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/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 struct kmem_cache
*skbuff_head_cache __read_mostly
;
75 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
77 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
78 struct pipe_buffer
*buf
)
83 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
84 struct pipe_buffer
*buf
)
89 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
90 struct pipe_buffer
*buf
)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops
= {
99 .map
= generic_pipe_buf_map
,
100 .unmap
= generic_pipe_buf_unmap
,
101 .confirm
= generic_pipe_buf_confirm
,
102 .release
= sock_pipe_buf_release
,
103 .steal
= sock_pipe_buf_steal
,
104 .get
= sock_pipe_buf_get
,
108 * skb_panic - private function for out-of-line support
112 * @msg: skb_over_panic or skb_under_panic
114 * Out-of-line support for skb_put() and skb_push().
115 * Called via the wrapper skb_over_panic() or skb_under_panic().
116 * Keep out of line to prevent kernel bloat.
117 * __builtin_return_address is not used because it is not always reliable.
119 static void skb_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
,
122 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
123 msg
, addr
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
129 static void skb_over_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
131 skb_panic(skb
, sz
, addr
, __func__
);
134 static void skb_under_panic(struct sk_buff
*skb
, unsigned int sz
, void *addr
)
136 skb_panic(skb
, sz
, addr
, __func__
);
140 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
141 * the caller if emergency pfmemalloc reserves are being used. If it is and
142 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
143 * may be used. Otherwise, the packet data may be discarded until enough
146 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
147 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
149 static void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
,
150 unsigned long ip
, bool *pfmemalloc
)
153 bool ret_pfmemalloc
= false;
156 * Try a regular allocation, when that fails and we're not entitled
157 * to the reserves, fail.
159 obj
= kmalloc_node_track_caller(size
,
160 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
162 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
165 /* Try again but now we are using pfmemalloc reserves */
166 ret_pfmemalloc
= true;
167 obj
= kmalloc_node_track_caller(size
, flags
, node
);
171 *pfmemalloc
= ret_pfmemalloc
;
176 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
177 * 'private' fields and also do memory statistics to find all the
183 * __alloc_skb - allocate a network buffer
184 * @size: size to allocate
185 * @gfp_mask: allocation mask
186 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
187 * instead of head cache and allocate a cloned (child) skb.
188 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
189 * allocations in case the data is required for writeback
190 * @node: numa node to allocate memory on
192 * Allocate a new &sk_buff. The returned buffer has no headroom and a
193 * tail room of at least size bytes. The object has a reference count
194 * of one. The return is the buffer. On a failure the return is %NULL.
196 * Buffers may only be allocated from interrupts using a @gfp_mask of
199 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
202 struct kmem_cache
*cache
;
203 struct skb_shared_info
*shinfo
;
208 cache
= (flags
& SKB_ALLOC_FCLONE
)
209 ? skbuff_fclone_cache
: skbuff_head_cache
;
211 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
212 gfp_mask
|= __GFP_MEMALLOC
;
215 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
220 /* We do our best to align skb_shared_info on a separate cache
221 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
222 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
223 * Both skb->head and skb_shared_info are cache line aligned.
225 size
= SKB_DATA_ALIGN(size
);
226 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
227 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
230 /* kmalloc(size) might give us more room than requested.
231 * Put skb_shared_info exactly at the end of allocated zone,
232 * to allow max possible filling before reallocation.
234 size
= SKB_WITH_OVERHEAD(ksize(data
));
235 prefetchw(data
+ size
);
238 * Only clear those fields we need to clear, not those that we will
239 * actually initialise below. Hence, don't put any more fields after
240 * the tail pointer in struct sk_buff!
242 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
243 /* Account for allocated memory : skb + skb->head */
244 skb
->truesize
= SKB_TRUESIZE(size
);
245 skb
->pfmemalloc
= pfmemalloc
;
246 atomic_set(&skb
->users
, 1);
249 skb_reset_tail_pointer(skb
);
250 skb
->end
= skb
->tail
+ size
;
251 #ifdef NET_SKBUFF_DATA_USES_OFFSET
252 skb
->mac_header
= ~0U;
253 skb
->transport_header
= ~0U;
256 /* make sure we initialize shinfo sequentially */
257 shinfo
= skb_shinfo(skb
);
258 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
259 atomic_set(&shinfo
->dataref
, 1);
260 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
262 if (flags
& SKB_ALLOC_FCLONE
) {
263 struct sk_buff
*child
= skb
+ 1;
264 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
266 kmemcheck_annotate_bitfield(child
, flags1
);
267 kmemcheck_annotate_bitfield(child
, flags2
);
268 skb
->fclone
= SKB_FCLONE_ORIG
;
269 atomic_set(fclone_ref
, 1);
271 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
272 child
->pfmemalloc
= pfmemalloc
;
277 kmem_cache_free(cache
, skb
);
281 EXPORT_SYMBOL(__alloc_skb
);
284 * build_skb - build a network buffer
285 * @data: data buffer provided by caller
286 * @frag_size: size of fragment, or 0 if head was kmalloced
288 * Allocate a new &sk_buff. Caller provides space holding head and
289 * skb_shared_info. @data must have been allocated by kmalloc()
290 * The return is the new skb buffer.
291 * On a failure the return is %NULL, and @data is not freed.
293 * Before IO, driver allocates only data buffer where NIC put incoming frame
294 * Driver should add room at head (NET_SKB_PAD) and
295 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
296 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
297 * before giving packet to stack.
298 * RX rings only contains data buffers, not full skbs.
300 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
302 struct skb_shared_info
*shinfo
;
304 unsigned int size
= frag_size
? : ksize(data
);
306 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
310 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
312 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
313 skb
->truesize
= SKB_TRUESIZE(size
);
314 skb
->head_frag
= frag_size
!= 0;
315 atomic_set(&skb
->users
, 1);
318 skb_reset_tail_pointer(skb
);
319 skb
->end
= skb
->tail
+ size
;
320 #ifdef NET_SKBUFF_DATA_USES_OFFSET
321 skb
->mac_header
= ~0U;
322 skb
->transport_header
= ~0U;
325 /* make sure we initialize shinfo sequentially */
326 shinfo
= skb_shinfo(skb
);
327 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
328 atomic_set(&shinfo
->dataref
, 1);
329 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
333 EXPORT_SYMBOL(build_skb
);
335 struct netdev_alloc_cache
{
336 struct page_frag frag
;
337 /* we maintain a pagecount bias, so that we dont dirty cache line
338 * containing page->_count every time we allocate a fragment.
340 unsigned int pagecnt_bias
;
342 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
344 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
346 struct netdev_alloc_cache
*nc
;
351 local_irq_save(flags
);
352 nc
= &__get_cpu_var(netdev_alloc_cache
);
353 if (unlikely(!nc
->frag
.page
)) {
355 for (order
= NETDEV_FRAG_PAGE_MAX_ORDER
; ;) {
356 gfp_t gfp
= gfp_mask
;
359 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
360 nc
->frag
.page
= alloc_pages(gfp
, order
);
361 if (likely(nc
->frag
.page
))
366 nc
->frag
.size
= PAGE_SIZE
<< order
;
368 atomic_set(&nc
->frag
.page
->_count
, NETDEV_PAGECNT_MAX_BIAS
);
369 nc
->pagecnt_bias
= NETDEV_PAGECNT_MAX_BIAS
;
373 if (nc
->frag
.offset
+ fragsz
> nc
->frag
.size
) {
374 /* avoid unnecessary locked operations if possible */
375 if ((atomic_read(&nc
->frag
.page
->_count
) == nc
->pagecnt_bias
) ||
376 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->frag
.page
->_count
))
381 data
= page_address(nc
->frag
.page
) + nc
->frag
.offset
;
382 nc
->frag
.offset
+= fragsz
;
385 local_irq_restore(flags
);
390 * netdev_alloc_frag - allocate a page fragment
391 * @fragsz: fragment size
393 * Allocates a frag from a page for receive buffer.
394 * Uses GFP_ATOMIC allocations.
396 void *netdev_alloc_frag(unsigned int fragsz
)
398 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
400 EXPORT_SYMBOL(netdev_alloc_frag
);
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @length: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has unspecified headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
413 * %NULL is returned if there is no free memory.
