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
{
345 unsigned int pagecnt_bias
;
347 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
349 #define NETDEV_PAGECNT_BIAS (PAGE_SIZE / SMP_CACHE_BYTES)
351 static void *__netdev_alloc_frag(unsigned int fragsz
, gfp_t gfp_mask
)
353 struct netdev_alloc_cache
*nc
;
357 local_irq_save(flags
);
358 nc
= &__get_cpu_var(netdev_alloc_cache
);
359 if (unlikely(!nc
->page
)) {
361 nc
->page
= alloc_page(gfp_mask
);
362 if (unlikely(!nc
->page
))
365 atomic_set(&nc
->page
->_count
, NETDEV_PAGECNT_BIAS
);
366 nc
->pagecnt_bias
= NETDEV_PAGECNT_BIAS
;
370 if (nc
->offset
+ fragsz
> PAGE_SIZE
) {
371 /* avoid unnecessary locked operations if possible */
372 if ((atomic_read(&nc
->page
->_count
) == nc
->pagecnt_bias
) ||
373 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->page
->_count
))
378 data
= page_address(nc
->page
) + nc
->offset
;
379 nc
->offset
+= fragsz
;
382 local_irq_restore(flags
);
387 * netdev_alloc_frag - allocate a page fragment
388 * @fragsz: fragment size
390 * Allocates a frag from a page for receive buffer.
391 * Uses GFP_ATOMIC allocations.
393 void *netdev_alloc_frag(unsigned int fragsz
)
395 return __netdev_alloc_frag(fragsz
, GFP_ATOMIC
| __GFP_COLD
);
397 EXPORT_SYMBOL(netdev_alloc_frag
);
400 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
401 * @dev: network device to receive on
402 * @length: length to allocate
403 * @gfp_mask: get_free_pages mask, passed to alloc_skb
405 * Allocate a new &sk_buff and assign it a usage count of one. The
406 * buffer has unspecified headroom built in. Users should allocate
407 * the headroom they think they need without accounting for the
408 * built in space. The built in space is used for optimisations.
410 * %NULL is returned if there is no free memory.
412 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
413 unsigned int length
, gfp_t gfp_mask
)
415 struct sk_buff
*skb
= NULL
;
416 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
417 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
419 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
422 if (sk_memalloc_socks())
423 gfp_mask
|= __GFP_MEMALLOC
;
425 data
= __netdev_alloc_frag(fragsz
, gfp_mask
);
428 skb
= build_skb(data
, fragsz
);
430 put_page(virt_to_head_page(data
));
433 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
,
434 SKB_ALLOC_RX
, NUMA_NO_NODE
);
437 skb_reserve(skb
, NET_SKB_PAD
);
442 EXPORT_SYMBOL(__netdev_alloc_skb
);
444 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
445 int size
, unsigned int truesize
)
447 skb_fill_page_desc(skb
, i
, page
, off
, size
);
449 skb
->data_len
+= size
;
450 skb
->truesize
+= truesize
;
452 EXPORT_SYMBOL(skb_add_rx_frag
);
454 static void skb_drop_list(struct sk_buff
**listp
)
456 struct sk_buff
*list
= *listp
;
461 struct sk_buff
*this = list
;
467 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
469 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
472 static void skb_clone_fraglist(struct sk_buff
*skb
)
474 struct sk_buff
*list
;
476 skb_walk_frags(skb
, list
)
480 static void skb_free_head(struct sk_buff
*skb
)
483 put_page(virt_to_head_page(skb
->head
));
488 static void skb_release_data(struct sk_buff
*skb
)
491 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
492 &skb_shinfo(skb
)->dataref
)) {
493 if (skb_shinfo(skb
)->nr_frags
) {
495 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
496 skb_frag_unref(skb
, i
);
500 * If skb buf is from userspace, we need to notify the caller
501 * the lower device DMA has done;
503 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
504 struct ubuf_info
*uarg
;
506 uarg
= skb_shinfo(skb
)->destructor_arg
;
508 uarg
->callback(uarg
);
511 if (skb_has_frag_list(skb
))
512 skb_drop_fraglist(skb
);
519 * Free an skbuff by memory without cleaning the state.
521 static void kfree_skbmem(struct sk_buff
*skb
)
523 struct sk_buff
*other
;
524 atomic_t
*fclone_ref
;
526 switch (skb
->fclone
) {
527 case SKB_FCLONE_UNAVAILABLE
:
528 kmem_cache_free(skbuff_head_cache
, skb
);
531 case SKB_FCLONE_ORIG
:
532 fclone_ref
= (atomic_t
*) (skb
+ 2);
533 if (atomic_dec_and_test(fclone_ref
))
534 kmem_cache_free(skbuff_fclone_cache
, skb
);
537 case SKB_FCLONE_CLONE
:
538 fclone_ref
= (atomic_t
*) (skb
+ 1);
541 /* The clone portion is available for
542 * fast-cloning again.
544 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
546 if (atomic_dec_and_test(fclone_ref
))
547 kmem_cache_free(skbuff_fclone_cache
, other
);
552 static void skb_release_head_state(struct sk_buff
*skb
)
556 secpath_put(skb
->sp
);
558 if (skb
->destructor
) {
560 skb
->destructor(skb
);
562 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
563 nf_conntrack_put(skb
->nfct
);
565 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
566 nf_conntrack_put_reasm(skb
->nfct_reasm
);
568 #ifdef CONFIG_BRIDGE_NETFILTER
569 nf_bridge_put(skb
->nf_bridge
);
571 /* XXX: IS this still necessary? - JHS */
572 #ifdef CONFIG_NET_SCHED
574 #ifdef CONFIG_NET_CLS_ACT
580 /* Free everything but the sk_buff shell. */
581 static void skb_release_all(struct sk_buff
*skb
)
583 skb_release_head_state(skb
);
584 skb_release_data(skb
);
588 * __kfree_skb - private function
591 * Free an sk_buff. Release anything attached to the buffer.
592 * Clean the state. This is an internal helper function. Users should
593 * always call kfree_skb
596 void __kfree_skb(struct sk_buff
*skb
)
598 skb_release_all(skb
);
601 EXPORT_SYMBOL(__kfree_skb
);
604 * kfree_skb - free an sk_buff
605 * @skb: buffer to free
607 * Drop a reference to the buffer and free it if the usage count has
610 void kfree_skb(struct sk_buff
*skb
)
614 if (likely(atomic_read(&skb
->users
) == 1))
616 else if (likely(!atomic_dec_and_test(&skb
->users
)))
618 trace_kfree_skb(skb
, __builtin_return_address(0));
621 EXPORT_SYMBOL(kfree_skb
);
624 * consume_skb - free an skbuff
625 * @skb: buffer to free
627 * Drop a ref to the buffer and free it if the usage count has hit zero
628 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
629 * is being dropped after a failure and notes that
631 void consume_skb(struct sk_buff
*skb
)
635 if (likely(atomic_read(&skb
->users
) == 1))
637 else if (likely(!atomic_dec_and_test(&skb
->users
)))
639 trace_consume_skb(skb
);
642 EXPORT_SYMBOL(consume_skb
);
645 * skb_recycle - clean up an skb for reuse
648 * Recycles the skb to be reused as a receive buffer. This
649 * function does any necessary reference count dropping, and
650 * cleans up the skbuff as if it just came from __alloc_skb().
652 void skb_recycle(struct sk_buff
*skb
)
654 struct skb_shared_info
*shinfo
;
656 skb_release_head_state(skb
);
658 shinfo
= skb_shinfo(skb
);
659 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
660 atomic_set(&shinfo
->dataref
, 1);
662 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
663 skb
->data
= skb
->head
+ NET_SKB_PAD
;
664 skb_reset_tail_pointer(skb
);
666 EXPORT_SYMBOL(skb_recycle
);
669 * skb_recycle_check - check if skb can be reused for receive
671 * @skb_size: minimum receive buffer size
673 * Checks that the skb passed in is not shared or cloned, and
674 * that it is linear and its head portion at least as large as
675 * skb_size so that it can be recycled as a receive buffer.
676 * If these conditions are met, this function does any necessary
677 * reference count dropping and cleans up the skbuff as if it
678 * just came from __alloc_skb().
