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 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__
, here
, skb
->len
, sz
, skb
->head
, skb
->data
,
125 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
126 skb
->dev
? skb
->dev
->name
: "<NULL>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__
, here
, skb
->len
, sz
, skb
->head
, skb
->data
,
143 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
144 skb
->dev
? skb
->dev
->name
: "<NULL>");
150 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
151 * the caller if emergency pfmemalloc reserves are being used. If it is and
152 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
153 * may be used. Otherwise, the packet data may be discarded until enough
156 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
157 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
158 void *__kmalloc_reserve(size_t size
, gfp_t flags
, int node
, unsigned long ip
,
162 bool ret_pfmemalloc
= false;
165 * Try a regular allocation, when that fails and we're not entitled
166 * to the reserves, fail.
168 obj
= kmalloc_node_track_caller(size
,
169 flags
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
171 if (obj
|| !(gfp_pfmemalloc_allowed(flags
)))
174 /* Try again but now we are using pfmemalloc reserves */
175 ret_pfmemalloc
= true;
176 obj
= kmalloc_node_track_caller(size
, flags
, node
);
180 *pfmemalloc
= ret_pfmemalloc
;
185 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
186 * 'private' fields and also do memory statistics to find all the
192 * __alloc_skb - allocate a network buffer
193 * @size: size to allocate
194 * @gfp_mask: allocation mask
195 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
196 * instead of head cache and allocate a cloned (child) skb.
197 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
198 * allocations in case the data is required for writeback
199 * @node: numa node to allocate memory on
201 * Allocate a new &sk_buff. The returned buffer has no headroom and a
202 * tail room of at least size bytes. The object has a reference count
203 * of one. The return is the buffer. On a failure the return is %NULL.
205 * Buffers may only be allocated from interrupts using a @gfp_mask of
208 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
211 struct kmem_cache
*cache
;
212 struct skb_shared_info
*shinfo
;
217 cache
= (flags
& SKB_ALLOC_FCLONE
)
218 ? skbuff_fclone_cache
: skbuff_head_cache
;
220 if (sk_memalloc_socks() && (flags
& SKB_ALLOC_RX
))
221 gfp_mask
|= __GFP_MEMALLOC
;
224 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
229 /* We do our best to align skb_shared_info on a separate cache
230 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
231 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
232 * Both skb->head and skb_shared_info are cache line aligned.
234 size
= SKB_DATA_ALIGN(size
);
235 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
236 data
= kmalloc_reserve(size
, gfp_mask
, node
, &pfmemalloc
);
239 /* kmalloc(size) might give us more room than requested.
240 * Put skb_shared_info exactly at the end of allocated zone,
241 * to allow max possible filling before reallocation.
243 size
= SKB_WITH_OVERHEAD(ksize(data
));
244 prefetchw(data
+ size
);
247 * Only clear those fields we need to clear, not those that we will
248 * actually initialise below. Hence, don't put any more fields after
249 * the tail pointer in struct sk_buff!
251 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
252 /* Account for allocated memory : skb + skb->head */
253 skb
->truesize
= SKB_TRUESIZE(size
);
254 skb
->pfmemalloc
= pfmemalloc
;
255 atomic_set(&skb
->users
, 1);
258 skb_reset_tail_pointer(skb
);
259 skb
->end
= skb
->tail
+ size
;
260 #ifdef NET_SKBUFF_DATA_USES_OFFSET
261 skb
->mac_header
= ~0U;
264 /* make sure we initialize shinfo sequentially */
265 shinfo
= skb_shinfo(skb
);
266 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
267 atomic_set(&shinfo
->dataref
, 1);
268 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
270 if (flags
& SKB_ALLOC_FCLONE
) {
271 struct sk_buff
*child
= skb
+ 1;
272 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
274 kmemcheck_annotate_bitfield(child
, flags1
);
275 kmemcheck_annotate_bitfield(child
, flags2
);
276 skb
->fclone
= SKB_FCLONE_ORIG
;
277 atomic_set(fclone_ref
, 1);
279 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
280 child
->pfmemalloc
= pfmemalloc
;
285 kmem_cache_free(cache
, skb
);
289 EXPORT_SYMBOL(__alloc_skb
);
292 * build_skb - build a network buffer
293 * @data: data buffer provided by caller
294 * @frag_size: size of fragment, or 0 if head was kmalloced
296 * Allocate a new &sk_buff. Caller provides space holding head and
297 * skb_shared_info. @data must have been allocated by kmalloc()
298 * The return is the new skb buffer.
299 * On a failure the return is %NULL, and @data is not freed.
301 * Before IO, driver allocates only data buffer where NIC put incoming frame
302 * Driver should add room at head (NET_SKB_PAD) and
303 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
304 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
305 * before giving packet to stack.
306 * RX rings only contains data buffers, not full skbs.
308 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
310 struct skb_shared_info
*shinfo
;
312 unsigned int size
= frag_size
? : ksize(data
);
314 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
318 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
320 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
321 skb
->truesize
= SKB_TRUESIZE(size
);
322 skb
->head_frag
= frag_size
!= 0;
323 atomic_set(&skb
->users
, 1);
326 skb_reset_tail_pointer(skb
);
327 skb
->end
= skb
->tail
+ size
;
328 #ifdef NET_SKBUFF_DATA_USES_OFFSET
329 skb
->mac_header
= ~0U;
332 /* make sure we initialize shinfo sequentially */
333 shinfo
= skb_shinfo(skb
);
334 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
335 atomic_set(&shinfo
->dataref
, 1);
336 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
340 EXPORT_SYMBOL(build_skb
);
342 struct netdev_alloc_cache
{
343 struct page_frag frag
;
344 /* we maintain a pagecount bias, so that we dont dirty cache line
345 * containing page->_count every time we allocate a fragment.
347 unsigned int pagecnt_bias
;
349 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
351 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
352 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
353 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
355 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
357 struct netdev_alloc_cache
*nc
;
362 local_irq_save(flags
);
363 nc
= &__get_cpu_var(netdev_alloc_cache
);
364 if (unlikely(!nc
->frag
.page
)) {
366 for (order
= NETDEV_FRAG_PAGE_MAX_ORDER
; ;) {
367 gfp_t gfp
= gfp_mask
;
370 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
371 nc
->frag
.page
= alloc_pages(gfp
, order
);
372 if (likely(nc
->frag
.page
))
377 nc
->frag
.size
= PAGE_SIZE
<< order
;
379 atomic_set(&nc
->frag
.page
->_count
, NETDEV_PAGECNT_MAX_BIAS
);
380 nc
->pagecnt_bias
= NETDEV_PAGECNT_MAX_BIAS
;
384 if (nc
->frag
.offset
+ fragsz
> nc
->frag
.size
) {
385 /* avoid unnecessary locked operations if possible */
386 if ((atomic_read(&nc
->frag
.page
->_count
) == nc
->pagecnt_bias
) ||
387 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->frag
.page
->_count
))
392 data
= page_address(nc
->frag
.page
) + nc
->frag
.offset
;
393 nc
->frag
.offset
+= fragsz
;
396 local_irq_restore(flags
);
401 * netdev_alloc_frag - allocate a page fragment
402 * @fragsz: fragment size
404 * Allocates a frag from a page for receive buffer.
405 * Uses GFP_ATOMIC allocations.
407 void *netdev_alloc_frag(unsigned int fragsz
)
409 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
411 EXPORT_SYMBOL(netdev_alloc_frag
);
414 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
415 * @dev: network device to receive on
416 * @length: length to allocate
417 * @gfp_mask: get_free_pages mask, passed to alloc_skb
419 * Allocate a new &sk_buff and assign it a usage count of one. The
420 * buffer has unspecified headroom built in. Users should allocate
421 * the headroom they think they need without accounting for the
422 * built in space. The built in space is used for optimisations.
424 * %NULL is returned if there is no free memory.
426 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
427 unsigned int length
, gfp_t gfp_mask
)
429 struct sk_buff
*skb
= NULL
;
430 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
431 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
433 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
436 if (sk_memalloc_socks())
437 gfp_mask
|= __GFP_MEMALLOC
;
439 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
442 skb
= build_skb(data
, fragsz
);
444 put_page(virt_to_head_page(data
));
447 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
448 SKB_ALLOC_RX
, NUMA_NO_NODE
);
451 skb_reserve(skb
, NET_SKB_PAD
);
456 EXPORT_SYMBOL(__netdev_alloc_skb
);
458 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
459 int size
, unsigned int truesize
)
461 skb_fill_page_desc(skb
, i
, page
, off
, size
);
463 skb
->data_len
+= size
;
464 skb
->truesize
+= truesize
;
466 EXPORT_SYMBOL(skb_add_rx_frag
);
468 static void skb_drop_list(struct sk_buff
**listp
)
470 struct sk_buff
*list
= *listp
;
475 struct sk_buff
*this = list
;
481 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
483 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
486 static void skb_clone_fraglist(struct sk_buff
*skb
)
488 struct sk_buff
*list
;
490 skb_walk_frags(skb
, list
)
494 static void skb_free_head(struct sk_buff
*skb
)
497 put_page(virt_to_head_page(skb
->head
));
502 static void skb_release_data(struct sk_buff
*skb
)
505 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
506 &skb_shinfo(skb
)->dataref
)) {
507 if (skb_shinfo(skb
)->nr_frags
) {
509 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
510 skb_frag_unref(skb
, i
);
514 * If skb buf is from userspace, we need to notify the caller
515 * the lower device DMA has done;
517 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
518 struct ubuf_info
*uarg
;
520 uarg
= skb_shinfo(skb
)->destructor_arg
;
522 uarg
->callback(uarg
, true);
525 if (skb_has_frag_list(skb
))
526 skb_drop_fraglist(skb
);
533 * Free an skbuff by memory without cleaning the state.