415 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
416 unsigned int length
, gfp_t gfp_mask
)
418 struct sk_buff
*skb
= NULL
;
419 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
420 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
422 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
425 if (sk_memalloc_socks())
426 gfp_mask
|= __GFP_MEMALLOC
;
428 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
431 skb
= build_skb(data
, fragsz
);
433 put_page(virt_to_head_page(data
));
436 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
437 SKB_ALLOC_RX
, NUMA_NO_NODE
);
440 skb_reserve(skb
, NET_SKB_PAD
);
445 EXPORT_SYMBOL(__netdev_alloc_skb
);
447 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
448 int size
, unsigned int truesize
)
450 skb_fill_page_desc(skb
, i
, page
, off
, size
);
452 skb
->data_len
+= size
;
453 skb
->truesize
+= truesize
;
455 EXPORT_SYMBOL(skb_add_rx_frag
);
457 static void skb_drop_list(struct sk_buff
**listp
)
459 struct sk_buff
*list
= *listp
;
464 struct sk_buff
*this = list
;
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 NET_SKBUFF_NF_DEFRAG_NEEDED
569 nf_conntrack_put_reasm(skb
->nfct_reasm
);
571 #ifdef CONFIG_BRIDGE_NETFILTER
572 nf_bridge_put(skb
->nf_bridge
);
574 /* XXX: IS this still necessary? - JHS */
575 #ifdef CONFIG_NET_SCHED
577 #ifdef CONFIG_NET_CLS_ACT
583 /* Free everything but the sk_buff shell. */
584 static void skb_release_all(struct sk_buff
*skb
)
586 skb_release_head_state(skb
);
587 skb_release_data(skb
);
591 * __kfree_skb - private function
594 * Free an sk_buff. Release anything attached to the buffer.
595 * Clean the state. This is an internal helper function. Users should
596 * always call kfree_skb
599 void __kfree_skb(struct sk_buff
*skb
)
601 skb_release_all(skb
);
604 EXPORT_SYMBOL(__kfree_skb
);
607 * kfree_skb - free an sk_buff
608 * @skb: buffer to free
610 * Drop a reference to the buffer and free it if the usage count has
613 void kfree_skb(struct sk_buff
*skb
)
617 if (likely(atomic_read(&skb
->users
) == 1))
619 else if (likely(!atomic_dec_and_test(&skb
->users
)))
621 trace_kfree_skb(skb
, __builtin_return_address(0));
624 EXPORT_SYMBOL(kfree_skb
);
627 * skb_tx_error - report an sk_buff xmit error
628 * @skb: buffer that triggered an error
630 * Report xmit error if a device callback is tracking this skb.
631 * skb must be freed afterwards.
633 void skb_tx_error(struct sk_buff
*skb
)
635 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
636 struct ubuf_info
*uarg
;
638 uarg
= skb_shinfo(skb
)->destructor_arg
;
640 uarg
->callback(uarg
, false);
641 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
644 EXPORT_SYMBOL(skb_tx_error
);
647 * consume_skb - free an skbuff
648 * @skb: buffer to free
650 * Drop a ref to the buffer and free it if the usage count has hit zero
651 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
652 * is being dropped after a failure and notes that
654 void consume_skb(struct sk_buff
*skb
)
658 if (likely(atomic_read(&skb
->users
) == 1))
660 else if (likely(!atomic_dec_and_test(&skb
->users
)))
662 trace_consume_skb(skb
);
665 EXPORT_SYMBOL(consume_skb
);
667 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
669 new->tstamp
= old
->tstamp
;
671 new->transport_header
= old
->transport_header
;
672 new->network_header
= old
->network_header
;
673 new->mac_header
= old
->mac_header
;
674 new->inner_transport_header
= old
->inner_transport_header
;
675 new->inner_network_header
= old
->inner_network_header
;
676 skb_dst_copy(new, old
);
677 new->rxhash
= old
->rxhash
;
678 new->ooo_okay
= old
->ooo_okay
;
679 new->l4_rxhash
= old
->l4_rxhash
;
680 new->no_fcs
= old
->no_fcs
;
681 new->encapsulation
= old
->encapsulation
;
683 new->sp
= secpath_get(old
->sp
);
685 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
686 new->csum
= old
->csum
;
687 new->local_df
= old
->local_df
;
688 new->pkt_type
= old
->pkt_type
;
689 new->ip_summed
= old
->ip_summed
;
690 skb_copy_queue_mapping(new, old
);
691 new->priority
= old
->priority
;
692 #if IS_ENABLED(CONFIG_IP_VS)
693 new->ipvs_property
= old
->ipvs_property
;
695 new->pfmemalloc
= old
->pfmemalloc
;
696 new->protocol
= old
->protocol
;
697 new->mark
= old
->mark
;
698 new->skb_iif
= old
->skb_iif
;
700 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
701 new->nf_trace
= old
->nf_trace
;
703 #ifdef CONFIG_NET_SCHED
704 new->tc_index
= old
->tc_index
;
705 #ifdef CONFIG_NET_CLS_ACT
706 new->tc_verd
= old
->tc_verd
;
709 new->vlan_tci
= old
->vlan_tci
;
711 skb_copy_secmark(new, old
);
715 * You should not add any new code to this function. Add it to
716 * __copy_skb_header above instead.
718 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
720 #define C(x) n->x = skb->x
722 n
->next
= n
->prev
= NULL
;
724 __copy_skb_header(n
, skb
);
729 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
732 n
->destructor
= NULL
;
739 atomic_set(&n
->users
, 1);
741 atomic_inc(&(skb_shinfo(skb
)->dataref
));
749 * skb_morph - morph one skb into another
750 * @dst: the skb to receive the contents
751 * @src: the skb to supply the contents
753 * This is identical to skb_clone except that the target skb is
754 * supplied by the user.
756 * The target skb is returned upon exit.
758 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
760 skb_release_all(dst
);
761 return __skb_clone(dst
, src
);
763 EXPORT_SYMBOL_GPL(skb_morph
);
766 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
767 * @skb: the skb to modify
768 * @gfp_mask: allocation priority
770 * This must be called on SKBTX_DEV_ZEROCOPY skb.
771 * It will copy all frags into kernel and drop the reference
772 * to userspace pages.
774 * If this function is called from an interrupt gfp_mask() must be
777 * Returns 0 on success or a negative error code on failure
778 * to allocate kernel memory to copy to.
780 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
783 int num_frags
= skb_shinfo(skb
)->nr_frags
;
784 struct page
*page
, *head
= NULL
;
785 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
787 for (i
= 0; i
< num_frags
; i
++) {
789 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
791 page
= alloc_page(gfp_mask
);
794 struct page
*next
= (struct page
*)head
->private;
800 vaddr
= kmap_atomic(skb_frag_page(f
));
801 memcpy(page_address(page
),
802 vaddr
+ f
->page_offset
, skb_frag_size(f
));
803 kunmap_atomic(vaddr
);
804 page
->private = (unsigned long)head
;
808 /* skb frags release userspace buffers */
809 for (i
= 0; i
< num_frags
; i
++)
810 skb_frag_unref(skb
, i
);
812 uarg
->callback(uarg
, false);
814 /* skb frags point to kernel buffers */
815 for (i
= num_frags
- 1; i
>= 0; i
--) {
816 __skb_fill_page_desc(skb
, i
, head
, 0,
817 skb_shinfo(skb
)->frags
[i
].size
);
818 head
= (struct page
*)head
->private;
821 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
824 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
827 * skb_clone - duplicate an sk_buff
828 * @skb: buffer to clone
829 * @gfp_mask: allocation priority
831 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
832 * copies share the same packet data but not structure. The new
833 * buffer has a reference count of 1. If the allocation fails the
834 * function returns %NULL otherwise the new buffer is returned.
836 * If this function is called from an interrupt gfp_mask() must be
840 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
844 if (skb_orphan_frags(skb
, gfp_mask
))
848 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
849 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
850 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
851 n
->fclone
= SKB_FCLONE_CLONE
;
852 atomic_inc(fclone_ref
);
854 if (skb_pfmemalloc(skb
))
855 gfp_mask
|= __GFP_MEMALLOC
;
857 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
861 kmemcheck_annotate_bitfield(n
, flags1
);
862 kmemcheck_annotate_bitfield(n
, flags2
);
863 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
866 return __skb_clone(n
, skb
);
868 EXPORT_SYMBOL(skb_clone
);
870 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
872 #ifndef NET_SKBUFF_DATA_USES_OFFSET
874 * Shift between the two data areas in bytes
876 unsigned long offset
= new->data
- old
->data
;
879 __copy_skb_header(new, old
);
881 #ifndef NET_SKBUFF_DATA_USES_OFFSET
882 /* {transport,network,mac}_header are relative to skb->head */
883 new->transport_header
+= offset
;
884 new->network_header
+= offset
;
885 if (skb_mac_header_was_set(new))
886 new->mac_header
+= offset
;
887 new->inner_transport_header
+= offset
;
888 new->inner_network_header
+= offset
;
890 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
891 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
892 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
895 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
897 if (skb_pfmemalloc(skb
))
903 * skb_copy - create private copy of an sk_buff
904 * @skb: buffer to copy
905 * @gfp_mask: allocation priority
907 * Make a copy of both an &sk_buff and its data. This is used when the
908 * caller wishes to modify the data and needs a private copy of the
909 * data to alter. Returns %NULL on failure or the pointer to the buffer
910 * on success. The returned buffer has a reference count of 1.
912 * As by-product this function converts non-linear &sk_buff to linear
913 * one, so that &sk_buff becomes completely private and caller is allowed
914 * to modify all the data of returned buffer. This means that this
915 * function is not recommended for use in circumstances when only
916 * header is going to be modified. Use pskb_copy() instead.