680 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
682 if (!skb_is_recycleable(skb
, skb_size
))
689 EXPORT_SYMBOL(skb_recycle_check
);
691 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
693 new->tstamp
= old
->tstamp
;
695 new->transport_header
= old
->transport_header
;
696 new->network_header
= old
->network_header
;
697 new->mac_header
= old
->mac_header
;
698 skb_dst_copy(new, old
);
699 new->rxhash
= old
->rxhash
;
700 new->ooo_okay
= old
->ooo_okay
;
701 new->l4_rxhash
= old
->l4_rxhash
;
702 new->no_fcs
= old
->no_fcs
;
704 new->sp
= secpath_get(old
->sp
);
706 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
707 new->csum
= old
->csum
;
708 new->local_df
= old
->local_df
;
709 new->pkt_type
= old
->pkt_type
;
710 new->ip_summed
= old
->ip_summed
;
711 skb_copy_queue_mapping(new, old
);
712 new->priority
= old
->priority
;
713 #if IS_ENABLED(CONFIG_IP_VS)
714 new->ipvs_property
= old
->ipvs_property
;
716 new->pfmemalloc
= old
->pfmemalloc
;
717 new->protocol
= old
->protocol
;
718 new->mark
= old
->mark
;
719 new->skb_iif
= old
->skb_iif
;
721 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
722 new->nf_trace
= old
->nf_trace
;
724 #ifdef CONFIG_NET_SCHED
725 new->tc_index
= old
->tc_index
;
726 #ifdef CONFIG_NET_CLS_ACT
727 new->tc_verd
= old
->tc_verd
;
730 new->vlan_tci
= old
->vlan_tci
;
732 skb_copy_secmark(new, old
);
736 * You should not add any new code to this function. Add it to
737 * __copy_skb_header above instead.
739 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
741 #define C(x) n->x = skb->x
743 n
->next
= n
->prev
= NULL
;
745 __copy_skb_header(n
, skb
);
750 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
753 n
->destructor
= NULL
;
760 atomic_set(&n
->users
, 1);
762 atomic_inc(&(skb_shinfo(skb
)->dataref
));
770 * skb_morph - morph one skb into another
771 * @dst: the skb to receive the contents
772 * @src: the skb to supply the contents
774 * This is identical to skb_clone except that the target skb is
775 * supplied by the user.
777 * The target skb is returned upon exit.
779 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
781 skb_release_all(dst
);
782 return __skb_clone(dst
, src
);
784 EXPORT_SYMBOL_GPL(skb_morph
);
787 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
788 * @skb: the skb to modify
789 * @gfp_mask: allocation priority
791 * This must be called on SKBTX_DEV_ZEROCOPY skb.
792 * It will copy all frags into kernel and drop the reference
793 * to userspace pages.
795 * If this function is called from an interrupt gfp_mask() must be
798 * Returns 0 on success or a negative error code on failure
799 * to allocate kernel memory to copy to.
801 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
804 int num_frags
= skb_shinfo(skb
)->nr_frags
;
805 struct page
*page
, *head
= NULL
;
806 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
808 for (i
= 0; i
< num_frags
; i
++) {
810 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
812 page
= alloc_page(gfp_mask
);
815 struct page
*next
= (struct page
*)head
->private;
821 vaddr
= kmap_atomic(skb_frag_page(f
));
822 memcpy(page_address(page
),
823 vaddr
+ f
->page_offset
, skb_frag_size(f
));
824 kunmap_atomic(vaddr
);
825 page
->private = (unsigned long)head
;
829 /* skb frags release userspace buffers */
830 for (i
= 0; i
< num_frags
; i
++)
831 skb_frag_unref(skb
, i
);
833 uarg
->callback(uarg
);
835 /* skb frags point to kernel buffers */
836 for (i
= num_frags
- 1; i
>= 0; i
--) {
837 __skb_fill_page_desc(skb
, i
, head
, 0,
838 skb_shinfo(skb
)->frags
[i
].size
);
839 head
= (struct page
*)head
->private;
842 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
845 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
848 * skb_clone - duplicate an sk_buff
849 * @skb: buffer to clone
850 * @gfp_mask: allocation priority
852 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
853 * copies share the same packet data but not structure. The new
854 * buffer has a reference count of 1. If the allocation fails the
855 * function returns %NULL otherwise the new buffer is returned.
857 * If this function is called from an interrupt gfp_mask() must be
861 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
865 if (skb_orphan_frags(skb
, gfp_mask
))
869 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
870 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
871 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
872 n
->fclone
= SKB_FCLONE_CLONE
;
873 atomic_inc(fclone_ref
);
875 if (skb_pfmemalloc(skb
))
876 gfp_mask
|= __GFP_MEMALLOC
;
878 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
882 kmemcheck_annotate_bitfield(n
, flags1
);
883 kmemcheck_annotate_bitfield(n
, flags2
);
884 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
887 return __skb_clone(n
, skb
);
889 EXPORT_SYMBOL(skb_clone
);
891 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
893 #ifndef NET_SKBUFF_DATA_USES_OFFSET
895 * Shift between the two data areas in bytes
897 unsigned long offset
= new->data
- old
->data
;
900 __copy_skb_header(new, old
);
902 #ifndef NET_SKBUFF_DATA_USES_OFFSET
903 /* {transport,network,mac}_header are relative to skb->head */
904 new->transport_header
+= offset
;
905 new->network_header
+= offset
;
906 if (skb_mac_header_was_set(new))
907 new->mac_header
+= offset
;
909 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
910 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
911 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
914 static inline int skb_alloc_rx_flag(const struct sk_buff
*skb
)
916 if (skb_pfmemalloc(skb
))
922 * skb_copy - create private copy of an sk_buff
923 * @skb: buffer to copy
924 * @gfp_mask: allocation priority
926 * Make a copy of both an &sk_buff and its data. This is used when the
927 * caller wishes to modify the data and needs a private copy of the
928 * data to alter. Returns %NULL on failure or the pointer to the buffer
929 * on success. The returned buffer has a reference count of 1.
931 * As by-product this function converts non-linear &sk_buff to linear
932 * one, so that &sk_buff becomes completely private and caller is allowed
933 * to modify all the data of returned buffer. This means that this
934 * function is not recommended for use in circumstances when only
935 * header is going to be modified. Use pskb_copy() instead.
938 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
940 int headerlen
= skb_headroom(skb
);
941 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
942 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
943 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
948 /* Set the data pointer */
949 skb_reserve(n
, headerlen
);
950 /* Set the tail pointer and length */
951 skb_put(n
, skb
->len
);
953 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
956 copy_skb_header(n
, skb
);
959 EXPORT_SYMBOL(skb_copy
);
962 * __pskb_copy - create copy of an sk_buff with private head.
963 * @skb: buffer to copy
964 * @headroom: headroom of new skb
965 * @gfp_mask: allocation priority
967 * Make a copy of both an &sk_buff and part of its data, located
968 * in header. Fragmented data remain shared. This is used when
969 * the caller wishes to modify only header of &sk_buff and needs
970 * private copy of the header to alter. Returns %NULL on failure
971 * or the pointer to the buffer on success.
972 * The returned buffer has a reference count of 1.
975 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
977 unsigned int size
= skb_headlen(skb
) + headroom
;
978 struct sk_buff
*n
= __alloc_skb(size
, gfp_mask
,
979 skb_alloc_rx_flag(skb
), NUMA_NO_NODE
);
984 /* Set the data pointer */
985 skb_reserve(n
, headroom
);
986 /* Set the tail pointer and length */
987 skb_put(n
, skb_headlen(skb
));
989 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
991 n
->truesize
+= skb
->data_len
;
992 n
->data_len
= skb
->data_len
;
995 if (skb_shinfo(skb
)->nr_frags
) {
998 if (skb_orphan_frags(skb
, gfp_mask
)) {
1003 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1004 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
1005 skb_frag_ref(skb
, i
);
1007 skb_shinfo(n
)->nr_frags
= i
;
1010 if (skb_has_frag_list(skb
)) {
1011 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
1012 skb_clone_fraglist(n
);
1015 copy_skb_header(n
, skb
);
1019 EXPORT_SYMBOL(__pskb_copy
);
1022 * pskb_expand_head - reallocate header of &sk_buff
1023 * @skb: buffer to reallocate
1024 * @nhead: room to add at head
1025 * @ntail: room to add at tail
1026 * @gfp_mask: allocation priority
1028 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1029 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1030 * reference count of 1. Returns zero in the case of success or error,
1031 * if expansion failed. In the last case, &sk_buff is not changed.
1033 * All the pointers pointing into skb header may change and must be
1034 * reloaded after call to this function.
1037 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
1042 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
1047 if (skb_shared(skb
))
1050 size
= SKB_DATA_ALIGN(size
);
1052 if (skb_pfmemalloc(skb
))
1053 gfp_mask
|= __GFP_MEMALLOC
;
1054 data
= kmalloc_reserve(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
1055 gfp_mask
, NUMA_NO_NODE
, NULL
);
1058 size
= SKB_WITH_OVERHEAD(ksize(data
));
1060 /* Copy only real data... and, alas, header. This should be
1061 * optimized for the cases when header is void.