535 static void kfree_skbmem(struct sk_buff
*skb
)
537 struct sk_buff
*other
;
538 atomic_t
*fclone_ref
;
540 switch (skb
->fclone
) {
541 case SKB_FCLONE_UNAVAILABLE
:
542 kmem_cache_free(skbuff_head_cache
, skb
);
545 case SKB_FCLONE_ORIG
:
546 fclone_ref
= (atomic_t
*) (skb
+ 2);
547 if (atomic_dec_and_test(fclone_ref
))
548 kmem_cache_free(skbuff_fclone_cache
, skb
);
551 case SKB_FCLONE_CLONE
:
552 fclone_ref
= (atomic_t
*) (skb
+ 1);
555 /* The clone portion is available for
556 * fast-cloning again.
558 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
560 if (atomic_dec_and_test(fclone_ref
))
561 kmem_cache_free(skbuff_fclone_cache
, other
);
566 static void skb_release_head_state(struct sk_buff
*skb
)
570 secpath_put(skb
->sp
);
572 if (skb
->destructor
) {
574 skb
->destructor(skb
);
576 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
577 nf_conntrack_put(skb
->nfct
);
579 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
580 nf_conntrack_put_reasm(skb
->nfct_reasm
);
582 #ifdef CONFIG_BRIDGE_NETFILTER
583 nf_bridge_put(skb
->nf_bridge
);
585 /* XXX: IS this still necessary? - JHS */
586 #ifdef CONFIG_NET_SCHED
588 #ifdef CONFIG_NET_CLS_ACT
594 /* Free everything but the sk_buff shell. */
595 static void skb_release_all(struct sk_buff
*skb
)
597 skb_release_head_state(skb
);
598 skb_release_data(skb
);
602 * __kfree_skb - private function
605 * Free an sk_buff. Release anything attached to the buffer.
606 * Clean the state. This is an internal helper function. Users should
607 * always call kfree_skb
610 void __kfree_skb(struct sk_buff
*skb
)
612 skb_release_all(skb
);
615 EXPORT_SYMBOL(__kfree_skb
);
618 * kfree_skb - free an sk_buff
619 * @skb: buffer to free
621 * Drop a reference to the buffer and free it if the usage count has
624 void kfree_skb(struct sk_buff
*skb
)
628 if (likely(atomic_read(&skb
->users
) == 1))
630 else if (likely(!atomic_dec_and_test(&skb
->users
)))
632 trace_kfree_skb(skb
, __builtin_return_address(0));
635 EXPORT_SYMBOL(kfree_skb
);
638 * skb_tx_error - report an sk_buff xmit error
639 * @skb: buffer that triggered an error
641 * Report xmit error if a device callback is tracking this skb.
642 * skb must be freed afterwards.
644 void skb_tx_error(struct sk_buff
*skb
)
646 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
647 struct ubuf_info
*uarg
;
649 uarg
= skb_shinfo(skb
)->destructor_arg
;
651 uarg
->callback(uarg
, false);
652 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
655 EXPORT_SYMBOL(skb_tx_error
);
658 * consume_skb - free an skbuff
659 * @skb: buffer to free
661 * Drop a ref to the buffer and free it if the usage count has hit zero
662 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
663 * is being dropped after a failure and notes that
665 void consume_skb(struct sk_buff
*skb
)
669 if (likely(atomic_read(&skb
->users
) == 1))
671 else if (likely(!atomic_dec_and_test(&skb
->users
)))
673 trace_consume_skb(skb
);
676 EXPORT_SYMBOL(consume_skb
);
678 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
680 new->tstamp
= old
->tstamp
;
682 new->transport_header
= old
->transport_header
;
683 new->network_header
= old
->network_header
;
684 new->mac_header
= old
->mac_header
;
685 new->inner_transport_header
= old
->inner_transport_header
;
686 new->inner_network_header
= old
->inner_transport_header
;
687 skb_dst_copy(new, old
);
688 new->rxhash
= old
->rxhash
;
689 new->ooo_okay
= old
->ooo_okay
;
690 new->l4_rxhash
= old
->l4_rxhash
;
691 new->no_fcs
= old
->no_fcs
;
692 new->encapsulation
= old
->encapsulation
;
694 new->sp
= secpath_get(old
->sp
);
696 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
697 new->csum
= old
->csum
;
698 new->local_df
= old
->local_df
;
699 new->pkt_type
= old
->pkt_type
;
700 new->ip_summed
= old
->ip_summed
;
701 skb_copy_queue_mapping(new, old
);
702 new->priority
= old
->priority
;
703 #if IS_ENABLED(CONFIG_IP_VS)
704 new->ipvs_property
= old
->ipvs_property
;
706 new->pfmemalloc
= old
->pfmemalloc
;
707 new->protocol
= old
->protocol
;
708 new->mark
= old
->mark
;
709 new->skb_iif
= old
->skb_iif
;
711 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
712 new->nf_trace
= old
->nf_trace
;
714 #ifdef CONFIG_NET_SCHED
715 new->tc_index
= old
->tc_index
;
716 #ifdef CONFIG_NET_CLS_ACT
717 new->tc_verd
= old
->tc_verd
;
720 new->vlan_tci
= old
->vlan_tci
;
722 skb_copy_secmark(new, old
);
726 * You should not add any new code to this function. Add it to
727 * __copy_skb_header above instead.
729 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
731 #define C(x) n->x = skb->x
733 n
->next
= n
->prev
= NULL
;
735 __copy_skb_header(n
, skb
);
740 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
743 n
->destructor
= NULL
;
750 atomic_set(&n
->users
, 1);
752 atomic_inc(&(skb_shinfo(skb
)->dataref
));
760 * skb_morph - morph one skb into another
761 * @dst: the skb to receive the contents
762 * @src: the skb to supply the contents
764 * This is identical to skb_clone except that the target skb is
765 * supplied by the user.
767 * The target skb is returned upon exit.
769 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
771 skb_release_all(dst
);
772 return __skb_clone(dst
, src
);
774 EXPORT_SYMBOL_GPL(skb_morph
);
777 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
778 * @skb: the skb to modify
779 * @gfp_mask: allocation priority
781 * This must be called on SKBTX_DEV_ZEROCOPY skb.
782 * It will copy all frags into kernel and drop the reference
783 * to userspace pages.
785 * If this function is called from an interrupt gfp_mask() must be
788 * Returns 0 on success or a negative error code on failure
789 * to allocate kernel memory to copy to.
791 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
794 int num_frags
= skb_shinfo(skb
)->nr_frags
;
795 struct page
*page
, *head
= NULL
;
796 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
798 for (i
= 0; i
< num_frags
; i
++) {
800 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
802 page
= alloc_page(gfp_mask
);
805 struct page
*next
= (struct page
*)head
->private;
811 vaddr
= kmap_atomic(skb_frag_page(f
));
812 memcpy(page_address(page
),
813 vaddr
+ f
->page_offset
, skb_frag_size(f
));
814 kunmap_atomic(vaddr
);
815 page
->private = (unsigned long)head
;
819 /* skb frags release userspace buffers */
820 for (i
= 0; i
< num_frags
; i
++)
821 skb_frag_unref(skb
, i
);
823 uarg
->callback(uarg
, false);
825 /* skb frags point to kernel buffers */
826 for (i
= num_frags
- 1; i
>= 0; i
--) {
827 __skb_fill_page_desc(skb
, i
, head
, 0,
828 skb_shinfo(skb
)->frags
[i
].size
);
829 head
= (struct page
*)head
->private;
832 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
835 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
838 * skb_clone - duplicate an sk_buff
839 * @skb: buffer to clone
840 * @gfp_mask: allocation priority
842 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
843 * copies share the same packet data but not structure. The new
844 * buffer has a reference count of 1. If the allocation fails the
845 * function returns %NULL otherwise the new buffer is returned.