919 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
921 int headerlen
= skb_headroom(skb
);
922 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
923 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
924 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
929 /* Set the data pointer */
930 skb_reserve(n
, headerlen
);
931 /* Set the tail pointer and length */
932 skb_put(n
, skb
->len
);
934 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
937 copy_skb_header(n
, skb
);
940 EXPORT_SYMBOL(skb_copy
);
943 * __pskb_copy - create copy of an sk_buff with private head.
944 * @skb: buffer to copy
945 * @headroom: headroom of new skb
946 * @gfp_mask: allocation priority
948 * Make a copy of both an &sk_buff and part of its data, located
949 * in header. Fragmented data remain shared. This is used when
950 * the caller wishes to modify only header of &sk_buff and needs
951 * private copy of the header to alter. Returns %NULL on failure
952 * or the pointer to the buffer on success.
953 * The returned buffer has a reference count of 1.
956 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
958 unsigned int size
= skb_headlen(skb
) + headroom
;
959 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
960 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
965 /* Set the data pointer */
966 skb_reserve(n
, headroom
);
967 /* Set the tail pointer and length */
968 skb_put(n
, skb_headlen(skb
));
970 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
972 n
->truesize
+= skb
->data_len
;
973 n
->data_len
= skb
->data_len
;
976 if (skb_shinfo(skb
)->nr_frags
) {
979 if (skb_orphan_frags(skb
, gfp_mask
)) {
984 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
985 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
986 skb_frag_ref(skb
, i
);
988 skb_shinfo(n
)->nr_frags
= i
;
991 if (skb_has_frag_list(skb
)) {
992 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
993 skb_clone_fraglist(n
);
996 copy_skb_header(n
, skb
);
1000 EXPORT_SYMBOL(__pskb_copy
);
1003 * pskb_expand_head - reallocate header of &sk_buff
1004 * @skb: buffer to reallocate
1005 * @nhead: room to add at head
1006 * @ntail: room to add at tail
1007 * @gfp_mask: allocation priority
1009 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1010 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1011 * reference count of 1. Returns zero in the case of success or error,
1012 * if expansion failed. In the last case, &sk_buff is not changed.
1014 * All the pointers pointing into skb header may change and must be
1015 * reloaded after call to this function.
1018 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1023 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1028 if (skb_shared(skb
))
1031 size
= SKB_DATA_ALIGN(size
);
1033 if (skb_pfmemalloc(skb
))
1034 gfp_mask
|= __GFP_MEMALLOC
;
1035 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1036 gfp_mask
, NUMA_NO_NODE
, NULL
);
1039 size
= SKB_WITH_OVERHEAD(ksize(data
));
1041 /* Copy only real data... and, alas, header. This should be
1042 * optimized for the cases when header is void.
1044 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1046 memcpy((struct skb_shared_info
*)(data
+ size
),
1048 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1051 * if shinfo is shared we must drop the old head gracefully, but if it
1052 * is not we can just drop the old head and let the existing refcount
1053 * be since all we did is relocate the values
1055 if (skb_cloned(skb
)) {
1056 /* copy this zero copy skb frags */
1057 if (skb_orphan_frags(skb
, gfp_mask
))
1059 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1060 skb_frag_ref(skb
, i
);
1062 if (skb_has_frag_list(skb
))
1063 skb_clone_fraglist(skb
);
1065 skb_release_data(skb
);
1069 off
= (data
+ nhead
) - skb
->head
;
1074 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1078 skb
->end
= skb
->head
+ size
;
1080 /* {transport,network,mac}_header and tail are relative to skb->head */
1082 skb
->transport_header
+= off
;
1083 skb
->network_header
+= off
;
1084 if (skb_mac_header_was_set(skb
))
1085 skb
->mac_header
+= off
;
1086 skb
->inner_transport_header
+= off
;
1087 skb
->inner_network_header
+= off
;
1088 /* Only adjust this if it actually is csum_start rather than csum */
1089 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1090 skb
->csum_start
+= nhead
;
1094 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1102 EXPORT_SYMBOL(pskb_expand_head
);
1104 /* Make private copy of skb with writable head and some headroom */
1106 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1108 struct sk_buff
*skb2
;
1109 int delta
= headroom
- skb_headroom(skb
);
1112 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1114 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1115 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1123 EXPORT_SYMBOL(skb_realloc_headroom
);
1126 * skb_copy_expand - copy and expand sk_buff
1127 * @skb: buffer to copy
1128 * @newheadroom: new free bytes at head
1129 * @newtailroom: new free bytes at tail
1130 * @gfp_mask: allocation priority
1132 * Make a copy of both an &sk_buff and its data and while doing so
1133 * allocate additional space.
1135 * This is used when the caller wishes to modify the data and needs a
1136 * private copy of the data to alter as well as more space for new fields.
1137 * Returns %NULL on failure or the pointer to the buffer
1138 * on success. The returned buffer has a reference count of 1.
1140 * You must pass %GFP_ATOMIC as the allocation priority if this function
1141 * is called from an interrupt.
1143 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1144 int newheadroom
, int newtailroom
,
1148 * Allocate the copy buffer
1150 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1151 gfp_mask
, skb_alloc_rx_flag(skb
),
1153 int oldheadroom
= skb_headroom(skb
);
1154 int head_copy_len
, head_copy_off
;
1160 skb_reserve(n
, newheadroom
);
1162 /* Set the tail pointer and length */
1163 skb_put(n
, skb
->len
);
1165 head_copy_len
= oldheadroom
;
1167 if (newheadroom
<= head_copy_len
)
1168 head_copy_len
= newheadroom
;
1170 head_copy_off
= newheadroom
- head_copy_len
;
1172 /* Copy the linear header and data. */
1173 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1174 skb
->len
+ head_copy_len
))
1177 copy_skb_header(n
, skb
);
1179 off
= newheadroom
- oldheadroom
;
1180 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1181 n
->csum_start
+= off
;
1182 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1183 n
->transport_header
+= off
;
1184 n
->network_header
+= off
;
1185 if (skb_mac_header_was_set(skb
))
1186 n
->mac_header
+= off
;
1187 n
->inner_transport_header
+= off
;
1188 n
->inner_network_header
+= off
;
1193 EXPORT_SYMBOL(skb_copy_expand
);
1196 * skb_pad - zero pad the tail of an skb
1197 * @skb: buffer to pad
1198 * @pad: space to pad
1200 * Ensure that a buffer is followed by a padding area that is zero
1201 * filled. Used by network drivers which may DMA or transfer data
1202 * beyond the buffer end onto the wire.
1204 * May return error in out of memory cases. The skb is freed on error.
1207 int skb_pad(struct sk_buff
*skb
, int pad
)
1212 /* If the skbuff is non linear tailroom is always zero.. */
1213 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1214 memset(skb
->data
+skb
->len
, 0, pad
);
1218 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1219 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1220 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1225 /* FIXME: The use of this function with non-linear skb's really needs
1228 err
= skb_linearize(skb
);
1232 memset(skb
->data
+ skb
->len
, 0, pad
);
1239 EXPORT_SYMBOL(skb_pad
);
1242 * skb_put - add data to a buffer
1243 * @skb: buffer to use
1244 * @len: amount of data to add
1246 * This function extends the used data area of the buffer. If this would
1247 * exceed the total buffer size the kernel will panic. A pointer to the
1248 * first byte of the extra data is returned.
1250 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1252 unsigned char *tmp
= skb_tail_pointer(skb
);
1253 SKB_LINEAR_ASSERT(skb
);
1256 if (unlikely(skb
->tail
> skb
->end
))
1257 skb_over_panic(skb
, len
, __builtin_return_address(0));
1260 EXPORT_SYMBOL(skb_put
);
1263 * skb_push - add data to the start of a buffer
1264 * @skb: buffer to use
1265 * @len: amount of data to add
1267 * This function extends the used data area of the buffer at the buffer
1268 * start. If this would exceed the total buffer headroom the kernel will
1269 * panic. A pointer to the first byte of the extra data is returned.
1271 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1275 if (unlikely(skb
->data
<skb
->head
))
1276 skb_under_panic(skb
, len
, __builtin_return_address(0));
1279 EXPORT_SYMBOL(skb_push
);
1282 * skb_pull - remove data from the start of a buffer
1283 * @skb: buffer to use
1284 * @len: amount of data to remove
1286 * This function removes data from the start of a buffer, returning
1287 * the memory to the headroom. A pointer to the next data in the buffer
1288 * is returned. Once the data has been pulled future pushes will overwrite
1291 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1293 return skb_pull_inline(skb
, len
);
1295 EXPORT_SYMBOL(skb_pull
);
1298 * skb_trim - remove end from a buffer
1299 * @skb: buffer to alter
1302 * Cut the length of a buffer down by removing data from the tail. If
1303 * the buffer is already under the length specified it is not modified.
1304 * The skb must be linear.
1306 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1309 __skb_trim(skb
, len
);
1311 EXPORT_SYMBOL(skb_trim
);
1313 /* Trims skb to length len. It can change skb pointers.
