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>");
148 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of at least size bytes. The object has a reference count
164 * of one. The return is the buffer. On a failure the return is %NULL.
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
170 int fclone
, int node
)
172 struct kmem_cache
*cache
;
173 struct skb_shared_info
*shinfo
;
177 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
180 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
190 size
= SKB_DATA_ALIGN(size
);
191 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
192 data
= kmalloc_node_track_caller(size
, gfp_mask
, node
);
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
199 size
= SKB_WITH_OVERHEAD(ksize(data
));
200 prefetchw(data
+ size
);
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
207 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
208 /* Account for allocated memory : skb + skb->head */
209 skb
->truesize
= SKB_TRUESIZE(size
);
210 atomic_set(&skb
->users
, 1);
213 skb_reset_tail_pointer(skb
);
214 skb
->end
= skb
->tail
+ size
;
215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb
->mac_header
= ~0U;
219 /* make sure we initialize shinfo sequentially */
220 shinfo
= skb_shinfo(skb
);
221 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
222 atomic_set(&shinfo
->dataref
, 1);
223 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
226 struct sk_buff
*child
= skb
+ 1;
227 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
229 kmemcheck_annotate_bitfield(child
, flags1
);
230 kmemcheck_annotate_bitfield(child
, flags2
);
231 skb
->fclone
= SKB_FCLONE_ORIG
;
232 atomic_set(fclone_ref
, 1);
234 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
239 kmem_cache_free(cache
, skb
);
243 EXPORT_SYMBOL(__alloc_skb
);
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
248 * @frag_size: size of fragment, or 0 if head was kmalloced
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
262 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
264 struct skb_shared_info
*shinfo
;
266 unsigned int size
= frag_size
? : ksize(data
);
268 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
272 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
274 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
275 skb
->truesize
= SKB_TRUESIZE(size
);
276 skb
->head_frag
= frag_size
!= 0;
277 atomic_set(&skb
->users
, 1);
280 skb_reset_tail_pointer(skb
);
281 skb
->end
= skb
->tail
+ size
;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb
->mac_header
= ~0U;
286 /* make sure we initialize shinfo sequentially */
287 shinfo
= skb_shinfo(skb
);
288 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
289 atomic_set(&shinfo
->dataref
, 1);
290 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
294 EXPORT_SYMBOL(build_skb
);
296 struct netdev_alloc_cache
{
299 unsigned int pagecnt_bias
;
301 static DEFINE_PER_CPU(struct netdev_alloc_cache
, netdev_alloc_cache
);
303 #define NETDEV_PAGECNT_BIAS (PAGE_SIZE / SMP_CACHE_BYTES)
306 * netdev_alloc_frag - allocate a page fragment
307 * @fragsz: fragment size
309 * Allocates a frag from a page for receive buffer.
310 * Uses GFP_ATOMIC allocations.
312 void *netdev_alloc_frag(unsigned int fragsz
)
314 struct netdev_alloc_cache
*nc
;
318 local_irq_save(flags
);
319 nc
= &__get_cpu_var(netdev_alloc_cache
);
320 if (unlikely(!nc
->page
)) {
322 nc
->page
= alloc_page(GFP_ATOMIC
| __GFP_COLD
);
323 if (unlikely(!nc
->page
))
326 atomic_set(&nc
->page
->_count
, NETDEV_PAGECNT_BIAS
);
327 nc
->pagecnt_bias
= NETDEV_PAGECNT_BIAS
;
331 if (nc
->offset
+ fragsz
> PAGE_SIZE
) {
332 /* avoid unnecessary locked operations if possible */
333 if ((atomic_read(&nc
->page
->_count
) == nc
->pagecnt_bias
) ||
334 atomic_sub_and_test(nc
->pagecnt_bias
, &nc
->page
->_count
))
339 data
= page_address(nc
->page
) + nc
->offset
;
340 nc
->offset
+= fragsz
;
343 local_irq_restore(flags
);
346 EXPORT_SYMBOL(netdev_alloc_frag
);
349 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
350 * @dev: network device to receive on
351 * @length: length to allocate
352 * @gfp_mask: get_free_pages mask, passed to alloc_skb
354 * Allocate a new &sk_buff and assign it a usage count of one. The
355 * buffer has unspecified headroom built in. Users should allocate
356 * the headroom they think they need without accounting for the
357 * built in space. The built in space is used for optimisations.
359 * %NULL is returned if there is no free memory.
