp_mask
, NUMA_NO_NODE
, NULL
);
1054 size
= SKB_WITH_OVERHEAD(ksize(data
));
1056 /* Copy only real data... and, alas, header. This should be
1057 * optimized for the cases when header is void.
1059 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1061 memcpy((struct skb_shared_info
*)(data
+ size
),
1063 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1066 * if shinfo is shared we must drop the old head gracefully, but if it
1067 * is not we can just drop the old head and let the existing refcount
1068 * be since all we did is relocate the values
1070 if (skb_cloned(skb
)) {
1071 /* copy this zero copy skb frags */
1072 if (skb_orphan_frags(skb
, gfp_mask
))
1074 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1075 skb_frag_ref(skb
, i
);
1077 if (skb_has_frag_list(skb
))
1078 skb_clone_fraglist(skb
);
1080 skb_release_data(skb
);
1084 off
= (data
+ nhead
) - skb
->head
;
1089 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1093 skb
->end
= skb
->head
+ size
;
1096 skb_headers_offset_update(skb
, nhead
);
1100 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1108 EXPORT_SYMBOL(pskb_expand_head
);
1110 /* Make private copy of skb with writable head and some headroom */
1112 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1114 struct sk_buff
*skb2
;
1115 int delta
= headroom
- skb_headroom(skb
);
1118 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1120 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1121 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1129 EXPORT_SYMBOL(skb_realloc_headroom
);
1132 * skb_copy_expand - copy and expand sk_buff
1133 * @skb: buffer to copy
1134 * @newheadroom: new free bytes at head
1135 * @newtailroom: new free bytes at tail
1136 * @gfp_mask: allocation priority
1138 * Make a copy of both an &sk_buff and its data and while doing so
1139 * allocate additional space.
1141 * This is used when the caller wishes to modify the data and needs a
1142 * private copy of the data to alter as well as more space for new fields.
1143 * Returns %NULL on failure or the pointer to the buffer
1144 * on success. The returned buffer has a reference count of 1.
1146 * You must pass %GFP_ATOMIC as the allocation priority if this function
1147 * is called from an interrupt.
1149 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1150 int newheadroom
, int newtailroom
,
1154 * Allocate the copy buffer
1156 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1157 gfp_mask
, skb_alloc_rx_flag(skb
),
1159 int oldheadroom
= skb_headroom(skb
);
1160 int head_copy_len
, head_copy_off
;
1165 skb_reserve(n
, newheadroom
);
1167 /* Set the tail pointer and length */
1168 skb_put(n
, skb
->len
);
1170 head_copy_len
= oldheadroom
;
1172 if (newheadroom
<= head_copy_len
)
1173 head_copy_len
= newheadroom
;
1175 head_copy_off
= newheadroom
- head_copy_len
;
1177 /* Copy the linear header and data. */
1178 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1179 skb
->len
+ head_copy_len
))
1182 copy_skb_header(n
, skb
);
1184 skb_headers_offset_update(n
, newheadroom
- oldheadroom
);
1188 EXPORT_SYMBOL(skb_copy_expand
);
1191 * skb_pad - zero pad the tail of an skb
1192 * @skb: buffer to pad
1193 * @pad: space to pad
1195 * Ensure that a buffer is followed by a padding area that is zero
1196 * filled. Used by network drivers which may DMA or transfer data
1197 * beyond the buffer end onto the wire.
1199 * May return error in out of memory cases. The skb is freed on error.
1202 int skb_pad(struct sk_buff
*skb
, int pad
)
1207 /* If the skbuff is non linear tailroom is always zero.. */
1208 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1209 memset(skb
->data
+skb
->len
, 0, pad
);
1213 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1214 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1215 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1220 /* FIXME: The use of this function with non-linear skb's really needs
1223 err
= skb_linearize(skb
);
1227 memset(skb
->data
+ skb
->len
, 0, pad
);
1234 EXPORT_SYMBOL(skb_pad
);
1237 * pskb_put - add data to the tail of a potentially fragmented buffer
1238 * @skb: start of the buffer to use
1239 * @tail: tail fragment of the buffer to use
1240 * @len: amount of data to add
1242 * This function extends the used data area of the potentially
1243 * fragmented buffer. @tail must be the last fragment of @skb -- or
1244 * @skb itself. If this would exceed the total buffer size the kernel
1245 * will panic. A pointer to the first byte of the extra data is
1249 unsigned char *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
)
1252 skb
->data_len
+= len
;
1255 return skb_put(tail
, len
);
1257 EXPORT_SYMBOL_GPL(pskb_put
);
1260 * skb_put - add data to a buffer
1261 * @skb: buffer to use
1262 * @len: amount of data to add
1264 * This function extends the used data area of the buffer. If this would
1265 * exceed the total buffer size the kernel will panic. A pointer to the
1266 * first byte of the extra data is returned.
1268 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1270 unsigned char *tmp
= skb_tail_pointer(skb
);
1271 SKB_LINEAR_ASSERT(skb
);
1274 if (unlikely(skb
->tail
> skb
->end
))
1275 skb_over_panic(skb
, len
, __builtin_return_address(0));
1278 EXPORT_SYMBOL(skb_put
);
1281 * skb_push - add data to the start of a buffer
1282 * @skb: buffer to use
1283 * @len: amount of data to add
1285 * This function extends the used data area of the buffer at the buffer
1286 * start. If this would exceed the total buffer headroom the kernel will
1287 * panic. A pointer to the first byte of the extra data is returned.
1289 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1293 if (unlikely(skb
->data
<skb
->head
))
1294 skb_under_panic(skb
, len
, __builtin_return_address(0));
1297 EXPORT_SYMBOL(skb_push
);
1300 * skb_pull - remove data from the start of a buffer
1301 * @skb: buffer to use
1302 * @len: amount of data to remove
1304 * This function removes data from the start of a buffer, returning
1305 * the memory to the headroom. A pointer to the next data in the buffer
1306 * is returned. Once the data has been pulled future pushes will overwrite
1309 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1311 return skb_pull_inline(skb
, len
);
1313 EXPORT_SYMBOL(skb_pull
);
1316 * skb_trim - remove end from a buffer
1317 * @skb: buffer to alter
1320 * Cut the length of a buffer down by removing data from the tail. If
1321 * the buffer is already under the length specified it is not modified.
1322 * The skb must be linear.
1324 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1327 __skb_trim(skb
, len
);
1329 EXPORT_SYMBOL(skb_trim
);
1331 /* Trims skb to length len. It can change skb pointers.
1334 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1336 struct sk_buff
**fragp
;
1337 struct sk_buff
*frag
;
1338 int offset
= skb_headlen(skb
);
1339 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1343 if (skb_cloned(skb
) &&
1344 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1351 for (; i
< nfrags
; i
++) {
1352 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1359 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1362 skb_shinfo(skb
)->nr_frags
= i
;
1364 for (; i
< nfrags
; i
++)
1365 skb_frag_unref(skb
, i
);
1367 if (skb_has_frag_list(skb
))
1368 skb_drop_fraglist(skb
);
1372 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1373 fragp
= &frag
->next
) {
1374 int end
= offset
+ frag
->len
;
1376 if (skb_shared(frag
)) {
1377 struct sk_buff
*nfrag
;
1379 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1380 if (unlikely(!nfrag
))
1383 nfrag
->next
= frag
->next
;
1395 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1399 skb_drop_list(&frag
->next
);
1404 if (len
> skb_headlen(skb
)) {
1405 skb
->data_len
-= skb
->len
- len
;
1410 skb_set_tail_pointer(skb
, len
);
1415 EXPORT_SYMBOL(___pskb_trim
);
1418 * __pskb_pull_tail - advance tail of skb header
1419 * @skb: buffer to reallocate
1420 * @delta: number of bytes to advance tail
1422 * The function makes a sense only on a fragmented &sk_buff,
1423 * it expands header moving its tail forward and copying necessary
1424 * data from fragmented part.
1426 * &sk_buff MUST have reference count of 1.
1428 * Returns %NULL (and &sk_buff does not change) if pull failed
1429 * or value of new tail of skb in the case of success.