1316 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1318 struct sk_buff
**fragp
;
1319 struct sk_buff
*frag
;
1320 int offset
= skb_headlen(skb
);
1321 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1325 if (skb_cloned(skb
) &&
1326 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1333 for (; i
< nfrags
; i
++) {
1334 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1341 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1344 skb_shinfo(skb
)->nr_frags
= i
;
1346 for (; i
< nfrags
; i
++)
1347 skb_frag_unref(skb
, i
);
1349 if (skb_has_frag_list(skb
))
1350 skb_drop_fraglist(skb
);
1354 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1355 fragp
= &frag
->next
) {
1356 int end
= offset
+ frag
->len
;
1358 if (skb_shared(frag
)) {
1359 struct sk_buff
*nfrag
;
1361 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1362 if (unlikely(!nfrag
))
1365 nfrag
->next
= frag
->next
;
1377 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1381 skb_drop_list(&frag
->next
);
1386 if (len
> skb_headlen(skb
)) {
1387 skb
->data_len
-= skb
->len
- len
;
1392 skb_set_tail_pointer(skb
, len
);
1397 EXPORT_SYMBOL(___pskb_trim
);
1400 * __pskb_pull_tail - advance tail of skb header
1401 * @skb: buffer to reallocate
1402 * @delta: number of bytes to advance tail
1404 * The function makes a sense only on a fragmented &sk_buff,
1405 * it expands header moving its tail forward and copying necessary
1406 * data from fragmented part.
1408 * &sk_buff MUST have reference count of 1.
1410 * Returns %NULL (and &sk_buff does not change) if pull failed
1411 * or value of new tail of skb in the case of success.
1413 * All the pointers pointing into skb header may change and must be
1414 * reloaded after call to this function.
1417 /* Moves tail of skb head forward, copying data from fragmented part,
1418 * when it is necessary.
1419 * 1. It may fail due to malloc failure.
1420 * 2. It may change skb pointers.
1422 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1424 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1426 /* If skb has not enough free space at tail, get new one
1427 * plus 128 bytes for future expansions. If we have enough
1428 * room at tail, reallocate without expansion only if skb is cloned.
1430 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1432 if (eat
> 0 || skb_cloned(skb
)) {
1433 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1438 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1441 /* Optimization: no fragments, no reasons to preestimate
1442 * size of pulled pages. Superb.
1444 if (!skb_has_frag_list(skb
))
1447 /* Estimate size of pulled pages. */
1449 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1450 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1457 /* If we need update frag list, we are in troubles.
1458 * Certainly, it possible to add an offset to skb data,
1459 * but taking into account that pulling is expected to
1460 * be very rare operation, it is worth to fight against
1461 * further bloating skb head and crucify ourselves here instead.
1462 * Pure masohism, indeed. 8)8)
1465 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1466 struct sk_buff
*clone
= NULL
;
1467 struct sk_buff
*insp
= NULL
;
1472 if (list
->len
<= eat
) {
1473 /* Eaten as whole. */
1478 /* Eaten partially. */
1480 if (skb_shared(list
)) {
1481 /* Sucks! We need to fork list. :-( */
1482 clone
= skb_clone(list
, GFP_ATOMIC
);
1488 /* This may be pulled without
1492 if (!pskb_pull(list
, eat
)) {
1500 /* Free pulled out fragments. */
1501 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1502 skb_shinfo(skb
)->frag_list
= list
->next
;
1505 /* And insert new clone at head. */
1508 skb_shinfo(skb
)->frag_list
= clone
;
1511 /* Success! Now we may commit changes to skb data. */
1516 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1517 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1520 skb_frag_unref(skb
, i
);
1523 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1525 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1526 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1532 skb_shinfo(skb
)->nr_frags
= k
;
1535 skb
->data_len
-= delta
;
1537 return skb_tail_pointer(skb
);
1539 EXPORT_SYMBOL(__pskb_pull_tail
);
1542 * skb_copy_bits - copy bits from skb to kernel buffer
1544 * @offset: offset in source
1545 * @to: destination buffer
1546 * @len: number of bytes to copy
1548 * Copy the specified number of bytes from the source skb to the
1549 * destination buffer.
1552 * If its prototype is ever changed,
1553 * check arch/{*}/net/{*}.S files,
1554 * since it is called from BPF assembly code.
1556 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1558 int start
= skb_headlen(skb
);
1559 struct sk_buff
*frag_iter
;
1562 if (offset
> (int)skb
->len
- len
)
1566 if ((copy
= start
- offset
) > 0) {
1569 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1570 if ((len
-= copy
) == 0)
1576 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1578 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1580 WARN_ON(start
> offset
+ len
);
1582 end
= start
+ skb_frag_size(f
);
1583 if ((copy
= end
- offset
) > 0) {
1589 vaddr
= kmap_atomic(skb_frag_page(f
));
1591 vaddr
+ f
->page_offset
+ offset
- start
,
1593 kunmap_atomic(vaddr
);
1595 if ((len
-= copy
) == 0)
1603 skb_walk_frags(skb
, frag_iter
) {
1606 WARN_ON(start
> offset
+ len
);
1608 end
= start
+ frag_iter
->len
;
1609 if ((copy
= end
- offset
) > 0) {
1612 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1614 if ((len
-= copy
) == 0)
1628 EXPORT_SYMBOL(skb_copy_bits
);
1631 * Callback from splice_to_pipe(), if we need to release some pages
1632 * at the end of the spd in case we error'ed out in filling the pipe.
1634 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1636 put_page(spd
->pages
[i
]);
1639 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1640 unsigned int *offset
,
1643 struct page_frag
*pfrag
= sk_page_frag(sk
);
1645 if (!sk_page_frag_refill(sk
, pfrag
))
1648 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1650 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1651 page_address(page
) + *offset
, *len
);
1652 *offset
= pfrag
->offset
;
1653 pfrag
->offset
+= *len
;
1658 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1660 unsigned int offset
)
1662 return spd
->nr_pages
&&
1663 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1664 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1665 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1669 * Fill page/offset/length into spd, if it can hold more pages.
1671 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1672 struct pipe_inode_info
*pipe
, struct page
*page
,
1673 unsigned int *len
, unsigned int offset
,
1677 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1681 page
= linear_to_page(page
, len
, &offset
, sk
);
1685 if (spd_can_coalesce(spd
, page
, offset
)) {
1686 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1690 spd
->pages
[spd
->nr_pages
] = page
;
1691 spd
->partial
[spd
->nr_pages
].len
= *len
;
1692 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1698 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1699 unsigned int plen
, unsigned int *off
,
1701 struct splice_pipe_desc
*spd
, bool linear
,
1703 struct pipe_inode_info
*pipe
)
1708 /* skip this segment if already processed */
1714 /* ignore any bits we already processed */
1720 unsigned int flen
= min(*len
, plen
);
1722 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
,
1728 } while (*len
&& plen
);
1734 * Map linear and fragment data from the skb to spd. It reports true if the
1735 * pipe is full or if we already spliced the requested length.
1737 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1738 unsigned int *offset
, unsigned int *len
,
1739 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1743 /* map the linear part :
1744 * If skb->head_frag is set, this 'linear' part is backed by a
1745 * fragment, and if the head is not shared with any clones then
1746 * we can avoid a copy since we own the head portion of this page.
1748 if (__splice_segment(virt_to_page(skb
->data
),
1749 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1752 skb_head_is_locked(skb
),
1757 * then map the fragments
1759 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1760 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1762 if (__splice_segment(skb_frag_page(f
),
1763 f
->page_offset
, skb_frag_size(f
),
1764 offset
, len
, spd
, false, sk
, pipe
))
1772 * Map data from the skb to a pipe. Should handle both the linear part,
1773 * the fragments, and the frag list. It does NOT handle frag lists within
1774 * the frag list, if such a thing exists. We'd probably need to recurse to
1775 * handle that cleanly.
1777 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1778 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1781 struct partial_page partial
[MAX_SKB_FRAGS
];
1782 struct page
*pages
[MAX_SKB_FRAGS
];
1783 struct splice_pipe_desc spd
= {
1786 .nr_pages_max
= MAX_SKB_FRAGS
,
1788 .ops
= &sock_pipe_buf_ops
,
1789 .spd_release
= sock_spd_release
,
1791 struct sk_buff
*frag_iter
;
1792 struct sock
*sk
= skb
->sk
;
1796 * __skb_splice_bits() only fails if the output has no room left,
1797 * so no point in going over the frag_list for the error case.
1799 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1805 * now see if we have a frag_list to map
1807 skb_walk_frags(skb
, frag_iter
) {
1810 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1817 * Drop the socket lock, otherwise we have reverse
1818 * locking dependencies between sk_lock and i_mutex
1819 * here as compared to sendfile(). We enter here
1820 * with the socket lock held, and splice_to_pipe() will
1821 * grab the pipe inode lock. For sendfile() emulation,
1822 * we call into ->sendpage() with the i_mutex lock held
1823 * and networking will grab the socket lock.
1826 ret
= splice_to_pipe(pipe
, &spd
);
1834 * skb_store_bits - store bits from kernel buffer to skb
1835 * @skb: destination buffer
1836 * @offset: offset in destination
1837 * @from: source buffer
1838 * @len: number of bytes to copy
1840 * Copy the specified number of bytes from the source buffer to the
1841 * destination skb. This function handles all the messy bits of
1842 * traversing fragment lists and such.