847 * If this function is called from an interrupt gfp_mask() must be
851 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
855 if (skb_orphan_frags(skb
, gfp_mask
))
859 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
860 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
861 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
862 n
->fclone
= SKB_FCLONE_CLONE
;
863 atomic_inc(fclone_ref
);
865 if (skb_pfmemalloc(skb
))
866 gfp_mask
|= __GFP_MEMALLOC
;
868 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
872 kmemcheck_annotate_bitfield(n
, flags1
);
873 kmemcheck_annotate_bitfield(n
, flags2
);
874 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
877 return __skb_clone(n
, skb
);
879 EXPORT_SYMBOL(skb_clone
);
881 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
883 #ifndef NET_SKBUFF_DATA_USES_OFFSET
885 * Shift between the two data areas in bytes
887 unsigned long offset
= new->data
- old
->data
;
890 __copy_skb_header(new, old
);
892 #ifndef NET_SKBUFF_DATA_USES_OFFSET
893 /* {transport,network,mac}_header are relative to skb->head */
894 new->transport_header
+= offset
;
895 new->network_header
+= offset
;
896 if (skb_mac_header_was_set(new))
897 new->mac_header
+= offset
;
898 new->inner_transport_header
+= offset
;
899 new->inner_network_header
+= offset
;
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 - 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
959 * Make a copy of both an &sk_buff and part of its data, located
960 * in header. Fragmented data remain shared. This is used when
961 * the caller wishes to modify only header of &sk_buff and needs
962 * private copy of the header to alter. Returns %NULL on failure
963 * or the pointer to the buffer on success.
964 * The returned buffer has a reference count of 1.
967 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
969 unsigned int size
= skb_headlen(skb
) + headroom
;
970 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
971 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
976 /* Set the data pointer */
977 skb_reserve(n
, headroom
);
978 /* Set the tail pointer and length */
979 skb_put(n
, skb_headlen(skb
));
981 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
983 n
->truesize
+= skb
->data_len
;
984 n
->data_len
= skb
->data_len
;
987 if (skb_shinfo(skb
)->nr_frags
) {
990 if (skb_orphan_frags(skb
, gfp_mask
)) {
995 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
996 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
997 skb_frag_ref(skb
, i
);
999 skb_shinfo(n
)->nr_frags
= i
;
1002 if (skb_has_frag_list(skb
)) {
1003 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1004 skb_clone_fraglist(n
);
1007 copy_skb_header(n
, skb
);
1011 EXPORT_SYMBOL(__pskb_copy
);
1014 * pskb_expand_head - reallocate header of &sk_buff
1015 * @skb: buffer to reallocate
1016 * @nhead: room to add at head
1017 * @ntail: room to add at tail
1018 * @gfp_mask: allocation priority
1020 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1021 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1022 * reference count of 1. Returns zero in the case of success or error,
1023 * if expansion failed. In the last case, &sk_buff is not changed.
1025 * All the pointers pointing into skb header may change and must be
1026 * reloaded after call to this function.
1029 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1034 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1039 if (skb_shared(skb
))
1042 size
= SKB_DATA_ALIGN(size
);
1044 if (skb_pfmemalloc(skb
))
1045 gfp_mask
|= __GFP_MEMALLOC
;
1046 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1047 gfp_mask
, NUMA_NO_NODE
, NULL
);
1050 size
= SKB_WITH_OVERHEAD(ksize(data
));
1052 /* Copy only real data... and, alas, header. This should be
1053 * optimized for the cases when header is void.
1055 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1057 memcpy((struct skb_shared_info
*)(data
+ size
),
1059 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1062 * if shinfo is shared we must drop the old head gracefully, but if it
1063 * is not we can just drop the old head and let the existing refcount
1064 * be since all we did is relocate the values
1066 if (skb_cloned(skb
)) {
1067 /* copy this zero copy skb frags */
1068 if (skb_orphan_frags(skb
, gfp_mask
))
1070 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1071 skb_frag_ref(skb
, i
);
1073 if (skb_has_frag_list(skb
))
1074 skb_clone_fraglist(skb
);
1076 skb_release_data(skb
);
1080 off
= (data
+ nhead
) - skb
->head
;
1085 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1089 skb
->end
= skb
->head
+ size
;
1091 /* {transport,network,mac}_header and tail are relative to skb->head */
1093 skb
->transport_header
+= off
;
1094 skb
->network_header
+= off
;
1095 if (skb_mac_header_was_set(skb
))
1096 skb
->mac_header
+= off
;
1097 skb
->inner_transport_header
+= off
;
1098 skb
->inner_network_header
+= off
;
1099 /* Only adjust this if it actually is csum_start rather than csum */
1100 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1101 skb
->csum_start
+= nhead
;
1105 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1113 EXPORT_SYMBOL(pskb_expand_head
);
1115 /* Make private copy of skb with writable head and some headroom */
1117 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1119 struct sk_buff
*skb2
;
1120 int delta
= headroom
- skb_headroom(skb
);
1123 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1125 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1126 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1134 EXPORT_SYMBOL(skb_realloc_headroom
);
1137 * skb_copy_expand - copy and expand sk_buff
1138 * @skb: buffer to copy
1139 * @newheadroom: new free bytes at head
1140 * @newtailroom: new free bytes at tail
1141 * @gfp_mask: allocation priority
1143 * Make a copy of both an &sk_buff and its data and while doing so
1144 * allocate additional space.
1146 * This is used when the caller wishes to modify the data and needs a
1147 * private copy of the data to alter as well as more space for new fields.
1148 * Returns %NULL on failure or the pointer to the buffer
1149 * on success. The returned buffer has a reference count of 1.
1151 * You must pass %GFP_ATOMIC as the allocation priority if this function
1152 * is called from an interrupt.
1154 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1155 int newheadroom
, int newtailroom
,
1159 * Allocate the copy buffer
1161 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1162 gfp_mask
, skb_alloc_rx_flag(skb
),
1164 int oldheadroom
= skb_headroom(skb
);
1165 int head_copy_len
, head_copy_off
;
1171 skb_reserve(n
, newheadroom
);
1173 /* Set the tail pointer and length */
1174 skb_put(n
, skb
->len
);
1176 head_copy_len
= oldheadroom
;
1178 if (newheadroom
<= head_copy_len
)
1179 head_copy_len
= newheadroom
;
1181 head_copy_off
= newheadroom
- head_copy_len
;
1183 /* Copy the linear header and data. */
1184 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1185 skb
->len
+ head_copy_len
))
1188 copy_skb_header(n
, skb
);
1190 off
= newheadroom
- oldheadroom
;
1191 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1192 n
->csum_start
+= off
;
1193 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1194 n
->transport_header
+= off
;
1195 n
->network_header
+= off
;
1196 if (skb_mac_header_was_set(skb
))
1197 n
->mac_header
+= off
;
1198 n
->inner_transport_header
+= off
;
1199 n
->inner_network_header
+= off
;
1204 EXPORT_SYMBOL(skb_copy_expand
);
1207 * skb_pad - zero pad the tail of an skb
1208 * @skb: buffer to pad
1209 * @pad: space to pad
1211 * Ensure that a buffer is followed by a padding area that is zero
1212 * filled. Used by network drivers which may DMA or transfer data
1213 * beyond the buffer end onto the wire.
1215 * May return error in out of memory cases. The skb is freed on error.
1218 int skb_pad(struct sk_buff
*skb
, int pad
)
1223 /* If the skbuff is non linear tailroom is always zero.. */
1224 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1225 memset(skb
->data
+skb
->len
, 0, pad
);
1229 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1230 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1231 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1236 /* FIXME: The use of this function with non-linear skb's really needs
1239 err
= skb_linearize(skb
);
1243 memset(skb
->data
+ skb
->len
, 0, pad
);
1250 EXPORT_SYMBOL(skb_pad
);
1253 * skb_put - add data to a buffer
1254 * @skb: buffer to use
1255 * @len: amount of data to add
1257 * This function extends the used data area of the buffer. If this would
1258 * exceed the total buffer size the kernel will panic. A pointer to the
1259 * first byte of the extra data is returned.
1261 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1263 unsigned char *tmp
= skb_tail_pointer(skb
);
1264 SKB_LINEAR_ASSERT(skb
);
1267 if (unlikely(skb
->tail
> skb
->end
))
1268 skb_over_panic(skb
, len
, __builtin_return_address(0));
1271 EXPORT_SYMBOL(skb_put
);
1274 * skb_push - add data to the start of a buffer
1275 * @skb: buffer to use
1276 * @len: amount of data to add
1278 * This function extends the used data area of the buffer at the buffer
1279 * start. If this would exceed the total buffer headroom the kernel will
1280 * panic. A pointer to the first byte of the extra data is returned.
1282 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1286 if (unlikely(skb
->data
<skb
->head
))
1287 skb_under_panic(skb
, len
, __builtin_return_address(0));
1290 EXPORT_SYMBOL(skb_push
);
1293 * skb_pull - remove data from the start of a buffer
1294 * @skb: buffer to use
1295 * @len: amount of data to remove
1297 * This function removes data from the start of a buffer, returning
1298 * the memory to the headroom. A pointer to the next data in the buffer
1299 * is returned. Once the data has been pulled future pushes will overwrite
1302 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1304 return skb_pull_inline(skb
, len
);
1306 EXPORT_SYMBOL(skb_pull
);
1309 * skb_trim - remove end from a buffer
1310 * @skb: buffer to alter
1313 * Cut the length of a buffer down by removing data from the tail. If
1314 * the buffer is already under the length specified it is not modified.