1431 * All the pointers pointing into skb header may change and must be
1432 * reloaded after call to this function.
1435 /* Moves tail of skb head forward, copying data from fragmented part,
1436 * when it is necessary.
1437 * 1. It may fail due to malloc failure.
1438 * 2. It may change skb pointers.
1440 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1442 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1444 /* If skb has not enough free space at tail, get new one
1445 * plus 128 bytes for future expansions. If we have enough
1446 * room at tail, reallocate without expansion only if skb is cloned.
1448 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1450 if (eat
> 0 || skb_cloned(skb
)) {
1451 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1456 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1459 /* Optimization: no fragments, no reasons to preestimate
1460 * size of pulled pages. Superb.
1462 if (!skb_has_frag_list(skb
))
1465 /* Estimate size of pulled pages. */
1467 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1468 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1475 /* If we need update frag list, we are in troubles.
1476 * Certainly, it possible to add an offset to skb data,
1477 * but taking into account that pulling is expected to
1478 * be very rare operation, it is worth to fight against
1479 * further bloating skb head and crucify ourselves here instead.
1480 * Pure masohism, indeed. 8)8)
1483 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1484 struct sk_buff
*clone
= NULL
;
1485 struct sk_buff
*insp
= NULL
;
1490 if (list
->len
<= eat
) {
1491 /* Eaten as whole. */
1496 /* Eaten partially. */
1498 if (skb_shared(list
)) {
1499 /* Sucks! We need to fork list. :-( */
1500 clone
= skb_clone(list
, GFP_ATOMIC
);
1506 /* This may be pulled without
1510 if (!pskb_pull(list
, eat
)) {
1518 /* Free pulled out fragments. */
1519 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1520 skb_shinfo(skb
)->frag_list
= list
->next
;
1523 /* And insert new clone at head. */
1526 skb_shinfo(skb
)->frag_list
= clone
;
1529 /* Success! Now we may commit changes to skb data. */
1534 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1535 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1538 skb_frag_unref(skb
, i
);
1541 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1543 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1544 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1550 skb_shinfo(skb
)->nr_frags
= k
;
1553 skb
->data_len
-= delta
;
1555 return skb_tail_pointer(skb
);
1557 EXPORT_SYMBOL(__pskb_pull_tail
);
1560 * skb_copy_bits - copy bits from skb to kernel buffer
1562 * @offset: offset in source
1563 * @to: destination buffer
1564 * @len: number of bytes to copy
1566 * Copy the specified number of bytes from the source skb to the
1567 * destination buffer.
1570 * If its prototype is ever changed,
1571 * check arch/{*}/net/{*}.S files,
1572 * since it is called from BPF assembly code.
1574 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1576 int start
= skb_headlen(skb
);
1577 struct sk_buff
*frag_iter
;
1580 if (offset
> (int)skb
->len
- len
)
1584 if ((copy
= start
- offset
) > 0) {
1587 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1588 if ((len
-= copy
) == 0)
1594 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1596 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1598 WARN_ON(start
> offset
+ len
);
1600 end
= start
+ skb_frag_size(f
);
1601 if ((copy
= end
- offset
) > 0) {
1607 vaddr
= kmap_atomic(skb_frag_page(f
));
1609 vaddr
+ f
->page_offset
+ offset
- start
,
1611 kunmap_atomic(vaddr
);
1613 if ((len
-= copy
) == 0)
1621 skb_walk_frags(skb
, frag_iter
) {
1624 WARN_ON(start
> offset
+ len
);
1626 end
= start
+ frag_iter
->len
;
1627 if ((copy
= end
- offset
) > 0) {
1630 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1632 if ((len
-= copy
) == 0)
1646 EXPORT_SYMBOL(skb_copy_bits
);
1649 * Callback from splice_to_pipe(), if we need to release some pages
1650 * at the end of the spd in case we error'ed out in filling the pipe.
1652 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1654 put_page(spd
->pages
[i
]);
1657 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1658 unsigned int *offset
,
1661 struct page_frag
*pfrag
= sk_page_frag(sk
);
1663 if (!sk_page_frag_refill(sk
, pfrag
))
1666 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1668 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1669 page_address(page
) + *offset
, *len
);
1670 *offset
= pfrag
->offset
;
1671 pfrag
->offset
+= *len
;
1676 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1678 unsigned int offset
)
1680 return spd
->nr_pages
&&
1681 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1682 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1683 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1687 * Fill page/offset/length into spd, if it can hold more pages.
1689 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1690 struct pipe_inode_info
*pipe
, struct page
*page
,
1691 unsigned int *len
, unsigned int offset
,
1695 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1699 page
= linear_to_page(page
, len
, &offset
, sk
);
1703 if (spd_can_coalesce(spd
, page
, offset
)) {
1704 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1708 spd
->pages
[spd
->nr_pages
] = page
;
1709 spd
->partial
[spd
->nr_pages
].len
= *len
;
1710 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1716 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1717 unsigned int plen
, unsigned int *off
,
1719 struct splice_pipe_desc
*spd
, bool linear
,
1721 struct pipe_inode_info
*pipe
)
1726 /* skip this segment if already processed */
1732 /* ignore any bits we already processed */
1738 unsigned int flen
= min(*len
, plen
);
1740 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1746 } while (*len
&& plen
);
1752 * Map linear and fragment data from the skb to spd. It reports true if the
1753 * pipe is full or if we already spliced the requested length.
1755 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1756 unsigned int *offset
, unsigned int *len
,
1757 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1761 /* map the linear part :
1762 * If skb->head_frag is set, this 'linear' part is backed by a
1763 * fragment, and if the head is not shared with any clones then
1764 * we can avoid a copy since we own the head portion of this page.
1766 if (__splice_segment(virt_to_page(skb
->data
),
1767 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1770 skb_head_is_locked(skb
),
1775 * then map the fragments
1777 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1778 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1780 if (__splice_segment(skb_frag_page(f
),
1781 f
->page_offset
, skb_frag_size(f
),
1782 offset
, len
, spd
, false, sk
, pipe
))
1790 * Map data from the skb to a pipe. Should handle both the linear part,
1791 * the fragments, and the frag list. It does NOT handle frag lists within
1792 * the frag list, if such a thing exists. We'd probably need to recurse to
1793 * handle that cleanly.
1795 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1796 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1799 struct partial_page partial
[MAX_SKB_FRAGS
];
1800 struct page
*pages
[MAX_SKB_FRAGS
];
1801 struct splice_pipe_desc spd
= {
1804 .nr_pages_max
= MAX_SKB_FRAGS
,
1806 .ops
= &nosteal_pipe_buf_ops
,
1807 .spd_release
= sock_spd_release
,
1809 struct sk_buff
*frag_iter
;
1810 struct sock
*sk
= skb
->sk
;
1814 * __skb_splice_bits() only fails if the output has no room left,
1815 * so no point in going over the frag_list for the error case.
1817 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1823 * now see if we have a frag_list to map
1825 skb_walk_frags(skb
, frag_iter
) {
1828 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1835 * Drop the socket lock, otherwise we have reverse
1836 * locking dependencies between sk_lock and i_mutex
1837 * here as compared to sendfile(). We enter here
1838 * with the socket lock held, and splice_to_pipe() will
1839 * grab the pipe inode lock. For sendfile() emulation,
1840 * we call into ->sendpage() with the i_mutex lock held
1841 * and networking will grab the socket lock.
1844 ret
= splice_to_pipe(pipe
, &spd
);
1852 * skb_store_bits - store bits from kernel buffer to skb
1853 * @skb: destination buffer
1854 * @offset: offset in destination
1855 * @from: source buffer
1856 * @len: number of bytes to copy
1858 * Copy the specified number of bytes from the source buffer to the
1859 * destination skb. This function handles all the messy bits of
1860 * traversing fragment lists and such.