1063 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
1065 memcpy((struct skb_shared_info
*)(data
+ size
),
1067 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
1070 * if shinfo is shared we must drop the old head gracefully, but if it
1071 * is not we can just drop the old head and let the existing refcount
1072 * be since all we did is relocate the values
1074 if (skb_cloned(skb
)) {
1075 /* copy this zero copy skb frags */
1076 if (skb_orphan_frags(skb
, gfp_mask
))
1078 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1079 skb_frag_ref(skb
, i
);
1081 if (skb_has_frag_list(skb
))
1082 skb_clone_fraglist(skb
);
1084 skb_release_data(skb
);
1088 off
= (data
+ nhead
) - skb
->head
;
1093 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1097 skb
->end
= skb
->head
+ size
;
1099 /* {transport,network,mac}_header and tail are relative to skb->head */
1101 skb
->transport_header
+= off
;
1102 skb
->network_header
+= off
;
1103 if (skb_mac_header_was_set(skb
))
1104 skb
->mac_header
+= off
;
1105 /* Only adjust this if it actually is csum_start rather than csum */
1106 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1107 skb
->csum_start
+= nhead
;
1111 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1119 EXPORT_SYMBOL(pskb_expand_head
);
1121 /* Make private copy of skb with writable head and some headroom */
1123 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1125 struct sk_buff
*skb2
;
1126 int delta
= headroom
- skb_headroom(skb
);
1129 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1131 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1132 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1140 EXPORT_SYMBOL(skb_realloc_headroom
);
1143 * skb_copy_expand - copy and expand sk_buff
1144 * @skb: buffer to copy
1145 * @newheadroom: new free bytes at head
1146 * @newtailroom: new free bytes at tail
1147 * @gfp_mask: allocation priority
1149 * Make a copy of both an &sk_buff and its data and while doing so
1150 * allocate additional space.
1152 * This is used when the caller wishes to modify the data and needs a
1153 * private copy of the data to alter as well as more space for new fields.
1154 * Returns %NULL on failure or the pointer to the buffer
1155 * on success. The returned buffer has a reference count of 1.
1157 * You must pass %GFP_ATOMIC as the allocation priority if this function
1158 * is called from an interrupt.
1160 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1161 int newheadroom
, int newtailroom
,
1165 * Allocate the copy buffer
1167 struct sk_buff
*n
= __alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1168 gfp_mask
, skb_alloc_rx_flag(skb
),
1170 int oldheadroom
= skb_headroom(skb
);
1171 int head_copy_len
, head_copy_off
;
1177 skb_reserve(n
, newheadroom
);
1179 /* Set the tail pointer and length */
1180 skb_put(n
, skb
->len
);
1182 head_copy_len
= oldheadroom
;
1184 if (newheadroom
<= head_copy_len
)
1185 head_copy_len
= newheadroom
;
1187 head_copy_off
= newheadroom
- head_copy_len
;
1189 /* Copy the linear header and data. */
1190 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1191 skb
->len
+ head_copy_len
))
1194 copy_skb_header(n
, skb
);
1196 off
= newheadroom
- oldheadroom
;
1197 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1198 n
->csum_start
+= off
;
1199 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1200 n
->transport_header
+= off
;
1201 n
->network_header
+= off
;
1202 if (skb_mac_header_was_set(skb
))
1203 n
->mac_header
+= off
;
1208 EXPORT_SYMBOL(skb_copy_expand
);
1211 * skb_pad - zero pad the tail of an skb
1212 * @skb: buffer to pad
1213 * @pad: space to pad
1215 * Ensure that a buffer is followed by a padding area that is zero
1216 * filled. Used by network drivers which may DMA or transfer data
1217 * beyond the buffer end onto the wire.
1219 * May return error in out of memory cases. The skb is freed on error.
1222 int skb_pad(struct sk_buff
*skb
, int pad
)
1227 /* If the skbuff is non linear tailroom is always zero.. */
1228 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1229 memset(skb
->data
+skb
->len
, 0, pad
);
1233 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1234 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1235 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1240 /* FIXME: The use of this function with non-linear skb's really needs
1243 err
= skb_linearize(skb
);
1247 memset(skb
->data
+ skb
->len
, 0, pad
);
1254 EXPORT_SYMBOL(skb_pad
);
1257 * skb_put - add data to a buffer
1258 * @skb: buffer to use
1259 * @len: amount of data to add
1261 * This function extends the used data area of the buffer. If this would
1262 * exceed the total buffer size the kernel will panic. A pointer to the
1263 * first byte of the extra data is returned.
1265 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1267 unsigned char *tmp
= skb_tail_pointer(skb
);
1268 SKB_LINEAR_ASSERT(skb
);
1271 if (unlikely(skb
->tail
> skb
->end
))
1272 skb_over_panic(skb
, len
, __builtin_return_address(0));
1275 EXPORT_SYMBOL(skb_put
);
1278 * skb_push - add data to the start of a buffer
1279 * @skb: buffer to use
1280 * @len: amount of data to add
1282 * This function extends the used data area of the buffer at the buffer
1283 * start. If this would exceed the total buffer headroom the kernel will
1284 * panic. A pointer to the first byte of the extra data is returned.
1286 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1290 if (unlikely(skb
->data
<skb
->head
))
1291 skb_under_panic(skb
, len
, __builtin_return_address(0));
1294 EXPORT_SYMBOL(skb_push
);
1297 * skb_pull - remove data from the start of a buffer
1298 * @skb: buffer to use
1299 * @len: amount of data to remove
1301 * This function removes data from the start of a buffer, returning
1302 * the memory to the headroom. A pointer to the next data in the buffer
1303 * is returned. Once the data has been pulled future pushes will overwrite
1306 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1308 return skb_pull_inline(skb
, len
);
1310 EXPORT_SYMBOL(skb_pull
);
1313 * skb_trim - remove end from a buffer
1314 * @skb: buffer to alter
1317 * Cut the length of a buffer down by removing data from the tail. If
1318 * the buffer is already under the length specified it is not modified.
1319 * The skb must be linear.
1321 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1324 __skb_trim(skb
, len
);
1326 EXPORT_SYMBOL(skb_trim
);
1328 /* Trims skb to length len. It can change skb pointers.
1331 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1333 struct sk_buff
**fragp
;
1334 struct sk_buff
*frag
;
1335 int offset
= skb_headlen(skb
);
1336 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1340 if (skb_cloned(skb
) &&
1341 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1348 for (; i
< nfrags
; i
++) {
1349 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1356 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1359 skb_shinfo(skb
)->nr_frags
= i
;
1361 for (; i
< nfrags
; i
++)
1362 skb_frag_unref(skb
, i
);
1364 if (skb_has_frag_list(skb
))
1365 skb_drop_fraglist(skb
);
1369 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1370 fragp
= &frag
->next
) {
1371 int end
= offset
+ frag
->len
;
1373 if (skb_shared(frag
)) {
1374 struct sk_buff
*nfrag
;
1376 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1377 if (unlikely(!nfrag
))
1380 nfrag
->next
= frag
->next
;
1392 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1396 skb_drop_list(&frag
->next
);
1401 if (len
> skb_headlen(skb
)) {
1402 skb
->data_len
-= skb
->len
- len
;
1407 skb_set_tail_pointer(skb
, len
);
1412 EXPORT_SYMBOL(___pskb_trim
);
1415 * __pskb_pull_tail - advance tail of skb header
1416 * @skb: buffer to reallocate
1417 * @delta: number of bytes to advance tail
1419 * The function makes a sense only on a fragmented &sk_buff,
1420 * it expands header moving its tail forward and copying necessary
1421 * data from fragmented part.
1423 * &sk_buff MUST have reference count of 1.
1425 * Returns %NULL (and &sk_buff does not change) if pull failed
1426 * or value of new tail of skb in the case of success.
1428 * All the pointers pointing into skb header may change and must be
1429 * reloaded after call to this function.
1432 /* Moves tail of skb head forward, copying data from fragmented part,
1433 * when it is necessary.
1434 * 1. It may fail due to malloc failure.
1435 * 2. It may change skb pointers.
1437 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1439 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1441 /* If skb has not enough free space at tail, get new one
1442 * plus 128 bytes for future expansions. If we have enough
1443 * room at tail, reallocate without expansion only if skb is cloned.
1445 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1447 if (eat
> 0 || skb_cloned(skb
)) {
1448 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1453 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1456 /* Optimization: no fragments, no reasons to preestimate
1457 * size of pulled pages. Superb.
1459 if (!skb_has_frag_list(skb
))
1462 /* Estimate size of pulled pages. */
1464 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1465 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1472 /* If we need update frag list, we are in troubles.
1473 * Certainly, it possible to add an offset to skb data,
1474 * but taking into account that pulling is expected to
1475 * be very rare operation, it is worth to fight against
1476 * further bloating skb head and crucify ourselves here instead.