1315 * The skb must be linear.
1317 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1320 __skb_trim(skb
, len
);
1322 EXPORT_SYMBOL(skb_trim
);
1324 /* Trims skb to length len. It can change skb pointers.
1327 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1329 struct sk_buff
**fragp
;
1330 struct sk_buff
*frag
;
1331 int offset
= skb_headlen(skb
);
1332 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1336 if (skb_cloned(skb
) &&
1337 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1344 for (; i
< nfrags
; i
++) {
1345 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1352 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1355 skb_shinfo(skb
)->nr_frags
= i
;
1357 for (; i
< nfrags
; i
++)
1358 skb_frag_unref(skb
, i
);
1360 if (skb_has_frag_list(skb
))
1361 skb_drop_fraglist(skb
);
1365 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1366 fragp
= &frag
->next
) {
1367 int end
= offset
+ frag
->len
;
1369 if (skb_shared(frag
)) {
1370 struct sk_buff
*nfrag
;
1372 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1373 if (unlikely(!nfrag
))
1376 nfrag
->next
= frag
->next
;
1388 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1392 skb_drop_list(&frag
->next
);
1397 if (len
> skb_headlen(skb
)) {
1398 skb
->data_len
-= skb
->len
- len
;
1403 skb_set_tail_pointer(skb
, len
);
1408 EXPORT_SYMBOL(___pskb_trim
);
1411 * __pskb_pull_tail - advance tail of skb header
1412 * @skb: buffer to reallocate
1413 * @delta: number of bytes to advance tail
1415 * The function makes a sense only on a fragmented &sk_buff,
1416 * it expands header moving its tail forward and copying necessary
1417 * data from fragmented part.
1419 * &sk_buff MUST have reference count of 1.
1421 * Returns %NULL (and &sk_buff does not change) if pull failed
1422 * or value of new tail of skb in the case of success.
1424 * All the pointers pointing into skb header may change and must be
1425 * reloaded after call to this function.
1428 /* Moves tail of skb head forward, copying data from fragmented part,
1429 * when it is necessary.
1430 * 1. It may fail due to malloc failure.
1431 * 2. It may change skb pointers.
1433 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1435 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1437 /* If skb has not enough free space at tail, get new one
1438 * plus 128 bytes for future expansions. If we have enough
1439 * room at tail, reallocate without expansion only if skb is cloned.
1441 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1443 if (eat
> 0 || skb_cloned(skb
)) {
1444 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1449 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1452 /* Optimization: no fragments, no reasons to preestimate
1453 * size of pulled pages. Superb.
1455 if (!skb_has_frag_list(skb
))
1458 /* Estimate size of pulled pages. */
1460 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1461 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1468 /* If we need update frag list, we are in troubles.
1469 * Certainly, it possible to add an offset to skb data,
1470 * but taking into account that pulling is expected to
1471 * be very rare operation, it is worth to fight against
1472 * further bloating skb head and crucify ourselves here instead.
1473 * Pure masohism, indeed. 8)8)
1476 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1477 struct sk_buff
*clone
= NULL
;
1478 struct sk_buff
*insp
= NULL
;
1483 if (list
->len
<= eat
) {
1484 /* Eaten as whole. */
1489 /* Eaten partially. */
1491 if (skb_shared(list
)) {
1492 /* Sucks! We need to fork list. :-( */
1493 clone
= skb_clone(list
, GFP_ATOMIC
);
1499 /* This may be pulled without
1503 if (!pskb_pull(list
, eat
)) {
1511 /* Free pulled out fragments. */
1512 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1513 skb_shinfo(skb
)->frag_list
= list
->next
;
1516 /* And insert new clone at head. */
1519 skb_shinfo(skb
)->frag_list
= clone
;
1522 /* Success! Now we may commit changes to skb data. */
1527 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1528 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1531 skb_frag_unref(skb
, i
);
1534 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1536 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1537 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1543 skb_shinfo(skb
)->nr_frags
= k
;
1546 skb
->data_len
-= delta
;
1548 return skb_tail_pointer(skb
);
1550 EXPORT_SYMBOL(__pskb_pull_tail
);
1553 * skb_copy_bits - copy bits from skb to kernel buffer
1555 * @offset: offset in source
1556 * @to: destination buffer
1557 * @len: number of bytes to copy
1559 * Copy the specified number of bytes from the source skb to the
1560 * destination buffer.
1563 * If its prototype is ever changed,
1564 * check arch/{*}/net/{*}.S files,
1565 * since it is called from BPF assembly code.
1567 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1569 int start
= skb_headlen(skb
);
1570 struct sk_buff
*frag_iter
;
1573 if (offset
> (int)skb
->len
- len
)
1577 if ((copy
= start
- offset
) > 0) {
1580 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1581 if ((len
-= copy
) == 0)
1587 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1589 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1591 WARN_ON(start
> offset
+ len
);
1593 end
= start
+ skb_frag_size(f
);
1594 if ((copy
= end
- offset
) > 0) {
1600 vaddr
= kmap_atomic(skb_frag_page(f
));
1602 vaddr
+ f
->page_offset
+ offset
- start
,
1604 kunmap_atomic(vaddr
);
1606 if ((len
-= copy
) == 0)
1614 skb_walk_frags(skb
, frag_iter
) {
1617 WARN_ON(start
> offset
+ len
);
1619 end
= start
+ frag_iter
->len
;
1620 if ((copy
= end
- offset
) > 0) {
1623 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1625 if ((len
-= copy
) == 0)
1639 EXPORT_SYMBOL(skb_copy_bits
);
1642 * Callback from splice_to_pipe(), if we need to release some pages
1643 * at the end of the spd in case we error'ed out in filling the pipe.
1645 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1647 put_page(spd
->pages
[i
]);
1650 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1651 unsigned int *offset
,
1652 struct sk_buff
*skb
, struct sock
*sk
)
1654 struct page_frag
*pfrag
= sk_page_frag(sk
);
1656 if (!sk_page_frag_refill(sk
, pfrag
))
1659 *len
= min_t(unsigned int, *len
, pfrag
->size
- pfrag
->offset
);
1661 memcpy(page_address(pfrag
->page
) + pfrag
->offset
,
1662 page_address(page
) + *offset
, *len
);
1663 *offset
= pfrag
->offset
;
1664 pfrag
->offset
+= *len
;
1669 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1671 unsigned int offset
)
1673 return spd
->nr_pages
&&
1674 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1675 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1676 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1680 * Fill page/offset/length into spd, if it can hold more pages.
1682 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1683 struct pipe_inode_info
*pipe
, struct page
*page
,
1684 unsigned int *len
, unsigned int offset
,
1685 struct sk_buff
*skb
, bool linear
,
1688 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1692 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1696 if (spd_can_coalesce(spd
, page
, offset
)) {
1697 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1701 spd
->pages
[spd
->nr_pages
] = page
;
1702 spd
->partial
[spd
->nr_pages
].len
= *len
;
1703 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1709 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1710 unsigned int *plen
, unsigned int off
)
1715 n
= *poff
/ PAGE_SIZE
;
1717 *page
= nth_page(*page
, n
);
1719 *poff
= *poff
% PAGE_SIZE
;
1723 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1724 unsigned int plen
, unsigned int *off
,
1725 unsigned int *len
, struct sk_buff
*skb
,
1726 struct splice_pipe_desc
*spd
, bool linear
,
1728 struct pipe_inode_info
*pipe
)
1733 /* skip this segment if already processed */
1739 /* ignore any bits we already processed */
1741 __segment_seek(&page
, &poff
, &plen
, *off
);
1746 unsigned int flen
= min(*len
, plen
);
1748 /* the linear region may spread across several pages */
1749 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1751 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1754 __segment_seek(&page
, &poff
, &plen
, flen
);
1757 } while (*len
&& plen
);
1763 * Map linear and fragment data from the skb to spd. It reports true if the
1764 * pipe is full or if we already spliced the requested length.
1766 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1767 unsigned int *offset
, unsigned int *len
,
1768 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1772 /* map the linear part :
1773 * If skb->head_frag is set, this 'linear' part is backed by a
1774 * fragment, and if the head is not shared with any clones then
1775 * we can avoid a copy since we own the head portion of this page.
1777 if (__splice_segment(virt_to_page(skb
->data
),
1778 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1780 offset
, len
, skb
, spd
,
1781 skb_head_is_locked(skb
),
1786 * then map the fragments
1788 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1789 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1791 if (__splice_segment(skb_frag_page(f
),
1792 f
->page_offset
, skb_frag_size(f
),
1793 offset
, len
, skb
, spd
, false, sk
, pipe
))
1801 * Map data from the skb to a pipe. Should handle both the linear part,
1802 * the fragments, and the frag list. It does NOT handle frag lists within
1803 * the frag list, if such a thing exists. We'd probably need to recurse to
1804 * handle that cleanly.