1863 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1865 int start
= skb_headlen(skb
);
1866 struct sk_buff
*frag_iter
;
1869 if (offset
> (int)skb
->len
- len
)
1872 if ((copy
= start
- offset
) > 0) {
1875 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1876 if ((len
-= copy
) == 0)
1882 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1883 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1886 WARN_ON(start
> offset
+ len
);
1888 end
= start
+ skb_frag_size(frag
);
1889 if ((copy
= end
- offset
) > 0) {
1895 vaddr
= kmap_atomic(skb_frag_page(frag
));
1896 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1898 kunmap_atomic(vaddr
);
1900 if ((len
-= copy
) == 0)
1908 skb_walk_frags(skb
, frag_iter
) {
1911 WARN_ON(start
> offset
+ len
);
1913 end
= start
+ frag_iter
->len
;
1914 if ((copy
= end
- offset
) > 0) {
1917 if (skb_store_bits(frag_iter
, offset
- start
,
1920 if ((len
-= copy
) == 0)
1933 EXPORT_SYMBOL(skb_store_bits
);
1935 /* Checksum skb data. */
1936 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
1937 __wsum csum
, const struct skb_checksum_ops
*ops
)
1939 int start
= skb_headlen(skb
);
1940 int i
, copy
= start
- offset
;
1941 struct sk_buff
*frag_iter
;
1944 /* Checksum header. */
1948 csum
= ops
->update(skb
->data
+ offset
, copy
, csum
);
1949 if ((len
-= copy
) == 0)
1955 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1957 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1959 WARN_ON(start
> offset
+ len
);
1961 end
= start
+ skb_frag_size(frag
);
1962 if ((copy
= end
- offset
) > 0) {
1968 vaddr
= kmap_atomic(skb_frag_page(frag
));
1969 csum2
= ops
->update(vaddr
+ frag
->page_offset
+
1970 offset
- start
, copy
, 0);
1971 kunmap_atomic(vaddr
);
1972 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
1981 skb_walk_frags(skb
, frag_iter
) {
1984 WARN_ON(start
> offset
+ len
);
1986 end
= start
+ frag_iter
->len
;
1987 if ((copy
= end
- offset
) > 0) {
1991 csum2
= __skb_checksum(frag_iter
, offset
- start
,
1993 csum
= ops
->combine(csum
, csum2
, pos
, copy
);
1994 if ((len
-= copy
) == 0)
2005 EXPORT_SYMBOL(__skb_checksum
);
2007 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
2008 int len
, __wsum csum
)
2010 const struct skb_checksum_ops ops
= {
2011 .update
= csum_partial_ext
,
2012 .combine
= csum_block_add_ext
,
2015 return __skb_checksum(skb
, offset
, len
, csum
, &ops
);
2017 EXPORT_SYMBOL(skb_checksum
);
2019 /* Both of above in one bottle. */
2021 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2022 u8
*to
, int len
, __wsum csum
)
2024 int start
= skb_headlen(skb
);
2025 int i
, copy
= start
- offset
;
2026 struct sk_buff
*frag_iter
;
2033 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2035 if ((len
-= copy
) == 0)
2042 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2045 WARN_ON(start
> offset
+ len
);
2047 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2048 if ((copy
= end
- offset
) > 0) {
2051 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2055 vaddr
= kmap_atomic(skb_frag_page(frag
));
2056 csum2
= csum_partial_copy_nocheck(vaddr
+
2060 kunmap_atomic(vaddr
);
2061 csum
= csum_block_add(csum
, csum2
, pos
);
2071 skb_walk_frags(skb
, frag_iter
) {
2075 WARN_ON(start
> offset
+ len
);
2077 end
= start
+ frag_iter
->len
;
2078 if ((copy
= end
- offset
) > 0) {
2081 csum2
= skb_copy_and_csum_bits(frag_iter
,
2084 csum
= csum_block_add(csum
, csum2
, pos
);
2085 if ((len
-= copy
) == 0)
2096 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2099 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2100 * @from: source buffer
2102 * Calculates the amount of linear headroom needed in the 'to' skb passed
2103 * into skb_zerocopy().
2106 skb_zerocopy_headlen(const struct sk_buff
*from
)
2108 unsigned int hlen
= 0;
2110 if (!from
->head_frag
||
2111 skb_headlen(from
) < L1_CACHE_BYTES
||
2112 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
2113 hlen
= skb_headlen(from
);
2115 if (skb_has_frag_list(from
))
2120 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen
);
2123 * skb_zerocopy - Zero copy skb to skb
2124 * @to: destination buffer
2125 * @from: source buffer
2126 * @len: number of bytes to copy from source buffer
2127 * @hlen: size of linear headroom in destination buffer
2129 * Copies up to `len` bytes from `from` to `to` by creating references
2130 * to the frags in the source buffer.
2132 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2133 * headroom in the `to` buffer.
2136 * 0: everything is OK
2137 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2138 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2141 skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
, int len
, int hlen
)
2144 int plen
= 0; /* length of skb->head fragment */
2147 unsigned int offset
;
2149 BUG_ON(!from
->head_frag
&& !hlen
);
2151 /* dont bother with small payloads */
2152 if (len
<= skb_tailroom(to
))
2153 return skb_copy_bits(from
, 0, skb_put(to
, len
), len
);
2156 ret
= skb_copy_bits(from
, 0, skb_put(to
, hlen
), hlen
);
2161 plen
= min_t(int, skb_headlen(from
), len
);
2163 page
= virt_to_head_page(from
->head
);
2164 offset
= from
->data
- (unsigned char *)page_address(page
);
2165 __skb_fill_page_desc(to
, 0, page
, offset
, plen
);
2172 to
->truesize
+= len
+ plen
;
2173 to
->len
+= len
+ plen
;
2174 to
->data_len
+= len
+ plen
;
2176 if (unlikely(skb_orphan_frags(from
, GFP_ATOMIC
))) {
2181 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++) {
2184 skb_shinfo(to
)->frags
[j
] = skb_shinfo(from
)->frags
[i
];
2185 skb_shinfo(to
)->frags
[j
].size
= min_t(int, skb_shinfo(to
)->frags
[j
].size
, len
);
2186 len
-= skb_shinfo(to
)->frags
[j
].size
;
2187 skb_frag_ref(to
, j
);
2190 skb_shinfo(to
)->nr_frags
= j
;
2194 EXPORT_SYMBOL_GPL(skb_zerocopy
);
2196 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2201 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2202 csstart
= skb_checksum_start_offset(skb
);
2204 csstart
= skb_headlen(skb
);
2206 BUG_ON(csstart
> skb_headlen(skb
));
2208 skb_copy_from_linear_data(skb
, to
, csstart
);
2211 if (csstart
!= skb
->len
)
2212 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2213 skb
->len
- csstart
, 0);
2215 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2216 long csstuff
= csstart
+ skb
->csum_offset
;
2218 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2221 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2224 * skb_dequeue - remove from the head of the queue
2225 * @list: list to dequeue from
2227 * Remove the head of the list. The list lock is taken so the function
2228 * may be used safely with other locking list functions. The head item is
2229 * returned or %NULL if the list is empty.
2232 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2234 unsigned long flags
;
2235 struct sk_buff
*result
;
2237 spin_lock_irqsave(&list
->lock
, flags
);
2238 result
= __skb_dequeue(list
);
2239 spin_unlock_irqrestore(&list
->lock
, flags
);
2242 EXPORT_SYMBOL(skb_dequeue
);
2245 * skb_dequeue_tail - remove from the tail of the queue
2246 * @list: list to dequeue from
2248 * Remove the tail of the list. The list lock is taken so the function
2249 * may be used safely with other locking list functions. The tail item is
2250 * returned or %NULL if the list is empty.
2252 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2254 unsigned long flags
;
2255 struct sk_buff
*result
;
2257 spin_lock_irqsave(&list
->lock
, flags
);
2258 result
= __skb_dequeue_tail(list
);
2259 spin_unlock_irqrestore(&list
->lock
, flags
);
2262 EXPORT_SYMBOL(skb_dequeue_tail
);
2265 * skb_queue_purge - empty a list
2266 * @list: list to empty
2268 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2269 * the list and one reference dropped. This function takes the list
2270 * lock and is atomic with respect to other list locking functions.