1477 * Pure masohism, indeed. 8)8)
1480 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1481 struct sk_buff
*clone
= NULL
;
1482 struct sk_buff
*insp
= NULL
;
1487 if (list
->len
<= eat
) {
1488 /* Eaten as whole. */
1493 /* Eaten partially. */
1495 if (skb_shared(list
)) {
1496 /* Sucks! We need to fork list. :-( */
1497 clone
= skb_clone(list
, GFP_ATOMIC
);
1503 /* This may be pulled without
1507 if (!pskb_pull(list
, eat
)) {
1515 /* Free pulled out fragments. */
1516 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1517 skb_shinfo(skb
)->frag_list
= list
->next
;
1520 /* And insert new clone at head. */
1523 skb_shinfo(skb
)->frag_list
= clone
;
1526 /* Success! Now we may commit changes to skb data. */
1531 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1532 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1535 skb_frag_unref(skb
, i
);
1538 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1540 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1541 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1547 skb_shinfo(skb
)->nr_frags
= k
;
1550 skb
->data_len
-= delta
;
1552 return skb_tail_pointer(skb
);
1554 EXPORT_SYMBOL(__pskb_pull_tail
);
1557 * skb_copy_bits - copy bits from skb to kernel buffer
1559 * @offset: offset in source
1560 * @to: destination buffer
1561 * @len: number of bytes to copy
1563 * Copy the specified number of bytes from the source skb to the
1564 * destination buffer.
1567 * If its prototype is ever changed,
1568 * check arch/{*}/net/{*}.S files,
1569 * since it is called from BPF assembly code.
1571 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1573 int start
= skb_headlen(skb
);
1574 struct sk_buff
*frag_iter
;
1577 if (offset
> (int)skb
->len
- len
)
1581 if ((copy
= start
- offset
) > 0) {
1584 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1585 if ((len
-= copy
) == 0)
1591 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1593 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1595 WARN_ON(start
> offset
+ len
);
1597 end
= start
+ skb_frag_size(f
);
1598 if ((copy
= end
- offset
) > 0) {
1604 vaddr
= kmap_atomic(skb_frag_page(f
));
1606 vaddr
+ f
->page_offset
+ offset
- start
,
1608 kunmap_atomic(vaddr
);
1610 if ((len
-= copy
) == 0)
1618 skb_walk_frags(skb
, frag_iter
) {
1621 WARN_ON(start
> offset
+ len
);
1623 end
= start
+ frag_iter
->len
;
1624 if ((copy
= end
- offset
) > 0) {
1627 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1629 if ((len
-= copy
) == 0)
1643 EXPORT_SYMBOL(skb_copy_bits
);
1646 * Callback from splice_to_pipe(), if we need to release some pages
1647 * at the end of the spd in case we error'ed out in filling the pipe.
1649 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1651 put_page(spd
->pages
[i
]);
1654 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1655 unsigned int *offset
,
1656 struct sk_buff
*skb
, struct sock
*sk
)
1658 struct page
*p
= sk
->sk_sndmsg_page
;
1663 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1667 off
= sk
->sk_sndmsg_off
= 0;
1668 /* hold one ref to this page until it's full */
1672 /* If we are the only user of the page, we can reset offset */
1673 if (page_count(p
) == 1)
1674 sk
->sk_sndmsg_off
= 0;
1675 off
= sk
->sk_sndmsg_off
;
1676 mlen
= PAGE_SIZE
- off
;
1677 if (mlen
< 64 && mlen
< *len
) {
1682 *len
= min_t(unsigned int, *len
, mlen
);
1685 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1686 sk
->sk_sndmsg_off
+= *len
;
1692 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1694 unsigned int offset
)
1696 return spd
->nr_pages
&&
1697 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1698 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1699 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1703 * Fill page/offset/length into spd, if it can hold more pages.
1705 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1706 struct pipe_inode_info
*pipe
, struct page
*page
,
1707 unsigned int *len
, unsigned int offset
,
1708 struct sk_buff
*skb
, bool linear
,
1711 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1715 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1719 if (spd_can_coalesce(spd
, page
, offset
)) {
1720 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1724 spd
->pages
[spd
->nr_pages
] = page
;
1725 spd
->partial
[spd
->nr_pages
].len
= *len
;
1726 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1732 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1733 unsigned int *plen
, unsigned int off
)
1738 n
= *poff
/ PAGE_SIZE
;
1740 *page
= nth_page(*page
, n
);
1742 *poff
= *poff
% PAGE_SIZE
;
1746 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1747 unsigned int plen
, unsigned int *off
,
1748 unsigned int *len
, struct sk_buff
*skb
,
1749 struct splice_pipe_desc
*spd
, bool linear
,
1751 struct pipe_inode_info
*pipe
)
1756 /* skip this segment if already processed */
1762 /* ignore any bits we already processed */
1764 __segment_seek(&page
, &poff
, &plen
, *off
);
1769 unsigned int flen
= min(*len
, plen
);
1771 /* the linear region may spread across several pages */
1772 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1774 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1777 __segment_seek(&page
, &poff
, &plen
, flen
);
1780 } while (*len
&& plen
);
1786 * Map linear and fragment data from the skb to spd. It reports true if the
1787 * pipe is full or if we already spliced the requested length.
1789 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1790 unsigned int *offset
, unsigned int *len
,
1791 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1795 /* map the linear part :
1796 * If skb->head_frag is set, this 'linear' part is backed by a
1797 * fragment, and if the head is not shared with any clones then
1798 * we can avoid a copy since we own the head portion of this page.
1800 if (__splice_segment(virt_to_page(skb
->data
),
1801 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1803 offset
, len
, skb
, spd
,
1804 skb_head_is_locked(skb
),
1809 * then map the fragments
1811 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1812 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1814 if (__splice_segment(skb_frag_page(f
),
1815 f
->page_offset
, skb_frag_size(f
),
1816 offset
, len
, skb
, spd
, false, sk
, pipe
))
1824 * Map data from the skb to a pipe. Should handle both the linear part,
1825 * the fragments, and the frag list. It does NOT handle frag lists within
1826 * the frag list, if such a thing exists. We'd probably need to recurse to
1827 * handle that cleanly.
1829 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1830 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1833 struct partial_page partial
[MAX_SKB_FRAGS
];
1834 struct page
*pages
[MAX_SKB_FRAGS
];
1835 struct splice_pipe_desc spd
= {
1838 .nr_pages_max
= MAX_SKB_FRAGS
,
1840 .ops
= &sock_pipe_buf_ops
,
1841 .spd_release
= sock_spd_release
,
1843 struct sk_buff
*frag_iter
;
1844 struct sock
*sk
= skb
->sk
;
1848 * __skb_splice_bits() only fails if the output has no room left,
1849 * so no point in going over the frag_list for the error case.
1851 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1857 * now see if we have a frag_list to map
1859 skb_walk_frags(skb
, frag_iter
) {
1862 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1869 * Drop the socket lock, otherwise we have reverse
1870 * locking dependencies between sk_lock and i_mutex
1871 * here as compared to sendfile(). We enter here
1872 * with the socket lock held, and splice_to_pipe() will
1873 * grab the pipe inode lock. For sendfile() emulation,
1874 * we call into ->sendpage() with the i_mutex lock held
1875 * and networking will grab the socket lock.
1878 ret
= splice_to_pipe(pipe
, &spd
);
1886 * skb_store_bits - store bits from kernel buffer to skb
1887 * @skb: destination buffer
1888 * @offset: offset in destination
1889 * @from: source buffer
1890 * @len: number of bytes to copy
1892 * Copy the specified number of bytes from the source buffer to the
1893 * destination skb. This function handles all the messy bits of
1894 * traversing fragment lists and such.