1806 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1807 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1810 struct partial_page partial
[MAX_SKB_FRAGS
];
1811 struct page
*pages
[MAX_SKB_FRAGS
];
1812 struct splice_pipe_desc spd
= {
1815 .nr_pages_max
= MAX_SKB_FRAGS
,
1817 .ops
= &sock_pipe_buf_ops
,
1818 .spd_release
= sock_spd_release
,
1820 struct sk_buff
*frag_iter
;
1821 struct sock
*sk
= skb
->sk
;
1825 * __skb_splice_bits() only fails if the output has no room left,
1826 * so no point in going over the frag_list for the error case.
1828 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1834 * now see if we have a frag_list to map
1836 skb_walk_frags(skb
, frag_iter
) {
1839 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1846 * Drop the socket lock, otherwise we have reverse
1847 * locking dependencies between sk_lock and i_mutex
1848 * here as compared to sendfile(). We enter here
1849 * with the socket lock held, and splice_to_pipe() will
1850 * grab the pipe inode lock. For sendfile() emulation,
1851 * we call into ->sendpage() with the i_mutex lock held
1852 * and networking will grab the socket lock.
1855 ret
= splice_to_pipe(pipe
, &spd
);
1863 * skb_store_bits - store bits from kernel buffer to skb
1864 * @skb: destination buffer
1865 * @offset: offset in destination
1866 * @from: source buffer
1867 * @len: number of bytes to copy
1869 * Copy the specified number of bytes from the source buffer to the
1870 * destination skb. This function handles all the messy bits of
1871 * traversing fragment lists and such.
1874 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1876 int start
= skb_headlen(skb
);
1877 struct sk_buff
*frag_iter
;
1880 if (offset
> (int)skb
->len
- len
)
1883 if ((copy
= start
- offset
) > 0) {
1886 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1887 if ((len
-= copy
) == 0)
1893 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1894 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1897 WARN_ON(start
> offset
+ len
);
1899 end
= start
+ skb_frag_size(frag
);
1900 if ((copy
= end
- offset
) > 0) {
1906 vaddr
= kmap_atomic(skb_frag_page(frag
));
1907 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1909 kunmap_atomic(vaddr
);
1911 if ((len
-= copy
) == 0)
1919 skb_walk_frags(skb
, frag_iter
) {
1922 WARN_ON(start
> offset
+ len
);
1924 end
= start
+ frag_iter
->len
;
1925 if ((copy
= end
- offset
) > 0) {
1928 if (skb_store_bits(frag_iter
, offset
- start
,
1931 if ((len
-= copy
) == 0)
1944 EXPORT_SYMBOL(skb_store_bits
);
1946 /* Checksum skb data. */
1948 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1949 int len
, __wsum csum
)
1951 int start
= skb_headlen(skb
);
1952 int i
, copy
= start
- offset
;
1953 struct sk_buff
*frag_iter
;
1956 /* Checksum header. */
1960 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1961 if ((len
-= copy
) == 0)
1967 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1969 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1971 WARN_ON(start
> offset
+ len
);
1973 end
= start
+ skb_frag_size(frag
);
1974 if ((copy
= end
- offset
) > 0) {
1980 vaddr
= kmap_atomic(skb_frag_page(frag
));
1981 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1982 offset
- start
, copy
, 0);
1983 kunmap_atomic(vaddr
);
1984 csum
= csum_block_add(csum
, csum2
, pos
);
1993 skb_walk_frags(skb
, frag_iter
) {
1996 WARN_ON(start
> offset
+ len
);
1998 end
= start
+ frag_iter
->len
;
1999 if ((copy
= end
- offset
) > 0) {
2003 csum2
= skb_checksum(frag_iter
, offset
- start
,
2005 csum
= csum_block_add(csum
, csum2
, pos
);
2006 if ((len
-= copy
) == 0)
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
);
2098 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2103 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2104 csstart
= skb_checksum_start_offset(skb
);
2106 csstart
= skb_headlen(skb
);
2108 BUG_ON(csstart
> skb_headlen(skb
));
2110 skb_copy_from_linear_data(skb
, to
, csstart
);
2113 if (csstart
!= skb
->len
)
2114 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2115 skb
->len
- csstart
, 0);
2117 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2118 long csstuff
= csstart
+ skb
->csum_offset
;
2120 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2123 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2126 * skb_dequeue - remove from the head of the queue
2127 * @list: list to dequeue from
2129 * Remove the head of the list. The list lock is taken so the function
2130 * may be used safely with other locking list functions. The head item is
2131 * returned or %NULL if the list is empty.
2134 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2136 unsigned long flags
;
2137 struct sk_buff
*result
;
2139 spin_lock_irqsave(&list
->lock
, flags
);
2140 result
= __skb_dequeue(list
);
2141 spin_unlock_irqrestore(&list
->lock
, flags
);
2144 EXPORT_SYMBOL(skb_dequeue
);
2147 * skb_dequeue_tail - remove from the tail of the queue
2148 * @list: list to dequeue from
2150 * Remove the tail of the list. The list lock is taken so the function
2151 * may be used safely with other locking list functions. The tail item is
2152 * returned or %NULL if the list is empty.
2154 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2156 unsigned long flags
;
2157 struct sk_buff
*result
;
2159 spin_lock_irqsave(&list
->lock
, flags
);
2160 result
= __skb_dequeue_tail(list
);
2161 spin_unlock_irqrestore(&list
->lock
, flags
);
2164 EXPORT_SYMBOL(skb_dequeue_tail
);
2167 * skb_queue_purge - empty a list
2168 * @list: list to empty
2170 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2171 * the list and one reference dropped. This function takes the list
2172 * lock and is atomic with respect to other list locking functions.
2174 void skb_queue_purge(struct sk_buff_head
*list
)
2176 struct sk_buff
*skb
;
2177 while ((skb
= skb_dequeue(list
)) != NULL
)
2180 EXPORT_SYMBOL(skb_queue_purge
);
2183 * skb_queue_head - queue a buffer at the list head
2184 * @list: list to use
2185 * @newsk: buffer to queue
2187 * Queue a buffer at the start of the list. This function takes the
2188 * list lock and can be used safely with other locking &sk_buff functions
2191 * A buffer cannot be placed on two lists at the same time.
2193 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2195 unsigned long flags
;
2197 spin_lock_irqsave(&list
->lock
, flags
);
2198 __skb_queue_head(list
, newsk
);
2199 spin_unlock_irqrestore(&list
->lock
, flags
);
2201 EXPORT_SYMBOL(skb_queue_head
);
2204 * skb_queue_tail - queue a buffer at the list tail
2205 * @list: list to use
2206 * @newsk: buffer to queue
2208 * Queue a buffer at the tail of the list. This function takes the
2209 * list lock and can be used safely with other locking &sk_buff functions
2212 * A buffer cannot be placed on two lists at the same time.
2214 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2216 unsigned long flags
;
2218 spin_lock_irqsave(&list
->lock
, flags
);
2219 __skb_queue_tail(list
, newsk
);
2220 spin_unlock_irqrestore(&list
->lock
, flags
);
2222 EXPORT_SYMBOL(skb_queue_tail
);
2225 * skb_unlink - remove a buffer from a list
2226 * @skb: buffer to remove
2227 * @list: list to use
2229 * Remove a packet from a list. The list locks are taken and this
2230 * function is atomic with respect to other list locked calls
2232 * You must know what list the SKB is on.
2234 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2236 unsigned long flags
;
2238 spin_lock_irqsave(&list
->lock
, flags
);
2239 __skb_unlink(skb
, list
);
2240 spin_unlock_irqrestore(&list
->lock
, flags
);
2242 EXPORT_SYMBOL(skb_unlink
);
2245 * skb_append - append a buffer
2246 * @old: buffer to insert after
2247 * @newsk: buffer to insert
2248 * @list: list to use
2250 * Place a packet after a given packet in a list. The list locks are taken
2251 * and this function is atomic with respect to other list locked calls.
2252 * A buffer cannot be placed on two lists at the same time.
2254 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2256 unsigned long flags
;
2258 spin_lock_irqsave(&list
->lock
, flags
);
2259 __skb_queue_after(list
, old
, newsk
);
2260 spin_unlock_irqrestore(&list
->lock
, flags
);
2262 EXPORT_SYMBOL(skb_append
);
2265 * skb_insert - insert a buffer
2266 * @old: buffer to insert before
2267 * @newsk: buffer to insert
2268 * @list: list to use
2270 * Place a packet before a given packet in a list. The list locks are
2271 * taken and this function is atomic with respect to other list locked
2274 * A buffer cannot be placed on two lists at the same time.