2272 void skb_queue_purge(struct sk_buff_head
*list
)
2274 struct sk_buff
*skb
;
2275 while ((skb
= skb_dequeue(list
)) != NULL
)
2278 EXPORT_SYMBOL(skb_queue_purge
);
2281 * skb_queue_head - queue a buffer at the list head
2282 * @list: list to use
2283 * @newsk: buffer to queue
2285 * Queue a buffer at the start of the list. This function takes the
2286 * list lock and can be used safely with other locking &sk_buff functions
2289 * A buffer cannot be placed on two lists at the same time.
2291 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2293 unsigned long flags
;
2295 spin_lock_irqsave(&list
->lock
, flags
);
2296 __skb_queue_head(list
, newsk
);
2297 spin_unlock_irqrestore(&list
->lock
, flags
);
2299 EXPORT_SYMBOL(skb_queue_head
);
2302 * skb_queue_tail - queue a buffer at the list tail
2303 * @list: list to use
2304 * @newsk: buffer to queue
2306 * Queue a buffer at the tail of the list. This function takes the
2307 * list lock and can be used safely with other locking &sk_buff functions
2310 * A buffer cannot be placed on two lists at the same time.
2312 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2314 unsigned long flags
;
2316 spin_lock_irqsave(&list
->lock
, flags
);
2317 __skb_queue_tail(list
, newsk
);
2318 spin_unlock_irqrestore(&list
->lock
, flags
);
2320 EXPORT_SYMBOL(skb_queue_tail
);
2323 * skb_unlink - remove a buffer from a list
2324 * @skb: buffer to remove
2325 * @list: list to use
2327 * Remove a packet from a list. The list locks are taken and this
2328 * function is atomic with respect to other list locked calls
2330 * You must know what list the SKB is on.
2332 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2334 unsigned long flags
;
2336 spin_lock_irqsave(&list
->lock
, flags
);
2337 __skb_unlink(skb
, list
);
2338 spin_unlock_irqrestore(&list
->lock
, flags
);
2340 EXPORT_SYMBOL(skb_unlink
);
2343 * skb_append - append a buffer
2344 * @old: buffer to insert after
2345 * @newsk: buffer to insert
2346 * @list: list to use
2348 * Place a packet after a given packet in a list. The list locks are taken
2349 * and this function is atomic with respect to other list locked calls.
2350 * A buffer cannot be placed on two lists at the same time.
2352 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2354 unsigned long flags
;
2356 spin_lock_irqsave(&list
->lock
, flags
);
2357 __skb_queue_after(list
, old
, newsk
);
2358 spin_unlock_irqrestore(&list
->lock
, flags
);
2360 EXPORT_SYMBOL(skb_append
);
2363 * skb_insert - insert a buffer
2364 * @old: buffer to insert before
2365 * @newsk: buffer to insert
2366 * @list: list to use
2368 * Place a packet before a given packet in a list. The list locks are
2369 * taken and this function is atomic with respect to other list locked
2372 * A buffer cannot be placed on two lists at the same time.
2374 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2376 unsigned long flags
;
2378 spin_lock_irqsave(&list
->lock
, flags
);
2379 __skb_insert(newsk
, old
->prev
, old
, list
);
2380 spin_unlock_irqrestore(&list
->lock
, flags
);
2382 EXPORT_SYMBOL(skb_insert
);
2384 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2385 struct sk_buff
* skb1
,
2386 const u32 len
, const int pos
)
2390 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2392 /* And move data appendix as is. */
2393 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2394 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2396 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2397 skb_shinfo(skb
)->nr_frags
= 0;
2398 skb1
->data_len
= skb
->data_len
;
2399 skb1
->len
+= skb1
->data_len
;
2402 skb_set_tail_pointer(skb
, len
);
2405 static inline void skb_split_no_header(struct sk_buff
*skb
,
2406 struct sk_buff
* skb1
,
2407 const u32 len
, int pos
)
2410 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2412 skb_shinfo(skb
)->nr_frags
= 0;
2413 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2415 skb
->data_len
= len
- pos
;
2417 for (i
= 0; i
< nfrags
; i
++) {
2418 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2420 if (pos
+ size
> len
) {
2421 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2425 * We have two variants in this case:
2426 * 1. Move all the frag to the second
2427 * part, if it is possible. F.e.
2428 * this approach is mandatory for TUX,
2429 * where splitting is expensive.
2430 * 2. Split is accurately. We make this.
2432 skb_frag_ref(skb
, i
);
2433 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2434 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2435 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2436 skb_shinfo(skb
)->nr_frags
++;
2440 skb_shinfo(skb
)->nr_frags
++;
2443 skb_shinfo(skb1
)->nr_frags
= k
;
2447 * skb_split - Split fragmented skb to two parts at length len.
2448 * @skb: the buffer to split
2449 * @skb1: the buffer to receive the second part
2450 * @len: new length for skb
2452 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2454 int pos
= skb_headlen(skb
);
2456 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2457 if (len
< pos
) /* Split line is inside header. */
2458 skb_split_inside_header(skb
, skb1
, len
, pos
);
2459 else /* Second chunk has no header, nothing to copy. */
2460 skb_split_no_header(skb
, skb1
, len
, pos
);
2462 EXPORT_SYMBOL(skb_split
);
2464 /* Shifting from/to a cloned skb is a no-go.
2466 * Caller cannot keep skb_shinfo related pointers past calling here!
2468 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2470 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2474 * skb_shift - Shifts paged data partially from skb to another
2475 * @tgt: buffer into which tail data gets added
2476 * @skb: buffer from which the paged data comes from
2477 * @shiftlen: shift up to this many bytes
2479 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2480 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2481 * It's up to caller to free skb if everything was shifted.
2483 * If @tgt runs out of frags, the whole operation is aborted.
2485 * Skb cannot include anything else but paged data while tgt is allowed
2486 * to have non-paged data as well.
2488 * TODO: full sized shift could be optimized but that would need
2489 * specialized skb free'er to handle frags without up-to-date nr_frags.
2491 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2493 int from
, to
, merge
, todo
;
2494 struct skb_frag_struct
*fragfrom
, *fragto
;
2496 BUG_ON(shiftlen
> skb
->len
);
2497 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2501 to
= skb_shinfo(tgt
)->nr_frags
;
2502 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2504 /* Actual merge is delayed until the point when we know we can
2505 * commit all, so that we don't have to undo partial changes
2508 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2509 fragfrom
->page_offset
)) {
2514 todo
-= skb_frag_size(fragfrom
);
2516 if (skb_prepare_for_shift(skb
) ||
2517 skb_prepare_for_shift(tgt
))
2520 /* All previous frag pointers might be stale! */
2521 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2522 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2524 skb_frag_size_add(fragto
, shiftlen
);
2525 skb_frag_size_sub(fragfrom
, shiftlen
);
2526 fragfrom
->page_offset
+= shiftlen
;
2534 /* Skip full, not-fitting skb to avoid expensive operations */
2535 if ((shiftlen
== skb
->len
) &&
2536 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2539 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2542 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2543 if (to
== MAX_SKB_FRAGS
)
2546 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2547 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2549 if (todo
>= skb_frag_size(fragfrom
)) {
2550 *fragto
= *fragfrom
;
2551 todo
-= skb_frag_size(fragfrom
);
2556 __skb_frag_ref(fragfrom
);
2557 fragto
->page
= fragfrom
->page
;
2558 fragto
->page_offset
= fragfrom
->page_offset
;
2559 skb_frag_size_set(fragto
, todo
);
2561 fragfrom
->page_offset
+= todo
;
2562 skb_frag_size_sub(fragfrom
, todo
);
2570 /* Ready to "commit" this state change to tgt */
2571 skb_shinfo(tgt
)->nr_frags
= to
;
2574 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2575 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2577 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2578 __skb_frag_unref(fragfrom
);
2581 /* Reposition in the original skb */
2583 while (from
< skb_shinfo(skb
)->nr_frags
)
2584 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2585 skb_shinfo(skb
)->nr_frags
= to
;
2587 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2590 /* Most likely the tgt won't ever need its checksum anymore, skb on
2591 * the other hand might need it if it needs to be resent
2593 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2594 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2596 /* Yak, is it really working this way? Some helper please? */
2597 skb
->len
-= shiftlen
;
2598 skb
->data_len
-= shiftlen
;
2599 skb
->truesize
-= shiftlen
;
2600 tgt
->len
+= shiftlen
;
2601 tgt
->data_len
+= shiftlen
;
2602 tgt
->truesize
+= shiftlen
;
2608 * skb_prepare_seq_read - Prepare a sequential read of skb data
2609 * @skb: the buffer to read
2610 * @from: lower offset of data to be read
2611 * @to: upper offset of data to be read
2612 * @st: state variable
2614 * Initializes the specified state variable. Must be called before
2615 * invoking skb_seq_read() for the first time.