1897 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1899 int start
= skb_headlen(skb
);
1900 struct sk_buff
*frag_iter
;
1903 if (offset
> (int)skb
->len
- len
)
1906 if ((copy
= start
- offset
) > 0) {
1909 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1910 if ((len
-= copy
) == 0)
1916 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1917 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1920 WARN_ON(start
> offset
+ len
);
1922 end
= start
+ skb_frag_size(frag
);
1923 if ((copy
= end
- offset
) > 0) {
1929 vaddr
= kmap_atomic(skb_frag_page(frag
));
1930 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1932 kunmap_atomic(vaddr
);
1934 if ((len
-= copy
) == 0)
1942 skb_walk_frags(skb
, frag_iter
) {
1945 WARN_ON(start
> offset
+ len
);
1947 end
= start
+ frag_iter
->len
;
1948 if ((copy
= end
- offset
) > 0) {
1951 if (skb_store_bits(frag_iter
, offset
- start
,
1954 if ((len
-= copy
) == 0)
1967 EXPORT_SYMBOL(skb_store_bits
);
1969 /* Checksum skb data. */
1971 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1972 int len
, __wsum csum
)
1974 int start
= skb_headlen(skb
);
1975 int i
, copy
= start
- offset
;
1976 struct sk_buff
*frag_iter
;
1979 /* Checksum header. */
1983 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1984 if ((len
-= copy
) == 0)
1990 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1992 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1994 WARN_ON(start
> offset
+ len
);
1996 end
= start
+ skb_frag_size(frag
);
1997 if ((copy
= end
- offset
) > 0) {
2003 vaddr
= kmap_atomic(skb_frag_page(frag
));
2004 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
2005 offset
- start
, copy
, 0);
2006 kunmap_atomic(vaddr
);
2007 csum
= csum_block_add(csum
, csum2
, pos
);
2016 skb_walk_frags(skb
, frag_iter
) {
2019 WARN_ON(start
> offset
+ len
);
2021 end
= start
+ frag_iter
->len
;
2022 if ((copy
= end
- offset
) > 0) {
2026 csum2
= skb_checksum(frag_iter
, offset
- start
,
2028 csum
= csum_block_add(csum
, csum2
, pos
);
2029 if ((len
-= copy
) == 0)
2040 EXPORT_SYMBOL(skb_checksum
);
2042 /* Both of above in one bottle. */
2044 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
2045 u8
*to
, int len
, __wsum csum
)
2047 int start
= skb_headlen(skb
);
2048 int i
, copy
= start
- offset
;
2049 struct sk_buff
*frag_iter
;
2056 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
2058 if ((len
-= copy
) == 0)
2065 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2068 WARN_ON(start
> offset
+ len
);
2070 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2071 if ((copy
= end
- offset
) > 0) {
2074 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2078 vaddr
= kmap_atomic(skb_frag_page(frag
));
2079 csum2
= csum_partial_copy_nocheck(vaddr
+
2083 kunmap_atomic(vaddr
);
2084 csum
= csum_block_add(csum
, csum2
, pos
);
2094 skb_walk_frags(skb
, frag_iter
) {
2098 WARN_ON(start
> offset
+ len
);
2100 end
= start
+ frag_iter
->len
;
2101 if ((copy
= end
- offset
) > 0) {
2104 csum2
= skb_copy_and_csum_bits(frag_iter
,
2107 csum
= csum_block_add(csum
, csum2
, pos
);
2108 if ((len
-= copy
) == 0)
2119 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2121 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2126 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2127 csstart
= skb_checksum_start_offset(skb
);
2129 csstart
= skb_headlen(skb
);
2131 BUG_ON(csstart
> skb_headlen(skb
));
2133 skb_copy_from_linear_data(skb
, to
, csstart
);
2136 if (csstart
!= skb
->len
)
2137 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2138 skb
->len
- csstart
, 0);
2140 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2141 long csstuff
= csstart
+ skb
->csum_offset
;
2143 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2146 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2149 * skb_dequeue - remove from the head of the queue
2150 * @list: list to dequeue from
2152 * Remove the head of the list. The list lock is taken so the function
2153 * may be used safely with other locking list functions. The head item is
2154 * returned or %NULL if the list is empty.
2157 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2159 unsigned long flags
;
2160 struct sk_buff
*result
;
2162 spin_lock_irqsave(&list
->lock
, flags
);
2163 result
= __skb_dequeue(list
);
2164 spin_unlock_irqrestore(&list
->lock
, flags
);
2167 EXPORT_SYMBOL(skb_dequeue
);
2170 * skb_dequeue_tail - remove from the tail of the queue
2171 * @list: list to dequeue from
2173 * Remove the tail of the list. The list lock is taken so the function
2174 * may be used safely with other locking list functions. The tail item is
2175 * returned or %NULL if the list is empty.
2177 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2179 unsigned long flags
;
2180 struct sk_buff
*result
;
2182 spin_lock_irqsave(&list
->lock
, flags
);
2183 result
= __skb_dequeue_tail(list
);
2184 spin_unlock_irqrestore(&list
->lock
, flags
);
2187 EXPORT_SYMBOL(skb_dequeue_tail
);
2190 * skb_queue_purge - empty a list
2191 * @list: list to empty
2193 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2194 * the list and one reference dropped. This function takes the list
2195 * lock and is atomic with respect to other list locking functions.
2197 void skb_queue_purge(struct sk_buff_head
*list
)
2199 struct sk_buff
*skb
;
2200 while ((skb
= skb_dequeue(list
)) != NULL
)
2203 EXPORT_SYMBOL(skb_queue_purge
);
2206 * skb_queue_head - queue a buffer at the list head
2207 * @list: list to use
2208 * @newsk: buffer to queue
2210 * Queue a buffer at the start of the list. This function takes the
2211 * list lock and can be used safely with other locking &sk_buff functions
2214 * A buffer cannot be placed on two lists at the same time.
2216 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2218 unsigned long flags
;
2220 spin_lock_irqsave(&list
->lock
, flags
);
2221 __skb_queue_head(list
, newsk
);
2222 spin_unlock_irqrestore(&list
->lock
, flags
);
2224 EXPORT_SYMBOL(skb_queue_head
);
2227 * skb_queue_tail - queue a buffer at the list tail
2228 * @list: list to use
2229 * @newsk: buffer to queue
2231 * Queue a buffer at the tail of the list. This function takes the
2232 * list lock and can be used safely with other locking &sk_buff functions
2235 * A buffer cannot be placed on two lists at the same time.
2237 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2239 unsigned long flags
;
2241 spin_lock_irqsave(&list
->lock
, flags
);
2242 __skb_queue_tail(list
, newsk
);
2243 spin_unlock_irqrestore(&list
->lock
, flags
);
2245 EXPORT_SYMBOL(skb_queue_tail
);
2248 * skb_unlink - remove a buffer from a list
2249 * @skb: buffer to remove
2250 * @list: list to use
2252 * Remove a packet from a list. The list locks are taken and this
2253 * function is atomic with respect to other list locked calls
2255 * You must know what list the SKB is on.
2257 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2259 unsigned long flags
;
2261 spin_lock_irqsave(&list
->lock
, flags
);
2262 __skb_unlink(skb
, list
);
2263 spin_unlock_irqrestore(&list
->lock
, flags
);
2265 EXPORT_SYMBOL(skb_unlink
);
2268 * skb_append - append a buffer
2269 * @old: buffer to insert after
2270 * @newsk: buffer to insert
2271 * @list: list to use
2273 * Place a packet after a given packet in a list. The list locks are taken
2274 * and this function is atomic with respect to other list locked calls.
2275 * A buffer cannot be placed on two lists at the same time.
2277 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2279 unsigned long flags
;
2281 spin_lock_irqsave(&list
->lock
, flags
);
2282 __skb_queue_after(list
, old
, newsk
);
2283 spin_unlock_irqrestore(&list
->lock
, flags
);
2285 EXPORT_SYMBOL(skb_append
);
2288 * skb_insert - insert a buffer
2289 * @old: buffer to insert before
2290 * @newsk: buffer to insert
2291 * @list: list to use
2293 * Place a packet before a given packet in a list. The list locks are
2294 * taken and this function is atomic with respect to other list locked
2297 * A buffer cannot be placed on two lists at the same time.
2299 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2301 unsigned long flags
;
2303 spin_lock_irqsave(&list
->lock
, flags
);
2304 __skb_insert(newsk
, old
->prev
, old
, list
);
2305 spin_unlock_irqrestore(&list
->lock
, flags
);
2307 EXPORT_SYMBOL(skb_insert
);
2309 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2310 struct sk_buff
* skb1
,
2311 const u32 len
, const int pos
)
2315 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2317 /* And move data appendix as is. */
2318 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2319 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2321 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2322 skb_shinfo(skb
)->nr_frags
= 0;
2323 skb1
->data_len
= skb
->data_len
;
2324 skb1
->len
+= skb1
->data_len
;
2327 skb_set_tail_pointer(skb
, len
);
2330 static inline void skb_split_no_header(struct sk_buff
*skb
,
2331 struct sk_buff
* skb1
,
2332 const u32 len
, int pos
)
2335 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2337 skb_shinfo(skb
)->nr_frags
= 0;
2338 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2340 skb
->data_len
= len
- pos
;
2342 for (i
= 0; i
< nfrags
; i
++) {
2343 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2345 if (pos
+ size
> len
) {
2346 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2350 * We have two variants in this case:
2351 * 1. Move all the frag to the second
2352 * part, if it is possible. F.e.
2353 * this approach is mandatory for TUX,
2354 * where splitting is expensive.