1845 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1847 int start
= skb_headlen(skb
);
1848 struct sk_buff
*frag_iter
;
1851 if (offset
> (int)skb
->len
- len
)
1854 if ((copy
= start
- offset
) > 0) {
1857 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1858 if ((len
-= copy
) == 0)
1864 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1865 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1868 WARN_ON(start
> offset
+ len
);
1870 end
= start
+ skb_frag_size(frag
);
1871 if ((copy
= end
- offset
) > 0) {
1877 vaddr
= kmap_atomic(skb_frag_page(frag
));
1878 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1880 kunmap_atomic(vaddr
);
1882 if ((len
-= copy
) == 0)
1890 skb_walk_frags(skb
, frag_iter
) {
1893 WARN_ON(start
> offset
+ len
);
1895 end
= start
+ frag_iter
->len
;
1896 if ((copy
= end
- offset
) > 0) {
1899 if (skb_store_bits(frag_iter
, offset
- start
,
1902 if ((len
-= copy
) == 0)
1915 EXPORT_SYMBOL(skb_store_bits
);
1917 /* Checksum skb data. */
1919 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1920 int len
, __wsum csum
)
1922 int start
= skb_headlen(skb
);
1923 int i
, copy
= start
- offset
;
1924 struct sk_buff
*frag_iter
;
1927 /* Checksum header. */
1931 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1932 if ((len
-= copy
) == 0)
1938 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1940 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1942 WARN_ON(start
> offset
+ len
);
1944 end
= start
+ skb_frag_size(frag
);
1945 if ((copy
= end
- offset
) > 0) {
1951 vaddr
= kmap_atomic(skb_frag_page(frag
));
1952 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1953 offset
- start
, copy
, 0);
1954 kunmap_atomic(vaddr
);
1955 csum
= csum_block_add(csum
, csum2
, pos
);
1964 skb_walk_frags(skb
, frag_iter
) {
1967 WARN_ON(start
> offset
+ len
);
1969 end
= start
+ frag_iter
->len
;
1970 if ((copy
= end
- offset
) > 0) {
1974 csum2
= skb_checksum(frag_iter
, offset
- start
,
1976 csum
= csum_block_add(csum
, csum2
, pos
);
1977 if ((len
-= copy
) == 0)
1988 EXPORT_SYMBOL(skb_checksum
);
1990 /* Both of above in one bottle. */
1992 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1993 u8
*to
, int len
, __wsum csum
)
1995 int start
= skb_headlen(skb
);
1996 int i
, copy
= start
- offset
;
1997 struct sk_buff
*frag_iter
;
2004 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2006 if ((len
-= copy
) == 0)
2013 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2016 WARN_ON(start
> offset
+ len
);
2018 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2019 if ((copy
= end
- offset
) > 0) {
2022 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2026 vaddr
= kmap_atomic(skb_frag_page(frag
));
2027 csum2
= csum_partial_copy_nocheck(vaddr
+
2031 kunmap_atomic(vaddr
);
2032 csum
= csum_block_add(csum
, csum2
, pos
);
2042 skb_walk_frags(skb
, frag_iter
) {
2046 WARN_ON(start
> offset
+ len
);
2048 end
= start
+ frag_iter
->len
;
2049 if ((copy
= end
- offset
) > 0) {
2052 csum2
= skb_copy_and_csum_bits(frag_iter
,
2055 csum
= csum_block_add(csum
, csum2
, pos
);
2056 if ((len
-= copy
) == 0)
2067 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2069 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2074 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2075 csstart
= skb_checksum_start_offset(skb
);
2077 csstart
= skb_headlen(skb
);
2079 BUG_ON(csstart
> skb_headlen(skb
));
2081 skb_copy_from_linear_data(skb
, to
, csstart
);
2084 if (csstart
!= skb
->len
)
2085 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2086 skb
->len
- csstart
, 0);
2088 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2089 long csstuff
= csstart
+ skb
->csum_offset
;
2091 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2094 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2097 * skb_dequeue - remove from the head of the queue
2098 * @list: list to dequeue from
2100 * Remove the head of the list. The list lock is taken so the function
2101 * may be used safely with other locking list functions. The head item is
2102 * returned or %NULL if the list is empty.
2105 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2107 unsigned long flags
;
2108 struct sk_buff
*result
;
2110 spin_lock_irqsave(&list
->lock
, flags
);
2111 result
= __skb_dequeue(list
);
2112 spin_unlock_irqrestore(&list
->lock
, flags
);
2115 EXPORT_SYMBOL(skb_dequeue
);
2118 * skb_dequeue_tail - remove from the tail of the queue
2119 * @list: list to dequeue from
2121 * Remove the tail of the list. The list lock is taken so the function
2122 * may be used safely with other locking list functions. The tail item is
2123 * returned or %NULL if the list is empty.
2125 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2127 unsigned long flags
;
2128 struct sk_buff
*result
;
2130 spin_lock_irqsave(&list
->lock
, flags
);
2131 result
= __skb_dequeue_tail(list
);
2132 spin_unlock_irqrestore(&list
->lock
, flags
);
2135 EXPORT_SYMBOL(skb_dequeue_tail
);
2138 * skb_queue_purge - empty a list
2139 * @list: list to empty
2141 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2142 * the list and one reference dropped. This function takes the list
2143 * lock and is atomic with respect to other list locking functions.
2145 void skb_queue_purge(struct sk_buff_head
*list
)
2147 struct sk_buff
*skb
;
2148 while ((skb
= skb_dequeue(list
)) != NULL
)
2151 EXPORT_SYMBOL(skb_queue_purge
);
2154 * skb_queue_head - queue a buffer at the list head
2155 * @list: list to use
2156 * @newsk: buffer to queue
2158 * Queue a buffer at the start of the list. This function takes the
2159 * list lock and can be used safely with other locking &sk_buff functions
2162 * A buffer cannot be placed on two lists at the same time.
2164 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2166 unsigned long flags
;
2168 spin_lock_irqsave(&list
->lock
, flags
);
2169 __skb_queue_head(list
, newsk
);
2170 spin_unlock_irqrestore(&list
->lock
, flags
);
2172 EXPORT_SYMBOL(skb_queue_head
);
2175 * skb_queue_tail - queue a buffer at the list tail
2176 * @list: list to use
2177 * @newsk: buffer to queue
2179 * Queue a buffer at the tail of the list. This function takes the
2180 * list lock and can be used safely with other locking &sk_buff functions
2183 * A buffer cannot be placed on two lists at the same time.
2185 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2187 unsigned long flags
;
2189 spin_lock_irqsave(&list
->lock
, flags
);
2190 __skb_queue_tail(list
, newsk
);
2191 spin_unlock_irqrestore(&list
->lock
, flags
);
2193 EXPORT_SYMBOL(skb_queue_tail
);
2196 * skb_unlink - remove a buffer from a list
2197 * @skb: buffer to remove
2198 * @list: list to use
2200 * Remove a packet from a list. The list locks are taken and this
2201 * function is atomic with respect to other list locked calls
2203 * You must know what list the SKB is on.
2205 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2207 unsigned long flags
;
2209 spin_lock_irqsave(&list
->lock
, flags
);
2210 __skb_unlink(skb
, list
);
2211 spin_unlock_irqrestore(&list
->lock
, flags
);
2213 EXPORT_SYMBOL(skb_unlink
);
2216 * skb_append - append a buffer
2217 * @old: buffer to insert after
2218 * @newsk: buffer to insert
2219 * @list: list to use
2221 * Place a packet after a given packet in a list. The list locks are taken
2222 * and this function is atomic with respect to other list locked calls.
2223 * A buffer cannot be placed on two lists at the same time.
2225 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2227 unsigned long flags
;
2229 spin_lock_irqsave(&list
->lock
, flags
);
2230 __skb_queue_after(list
, old
, newsk
);
2231 spin_unlock_irqrestore(&list
->lock
, flags
);
2233 EXPORT_SYMBOL(skb_append
);
2236 * skb_insert - insert a buffer
2237 * @old: buffer to insert before
2238 * @newsk: buffer to insert
2239 * @list: list to use
2241 * Place a packet before a given packet in a list. The list locks are
2242 * taken and this function is atomic with respect to other list locked
2245 * A buffer cannot be placed on two lists at the same time.