2617 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2618 unsigned int to
, struct skb_seq_state
*st
)
2620 st
->lower_offset
= from
;
2621 st
->upper_offset
= to
;
2622 st
->root_skb
= st
->cur_skb
= skb
;
2623 st
->frag_idx
= st
->stepped_offset
= 0;
2624 st
->frag_data
= NULL
;
2626 EXPORT_SYMBOL(skb_prepare_seq_read
);
2629 * skb_seq_read - Sequentially read skb data
2630 * @consumed: number of bytes consumed by the caller so far
2631 * @data: destination pointer for data to be returned
2632 * @st: state variable
2634 * Reads a block of skb data at @consumed relative to the
2635 * lower offset specified to skb_prepare_seq_read(). Assigns
2636 * the head of the data block to @data and returns the length
2637 * of the block or 0 if the end of the skb data or the upper
2638 * offset has been reached.
2640 * The caller is not required to consume all of the data
2641 * returned, i.e. @consumed is typically set to the number
2642 * of bytes already consumed and the next call to
2643 * skb_seq_read() will return the remaining part of the block.
2645 * Note 1: The size of each block of data returned can be arbitrary,
2646 * this limitation is the cost for zerocopy seqeuental
2647 * reads of potentially non linear data.
2649 * Note 2: Fragment lists within fragments are not implemented
2650 * at the moment, state->root_skb could be replaced with
2651 * a stack for this purpose.
2653 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2654 struct skb_seq_state
*st
)
2656 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2659 if (unlikely(abs_offset
>= st
->upper_offset
)) {
2660 if (st
->frag_data
) {
2661 kunmap_atomic(st
->frag_data
);
2662 st
->frag_data
= NULL
;
2668 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2670 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2671 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2672 return block_limit
- abs_offset
;
2675 if (st
->frag_idx
== 0 && !st
->frag_data
)
2676 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2678 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2679 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2680 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2682 if (abs_offset
< block_limit
) {
2684 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2686 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2687 (abs_offset
- st
->stepped_offset
);
2689 return block_limit
- abs_offset
;
2692 if (st
->frag_data
) {
2693 kunmap_atomic(st
->frag_data
);
2694 st
->frag_data
= NULL
;
2698 st
->stepped_offset
+= skb_frag_size(frag
);
2701 if (st
->frag_data
) {
2702 kunmap_atomic(st
->frag_data
);
2703 st
->frag_data
= NULL
;
2706 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2707 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2710 } else if (st
->cur_skb
->next
) {
2711 st
->cur_skb
= st
->cur_skb
->next
;
2718 EXPORT_SYMBOL(skb_seq_read
);
2721 * skb_abort_seq_read - Abort a sequential read of skb data
2722 * @st: state variable
2724 * Must be called if skb_seq_read() was not called until it
2727 void skb_abort_seq_read(struct skb_seq_state
*st
)
2730 kunmap_atomic(st
->frag_data
);
2732 EXPORT_SYMBOL(skb_abort_seq_read
);
2734 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2736 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2737 struct ts_config
*conf
,
2738 struct ts_state
*state
)
2740 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2743 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2745 skb_abort_seq_read(TS_SKB_CB(state
));
2749 * skb_find_text - Find a text pattern in skb data
2750 * @skb: the buffer to look in
2751 * @from: search offset
2753 * @config: textsearch configuration
2754 * @state: uninitialized textsearch state variable
2756 * Finds a pattern in the skb data according to the specified
2757 * textsearch configuration. Use textsearch_next() to retrieve
2758 * subsequent occurrences of the pattern. Returns the offset
2759 * to the first occurrence or UINT_MAX if no match was found.
2761 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2762 unsigned int to
, struct ts_config
*config
,
2763 struct ts_state
*state
)
2767 config
->get_next_block
= skb_ts_get_next_block
;
2768 config
->finish
= skb_ts_finish
;
2770 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2772 ret
= textsearch_find(config
, state
);
2773 return (ret
<= to
- from
? ret
: UINT_MAX
);
2775 EXPORT_SYMBOL(skb_find_text
);
2778 * skb_append_datato_frags - append the user data to a skb
2779 * @sk: sock structure
2780 * @skb: skb structure to be appened with user data.
2781 * @getfrag: call back function to be used for getting the user data
2782 * @from: pointer to user message iov
2783 * @length: length of the iov message
2785 * Description: This procedure append the user data in the fragment part
2786 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2788 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2789 int (*getfrag
)(void *from
, char *to
, int offset
,
2790 int len
, int odd
, struct sk_buff
*skb
),
2791 void *from
, int length
)
2793 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2797 struct page_frag
*pfrag
= ¤t
->task_frag
;
2800 /* Return error if we don't have space for new frag */
2801 if (frg_cnt
>= MAX_SKB_FRAGS
)
2804 if (!sk_page_frag_refill(sk
, pfrag
))
2807 /* copy the user data to page */
2808 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2810 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2811 offset
, copy
, 0, skb
);
2815 /* copy was successful so update the size parameters */
2816 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
2819 pfrag
->offset
+= copy
;
2820 get_page(pfrag
->page
);
2822 skb
->truesize
+= copy
;
2823 atomic_add(copy
, &sk
->sk_wmem_alloc
);
2825 skb
->data_len
+= copy
;
2829 } while (length
> 0);
2833 EXPORT_SYMBOL(skb_append_datato_frags
);
2836 * skb_pull_rcsum - pull skb and update receive checksum
2837 * @skb: buffer to update
2838 * @len: length of data pulled
2840 * This function performs an skb_pull on the packet and updates
2841 * the CHECKSUM_COMPLETE checksum. It should be used on
2842 * receive path processing instead of skb_pull unless you know
2843 * that the checksum difference is zero (e.g., a valid IP header)
2844 * or you are setting ip_summed to CHECKSUM_NONE.
2846 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2848 BUG_ON(len
> skb
->len
);
2850 BUG_ON(skb
->len
< skb
->data_len
);
2851 skb_postpull_rcsum(skb
, skb
->data
, len
);
2852 return skb
->data
+= len
;
2854 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2857 * skb_segment - Perform protocol segmentation on skb.
2858 * @head_skb: buffer to segment
2859 * @features: features for the output path (see dev->features)
2861 * This function performs segmentation on the given skb. It returns
2862 * a pointer to the first in a list of new skbs for the segments.
2863 * In case of error it returns ERR_PTR(err).