2276 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2278 unsigned long flags
;
2280 spin_lock_irqsave(&list
->lock
, flags
);
2281 __skb_insert(newsk
, old
->prev
, old
, list
);
2282 spin_unlock_irqrestore(&list
->lock
, flags
);
2284 EXPORT_SYMBOL(skb_insert
);
2286 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2287 struct sk_buff
* skb1
,
2288 const u32 len
, const int pos
)
2292 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2294 /* And move data appendix as is. */
2295 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2296 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2298 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2299 skb_shinfo(skb
)->nr_frags
= 0;
2300 skb1
->data_len
= skb
->data_len
;
2301 skb1
->len
+= skb1
->data_len
;
2304 skb_set_tail_pointer(skb
, len
);
2307 static inline void skb_split_no_header(struct sk_buff
*skb
,
2308 struct sk_buff
* skb1
,
2309 const u32 len
, int pos
)
2312 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2314 skb_shinfo(skb
)->nr_frags
= 0;
2315 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2317 skb
->data_len
= len
- pos
;
2319 for (i
= 0; i
< nfrags
; i
++) {
2320 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2322 if (pos
+ size
> len
) {
2323 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2327 * We have two variants in this case:
2328 * 1. Move all the frag to the second
2329 * part, if it is possible. F.e.
2330 * this approach is mandatory for TUX,
2331 * where splitting is expensive.
2332 * 2. Split is accurately. We make this.
2334 skb_frag_ref(skb
, i
);
2335 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2336 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2337 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2338 skb_shinfo(skb
)->nr_frags
++;
2342 skb_shinfo(skb
)->nr_frags
++;
2345 skb_shinfo(skb1
)->nr_frags
= k
;
2349 * skb_split - Split fragmented skb to two parts at length len.
2350 * @skb: the buffer to split
2351 * @skb1: the buffer to receive the second part
2352 * @len: new length for skb
2354 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2356 int pos
= skb_headlen(skb
);
2358 if (len
< pos
) /* Split line is inside header. */
2359 skb_split_inside_header(skb
, skb1
, len
, pos
);
2360 else /* Second chunk has no header, nothing to copy. */
2361 skb_split_no_header(skb
, skb1
, len
, pos
);
2363 EXPORT_SYMBOL(skb_split
);
2365 /* Shifting from/to a cloned skb is a no-go.
2367 * Caller cannot keep skb_shinfo related pointers past calling here!
2369 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2371 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2375 * skb_shift - Shifts paged data partially from skb to another
2376 * @tgt: buffer into which tail data gets added
2377 * @skb: buffer from which the paged data comes from
2378 * @shiftlen: shift up to this many bytes
2380 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2381 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2382 * It's up to caller to free skb if everything was shifted.
2384 * If @tgt runs out of frags, the whole operation is aborted.
2386 * Skb cannot include anything else but paged data while tgt is allowed
2387 * to have non-paged data as well.
2389 * TODO: full sized shift could be optimized but that would need
2390 * specialized skb free'er to handle frags without up-to-date nr_frags.
2392 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2394 int from
, to
, merge
, todo
;
2395 struct skb_frag_struct
*fragfrom
, *fragto
;
2397 BUG_ON(shiftlen
> skb
->len
);
2398 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2402 to
= skb_shinfo(tgt
)->nr_frags
;
2403 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2405 /* Actual merge is delayed until the point when we know we can
2406 * commit all, so that we don't have to undo partial changes
2409 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2410 fragfrom
->page_offset
)) {
2415 todo
-= skb_frag_size(fragfrom
);
2417 if (skb_prepare_for_shift(skb
) ||
2418 skb_prepare_for_shift(tgt
))
2421 /* All previous frag pointers might be stale! */
2422 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2423 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2425 skb_frag_size_add(fragto
, shiftlen
);
2426 skb_frag_size_sub(fragfrom
, shiftlen
);
2427 fragfrom
->page_offset
+= shiftlen
;
2435 /* Skip full, not-fitting skb to avoid expensive operations */
2436 if ((shiftlen
== skb
->len
) &&
2437 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2440 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2443 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2444 if (to
== MAX_SKB_FRAGS
)
2447 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2448 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2450 if (todo
>= skb_frag_size(fragfrom
)) {
2451 *fragto
= *fragfrom
;
2452 todo
-= skb_frag_size(fragfrom
);
2457 __skb_frag_ref(fragfrom
);
2458 fragto
->page
= fragfrom
->page
;
2459 fragto
->page_offset
= fragfrom
->page_offset
;
2460 skb_frag_size_set(fragto
, todo
);
2462 fragfrom
->page_offset
+= todo
;
2463 skb_frag_size_sub(fragfrom
, todo
);
2471 /* Ready to "commit" this state change to tgt */
2472 skb_shinfo(tgt
)->nr_frags
= to
;
2475 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2476 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2478 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2479 __skb_frag_unref(fragfrom
);
2482 /* Reposition in the original skb */
2484 while (from
< skb_shinfo(skb
)->nr_frags
)
2485 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2486 skb_shinfo(skb
)->nr_frags
= to
;
2488 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2491 /* Most likely the tgt won't ever need its checksum anymore, skb on
2492 * the other hand might need it if it needs to be resent
2494 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2495 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2497 /* Yak, is it really working this way? Some helper please? */
2498 skb
->len
-= shiftlen
;
2499 skb
->data_len
-= shiftlen
;
2500 skb
->truesize
-= shiftlen
;
2501 tgt
->len
+= shiftlen
;
2502 tgt
->data_len
+= shiftlen
;
2503 tgt
->truesize
+= shiftlen
;
2509 * skb_prepare_seq_read - Prepare a sequential read of skb data
2510 * @skb: the buffer to read
2511 * @from: lower offset of data to be read
2512 * @to: upper offset of data to be read
2513 * @st: state variable
2515 * Initializes the specified state variable. Must be called before
2516 * invoking skb_seq_read() for the first time.
2518 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2519 unsigned int to
, struct skb_seq_state
*st
)
2521 st
->lower_offset
= from
;
2522 st
->upper_offset
= to
;
2523 st
->root_skb
= st
->cur_skb
= skb
;
2524 st
->frag_idx
= st
->stepped_offset
= 0;
2525 st
->frag_data
= NULL
;
2527 EXPORT_SYMBOL(skb_prepare_seq_read
);
2530 * skb_seq_read - Sequentially read skb data
2531 * @consumed: number of bytes consumed by the caller so far
2532 * @data: destination pointer for data to be returned
2533 * @st: state variable
2535 * Reads a block of skb data at &consumed relative to the
2536 * lower offset specified to skb_prepare_seq_read(). Assigns
2537 * the head of the data block to &data and returns the length
2538 * of the block or 0 if the end of the skb data or the upper
2539 * offset has been reached.
2541 * The caller is not required to consume all of the data
2542 * returned, i.e. &consumed is typically set to the number
2543 * of bytes already consumed and the next call to
2544 * skb_seq_read() will return the remaining part of the block.
2546 * Note 1: The size of each block of data returned can be arbitrary,
2547 * this limitation is the cost for zerocopy seqeuental
2548 * reads of potentially non linear data.
2550 * Note 2: Fragment lists within fragments are not implemented
2551 * at the moment, state->root_skb could be replaced with
2552 * a stack for this purpose.
2554 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2555 struct skb_seq_state
*st
)
2557 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2560 if (unlikely(abs_offset
>= st
->upper_offset
))
2564 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2566 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2567 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2568 return block_limit
- abs_offset
;
2571 if (st
->frag_idx
== 0 && !st
->frag_data
)
2572 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2574 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2575 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2576 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2578 if (abs_offset
< block_limit
) {
2580 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2582 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2583 (abs_offset
- st
->stepped_offset
);
2585 return block_limit
- abs_offset
;
2588 if (st
->frag_data
) {
2589 kunmap_atomic(st
->frag_data
);
2590 st
->frag_data
= NULL
;
2594 st
->stepped_offset
+= skb_frag_size(frag
);
2597 if (st
->frag_data
) {
2598 kunmap_atomic(st
->frag_data
);
2599 st
->frag_data
= NULL
;
2602 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2603 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2606 } else if (st
->cur_skb
->next
) {
2607 st
->cur_skb
= st
->cur_skb
->next
;
2614 EXPORT_SYMBOL(skb_seq_read
);
2617 * skb_abort_seq_read - Abort a sequential read of skb data
2618 * @st: state variable
2620 * Must be called if skb_seq_read() was not called until it
2623 void skb_abort_seq_read(struct skb_seq_state
*st
)
2626 kunmap_atomic(st
->frag_data
);
2628 EXPORT_SYMBOL(skb_abort_seq_read
);
2630 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2632 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2633 struct ts_config
*conf
,
2634 struct ts_state
*state
)
2636 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2639 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2641 skb_abort_seq_read(TS_SKB_CB(state
));
2645 * skb_find_text - Find a text pattern in skb data
2646 * @skb: the buffer to look in
2647 * @from: search offset
2649 * @config: textsearch configuration
2650 * @state: uninitialized textsearch state variable
2652 * Finds a pattern in the skb data according to the specified
2653 * textsearch configuration. Use textsearch_next() to retrieve
2654 * subsequent occurrences of the pattern. Returns the offset
2655 * to the first occurrence or UINT_MAX if no match was found.