2247 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2249 unsigned long flags
;
2251 spin_lock_irqsave(&list
->lock
, flags
);
2252 __skb_insert(newsk
, old
->prev
, old
, list
);
2253 spin_unlock_irqrestore(&list
->lock
, flags
);
2255 EXPORT_SYMBOL(skb_insert
);
2257 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2258 struct sk_buff
* skb1
,
2259 const u32 len
, const int pos
)
2263 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2265 /* And move data appendix as is. */
2266 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2267 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2269 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2270 skb_shinfo(skb
)->nr_frags
= 0;
2271 skb1
->data_len
= skb
->data_len
;
2272 skb1
->len
+= skb1
->data_len
;
2275 skb_set_tail_pointer(skb
, len
);
2278 static inline void skb_split_no_header(struct sk_buff
*skb
,
2279 struct sk_buff
* skb1
,
2280 const u32 len
, int pos
)
2283 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2285 skb_shinfo(skb
)->nr_frags
= 0;
2286 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2288 skb
->data_len
= len
- pos
;
2290 for (i
= 0; i
< nfrags
; i
++) {
2291 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2293 if (pos
+ size
> len
) {
2294 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2298 * We have two variants in this case:
2299 * 1. Move all the frag to the second
2300 * part, if it is possible. F.e.
2301 * this approach is mandatory for TUX,
2302 * where splitting is expensive.
2303 * 2. Split is accurately. We make this.
2305 skb_frag_ref(skb
, i
);
2306 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2307 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2308 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2309 skb_shinfo(skb
)->nr_frags
++;
2313 skb_shinfo(skb
)->nr_frags
++;
2316 skb_shinfo(skb1
)->nr_frags
= k
;
2320 * skb_split - Split fragmented skb to two parts at length len.
2321 * @skb: the buffer to split
2322 * @skb1: the buffer to receive the second part
2323 * @len: new length for skb
2325 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2327 int pos
= skb_headlen(skb
);
2329 skb_shinfo(skb1
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2330 if (len
< pos
) /* Split line is inside header. */
2331 skb_split_inside_header(skb
, skb1
, len
, pos
);
2332 else /* Second chunk has no header, nothing to copy. */
2333 skb_split_no_header(skb
, skb1
, len
, pos
);
2335 EXPORT_SYMBOL(skb_split
);
2337 /* Shifting from/to a cloned skb is a no-go.
2339 * Caller cannot keep skb_shinfo related pointers past calling here!
2341 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2343 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2347 * skb_shift - Shifts paged data partially from skb to another
2348 * @tgt: buffer into which tail data gets added
2349 * @skb: buffer from which the paged data comes from
2350 * @shiftlen: shift up to this many bytes
2352 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2353 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2354 * It's up to caller to free skb if everything was shifted.
2356 * If @tgt runs out of frags, the whole operation is aborted.
2358 * Skb cannot include anything else but paged data while tgt is allowed
2359 * to have non-paged data as well.
2361 * TODO: full sized shift could be optimized but that would need
2362 * specialized skb free'er to handle frags without up-to-date nr_frags.
2364 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2366 int from
, to
, merge
, todo
;
2367 struct skb_frag_struct
*fragfrom
, *fragto
;
2369 BUG_ON(shiftlen
> skb
->len
);
2370 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2374 to
= skb_shinfo(tgt
)->nr_frags
;
2375 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2377 /* Actual merge is delayed until the point when we know we can
2378 * commit all, so that we don't have to undo partial changes
2381 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2382 fragfrom
->page_offset
)) {
2387 todo
-= skb_frag_size(fragfrom
);
2389 if (skb_prepare_for_shift(skb
) ||
2390 skb_prepare_for_shift(tgt
))
2393 /* All previous frag pointers might be stale! */
2394 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2395 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2397 skb_frag_size_add(fragto
, shiftlen
);
2398 skb_frag_size_sub(fragfrom
, shiftlen
);
2399 fragfrom
->page_offset
+= shiftlen
;
2407 /* Skip full, not-fitting skb to avoid expensive operations */
2408 if ((shiftlen
== skb
->len
) &&
2409 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2412 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2415 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2416 if (to
== MAX_SKB_FRAGS
)
2419 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2420 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2422 if (todo
>= skb_frag_size(fragfrom
)) {
2423 *fragto
= *fragfrom
;
2424 todo
-= skb_frag_size(fragfrom
);
2429 __skb_frag_ref(fragfrom
);
2430 fragto
->page
= fragfrom
->page
;
2431 fragto
->page_offset
= fragfrom
->page_offset
;
2432 skb_frag_size_set(fragto
, todo
);
2434 fragfrom
->page_offset
+= todo
;
2435 skb_frag_size_sub(fragfrom
, todo
);
2443 /* Ready to "commit" this state change to tgt */
2444 skb_shinfo(tgt
)->nr_frags
= to
;
2447 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2448 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2450 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2451 __skb_frag_unref(fragfrom
);
2454 /* Reposition in the original skb */
2456 while (from
< skb_shinfo(skb
)->nr_frags
)
2457 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2458 skb_shinfo(skb
)->nr_frags
= to
;
2460 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2463 /* Most likely the tgt won't ever need its checksum anymore, skb on
2464 * the other hand might need it if it needs to be resent
2466 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2467 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2469 /* Yak, is it really working this way? Some helper please? */
2470 skb
->len
-= shiftlen
;
2471 skb
->data_len
-= shiftlen
;
2472 skb
->truesize
-= shiftlen
;
2473 tgt
->len
+= shiftlen
;
2474 tgt
->data_len
+= shiftlen
;
2475 tgt
->truesize
+= shiftlen
;
2481 * skb_prepare_seq_read - Prepare a sequential read of skb data
2482 * @skb: the buffer to read
2483 * @from: lower offset of data to be read
2484 * @to: upper offset of data to be read
2485 * @st: state variable
2487 * Initializes the specified state variable. Must be called before
2488 * invoking skb_seq_read() for the first time.
2490 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2491 unsigned int to
, struct skb_seq_state
*st
)
2493 st
->lower_offset
= from
;
2494 st
->upper_offset
= to
;
2495 st
->root_skb
= st
->cur_skb
= skb
;
2496 st
->frag_idx
= st
->stepped_offset
= 0;
2497 st
->frag_data
= NULL
;
2499 EXPORT_SYMBOL(skb_prepare_seq_read
);
2502 * skb_seq_read - Sequentially read skb data
2503 * @consumed: number of bytes consumed by the caller so far
2504 * @data: destination pointer for data to be returned
2505 * @st: state variable
2507 * Reads a block of skb data at &consumed relative to the
2508 * lower offset specified to skb_prepare_seq_read(). Assigns
2509 * the head of the data block to &data and returns the length
2510 * of the block or 0 if the end of the skb data or the upper
2511 * offset has been reached.
2513 * The caller is not required to consume all of the data
2514 * returned, i.e. &consumed is typically set to the number
2515 * of bytes already consumed and the next call to
2516 * skb_seq_read() will return the remaining part of the block.
2518 * Note 1: The size of each block of data returned can be arbitrary,
2519 * this limitation is the cost for zerocopy seqeuental
2520 * reads of potentially non linear data.
2522 * Note 2: Fragment lists within fragments are not implemented
2523 * at the moment, state->root_skb could be replaced with
2524 * a stack for this purpose.
2526 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2527 struct skb_seq_state
*st
)
2529 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2532 if (unlikely(abs_offset
>= st
->upper_offset
))
2536 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2538 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2539 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2540 return block_limit
- abs_offset
;
2543 if (st
->frag_idx
== 0 && !st
->frag_data
)
2544 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2546 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2547 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2548 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2550 if (abs_offset
< block_limit
) {
2552 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2554 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2555 (abs_offset
- st
->stepped_offset
);
2557 return block_limit
- abs_offset
;
2560 if (st
->frag_data
) {
2561 kunmap_atomic(st
->frag_data
);
2562 st
->frag_data
= NULL
;
2566 st
->stepped_offset
+= skb_frag_size(frag
);
2569 if (st
->frag_data
) {
2570 kunmap_atomic(st
->frag_data
);
2571 st
->frag_data
= NULL
;
2574 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2575 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2578 } else if (st
->cur_skb
->next
) {
2579 st
->cur_skb
= st
->cur_skb
->next
;
2586 EXPORT_SYMBOL(skb_seq_read
);
2589 * skb_abort_seq_read - Abort a sequential read of skb data
2590 * @st: state variable
2592 * Must be called if skb_seq_read() was not called until it
2595 void skb_abort_seq_read(struct skb_seq_state
*st
)
2598 kunmap_atomic(st
->frag_data
);
2600 EXPORT_SYMBOL(skb_abort_seq_read
);
2602 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2604 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2605 struct ts_config
*conf
,
2606 struct ts_state
*state
)
2608 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2611 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2613 skb_abort_seq_read(TS_SKB_CB(state
));
2617 * skb_find_text - Find a text pattern in skb data
2618 * @skb: the buffer to look in
2619 * @from: search offset
2621 * @config: textsearch configuration
2622 * @state: uninitialized textsearch state variable
2624 * Finds a pattern in the skb data according to the specified
2625 * textsearch configuration. Use textsearch_next() to retrieve
2626 * subsequent occurrences of the pattern. Returns the offset
2627 * to the first occurrence or UINT_MAX if no match was found.
2629 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2630 unsigned int to
, struct ts_config
*config
,
2631 struct ts_state
*state
)
2635 config
->get_next_block
= skb_ts_get_next_block
;
2636 config
->finish
= skb_ts_finish
;
2638 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2640 ret
= textsearch_find(config
, state
);
2641 return (ret
<= to
- from
? ret
: UINT_MAX
);
2643 EXPORT_SYMBOL(skb_find_text
);
2646 * skb_append_datato_frags - append the user data to a skb
2647 * @sk: sock structure
2648 * @skb: skb structure to be appened with user data.