2865 struct sk_buff
*skb_segment(struct sk_buff
*head_skb
,
2866 netdev_features_t features
)
2868 struct sk_buff
*segs
= NULL
;
2869 struct sk_buff
*tail
= NULL
;
2870 struct sk_buff
*list_skb
= skb_shinfo(head_skb
)->frag_list
;
2871 skb_frag_t
*frag
= skb_shinfo(head_skb
)->frags
;
2872 unsigned int mss
= skb_shinfo(head_skb
)->gso_size
;
2873 unsigned int doffset
= head_skb
->data
- skb_mac_header(head_skb
);
2874 struct sk_buff
*frag_skb
= head_skb
;
2875 unsigned int offset
= doffset
;
2876 unsigned int tnl_hlen
= skb_tnl_header_len(head_skb
);
2877 unsigned int headroom
;
2881 int sg
= !!(features
& NETIF_F_SG
);
2882 int nfrags
= skb_shinfo(head_skb
)->nr_frags
;
2888 proto
= skb_network_protocol(head_skb
, &dummy
);
2889 if (unlikely(!proto
))
2890 return ERR_PTR(-EINVAL
);
2892 csum
= !head_skb
->encap_hdr_csum
&&
2893 !!can_checksum_protocol(features
, proto
);
2895 __skb_push(head_skb
, doffset
);
2896 headroom
= skb_headroom(head_skb
);
2897 pos
= skb_headlen(head_skb
);
2900 struct sk_buff
*nskb
;
2901 skb_frag_t
*nskb_frag
;
2905 len
= head_skb
->len
- offset
;
2909 hsize
= skb_headlen(head_skb
) - offset
;
2912 if (hsize
> len
|| !sg
)
2915 if (!hsize
&& i
>= nfrags
&& skb_headlen(list_skb
) &&
2916 (skb_headlen(list_skb
) == len
|| sg
)) {
2917 BUG_ON(skb_headlen(list_skb
) > len
);
2920 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
2921 frag
= skb_shinfo(list_skb
)->frags
;
2922 frag_skb
= list_skb
;
2923 pos
+= skb_headlen(list_skb
);
2925 while (pos
< offset
+ len
) {
2926 BUG_ON(i
>= nfrags
);
2928 size
= skb_frag_size(frag
);
2929 if (pos
+ size
> offset
+ len
)
2937 nskb
= skb_clone(list_skb
, GFP_ATOMIC
);
2938 list_skb
= list_skb
->next
;
2940 if (unlikely(!nskb
))
2943 if (unlikely(pskb_trim(nskb
, len
))) {
2948 hsize
= skb_end_offset(nskb
);
2949 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2954 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2955 skb_release_head_state(nskb
);
2956 __skb_push(nskb
, doffset
);
2958 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2959 GFP_ATOMIC
, skb_alloc_rx_flag(head_skb
),
2962 if (unlikely(!nskb
))
2965 skb_reserve(nskb
, headroom
);
2966 __skb_put(nskb
, doffset
);
2975 __copy_skb_header(nskb
, head_skb
);
2976 nskb
->mac_len
= head_skb
->mac_len
;
2978 skb_headers_offset_update(nskb
, skb_headroom(nskb
) - headroom
);
2980 skb_copy_from_linear_data_offset(head_skb
, -tnl_hlen
,
2981 nskb
->data
- tnl_hlen
,
2982 doffset
+ tnl_hlen
);
2984 if (nskb
->len
== len
+ doffset
)
2985 goto perform_csum_check
;
2988 nskb
->ip_summed
= CHECKSUM_NONE
;
2989 nskb
->csum
= skb_copy_and_csum_bits(head_skb
, offset
,
2992 SKB_GSO_CB(nskb
)->csum_start
=
2993 skb_headroom(nskb
) + offset
;
2997 nskb_frag
= skb_shinfo(nskb
)->frags
;
2999 skb_copy_from_linear_data_offset(head_skb
, offset
,
3000 skb_put(nskb
, hsize
), hsize
);
3002 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(head_skb
)->tx_flags
&
3005 while (pos
< offset
+ len
) {
3007 BUG_ON(skb_headlen(list_skb
));
3010 nfrags
= skb_shinfo(list_skb
)->nr_frags
;
3011 frag
= skb_shinfo(list_skb
)->frags
;
3012 frag_skb
= list_skb
;
3016 list_skb
= list_skb
->next
;
3019 if (unlikely(skb_shinfo(nskb
)->nr_frags
>=
3021 net_warn_ratelimited(
3022 "skb_segment: too many frags: %u %u\n",
3027 if (unlikely(skb_orphan_frags(frag_skb
, GFP_ATOMIC
)))
3031 __skb_frag_ref(nskb_frag
);
3032 size
= skb_frag_size(nskb_frag
);
3035 nskb_frag
->page_offset
+= offset
- pos
;
3036 skb_frag_size_sub(nskb_frag
, offset
- pos
);
3039 skb_shinfo(nskb
)->nr_frags
++;
3041 if (pos
+ size
<= offset
+ len
) {
3046 skb_frag_size_sub(nskb_frag
, pos
+ size
- (offset
+ len
));
3054 nskb
->data_len
= len
- hsize
;
3055 nskb
->len
+= nskb
->data_len
;
3056 nskb
->truesize
+= nskb
->data_len
;
3060 nskb
->csum
= skb_checksum(nskb
, doffset
,
3061 nskb
->len
- doffset
, 0);
3062 nskb
->ip_summed
= CHECKSUM_NONE
;
3063 SKB_GSO_CB(nskb
)->csum_start
=
3064 skb_headroom(nskb
) + doffset
;
3066 } while ((offset
+= len
) < head_skb
->len
);
3071 kfree_skb_list(segs
);
3072 return ERR_PTR(err
);
3074 EXPORT_SYMBOL_GPL(skb_segment
);
3076 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
3078 struct skb_shared_info
*pinfo
, *skbinfo
= skb_shinfo(skb
);
3079 unsigned int offset
= skb_gro_offset(skb
);
3080 unsigned int headlen
= skb_headlen(skb
);
3081 struct sk_buff
*nskb
, *lp
, *p
= *head
;
3082 unsigned int len
= skb_gro_len(skb
);
3083 unsigned int delta_truesize
;
3084 unsigned int headroom
;
3086 if (unlikely(p
->len
+ len
>= 65536))
3089 lp
= NAPI_GRO_CB(p
)->last
;
3090 pinfo
= skb_shinfo(lp
);
3092 if (headlen
<= offset
) {
3095 int i
= skbinfo
->nr_frags
;
3096 int nr_frags
= pinfo
->nr_frags
+ i
;
3098 if (nr_frags
> MAX_SKB_FRAGS
)
3102 pinfo
->nr_frags
= nr_frags
;
3103 skbinfo
->nr_frags
= 0;
3105 frag
= pinfo
->frags
+ nr_frags
;
3106 frag2
= skbinfo
->frags
+ i
;
3111 frag
->page_offset
+= offset
;
3112 skb_frag_size_sub(frag
, offset
);
3114 /* all fragments truesize : remove (head size + sk_buff) */
3115 delta_truesize
= skb
->truesize
-
3116 SKB_TRUESIZE(skb_end_offset(skb
));
3118 skb
->truesize
-= skb
->data_len
;
3119 skb
->len
-= skb
->data_len
;
3122 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3124 } else if (skb
->head_frag
) {
3125 int nr_frags
= pinfo
->nr_frags
;
3126 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3127 struct page
*page
= virt_to_head_page(skb
->head
);
3128 unsigned int first_size
= headlen
- offset
;
3129 unsigned int first_offset
;
3131 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3134 first_offset
= skb
->data
-
3135 (unsigned char *)page_address(page
) +
3138 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3140 frag
->page
.p
= page
;
3141 frag
->page_offset
= first_offset
;
3142 skb_frag_size_set(frag
, first_size
);
3144 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3145 /* We dont need to clear skbinfo->nr_frags here */
3147 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3148 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3151 if (pinfo
->frag_list
)
3153 if (skb_gro_len(p
) != pinfo
->gso_size
)
3156 headroom
= skb_headroom(p
);
3157 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
3158 if (unlikely(!nskb
))
3161 __copy_skb_header(nskb
, p
);
3162 nskb
->mac_len
= p
->mac_len
;
3164 skb_reserve(nskb
, headroom
);
3165 __skb_put(nskb
, skb_gro_offset(p
));
3167 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
3168 skb_set_network_header(nskb
, skb_network_offset(p
));
3169 skb_set_transport_header(nskb
, skb_transport_offset(p
));
3171 __skb_pull(p
, skb_gro_offset(p
));
3172 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
3173 p
->data
- skb_mac_header(p
));
3175 skb_shinfo(nskb
)->frag_list
= p
;
3176 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3177 pinfo
->gso_size
= 0;
3178 skb_header_release(p
);
3179 NAPI_GRO_CB(nskb
)->last
= p
;
3181 nskb
->data_len
+= p
->len
;
3182 nskb
->truesize
+= p
->truesize
;
3183 nskb
->len
+= p
->len
;
3186 nskb
->next
= p
->next
;
3192 delta_truesize
= skb
->truesize
;
3193 if (offset
> headlen
) {
3194 unsigned int eat
= offset
- headlen
;
3196 skbinfo
->frags
[0].page_offset
+= eat
;
3197 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3198 skb
->data_len
-= eat
;
3203 __skb_pull(skb
, offset
);
3205 if (NAPI_GRO_CB(p
)->last
== p
)
3206 skb_shinfo(p
)->frag_list
= skb
;
3208 NAPI_GRO_CB(p
)->last
->next
= skb
;
3209 NAPI_GRO_CB(p
)->last
= skb
;
3210 skb_header_release(skb
);
3214 NAPI_GRO_CB(p
)->count
++;
3216 p
->truesize
+= delta_truesize
;
3219 lp
->data_len
+= len
;
3220 lp
->truesize
+= delta_truesize
;
3223 NAPI_GRO_CB(skb
)->same_flow
= 1;
3226 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3228 void __init
skb_init(void)
3230 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3231 sizeof(struct sk_buff
),
3233 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3235 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3236 (2*sizeof(struct sk_buff
)) +
3239 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3244 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3245 * @skb: Socket buffer containing the buffers to be mapped
3246 * @sg: The scatter-gather list to map into
3247 * @offset: The offset into the buffer's contents to start mapping
3248 * @len: Length of buffer space to be mapped
3250 * Fill the specified scatter-gather list with mappings/pointers into a
3251 * region of the buffer space attached to a socket buffer.