2657 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2658 unsigned int to
, struct ts_config
*config
,
2659 struct ts_state
*state
)
2663 config
->get_next_block
= skb_ts_get_next_block
;
2664 config
->finish
= skb_ts_finish
;
2666 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2668 ret
= textsearch_find(config
, state
);
2669 return (ret
<= to
- from
? ret
: UINT_MAX
);
2671 EXPORT_SYMBOL(skb_find_text
);
2674 * skb_append_datato_frags - append the user data to a skb
2675 * @sk: sock structure
2676 * @skb: skb structure to be appened with user data.
2677 * @getfrag: call back function to be used for getting the user data
2678 * @from: pointer to user message iov
2679 * @length: length of the iov message
2681 * Description: This procedure append the user data in the fragment part
2682 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2684 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2685 int (*getfrag
)(void *from
, char *to
, int offset
,
2686 int len
, int odd
, struct sk_buff
*skb
),
2687 void *from
, int length
)
2690 skb_frag_t
*frag
= NULL
;
2691 struct page
*page
= NULL
;
2697 /* Return error if we don't have space for new frag */
2698 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2699 if (frg_cnt
>= MAX_SKB_FRAGS
)
2702 /* allocate a new page for next frag */
2703 page
= alloc_pages(sk
->sk_allocation
, 0);
2705 /* If alloc_page fails just return failure and caller will
2706 * free previous allocated pages by doing kfree_skb()
2711 /* initialize the next frag */
2712 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2713 skb
->truesize
+= PAGE_SIZE
;
2714 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2716 /* get the new initialized frag */
2717 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2718 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2720 /* copy the user data to page */
2721 left
= PAGE_SIZE
- frag
->page_offset
;
2722 copy
= (length
> left
)? left
: length
;
2724 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2725 offset
, copy
, 0, skb
);
2729 /* copy was successful so update the size parameters */
2730 skb_frag_size_add(frag
, copy
);
2732 skb
->data_len
+= copy
;
2736 } while (length
> 0);
2740 EXPORT_SYMBOL(skb_append_datato_frags
);
2743 * skb_pull_rcsum - pull skb and update receive checksum
2744 * @skb: buffer to update
2745 * @len: length of data pulled
2747 * This function performs an skb_pull on the packet and updates
2748 * the CHECKSUM_COMPLETE checksum. It should be used on
2749 * receive path processing instead of skb_pull unless you know
2750 * that the checksum difference is zero (e.g., a valid IP header)
2751 * or you are setting ip_summed to CHECKSUM_NONE.
2753 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2755 BUG_ON(len
> skb
->len
);
2757 BUG_ON(skb
->len
< skb
->data_len
);
2758 skb_postpull_rcsum(skb
, skb
->data
, len
);
2759 return skb
->data
+= len
;
2761 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2764 * skb_segment - Perform protocol segmentation on skb.
2765 * @skb: buffer to segment
2766 * @features: features for the output path (see dev->features)
2768 * This function performs segmentation on the given skb. It returns
2769 * a pointer to the first in a list of new skbs for the segments.
2770 * In case of error it returns ERR_PTR(err).
2772 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2774 struct sk_buff
*segs
= NULL
;
2775 struct sk_buff
*tail
= NULL
;
2776 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2777 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2778 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2779 unsigned int offset
= doffset
;
2780 unsigned int headroom
;
2782 int sg
= !!(features
& NETIF_F_SG
);
2783 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2788 __skb_push(skb
, doffset
);
2789 headroom
= skb_headroom(skb
);
2790 pos
= skb_headlen(skb
);
2793 struct sk_buff
*nskb
;
2798 len
= skb
->len
- offset
;
2802 hsize
= skb_headlen(skb
) - offset
;
2805 if (hsize
> len
|| !sg
)
2808 if (!hsize
&& i
>= nfrags
) {
2809 BUG_ON(fskb
->len
!= len
);
2812 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2815 if (unlikely(!nskb
))
2818 hsize
= skb_end_offset(nskb
);
2819 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2824 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2825 skb_release_head_state(nskb
);
2826 __skb_push(nskb
, doffset
);
2828 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2829 GFP_ATOMIC
, skb_alloc_rx_flag(skb
),
2832 if (unlikely(!nskb
))
2835 skb_reserve(nskb
, headroom
);
2836 __skb_put(nskb
, doffset
);
2845 __copy_skb_header(nskb
, skb
);
2846 nskb
->mac_len
= skb
->mac_len
;
2848 /* nskb and skb might have different headroom */
2849 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2850 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2852 skb_reset_mac_header(nskb
);
2853 skb_set_network_header(nskb
, skb
->mac_len
);
2854 nskb
->transport_header
= (nskb
->network_header
+
2855 skb_network_header_len(skb
));
2856 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2858 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2862 nskb
->ip_summed
= CHECKSUM_NONE
;
2863 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2869 frag
= skb_shinfo(nskb
)->frags
;
2871 skb_copy_from_linear_data_offset(skb
, offset
,
2872 skb_put(nskb
, hsize
), hsize
);
2874 while (pos
< offset
+ len
&& i
< nfrags
) {
2875 *frag
= skb_shinfo(skb
)->frags
[i
];
2876 __skb_frag_ref(frag
);
2877 size
= skb_frag_size(frag
);
2880 frag
->page_offset
+= offset
- pos
;
2881 skb_frag_size_sub(frag
, offset
- pos
);
2884 skb_shinfo(nskb
)->nr_frags
++;
2886 if (pos
+ size
<= offset
+ len
) {
2890 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2897 if (pos
< offset
+ len
) {
2898 struct sk_buff
*fskb2
= fskb
;
2900 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2906 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2912 SKB_FRAG_ASSERT(nskb
);
2913 skb_shinfo(nskb
)->frag_list
= fskb2
;
2917 nskb
->data_len
= len
- hsize
;
2918 nskb
->len
+= nskb
->data_len
;
2919 nskb
->truesize
+= nskb
->data_len
;
2920 } while ((offset
+= len
) < skb
->len
);
2925 while ((skb
= segs
)) {
2929 return ERR_PTR(err
);
2931 EXPORT_SYMBOL_GPL(skb_segment
);
2933 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2935 struct sk_buff
*p
= *head
;
2936 struct sk_buff
*nskb
;
2937 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2938 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2939 unsigned int headroom
;
2940 unsigned int len
= skb_gro_len(skb
);
2941 unsigned int offset
= skb_gro_offset(skb
);
2942 unsigned int headlen
= skb_headlen(skb
);
2943 unsigned int delta_truesize
;
2945 if (p
->len
+ len
>= 65536)
2948 if (pinfo
->frag_list
)
2950 else if (headlen
<= offset
) {
2953 int i
= skbinfo
->nr_frags
;
2954 int nr_frags
= pinfo
->nr_frags
+ i
;
2958 if (nr_frags
> MAX_SKB_FRAGS
)
2961 pinfo
->nr_frags
= nr_frags
;
2962 skbinfo
->nr_frags
= 0;
2964 frag
= pinfo
->frags
+ nr_frags
;
2965 frag2
= skbinfo
->frags
+ i
;
2970 frag
->page_offset
+= offset
;
2971 skb_frag_size_sub(frag
, offset
);
2973 /* all fragments truesize : remove (head size + sk_buff) */
2974 delta_truesize
= skb
->truesize
-
2975 SKB_TRUESIZE(skb_end_offset(skb
));
2977 skb
->truesize
-= skb
->data_len
;
2978 skb
->len
-= skb
->data_len
;
2981 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2983 } else if (skb
->head_frag
) {
2984 int nr_frags
= pinfo
->nr_frags
;
2985 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2986 struct page
*page
= virt_to_head_page(skb
->head
);
2987 unsigned int first_size
= headlen
- offset
;
2988 unsigned int first_offset
;
2990 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2993 first_offset
= skb
->data
-
2994 (unsigned char *)page_address(page
) +
2997 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
2999 frag
->page
.p
= page
;
3000 frag
->page_offset
= first_offset
;
3001 skb_frag_size_set(frag
, first_size
);
3003 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3004 /* We dont need to clear skbinfo->nr_frags here */
3006 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3007 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3009 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
3012 headroom
= skb_headroom(p
);
3013 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
3014 if (unlikely(!nskb
))
3017 __copy_skb_header(nskb
, p
);
3018 nskb
->mac_len
= p
->mac_len
;
3020 skb_reserve(nskb
, headroom
);
3021 __skb_put(nskb
, skb_gro_offset(p
));
3023 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
3024 skb_set_network_header(nskb
, skb_network_offset(p
));
3025 skb_set_transport_header(nskb
, skb_transport_offset(p
));
3027 __skb_pull(p
, skb_gro_offset(p
));
3028 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
3029 p
->data
- skb_mac_header(p
));
3031 skb_shinfo(nskb
)->frag_list
= p
;
3032 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3033 pinfo
->gso_size
= 0;
3034 skb_header_release(p
);
3035 NAPI_GRO_CB(nskb
)->last
= p
;
3037 nskb
->data_len
+= p
->len
;
3038 nskb
->truesize
+= p
->truesize
;
3039 nskb
->len
+= p
->len
;
3042 nskb
->next
= p
->next
;
3048 delta_truesize
= skb
->truesize
;
3049 if (offset
> headlen
) {
3050 unsigned int eat
= offset
- headlen
;
3052 skbinfo
->frags
[0].page_offset
+= eat
;
3053 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3054 skb
->data_len
-= eat
;
3059 __skb_pull(skb
, offset
);
3061 NAPI_GRO_CB(p
)->last
->next
= skb
;
3062 NAPI_GRO_CB(p
)->last
= skb
;
3063 skb_header_release(skb
);
3066 NAPI_GRO_CB(p
)->count
++;
3068 p
->truesize
+= delta_truesize
;
3071 NAPI_GRO_CB(skb
)->same_flow
= 1;
3074 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3076 void __init
skb_init(void)
3078 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3079 sizeof(struct sk_buff
),
3081 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3083 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3084 (2*sizeof(struct sk_buff
)) +
3087 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3092 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3093 * @skb: Socket buffer containing the buffers to be mapped
3094 * @sg: The scatter-gather list to map into
3095 * @offset: The offset into the buffer's contents to start mapping
3096 * @len: Length of buffer space to be mapped
3098 * Fill the specified scatter-gather list with mappings/pointers into a
3099 * region of the buffer space attached to a socket buffer.