2355 * 2. Split is accurately. We make this.
2357 skb_frag_ref(skb
, i
);
2358 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2359 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2360 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2361 skb_shinfo(skb
)->nr_frags
++;
2365 skb_shinfo(skb
)->nr_frags
++;
2368 skb_shinfo(skb1
)->nr_frags
= k
;
2372 * skb_split - Split fragmented skb to two parts at length len.
2373 * @skb: the buffer to split
2374 * @skb1: the buffer to receive the second part
2375 * @len: new length for skb
2377 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2379 int pos
= skb_headlen(skb
);
2381 if (len
< pos
) /* Split line is inside header. */
2382 skb_split_inside_header(skb
, skb1
, len
, pos
);
2383 else /* Second chunk has no header, nothing to copy. */
2384 skb_split_no_header(skb
, skb1
, len
, pos
);
2386 EXPORT_SYMBOL(skb_split
);
2388 /* Shifting from/to a cloned skb is a no-go.
2390 * Caller cannot keep skb_shinfo related pointers past calling here!
2392 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2394 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2398 * skb_shift - Shifts paged data partially from skb to another
2399 * @tgt: buffer into which tail data gets added
2400 * @skb: buffer from which the paged data comes from
2401 * @shiftlen: shift up to this many bytes
2403 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2404 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2405 * It's up to caller to free skb if everything was shifted.
2407 * If @tgt runs out of frags, the whole operation is aborted.
2409 * Skb cannot include anything else but paged data while tgt is allowed
2410 * to have non-paged data as well.
2412 * TODO: full sized shift could be optimized but that would need
2413 * specialized skb free'er to handle frags without up-to-date nr_frags.
2415 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2417 int from
, to
, merge
, todo
;
2418 struct skb_frag_struct
*fragfrom
, *fragto
;
2420 BUG_ON(shiftlen
> skb
->len
);
2421 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2425 to
= skb_shinfo(tgt
)->nr_frags
;
2426 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2428 /* Actual merge is delayed until the point when we know we can
2429 * commit all, so that we don't have to undo partial changes
2432 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2433 fragfrom
->page_offset
)) {
2438 todo
-= skb_frag_size(fragfrom
);
2440 if (skb_prepare_for_shift(skb
) ||
2441 skb_prepare_for_shift(tgt
))
2444 /* All previous frag pointers might be stale! */
2445 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2446 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2448 skb_frag_size_add(fragto
, shiftlen
);
2449 skb_frag_size_sub(fragfrom
, shiftlen
);
2450 fragfrom
->page_offset
+= shiftlen
;
2458 /* Skip full, not-fitting skb to avoid expensive operations */
2459 if ((shiftlen
== skb
->len
) &&
2460 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2463 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2466 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2467 if (to
== MAX_SKB_FRAGS
)
2470 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2471 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2473 if (todo
>= skb_frag_size(fragfrom
)) {
2474 *fragto
= *fragfrom
;
2475 todo
-= skb_frag_size(fragfrom
);
2480 __skb_frag_ref(fragfrom
);
2481 fragto
->page
= fragfrom
->page
;
2482 fragto
->page_offset
= fragfrom
->page_offset
;
2483 skb_frag_size_set(fragto
, todo
);
2485 fragfrom
->page_offset
+= todo
;
2486 skb_frag_size_sub(fragfrom
, todo
);
2494 /* Ready to "commit" this state change to tgt */
2495 skb_shinfo(tgt
)->nr_frags
= to
;
2498 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2499 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2501 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2502 __skb_frag_unref(fragfrom
);
2505 /* Reposition in the original skb */
2507 while (from
< skb_shinfo(skb
)->nr_frags
)
2508 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2509 skb_shinfo(skb
)->nr_frags
= to
;
2511 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2514 /* Most likely the tgt won't ever need its checksum anymore, skb on
2515 * the other hand might need it if it needs to be resent
2517 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2518 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2520 /* Yak, is it really working this way? Some helper please? */
2521 skb
->len
-= shiftlen
;
2522 skb
->data_len
-= shiftlen
;
2523 skb
->truesize
-= shiftlen
;
2524 tgt
->len
+= shiftlen
;
2525 tgt
->data_len
+= shiftlen
;
2526 tgt
->truesize
+= shiftlen
;
2532 * skb_prepare_seq_read - Prepare a sequential read of skb data
2533 * @skb: the buffer to read
2534 * @from: lower offset of data to be read
2535 * @to: upper offset of data to be read
2536 * @st: state variable
2538 * Initializes the specified state variable. Must be called before
2539 * invoking skb_seq_read() for the first time.
2541 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2542 unsigned int to
, struct skb_seq_state
*st
)
2544 st
->lower_offset
= from
;
2545 st
->upper_offset
= to
;
2546 st
->root_skb
= st
->cur_skb
= skb
;
2547 st
->frag_idx
= st
->stepped_offset
= 0;
2548 st
->frag_data
= NULL
;
2550 EXPORT_SYMBOL(skb_prepare_seq_read
);
2553 * skb_seq_read - Sequentially read skb data
2554 * @consumed: number of bytes consumed by the caller so far
2555 * @data: destination pointer for data to be returned
2556 * @st: state variable
2558 * Reads a block of skb data at &consumed relative to the
2559 * lower offset specified to skb_prepare_seq_read(). Assigns
2560 * the head of the data block to &data and returns the length
2561 * of the block or 0 if the end of the skb data or the upper
2562 * offset has been reached.
2564 * The caller is not required to consume all of the data
2565 * returned, i.e. &consumed is typically set to the number
2566 * of bytes already consumed and the next call to
2567 * skb_seq_read() will return the remaining part of the block.
2569 * Note 1: The size of each block of data returned can be arbitrary,
2570 * this limitation is the cost for zerocopy seqeuental
2571 * reads of potentially non linear data.
2573 * Note 2: Fragment lists within fragments are not implemented
2574 * at the moment, state->root_skb could be replaced with
2575 * a stack for this purpose.
2577 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2578 struct skb_seq_state
*st
)
2580 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2583 if (unlikely(abs_offset
>= st
->upper_offset
))
2587 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2589 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2590 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2591 return block_limit
- abs_offset
;
2594 if (st
->frag_idx
== 0 && !st
->frag_data
)
2595 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2597 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2598 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2599 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2601 if (abs_offset
< block_limit
) {
2603 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2605 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2606 (abs_offset
- st
->stepped_offset
);
2608 return block_limit
- abs_offset
;
2611 if (st
->frag_data
) {
2612 kunmap_atomic(st
->frag_data
);
2613 st
->frag_data
= NULL
;
2617 st
->stepped_offset
+= skb_frag_size(frag
);
2620 if (st
->frag_data
) {
2621 kunmap_atomic(st
->frag_data
);
2622 st
->frag_data
= NULL
;
2625 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2626 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2629 } else if (st
->cur_skb
->next
) {
2630 st
->cur_skb
= st
->cur_skb
->next
;
2637 EXPORT_SYMBOL(skb_seq_read
);
2640 * skb_abort_seq_read - Abort a sequential read of skb data
2641 * @st: state variable
2643 * Must be called if skb_seq_read() was not called until it
2646 void skb_abort_seq_read(struct skb_seq_state
*st
)
2649 kunmap_atomic(st
->frag_data
);
2651 EXPORT_SYMBOL(skb_abort_seq_read
);
2653 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2655 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2656 struct ts_config
*conf
,
2657 struct ts_state
*state
)
2659 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2662 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2664 skb_abort_seq_read(TS_SKB_CB(state
));
2668 * skb_find_text - Find a text pattern in skb data
2669 * @skb: the buffer to look in
2670 * @from: search offset
2672 * @config: textsearch configuration
2673 * @state: uninitialized textsearch state variable
2675 * Finds a pattern in the skb data according to the specified
2676 * textsearch configuration. Use textsearch_next() to retrieve
2677 * subsequent occurrences of the pattern. Returns the offset
2678 * to the first occurrence or UINT_MAX if no match was found.
2680 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2681 unsigned int to
, struct ts_config
*config
,
2682 struct ts_state
*state
)
2686 config
->get_next_block
= skb_ts_get_next_block
;
2687 config
->finish
= skb_ts_finish
;
2689 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2691 ret
= textsearch_find(config
, state
);
2692 return (ret
<= to
- from
? ret
: UINT_MAX
);
2694 EXPORT_SYMBOL(skb_find_text
);
2697 * skb_append_datato_frags - append the user data to a skb
2698 * @sk: sock structure
2699 * @skb: skb structure to be appened with user data.