2649 * @getfrag: call back function to be used for getting the user data
2650 * @from: pointer to user message iov
2651 * @length: length of the iov message
2653 * Description: This procedure append the user data in the fragment part
2654 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2656 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2657 int (*getfrag
)(void *from
, char *to
, int offset
,
2658 int len
, int odd
, struct sk_buff
*skb
),
2659 void *from
, int length
)
2661 int frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2665 struct page_frag
*pfrag
= ¤t
->task_frag
;
2668 /* Return error if we don't have space for new frag */
2669 if (frg_cnt
>= MAX_SKB_FRAGS
)
2672 if (!sk_page_frag_refill(sk
, pfrag
))
2675 /* copy the user data to page */
2676 copy
= min_t(int, length
, pfrag
->size
- pfrag
->offset
);
2678 ret
= getfrag(from
, page_address(pfrag
->page
) + pfrag
->offset
,
2679 offset
, copy
, 0, skb
);
2683 /* copy was successful so update the size parameters */
2684 skb_fill_page_desc(skb
, frg_cnt
, pfrag
->page
, pfrag
->offset
,
2687 pfrag
->offset
+= copy
;
2688 get_page(pfrag
->page
);
2690 skb
->truesize
+= copy
;
2691 atomic_add(copy
, &sk
->sk_wmem_alloc
);
2693 skb
->data_len
+= copy
;
2697 } while (length
> 0);
2701 EXPORT_SYMBOL(skb_append_datato_frags
);
2704 * skb_pull_rcsum - pull skb and update receive checksum
2705 * @skb: buffer to update
2706 * @len: length of data pulled
2708 * This function performs an skb_pull on the packet and updates
2709 * the CHECKSUM_COMPLETE checksum. It should be used on
2710 * receive path processing instead of skb_pull unless you know
2711 * that the checksum difference is zero (e.g., a valid IP header)
2712 * or you are setting ip_summed to CHECKSUM_NONE.
2714 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2716 BUG_ON(len
> skb
->len
);
2718 BUG_ON(skb
->len
< skb
->data_len
);
2719 skb_postpull_rcsum(skb
, skb
->data
, len
);
2720 return skb
->data
+= len
;
2722 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2725 * skb_segment - Perform protocol segmentation on skb.
2726 * @skb: buffer to segment
2727 * @features: features for the output path (see dev->features)
2729 * This function performs segmentation on the given skb. It returns
2730 * a pointer to the first in a list of new skbs for the segments.
2731 * In case of error it returns ERR_PTR(err).
2733 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2735 struct sk_buff
*segs
= NULL
;
2736 struct sk_buff
*tail
= NULL
;
2737 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2738 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2739 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2740 unsigned int offset
= doffset
;
2741 unsigned int tnl_hlen
= skb_tnl_header_len(skb
);
2742 unsigned int headroom
;
2744 int sg
= !!(features
& NETIF_F_SG
);
2745 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2750 __skb_push(skb
, doffset
);
2751 headroom
= skb_headroom(skb
);
2752 pos
= skb_headlen(skb
);
2755 struct sk_buff
*nskb
;
2760 len
= skb
->len
- offset
;
2764 hsize
= skb_headlen(skb
) - offset
;
2767 if (hsize
> len
|| !sg
)
2770 if (!hsize
&& i
>= nfrags
) {
2771 BUG_ON(fskb
->len
!= len
);
2774 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2777 if (unlikely(!nskb
))
2780 hsize
= skb_end_offset(nskb
);
2781 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2786 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2787 skb_release_head_state(nskb
);
2788 __skb_push(nskb
, doffset
);
2790 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2791 GFP_ATOMIC
, skb_alloc_rx_flag(skb
),
2794 if (unlikely(!nskb
))
2797 skb_reserve(nskb
, headroom
);
2798 __skb_put(nskb
, doffset
);
2807 __copy_skb_header(nskb
, skb
);
2808 nskb
->mac_len
= skb
->mac_len
;
2810 /* nskb and skb might have different headroom */
2811 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2812 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2814 skb_reset_mac_header(nskb
);
2815 skb_set_network_header(nskb
, skb
->mac_len
);
2816 nskb
->transport_header
= (nskb
->network_header
+
2817 skb_network_header_len(skb
));
2819 skb_copy_from_linear_data_offset(skb
, -tnl_hlen
,
2820 nskb
->data
- tnl_hlen
,
2821 doffset
+ tnl_hlen
);
2823 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2827 nskb
->ip_summed
= CHECKSUM_NONE
;
2828 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2834 frag
= skb_shinfo(nskb
)->frags
;
2836 skb_copy_from_linear_data_offset(skb
, offset
,
2837 skb_put(nskb
, hsize
), hsize
);
2839 skb_shinfo(nskb
)->tx_flags
= skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2841 while (pos
< offset
+ len
&& i
< nfrags
) {
2842 *frag
= skb_shinfo(skb
)->frags
[i
];
2843 __skb_frag_ref(frag
);
2844 size
= skb_frag_size(frag
);
2847 frag
->page_offset
+= offset
- pos
;
2848 skb_frag_size_sub(frag
, offset
- pos
);
2851 skb_shinfo(nskb
)->nr_frags
++;
2853 if (pos
+ size
<= offset
+ len
) {
2857 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2864 if (pos
< offset
+ len
) {
2865 struct sk_buff
*fskb2
= fskb
;
2867 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2873 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2879 SKB_FRAG_ASSERT(nskb
);
2880 skb_shinfo(nskb
)->frag_list
= fskb2
;
2884 nskb
->data_len
= len
- hsize
;
2885 nskb
->len
+= nskb
->data_len
;
2886 nskb
->truesize
+= nskb
->data_len
;
2887 } while ((offset
+= len
) < skb
->len
);
2892 while ((skb
= segs
)) {
2896 return ERR_PTR(err
);
2898 EXPORT_SYMBOL_GPL(skb_segment
);
2900 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2902 struct sk_buff
*p
= *head
;
2903 struct sk_buff
*nskb
;
2904 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2905 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2906 unsigned int headroom
;
2907 unsigned int len
= skb_gro_len(skb
);
2908 unsigned int offset
= skb_gro_offset(skb
);
2909 unsigned int headlen
= skb_headlen(skb
);
2910 unsigned int delta_truesize
;
2912 if (p
->len
+ len
>= 65536)
2915 if (pinfo
->frag_list
)
2917 else if (headlen
<= offset
) {
2920 int i
= skbinfo
->nr_frags
;
2921 int nr_frags
= pinfo
->nr_frags
+ i
;
2925 if (nr_frags
> MAX_SKB_FRAGS
)
2928 pinfo
->nr_frags
= nr_frags
;
2929 skbinfo
->nr_frags
= 0;
2931 frag
= pinfo
->frags
+ nr_frags
;
2932 frag2
= skbinfo
->frags
+ i
;
2937 frag
->page_offset
+= offset
;
2938 skb_frag_size_sub(frag
, offset
);
2940 /* all fragments truesize : remove (head size + sk_buff) */
2941 delta_truesize
= skb
->truesize
-
2942 SKB_TRUESIZE(skb_end_offset(skb
));
2944 skb
->truesize
-= skb
->data_len
;
2945 skb
->len
-= skb
->data_len
;
2948 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2950 } else if (skb
->head_frag
) {
2951 int nr_frags
= pinfo
->nr_frags
;
2952 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2953 struct page
*page
= virt_to_head_page(skb
->head
);
2954 unsigned int first_size
= headlen
- offset
;
2955 unsigned int first_offset
;
2957 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2960 first_offset
= skb
->data
-
2961 (unsigned char *)page_address(page
) +
2964 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
2966 frag
->page
.p
= page
;
2967 frag
->page_offset
= first_offset
;
2968 skb_frag_size_set(frag
, first_size
);
2970 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
2971 /* We dont need to clear skbinfo->nr_frags here */
2973 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
2974 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
2976 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2979 headroom
= skb_headroom(p
);
2980 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2981 if (unlikely(!nskb
))
2984 __copy_skb_header(nskb
, p
);
2985 nskb
->mac_len
= p
->mac_len
;
2987 skb_reserve(nskb
, headroom
);
2988 __skb_put(nskb
, skb_gro_offset(p
));
2990 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2991 skb_set_network_header(nskb
, skb_network_offset(p
));
2992 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2994 __skb_pull(p
, skb_gro_offset(p
));
2995 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2996 p
->data
- skb_mac_header(p
));
2998 skb_shinfo(nskb
)->frag_list
= p
;
2999 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3000 pinfo
->gso_size
= 0;
3001 skb_header_release(p
);
3002 NAPI_GRO_CB(nskb
)->last
= p
;
3004 nskb
->data_len
+= p
->len
;
3005 nskb
->truesize
+= p
->truesize
;
3006 nskb
->len
+= p
->len
;
3009 nskb
->next
= p
->next
;
3015 delta_truesize
= skb
->truesize
;
3016 if (offset
> headlen
) {
3017 unsigned int eat
= offset
- headlen
;
3019 skbinfo
->frags
[0].page_offset
+= eat
;
3020 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3021 skb
->data_len
-= eat
;
3026 __skb_pull(skb
, offset
);
3028 NAPI_GRO_CB(p
)->last
->next
= skb
;
3029 NAPI_GRO_CB(p
)->last
= skb
;
3030 skb_header_release(skb
);
3033 NAPI_GRO_CB(p
)->count
++;
3035 p
->truesize
+= delta_truesize
;
3038 NAPI_GRO_CB(skb
)->same_flow
= 1;
3041 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3043 void __init
skb_init(void)
3045 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3046 sizeof(struct sk_buff
),
3048 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3050 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3051 (2*sizeof(struct sk_buff
)) +
3054 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3059 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3060 * @skb: Socket buffer containing the buffers to be mapped
3061 * @sg: The scatter-gather list to map into
3062 * @offset: The offset into the buffer's contents to start mapping
3063 * @len: Length of buffer space to be mapped
3065 * Fill the specified scatter-gather list with mappings/pointers into a
3066 * region of the buffer space attached to a socket buffer.