3254 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3256 int start
= skb_headlen(skb
);
3257 int i
, copy
= start
- offset
;
3258 struct sk_buff
*frag_iter
;
3264 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3266 if ((len
-= copy
) == 0)
3271 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3274 WARN_ON(start
> offset
+ len
);
3276 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3277 if ((copy
= end
- offset
) > 0) {
3278 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3282 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3283 frag
->page_offset
+offset
-start
);
3292 skb_walk_frags(skb
, frag_iter
) {
3295 WARN_ON(start
> offset
+ len
);
3297 end
= start
+ frag_iter
->len
;
3298 if ((copy
= end
- offset
) > 0) {
3301 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3303 if ((len
-= copy
) == 0)
3313 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3314 * sglist without mark the sg which contain last skb data as the end.
3315 * So the caller can mannipulate sg list as will when padding new data after
3316 * the first call without calling sg_unmark_end to expend sg list.
3318 * Scenario to use skb_to_sgvec_nomark:
3320 * 2. skb_to_sgvec_nomark(payload1)
3321 * 3. skb_to_sgvec_nomark(payload2)
3323 * This is equivalent to:
3325 * 2. skb_to_sgvec(payload1)
3327 * 4. skb_to_sgvec(payload2)
3329 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3330 * is more preferable.
3332 int skb_to_sgvec_nomark(struct sk_buff
*skb
, struct scatterlist
*sg
,
3333 int offset
, int len
)
3335 return __skb_to_sgvec(skb
, sg
, offset
, len
);
3337 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark
);
3339 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3341 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3343 sg_mark_end(&sg
[nsg
- 1]);
3347 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3350 * skb_cow_data - Check that a socket buffer's data buffers are writable
3351 * @skb: The socket buffer to check.
3352 * @tailbits: Amount of trailing space to be added
3353 * @trailer: Returned pointer to the skb where the @tailbits space begins
3355 * Make sure that the data buffers attached to a socket buffer are
3356 * writable. If they are not, private copies are made of the data buffers
3357 * and the socket buffer is set to use these instead.
3359 * If @tailbits is given, make sure that there is space to write @tailbits
3360 * bytes of data beyond current end of socket buffer. @trailer will be
3361 * set to point to the skb in which this space begins.
3363 * The number of scatterlist elements required to completely map the
3364 * COW'd and extended socket buffer will be returned.
3366 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3370 struct sk_buff
*skb1
, **skb_p
;
3372 /* If skb is cloned or its head is paged, reallocate
3373 * head pulling out all the pages (pages are considered not writable
3374 * at the moment even if they are anonymous).
3376 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3377 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3380 /* Easy case. Most of packets will go this way. */
3381 if (!skb_has_frag_list(skb
)) {
3382 /* A little of trouble, not enough of space for trailer.
3383 * This should not happen, when stack is tuned to generate
3384 * good frames. OK, on miss we reallocate and reserve even more
3385 * space, 128 bytes is fair. */
3387 if (skb_tailroom(skb
) < tailbits
&&
3388 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3396 /* Misery. We are in troubles, going to mincer fragments... */
3399 skb_p
= &skb_shinfo(skb
)->frag_list
;
3402 while ((skb1
= *skb_p
) != NULL
) {
3405 /* The fragment is partially pulled by someone,
3406 * this can happen on input. Copy it and everything
3409 if (skb_shared(skb1
))
3412 /* If the skb is the last, worry about trailer. */
3414 if (skb1
->next
== NULL
&& tailbits
) {
3415 if (skb_shinfo(skb1
)->nr_frags
||
3416 skb_has_frag_list(skb1
) ||
3417 skb_tailroom(skb1
) < tailbits
)
3418 ntail
= tailbits
+ 128;
3424 skb_shinfo(skb1
)->nr_frags
||
3425 skb_has_frag_list(skb1
)) {
3426 struct sk_buff
*skb2
;
3428 /* Fuck, we are miserable poor guys... */
3430 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3432 skb2
= skb_copy_expand(skb1
,
3436 if (unlikely(skb2
== NULL
))
3440 skb_set_owner_w(skb2
, skb1
->sk
);
3442 /* Looking around. Are we still alive?
3443 * OK, link new skb, drop old one */
3445 skb2
->next
= skb1
->next
;
3452 skb_p
= &skb1
->next
;
3457 EXPORT_SYMBOL_GPL(skb_cow_data
);
3459 static void sock_rmem_free(struct sk_buff
*skb
)
3461 struct sock
*sk
= skb
->sk
;
3463 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3467 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3469 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3471 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3472 (unsigned int)sk
->sk_rcvbuf
)
3477 skb
->destructor
= sock_rmem_free
;
3478 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3480 /* before exiting rcu section, make sure dst is refcounted */
3483 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3484 if (!sock_flag(sk
, SOCK_DEAD
))
3485 sk
->sk_data_ready(sk
);
3488 EXPORT_SYMBOL(sock_queue_err_skb
);
3490 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3491 struct skb_shared_hwtstamps
*hwtstamps
)
3493 struct sock
*sk
= orig_skb
->sk
;
3494 struct sock_exterr_skb
*serr
;
3495 struct sk_buff
*skb
;
3502 *skb_hwtstamps(orig_skb
) =
3506 * no hardware time stamps available,
3507 * so keep the shared tx_flags and only
3508 * store software time stamp
3510 orig_skb
->tstamp
= ktime_get_real();
3513 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3517 serr
= SKB_EXT_ERR(skb
);
3518 memset(serr
, 0, sizeof(*serr
));
3519 serr
->ee
.ee_errno
= ENOMSG
;
3520 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3522 err
= sock_queue_err_skb(sk
, skb
);
3527 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3529 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3531 struct sock
*sk
= skb
->sk
;
3532 struct sock_exterr_skb
*serr
;
3535 skb
->wifi_acked_valid
= 1;
3536 skb
->wifi_acked
= acked
;
3538 serr
= SKB_EXT_ERR(skb
);
3539 memset(serr
, 0, sizeof(*serr
));
3540 serr
->ee
.ee_errno
= ENOMSG
;
3541 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3543 err
= sock_queue_err_skb(sk
, skb
);
3547 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3551 * skb_partial_csum_set - set up and verify partial csum values for packet
3552 * @skb: the skb to set
3553 * @start: the number of bytes after skb->data to start checksumming.
3554 * @off: the offset from start to place the checksum.
3556 * For untrusted partially-checksummed packets, we need to make sure the values
3557 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3559 * This function checks and sets those values and skb->ip_summed: if this
3560 * returns false you should drop the packet.
3562 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3564 if (unlikely(start
> skb_headlen(skb
)) ||
3565 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3566 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3567 start
, off
, skb_headlen(skb
));
3570 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3571 skb
->csum_start
= skb_headroom(skb
) + start
;
3572 skb
->csum_offset
= off
;
3573 skb_set_transport_header(skb
, start
);
3576 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3578 static int skb_maybe_pull_tail(struct sk_buff
*skb
, unsigned int len
,
3581 if (skb_headlen(skb
) >= len
)
3584 /* If we need to pullup then pullup to the max, so we
3585 * won't need to do it again.