2700 * @getfrag: call back function to be used for getting the user data
2701 * @from: pointer to user message iov
2702 * @length: length of the iov message
2704 * Description: This procedure append the user data in the fragment part
2705 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2707 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2708 int (*getfrag
)(void *from
, char *to
, int offset
,
2709 int len
, int odd
, struct sk_buff
*skb
),
2710 void *from
, int length
)
2713 skb_frag_t
*frag
= NULL
;
2714 struct page
*page
= NULL
;
2720 /* Return error if we don't have space for new frag */
2721 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2722 if (frg_cnt
>= MAX_SKB_FRAGS
)
2725 /* allocate a new page for next frag */
2726 page
= alloc_pages(sk
->sk_allocation
, 0);
2728 /* If alloc_page fails just return failure and caller will
2729 * free previous allocated pages by doing kfree_skb()
2734 /* initialize the next frag */
2735 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2736 skb
->truesize
+= PAGE_SIZE
;
2737 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2739 /* get the new initialized frag */
2740 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2741 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2743 /* copy the user data to page */
2744 left
= PAGE_SIZE
- frag
->page_offset
;
2745 copy
= (length
> left
)? left
: length
;
2747 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2748 offset
, copy
, 0, skb
);
2752 /* copy was successful so update the size parameters */
2753 skb_frag_size_add(frag
, copy
);
2755 skb
->data_len
+= copy
;
2759 } while (length
> 0);
2763 EXPORT_SYMBOL(skb_append_datato_frags
);
2766 * skb_pull_rcsum - pull skb and update receive checksum
2767 * @skb: buffer to update
2768 * @len: length of data pulled
2770 * This function performs an skb_pull on the packet and updates
2771 * the CHECKSUM_COMPLETE checksum. It should be used on
2772 * receive path processing instead of skb_pull unless you know
2773 * that the checksum difference is zero (e.g., a valid IP header)
2774 * or you are setting ip_summed to CHECKSUM_NONE.
2776 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2778 BUG_ON(len
> skb
->len
);
2780 BUG_ON(skb
->len
< skb
->data_len
);
2781 skb_postpull_rcsum(skb
, skb
->data
, len
);
2782 return skb
->data
+= len
;
2784 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2787 * skb_segment - Perform protocol segmentation on skb.
2788 * @skb: buffer to segment
2789 * @features: features for the output path (see dev->features)
2791 * This function performs segmentation on the given skb. It returns
2792 * a pointer to the first in a list of new skbs for the segments.
2793 * In case of error it returns ERR_PTR(err).
2795 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2797 struct sk_buff
*segs
= NULL
;
2798 struct sk_buff
*tail
= NULL
;
2799 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2800 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2801 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2802 unsigned int offset
= doffset
;
2803 unsigned int headroom
;
2805 int sg
= !!(features
& NETIF_F_SG
);
2806 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2811 __skb_push(skb
, doffset
);
2812 headroom
= skb_headroom(skb
);
2813 pos
= skb_headlen(skb
);
2816 struct sk_buff
*nskb
;
2821 len
= skb
->len
- offset
;
2825 hsize
= skb_headlen(skb
) - offset
;
2828 if (hsize
> len
|| !sg
)
2831 if (!hsize
&& i
>= nfrags
) {
2832 BUG_ON(fskb
->len
!= len
);
2835 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2838 if (unlikely(!nskb
))
2841 hsize
= skb_end_offset(nskb
);
2842 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2847 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2848 skb_release_head_state(nskb
);
2849 __skb_push(nskb
, doffset
);
2851 nskb
= __alloc_skb(hsize
+ doffset
+ headroom
,
2852 GFP_ATOMIC
, skb_alloc_rx_flag(skb
),
2855 if (unlikely(!nskb
))
2858 skb_reserve(nskb
, headroom
);
2859 __skb_put(nskb
, doffset
);
2868 __copy_skb_header(nskb
, skb
);
2869 nskb
->mac_len
= skb
->mac_len
;
2871 /* nskb and skb might have different headroom */
2872 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2873 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2875 skb_reset_mac_header(nskb
);
2876 skb_set_network_header(nskb
, skb
->mac_len
);
2877 nskb
->transport_header
= (nskb
->network_header
+
2878 skb_network_header_len(skb
));
2879 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2881 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2885 nskb
->ip_summed
= CHECKSUM_NONE
;
2886 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2892 frag
= skb_shinfo(nskb
)->frags
;
2894 skb_copy_from_linear_data_offset(skb
, offset
,
2895 skb_put(nskb
, hsize
), hsize
);
2897 while (pos
< offset
+ len
&& i
< nfrags
) {
2898 *frag
= skb_shinfo(skb
)->frags
[i
];
2899 __skb_frag_ref(frag
);
2900 size
= skb_frag_size(frag
);
2903 frag
->page_offset
+= offset
- pos
;
2904 skb_frag_size_sub(frag
, offset
- pos
);
2907 skb_shinfo(nskb
)->nr_frags
++;
2909 if (pos
+ size
<= offset
+ len
) {
2913 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2920 if (pos
< offset
+ len
) {
2921 struct sk_buff
*fskb2
= fskb
;
2923 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2929 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2935 SKB_FRAG_ASSERT(nskb
);
2936 skb_shinfo(nskb
)->frag_list
= fskb2
;
2940 nskb
->data_len
= len
- hsize
;
2941 nskb
->len
+= nskb
->data_len
;
2942 nskb
->truesize
+= nskb
->data_len
;
2943 } while ((offset
+= len
) < skb
->len
);
2948 while ((skb
= segs
)) {
2952 return ERR_PTR(err
);
2954 EXPORT_SYMBOL_GPL(skb_segment
);
2956 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2958 struct sk_buff
*p
= *head
;
2959 struct sk_buff
*nskb
;
2960 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2961 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2962 unsigned int headroom
;
2963 unsigned int len
= skb_gro_len(skb
);
2964 unsigned int offset
= skb_gro_offset(skb
);
2965 unsigned int headlen
= skb_headlen(skb
);
2966 unsigned int delta_truesize
;
2968 if (p
->len
+ len
>= 65536)
2971 if (pinfo
->frag_list
)
2973 else if (headlen
<= offset
) {
2976 int i
= skbinfo
->nr_frags
;
2977 int nr_frags
= pinfo
->nr_frags
+ i
;
2981 if (nr_frags
> MAX_SKB_FRAGS
)
2984 pinfo
->nr_frags
= nr_frags
;
2985 skbinfo
->nr_frags
= 0;
2987 frag
= pinfo
->frags
+ nr_frags
;
2988 frag2
= skbinfo
->frags
+ i
;
2993 frag
->page_offset
+= offset
;
2994 skb_frag_size_sub(frag
, offset
);
2996 /* all fragments truesize : remove (head size + sk_buff) */
2997 delta_truesize
= skb
->truesize
-
2998 SKB_TRUESIZE(skb_end_offset(skb
));
3000 skb
->truesize
-= skb
->data_len
;
3001 skb
->len
-= skb
->data_len
;
3004 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
3006 } else if (skb
->head_frag
) {
3007 int nr_frags
= pinfo
->nr_frags
;
3008 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
3009 struct page
*page
= virt_to_head_page(skb
->head
);
3010 unsigned int first_size
= headlen
- offset
;
3011 unsigned int first_offset
;
3013 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
3016 first_offset
= skb
->data
-
3017 (unsigned char *)page_address(page
) +
3020 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
3022 frag
->page
.p
= page
;
3023 frag
->page_offset
= first_offset
;
3024 skb_frag_size_set(frag
, first_size
);
3026 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
3027 /* We dont need to clear skbinfo->nr_frags here */
3029 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3030 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
3032 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
3035 headroom
= skb_headroom(p
);
3036 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
3037 if (unlikely(!nskb
))
3040 __copy_skb_header(nskb
, p
);
3041 nskb
->mac_len
= p
->mac_len
;
3043 skb_reserve(nskb
, headroom
);
3044 __skb_put(nskb
, skb_gro_offset(p
));
3046 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
3047 skb_set_network_header(nskb
, skb_network_offset(p
));
3048 skb_set_transport_header(nskb
, skb_transport_offset(p
));
3050 __skb_pull(p
, skb_gro_offset(p
));
3051 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
3052 p
->data
- skb_mac_header(p
));
3054 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
3055 skb_shinfo(nskb
)->frag_list
= p
;
3056 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
3057 pinfo
->gso_size
= 0;
3058 skb_header_release(p
);
3061 nskb
->data_len
+= p
->len
;
3062 nskb
->truesize
+= p
->truesize
;
3063 nskb
->len
+= p
->len
;
3066 nskb
->next
= p
->next
;
3072 delta_truesize
= skb
->truesize
;
3073 if (offset
> headlen
) {
3074 unsigned int eat
= offset
- headlen
;
3076 skbinfo
->frags
[0].page_offset
+= eat
;
3077 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3078 skb
->data_len
-= eat
;
3083 __skb_pull(skb
, offset
);
3085 p
->prev
->next
= skb
;
3087 skb_header_release(skb
);
3090 NAPI_GRO_CB(p
)->count
++;
3092 p
->truesize
+= delta_truesize
;
3095 NAPI_GRO_CB(skb
)->same_flow
= 1;
3098 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3100 void __init
skb_init(void)
3102 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3103 sizeof(struct sk_buff
),
3105 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3107 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3108 (2*sizeof(struct sk_buff
)) +
3111 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3116 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3117 * @skb: Socket buffer containing the buffers to be mapped
3118 * @sg: The scatter-gather list to map into
3119 * @offset: The offset into the buffer's contents to start mapping
3120 * @len: Length of buffer space to be mapped
3122 * Fill the specified scatter-gather list with mappings/pointers into a
3123 * region of the buffer space attached to a socket buffer.