3069 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3071 int start
= skb_headlen(skb
);
3072 int i
, copy
= start
- offset
;
3073 struct sk_buff
*frag_iter
;
3079 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3081 if ((len
-= copy
) == 0)
3086 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3089 WARN_ON(start
> offset
+ len
);
3091 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3092 if ((copy
= end
- offset
) > 0) {
3093 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3097 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3098 frag
->page_offset
+offset
-start
);
3107 skb_walk_frags(skb
, frag_iter
) {
3110 WARN_ON(start
> offset
+ len
);
3112 end
= start
+ frag_iter
->len
;
3113 if ((copy
= end
- offset
) > 0) {
3116 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3118 if ((len
-= copy
) == 0)
3128 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3130 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3132 sg_mark_end(&sg
[nsg
- 1]);
3136 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3139 * skb_cow_data - Check that a socket buffer's data buffers are writable
3140 * @skb: The socket buffer to check.
3141 * @tailbits: Amount of trailing space to be added
3142 * @trailer: Returned pointer to the skb where the @tailbits space begins
3144 * Make sure that the data buffers attached to a socket buffer are
3145 * writable. If they are not, private copies are made of the data buffers
3146 * and the socket buffer is set to use these instead.
3148 * If @tailbits is given, make sure that there is space to write @tailbits
3149 * bytes of data beyond current end of socket buffer. @trailer will be
3150 * set to point to the skb in which this space begins.
3152 * The number of scatterlist elements required to completely map the
3153 * COW'd and extended socket buffer will be returned.
3155 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3159 struct sk_buff
*skb1
, **skb_p
;
3161 /* If skb is cloned or its head is paged, reallocate
3162 * head pulling out all the pages (pages are considered not writable
3163 * at the moment even if they are anonymous).
3165 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3166 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3169 /* Easy case. Most of packets will go this way. */
3170 if (!skb_has_frag_list(skb
)) {
3171 /* A little of trouble, not enough of space for trailer.
3172 * This should not happen, when stack is tuned to generate
3173 * good frames. OK, on miss we reallocate and reserve even more
3174 * space, 128 bytes is fair. */
3176 if (skb_tailroom(skb
) < tailbits
&&
3177 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3185 /* Misery. We are in troubles, going to mincer fragments... */
3188 skb_p
= &skb_shinfo(skb
)->frag_list
;
3191 while ((skb1
= *skb_p
) != NULL
) {
3194 /* The fragment is partially pulled by someone,
3195 * this can happen on input. Copy it and everything
3198 if (skb_shared(skb1
))
3201 /* If the skb is the last, worry about trailer. */
3203 if (skb1
->next
== NULL
&& tailbits
) {
3204 if (skb_shinfo(skb1
)->nr_frags
||
3205 skb_has_frag_list(skb1
) ||
3206 skb_tailroom(skb1
) < tailbits
)
3207 ntail
= tailbits
+ 128;
3213 skb_shinfo(skb1
)->nr_frags
||
3214 skb_has_frag_list(skb1
)) {
3215 struct sk_buff
*skb2
;
3217 /* Fuck, we are miserable poor guys... */
3219 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3221 skb2
= skb_copy_expand(skb1
,
3225 if (unlikely(skb2
== NULL
))
3229 skb_set_owner_w(skb2
, skb1
->sk
);
3231 /* Looking around. Are we still alive?
3232 * OK, link new skb, drop old one */
3234 skb2
->next
= skb1
->next
;
3241 skb_p
= &skb1
->next
;
3246 EXPORT_SYMBOL_GPL(skb_cow_data
);
3248 static void sock_rmem_free(struct sk_buff
*skb
)
3250 struct sock
*sk
= skb
->sk
;
3252 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3256 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3258 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3262 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3263 (unsigned int)sk
->sk_rcvbuf
)
3268 skb
->destructor
= sock_rmem_free
;
3269 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3271 /* before exiting rcu section, make sure dst is refcounted */
3274 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3275 if (!sock_flag(sk
, SOCK_DEAD
))
3276 sk
->sk_data_ready(sk
, len
);
3279 EXPORT_SYMBOL(sock_queue_err_skb
);
3281 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3282 struct skb_shared_hwtstamps
*hwtstamps
)
3284 struct sock
*sk
= orig_skb
->sk
;
3285 struct sock_exterr_skb
*serr
;
3286 struct sk_buff
*skb
;
3292 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3297 *skb_hwtstamps(skb
) =
3301 * no hardware time stamps available,
3302 * so keep the shared tx_flags and only
3303 * store software time stamp
3305 skb
->tstamp
= ktime_get_real();
3308 serr
= SKB_EXT_ERR(skb
);
3309 memset(serr
, 0, sizeof(*serr
));
3310 serr
->ee
.ee_errno
= ENOMSG
;
3311 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3313 err
= sock_queue_err_skb(sk
, skb
);
3318 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3320 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3322 struct sock
*sk
= skb
->sk
;
3323 struct sock_exterr_skb
*serr
;
3326 skb
->wifi_acked_valid
= 1;
3327 skb
->wifi_acked
= acked
;
3329 serr
= SKB_EXT_ERR(skb
);
3330 memset(serr
, 0, sizeof(*serr
));
3331 serr
->ee
.ee_errno
= ENOMSG
;
3332 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3334 err
= sock_queue_err_skb(sk
, skb
);
3338 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3342 * skb_partial_csum_set - set up and verify partial csum values for packet
3343 * @skb: the skb to set
3344 * @start: the number of bytes after skb->data to start checksumming.
3345 * @off: the offset from start to place the checksum.
3347 * For untrusted partially-checksummed packets, we need to make sure the values
3348 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3350 * This function checks and sets those values and skb->ip_summed: if this
3351 * returns false you should drop the packet.
3353 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3355 if (unlikely(start
> skb_headlen(skb
)) ||
3356 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3357 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3358 start
, off
, skb_headlen(skb
));
3361 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3362 skb
->csum_start
= skb_headroom(skb
) + start
;
3363 skb
->csum_offset
= off
;
3366 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3368 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3370 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3373 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3375 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3378 skb_release_head_state(skb
);
3379 kmem_cache_free(skbuff_head_cache
, skb
);
3384 EXPORT_SYMBOL(kfree_skb_partial
);
3387 * skb_try_coalesce - try to merge skb to prior one
3389 * @from: buffer to add
3390 * @fragstolen: pointer to boolean
3391 * @delta_truesize: how much more was allocated than was requested
3393 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3394 bool *fragstolen
, int *delta_truesize
)
3396 int i
, delta
, len
= from
->len
;
3398 *fragstolen
= false;
3403 if (len
<= skb_tailroom(to
)) {
3404 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3405 *delta_truesize
= 0;
3409 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3412 if (skb_headlen(from
) != 0) {
3414 unsigned int offset
;
3416 if (skb_shinfo(to
)->nr_frags
+
3417 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3420 if (skb_head_is_locked(from
))
3423 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3425 page
= virt_to_head_page(from
->head
);
3426 offset
= from
->data
- (unsigned char *)page_address(page
);
3428 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3429 page
, offset
, skb_headlen(from
));
3432 if (skb_shinfo(to
)->nr_frags
+
3433 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3436 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
3439 WARN_ON_ONCE(delta
< len
);
3441 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3442 skb_shinfo(from
)->frags
,
3443 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3444 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3446 if (!skb_cloned(from
))
3447 skb_shinfo(from
)->nr_frags
= 0;
3449 /* if the skb is not cloned this does nothing
3450 * since we set nr_frags to 0.
3452 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3453 skb_frag_ref(from
, i
);
3455 to
->truesize
+= delta
;
3457 to
->data_len
+= len
;
3459 *delta_truesize
= delta
;
3462 EXPORT_SYMBOL(skb_try_coalesce
);