3590 if (__pskb_pull_tail(skb
, max
- skb_headlen(skb
)) == NULL
)
3593 if (skb_headlen(skb
) < len
)
3599 #define MAX_TCP_HDR_LEN (15 * 4)
3601 static __sum16
*skb_checksum_setup_ip(struct sk_buff
*skb
,
3602 typeof(IPPROTO_IP
) proto
,
3609 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct tcphdr
),
3610 off
+ MAX_TCP_HDR_LEN
);
3611 if (!err
&& !skb_partial_csum_set(skb
, off
,
3612 offsetof(struct tcphdr
,
3615 return err
? ERR_PTR(err
) : &tcp_hdr(skb
)->check
;
3618 err
= skb_maybe_pull_tail(skb
, off
+ sizeof(struct udphdr
),
3619 off
+ sizeof(struct udphdr
));
3620 if (!err
&& !skb_partial_csum_set(skb
, off
,
3621 offsetof(struct udphdr
,
3624 return err
? ERR_PTR(err
) : &udp_hdr(skb
)->check
;
3627 return ERR_PTR(-EPROTO
);
3630 /* This value should be large enough to cover a tagged ethernet header plus
3631 * maximally sized IP and TCP or UDP headers.
3633 #define MAX_IP_HDR_LEN 128
3635 static int skb_checksum_setup_ipv4(struct sk_buff
*skb
, bool recalculate
)
3644 err
= skb_maybe_pull_tail(skb
,
3645 sizeof(struct iphdr
),
3650 if (ip_hdr(skb
)->frag_off
& htons(IP_OFFSET
| IP_MF
))
3653 off
= ip_hdrlen(skb
);
3660 csum
= skb_checksum_setup_ip(skb
, ip_hdr(skb
)->protocol
, off
);
3662 return PTR_ERR(csum
);
3665 *csum
= ~csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
3668 ip_hdr(skb
)->protocol
, 0);
3675 /* This value should be large enough to cover a tagged ethernet header plus
3676 * an IPv6 header, all options, and a maximal TCP or UDP header.
3678 #define MAX_IPV6_HDR_LEN 256
3680 #define OPT_HDR(type, skb, off) \
3681 (type *)(skb_network_header(skb) + (off))
3683 static int skb_checksum_setup_ipv6(struct sk_buff
*skb
, bool recalculate
)
3696 off
= sizeof(struct ipv6hdr
);
3698 err
= skb_maybe_pull_tail(skb
, off
, MAX_IPV6_HDR_LEN
);
3702 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
3704 len
= sizeof(struct ipv6hdr
) + ntohs(ipv6_hdr(skb
)->payload_len
);
3705 while (off
<= len
&& !done
) {
3707 case IPPROTO_DSTOPTS
:
3708 case IPPROTO_HOPOPTS
:
3709 case IPPROTO_ROUTING
: {
3710 struct ipv6_opt_hdr
*hp
;
3712 err
= skb_maybe_pull_tail(skb
,
3714 sizeof(struct ipv6_opt_hdr
),
3719 hp
= OPT_HDR(struct ipv6_opt_hdr
, skb
, off
);
3720 nexthdr
= hp
->nexthdr
;
3721 off
+= ipv6_optlen(hp
);
3725 struct ip_auth_hdr
*hp
;
3727 err
= skb_maybe_pull_tail(skb
,
3729 sizeof(struct ip_auth_hdr
),
3734 hp
= OPT_HDR(struct ip_auth_hdr
, skb
, off
);
3735 nexthdr
= hp
->nexthdr
;
3736 off
+= ipv6_authlen(hp
);
3739 case IPPROTO_FRAGMENT
: {
3740 struct frag_hdr
*hp
;
3742 err
= skb_maybe_pull_tail(skb
,
3744 sizeof(struct frag_hdr
),
3749 hp
= OPT_HDR(struct frag_hdr
, skb
, off
);
3751 if (hp
->frag_off
& htons(IP6_OFFSET
| IP6_MF
))
3754 nexthdr
= hp
->nexthdr
;
3755 off
+= sizeof(struct frag_hdr
);
3766 if (!done
|| fragment
)
3769 csum
= skb_checksum_setup_ip(skb
, nexthdr
, off
);
3771 return PTR_ERR(csum
);
3774 *csum
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
3775 &ipv6_hdr(skb
)->daddr
,
3776 skb
->len
- off
, nexthdr
, 0);
3784 * skb_checksum_setup - set up partial checksum offset
3785 * @skb: the skb to set up
3786 * @recalculate: if true the pseudo-header checksum will be recalculated
3788 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
)
3792 switch (skb
->protocol
) {
3793 case htons(ETH_P_IP
):
3794 err
= skb_checksum_setup_ipv4(skb
, recalculate
);
3797 case htons(ETH_P_IPV6
):
3798 err
= skb_checksum_setup_ipv6(skb
, recalculate
);
3808 EXPORT_SYMBOL(skb_checksum_setup
);
3810 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3812 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3815 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3817 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3820 skb_release_head_state(skb
);
3821 kmem_cache_free(skbuff_head_cache
, skb
);
3826 EXPORT_SYMBOL(kfree_skb_partial
);
3829 * skb_try_coalesce - try to merge skb to prior one
3831 * @from: buffer to add
3832 * @fragstolen: pointer to boolean
3833 * @delta_truesize: how much more was allocated than was requested
3835 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3836 bool *fragstolen
, int *delta_truesize
)
3838 int i
, delta
, len
= from
->len
;
3840 *fragstolen
= false;
3845 if (len
<= skb_tailroom(to
)) {
3846 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3847 *delta_truesize
= 0;
3851 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3854 if (skb_headlen(from
) != 0) {
3856 unsigned int offset
;
3858 if (skb_shinfo(to
)->nr_frags
+
3859 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3862 if (skb_head_is_locked(from
))
3865 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3867 page
= virt_to_head_page(from
->head
);
3868 offset
= from
->data
- (unsigned char *)page_address(page
);
3870 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3871 page
, offset
, skb_headlen(from
));
3874 if (skb_shinfo(to
)->nr_frags
+
3875 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3878 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
3881 WARN_ON_ONCE(delta
< len
);
3883 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3884 skb_shinfo(from
)->frags
,
3885 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3886 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3888 if (!skb_cloned(from
))
3889 skb_shinfo(from
)->nr_frags
= 0;
3891 /* if the skb is not cloned this does nothing
3892 * since we set nr_frags to 0.
3894 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3895 skb_frag_ref(from
, i
);
3897 to
->truesize
+= delta
;
3899 to
->data_len
+= len
;
3901 *delta_truesize
= delta
;
3904 EXPORT_SYMBOL(skb_try_coalesce
);
3907 * skb_scrub_packet - scrub an skb
3909 * @skb: buffer to clean
3910 * @xnet: packet is crossing netns
3912 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3913 * into/from a tunnel. Some information have to be cleared during these
3915 * skb_scrub_packet can also be used to clean a skb before injecting it in
3916 * another namespace (@xnet == true). We have to clear all information in the
3917 * skb that could impact namespace isolation.
3919 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
)
3923 skb
->tstamp
.tv64
= 0;
3924 skb
->pkt_type
= PACKET_HOST
;
3931 nf_reset_trace(skb
);
3933 EXPORT_SYMBOL_GPL(skb_scrub_packet
);
3936 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3940 * skb_gso_transport_seglen is used to determine the real size of the
3941 * individual segments, including Layer4 headers (TCP/UDP).
3943 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3945 unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
)
3947 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
3949 if (likely(shinfo
->gso_type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
)))
3950 return tcp_hdrlen(skb
) + shinfo
->gso_size
;
3952 /* UFO sets gso_size to the size of the fragmentation
3953 * payload, i.e. the size of the L4 (UDP) header is already
3956 return shinfo
->gso_size
;
3958 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen
);