3126 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3128 int start
= skb_headlen(skb
);
3129 int i
, copy
= start
- offset
;
3130 struct sk_buff
*frag_iter
;
3136 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3138 if ((len
-= copy
) == 0)
3143 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3146 WARN_ON(start
> offset
+ len
);
3148 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3149 if ((copy
= end
- offset
) > 0) {
3150 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3154 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3155 frag
->page_offset
+offset
-start
);
3164 skb_walk_frags(skb
, frag_iter
) {
3167 WARN_ON(start
> offset
+ len
);
3169 end
= start
+ frag_iter
->len
;
3170 if ((copy
= end
- offset
) > 0) {
3173 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3175 if ((len
-= copy
) == 0)
3185 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3187 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3189 sg_mark_end(&sg
[nsg
- 1]);
3193 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3196 * skb_cow_data - Check that a socket buffer's data buffers are writable
3197 * @skb: The socket buffer to check.
3198 * @tailbits: Amount of trailing space to be added
3199 * @trailer: Returned pointer to the skb where the @tailbits space begins
3201 * Make sure that the data buffers attached to a socket buffer are
3202 * writable. If they are not, private copies are made of the data buffers
3203 * and the socket buffer is set to use these instead.
3205 * If @tailbits is given, make sure that there is space to write @tailbits
3206 * bytes of data beyond current end of socket buffer. @trailer will be
3207 * set to point to the skb in which this space begins.
3209 * The number of scatterlist elements required to completely map the
3210 * COW'd and extended socket buffer will be returned.
3212 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3216 struct sk_buff
*skb1
, **skb_p
;
3218 /* If skb is cloned or its head is paged, reallocate
3219 * head pulling out all the pages (pages are considered not writable
3220 * at the moment even if they are anonymous).
3222 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3223 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3226 /* Easy case. Most of packets will go this way. */
3227 if (!skb_has_frag_list(skb
)) {
3228 /* A little of trouble, not enough of space for trailer.
3229 * This should not happen, when stack is tuned to generate
3230 * good frames. OK, on miss we reallocate and reserve even more
3231 * space, 128 bytes is fair. */
3233 if (skb_tailroom(skb
) < tailbits
&&
3234 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3242 /* Misery. We are in troubles, going to mincer fragments... */
3245 skb_p
= &skb_shinfo(skb
)->frag_list
;
3248 while ((skb1
= *skb_p
) != NULL
) {
3251 /* The fragment is partially pulled by someone,
3252 * this can happen on input. Copy it and everything
3255 if (skb_shared(skb1
))
3258 /* If the skb is the last, worry about trailer. */
3260 if (skb1
->next
== NULL
&& tailbits
) {
3261 if (skb_shinfo(skb1
)->nr_frags
||
3262 skb_has_frag_list(skb1
) ||
3263 skb_tailroom(skb1
) < tailbits
)
3264 ntail
= tailbits
+ 128;
3270 skb_shinfo(skb1
)->nr_frags
||
3271 skb_has_frag_list(skb1
)) {
3272 struct sk_buff
*skb2
;
3274 /* Fuck, we are miserable poor guys... */
3276 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3278 skb2
= skb_copy_expand(skb1
,
3282 if (unlikely(skb2
== NULL
))
3286 skb_set_owner_w(skb2
, skb1
->sk
);
3288 /* Looking around. Are we still alive?
3289 * OK, link new skb, drop old one */
3291 skb2
->next
= skb1
->next
;
3298 skb_p
= &skb1
->next
;
3303 EXPORT_SYMBOL_GPL(skb_cow_data
);
3305 static void sock_rmem_free(struct sk_buff
*skb
)
3307 struct sock
*sk
= skb
->sk
;
3309 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3313 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3315 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3319 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3320 (unsigned int)sk
->sk_rcvbuf
)
3325 skb
->destructor
= sock_rmem_free
;
3326 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3328 /* before exiting rcu section, make sure dst is refcounted */
3331 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3332 if (!sock_flag(sk
, SOCK_DEAD
))
3333 sk
->sk_data_ready(sk
, len
);
3336 EXPORT_SYMBOL(sock_queue_err_skb
);
3338 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3339 struct skb_shared_hwtstamps
*hwtstamps
)
3341 struct sock
*sk
= orig_skb
->sk
;
3342 struct sock_exterr_skb
*serr
;
3343 struct sk_buff
*skb
;
3349 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3354 *skb_hwtstamps(skb
) =
3358 * no hardware time stamps available,
3359 * so keep the shared tx_flags and only
3360 * store software time stamp
3362 skb
->tstamp
= ktime_get_real();
3365 serr
= SKB_EXT_ERR(skb
);
3366 memset(serr
, 0, sizeof(*serr
));
3367 serr
->ee
.ee_errno
= ENOMSG
;
3368 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3370 err
= sock_queue_err_skb(sk
, skb
);
3375 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3377 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3379 struct sock
*sk
= skb
->sk
;
3380 struct sock_exterr_skb
*serr
;
3383 skb
->wifi_acked_valid
= 1;
3384 skb
->wifi_acked
= acked
;
3386 serr
= SKB_EXT_ERR(skb
);
3387 memset(serr
, 0, sizeof(*serr
));
3388 serr
->ee
.ee_errno
= ENOMSG
;
3389 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3391 err
= sock_queue_err_skb(sk
, skb
);
3395 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3399 * skb_partial_csum_set - set up and verify partial csum values for packet
3400 * @skb: the skb to set
3401 * @start: the number of bytes after skb->data to start checksumming.
3402 * @off: the offset from start to place the checksum.
3404 * For untrusted partially-checksummed packets, we need to make sure the values
3405 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3407 * This function checks and sets those values and skb->ip_summed: if this
3408 * returns false you should drop the packet.
3410 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3412 if (unlikely(start
> skb_headlen(skb
)) ||
3413 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3414 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3415 start
, off
, skb_headlen(skb
));
3418 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3419 skb
->csum_start
= skb_headroom(skb
) + start
;
3420 skb
->csum_offset
= off
;
3423 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3425 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3427 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3430 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3432 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3435 kmem_cache_free(skbuff_head_cache
, skb
);
3439 EXPORT_SYMBOL(kfree_skb_partial
);
3442 * skb_try_coalesce - try to merge skb to prior one
3444 * @from: buffer to add
3445 * @fragstolen: pointer to boolean
3446 * @delta_truesize: how much more was allocated than was requested
3448 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3449 bool *fragstolen
, int *delta_truesize
)
3451 int i
, delta
, len
= from
->len
;
3453 *fragstolen
= false;
3458 if (len
<= skb_tailroom(to
)) {
3459 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3460 *delta_truesize
= 0;
3464 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3467 if (skb_headlen(from
) != 0) {
3469 unsigned int offset
;
3471 if (skb_shinfo(to
)->nr_frags
+
3472 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3475 if (skb_head_is_locked(from
))
3478 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3480 page
= virt_to_head_page(from
->head
);
3481 offset
= from
->data
- (unsigned char *)page_address(page
);
3483 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3484 page
, offset
, skb_headlen(from
));
3487 if (skb_shinfo(to
)->nr_frags
+
3488 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3491 delta
= from
->truesize
-
3492 SKB_TRUESIZE(skb_end_pointer(from
) - from
->head
);
3495 WARN_ON_ONCE(delta
< len
);
3497 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3498 skb_shinfo(from
)->frags
,
3499 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3500 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3502 if (!skb_cloned(from
))
3503 skb_shinfo(from
)->nr_frags
= 0;
3505 /* if the skb is not cloned this does nothing
3506 * since we set nr_frags to 0.
3508 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3509 skb_frag_ref(from
, i
);
3511 to
->truesize
+= delta
;
3513 to
->data_len
+= len
;
3515 *delta_truesize
= delta
;
3518 EXPORT_SYMBOL(skb_try_coalesce
);