3102 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3104 int start
= skb_headlen(skb
);
3105 int i
, copy
= start
- offset
;
3106 struct sk_buff
*frag_iter
;
3112 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3114 if ((len
-= copy
) == 0)
3119 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3122 WARN_ON(start
> offset
+ len
);
3124 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3125 if ((copy
= end
- offset
) > 0) {
3126 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3130 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3131 frag
->page_offset
+offset
-start
);
3140 skb_walk_frags(skb
, frag_iter
) {
3143 WARN_ON(start
> offset
+ len
);
3145 end
= start
+ frag_iter
->len
;
3146 if ((copy
= end
- offset
) > 0) {
3149 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3151 if ((len
-= copy
) == 0)
3161 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3163 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3165 sg_mark_end(&sg
[nsg
- 1]);
3169 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3172 * skb_cow_data - Check that a socket buffer's data buffers are writable
3173 * @skb: The socket buffer to check.
3174 * @tailbits: Amount of trailing space to be added
3175 * @trailer: Returned pointer to the skb where the @tailbits space begins
3177 * Make sure that the data buffers attached to a socket buffer are
3178 * writable. If they are not, private copies are made of the data buffers
3179 * and the socket buffer is set to use these instead.
3181 * If @tailbits is given, make sure that there is space to write @tailbits
3182 * bytes of data beyond current end of socket buffer. @trailer will be
3183 * set to point to the skb in which this space begins.
3185 * The number of scatterlist elements required to completely map the
3186 * COW'd and extended socket buffer will be returned.
3188 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3192 struct sk_buff
*skb1
, **skb_p
;
3194 /* If skb is cloned or its head is paged, reallocate
3195 * head pulling out all the pages (pages are considered not writable
3196 * at the moment even if they are anonymous).
3198 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3199 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3202 /* Easy case. Most of packets will go this way. */
3203 if (!skb_has_frag_list(skb
)) {
3204 /* A little of trouble, not enough of space for trailer.
3205 * This should not happen, when stack is tuned to generate
3206 * good frames. OK, on miss we reallocate and reserve even more
3207 * space, 128 bytes is fair. */
3209 if (skb_tailroom(skb
) < tailbits
&&
3210 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3218 /* Misery. We are in troubles, going to mincer fragments... */
3221 skb_p
= &skb_shinfo(skb
)->frag_list
;
3224 while ((skb1
= *skb_p
) != NULL
) {
3227 /* The fragment is partially pulled by someone,
3228 * this can happen on input. Copy it and everything
3231 if (skb_shared(skb1
))
3234 /* If the skb is the last, worry about trailer. */
3236 if (skb1
->next
== NULL
&& tailbits
) {
3237 if (skb_shinfo(skb1
)->nr_frags
||
3238 skb_has_frag_list(skb1
) ||
3239 skb_tailroom(skb1
) < tailbits
)
3240 ntail
= tailbits
+ 128;
3246 skb_shinfo(skb1
)->nr_frags
||
3247 skb_has_frag_list(skb1
)) {
3248 struct sk_buff
*skb2
;
3250 /* Fuck, we are miserable poor guys... */
3252 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3254 skb2
= skb_copy_expand(skb1
,
3258 if (unlikely(skb2
== NULL
))
3262 skb_set_owner_w(skb2
, skb1
->sk
);
3264 /* Looking around. Are we still alive?
3265 * OK, link new skb, drop old one */
3267 skb2
->next
= skb1
->next
;
3274 skb_p
= &skb1
->next
;
3279 EXPORT_SYMBOL_GPL(skb_cow_data
);
3281 static void sock_rmem_free(struct sk_buff
*skb
)
3283 struct sock
*sk
= skb
->sk
;
3285 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3289 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3291 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3295 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3296 (unsigned int)sk
->sk_rcvbuf
)
3301 skb
->destructor
= sock_rmem_free
;
3302 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3304 /* before exiting rcu section, make sure dst is refcounted */
3307 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3308 if (!sock_flag(sk
, SOCK_DEAD
))
3309 sk
->sk_data_ready(sk
, len
);
3312 EXPORT_SYMBOL(sock_queue_err_skb
);
3314 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3315 struct skb_shared_hwtstamps
*hwtstamps
)
3317 struct sock
*sk
= orig_skb
->sk
;
3318 struct sock_exterr_skb
*serr
;
3319 struct sk_buff
*skb
;
3325 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3330 *skb_hwtstamps(skb
) =
3334 * no hardware time stamps available,
3335 * so keep the shared tx_flags and only
3336 * store software time stamp
3338 skb
->tstamp
= ktime_get_real();
3341 serr
= SKB_EXT_ERR(skb
);
3342 memset(serr
, 0, sizeof(*serr
));
3343 serr
->ee
.ee_errno
= ENOMSG
;
3344 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3346 err
= sock_queue_err_skb(sk
, skb
);
3351 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3353 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3355 struct sock
*sk
= skb
->sk
;
3356 struct sock_exterr_skb
*serr
;
3359 skb
->wifi_acked_valid
= 1;
3360 skb
->wifi_acked
= acked
;
3362 serr
= SKB_EXT_ERR(skb
);
3363 memset(serr
, 0, sizeof(*serr
));
3364 serr
->ee
.ee_errno
= ENOMSG
;
3365 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3367 err
= sock_queue_err_skb(sk
, skb
);
3371 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3375 * skb_partial_csum_set - set up and verify partial csum values for packet
3376 * @skb: the skb to set
3377 * @start: the number of bytes after skb->data to start checksumming.
3378 * @off: the offset from start to place the checksum.
3380 * For untrusted partially-checksummed packets, we need to make sure the values
3381 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3383 * This function checks and sets those values and skb->ip_summed: if this
3384 * returns false you should drop the packet.
3386 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3388 if (unlikely(start
> skb_headlen(skb
)) ||
3389 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3390 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3391 start
, off
, skb_headlen(skb
));
3394 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3395 skb
->csum_start
= skb_headroom(skb
) + start
;
3396 skb
->csum_offset
= off
;
3399 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3401 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3403 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3406 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3408 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3411 skb_release_head_state(skb
);
3412 kmem_cache_free(skbuff_head_cache
, skb
);
3417 EXPORT_SYMBOL(kfree_skb_partial
);
3420 * skb_try_coalesce - try to merge skb to prior one
3422 * @from: buffer to add
3423 * @fragstolen: pointer to boolean
3424 * @delta_truesize: how much more was allocated than was requested
3426 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3427 bool *fragstolen
, int *delta_truesize
)
3429 int i
, delta
, len
= from
->len
;
3431 *fragstolen
= false;
3436 if (len
<= skb_tailroom(to
)) {
3437 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3438 *delta_truesize
= 0;
3442 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3445 if (skb_headlen(from
) != 0) {
3447 unsigned int offset
;
3449 if (skb_shinfo(to
)->nr_frags
+
3450 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3453 if (skb_head_is_locked(from
))
3456 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3458 page
= virt_to_head_page(from
->head
);
3459 offset
= from
->data
- (unsigned char *)page_address(page
);
3461 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3462 page
, offset
, skb_headlen(from
));
3465 if (skb_shinfo(to
)->nr_frags
+
3466 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3469 delta
= from
->truesize
- SKB_TRUESIZE(skb_end_offset(from
));
3472 WARN_ON_ONCE(delta
< len
);
3474 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3475 skb_shinfo(from
)->frags
,
3476 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3477 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3479 if (!skb_cloned(from
))
3480 skb_shinfo(from
)->nr_frags
= 0;
3482 /* if the skb is not cloned this does nothing
3483 * since we set nr_frags to 0.
3485 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3486 skb_frag_ref(from
, i
);
3488 to
->truesize
+= delta
;
3490 to
->data_len
+= len
;
3492 *delta_truesize
= delta
;
3495 EXPORT_SYMBOL(skb_try_coalesce
);