361 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
362 unsigned int length
, gfp_t gfp_mask
)
364 struct sk_buff
*skb
= NULL
;
365 unsigned int fragsz
= SKB_DATA_ALIGN(length
+ NET_SKB_PAD
) +
366 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
368 if (fragsz
<= PAGE_SIZE
&& !(gfp_mask
& (__GFP_WAIT
| GFP_DMA
))) {
369 void *data
= netdev_alloc_frag(fragsz
);
372 skb
= build_skb(data
, fragsz
);
374 put_page(virt_to_head_page(data
));
377 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
380 skb_reserve(skb
, NET_SKB_PAD
);
385 EXPORT_SYMBOL(__netdev_alloc_skb
);
387 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
388 int size
, unsigned int truesize
)
390 skb_fill_page_desc(skb
, i
, page
, off
, size
);
392 skb
->data_len
+= size
;
393 skb
->truesize
+= truesize
;
395 EXPORT_SYMBOL(skb_add_rx_frag
);
397 static void skb_drop_list(struct sk_buff
**listp
)
399 struct sk_buff
*list
= *listp
;
404 struct sk_buff
*this = list
;
410 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
412 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
415 static void skb_clone_fraglist(struct sk_buff
*skb
)
417 struct sk_buff
*list
;
419 skb_walk_frags(skb
, list
)
423 static void skb_free_head(struct sk_buff
*skb
)
426 put_page(virt_to_head_page(skb
->head
));
431 static void skb_release_data(struct sk_buff
*skb
)
434 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
435 &skb_shinfo(skb
)->dataref
)) {
436 if (skb_shinfo(skb
)->nr_frags
) {
438 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
439 skb_frag_unref(skb
, i
);
443 * If skb buf is from userspace, we need to notify the caller
444 * the lower device DMA has done;
446 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
447 struct ubuf_info
*uarg
;
449 uarg
= skb_shinfo(skb
)->destructor_arg
;
451 uarg
->callback(uarg
);
454 if (skb_has_frag_list(skb
))
455 skb_drop_fraglist(skb
);
462 * Free an skbuff by memory without cleaning the state.
464 static void kfree_skbmem(struct sk_buff
*skb
)
466 struct sk_buff
*other
;
467 atomic_t
*fclone_ref
;
469 switch (skb
->fclone
) {
470 case SKB_FCLONE_UNAVAILABLE
:
471 kmem_cache_free(skbuff_head_cache
, skb
);
474 case SKB_FCLONE_ORIG
:
475 fclone_ref
= (atomic_t
*) (skb
+ 2);
476 if (atomic_dec_and_test(fclone_ref
))
477 kmem_cache_free(skbuff_fclone_cache
, skb
);
480 case SKB_FCLONE_CLONE
:
481 fclone_ref
= (atomic_t
*) (skb
+ 1);
484 /* The clone portion is available for
485 * fast-cloning again.
487 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
489 if (atomic_dec_and_test(fclone_ref
))
490 kmem_cache_free(skbuff_fclone_cache
, other
);
495 static void skb_release_head_state(struct sk_buff
*skb
)
499 secpath_put(skb
->sp
);
501 if (skb
->destructor
) {
503 skb
->destructor(skb
);
505 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
506 nf_conntrack_put(skb
->nfct
);
508 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
509 nf_conntrack_put_reasm(skb
->nfct_reasm
);
511 #ifdef CONFIG_BRIDGE_NETFILTER
512 nf_bridge_put(skb
->nf_bridge
);
514 /* XXX: IS this still necessary? - JHS */
515 #ifdef CONFIG_NET_SCHED
517 #ifdef CONFIG_NET_CLS_ACT
523 /* Free everything but the sk_buff shell. */
524 static void skb_release_all(struct sk_buff
*skb
)
526 skb_release_head_state(skb
);
527 skb_release_data(skb
);
531 * __kfree_skb - private function
534 * Free an sk_buff. Release anything attached to the buffer.
535 * Clean the state. This is an internal helper function. Users should
536 * always call kfree_skb
539 void __kfree_skb(struct sk_buff
*skb
)
541 skb_release_all(skb
);
544 EXPORT_SYMBOL(__kfree_skb
);
547 * kfree_skb - free an sk_buff
548 * @skb: buffer to free
550 * Drop a reference to the buffer and free it if the usage count has
553 void kfree_skb(struct sk_buff
*skb
)
557 if (likely(atomic_read(&skb
->users
) == 1))
559 else if (likely(!atomic_dec_and_test(&skb
->users
)))
561 trace_kfree_skb(skb
, __builtin_return_address(0));
564 EXPORT_SYMBOL(kfree_skb
);
567 * consume_skb - free an skbuff
568 * @skb: buffer to free
570 * Drop a ref to the buffer and free it if the usage count has hit zero
571 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
572 * is being dropped after a failure and notes that
574 void consume_skb(struct sk_buff
*skb
)
578 if (likely(atomic_read(&skb
->users
) == 1))
580 else if (likely(!atomic_dec_and_test(&skb
->users
)))
582 trace_consume_skb(skb
);
585 EXPORT_SYMBOL(consume_skb
);
588 * skb_recycle - clean up an skb for reuse
591 * Recycles the skb to be reused as a receive buffer. This
592 * function does any necessary reference count dropping, and
593 * cleans up the skbuff as if it just came from __alloc_skb().
595 void skb_recycle(struct sk_buff
*skb
)
597 struct skb_shared_info
*shinfo
;
599 skb_release_head_state(skb
);
601 shinfo
= skb_shinfo(skb
);
602 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
603 atomic_set(&shinfo
->dataref
, 1);
605 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
606 skb
->data
= skb
->head
+ NET_SKB_PAD
;
607 skb_reset_tail_pointer(skb
);
609 EXPORT_SYMBOL(skb_recycle
);
612 * skb_recycle_check - check if skb can be reused for receive
614 * @skb_size: minimum receive buffer size
616 * Checks that the skb passed in is not shared or cloned, and
617 * that it is linear and its head portion at least as large as
618 * skb_size so that it can be recycled as a receive buffer.
619 * If these conditions are met, this function does any necessary
620 * reference count dropping and cleans up the skbuff as if it
621 * just came from __alloc_skb().
623 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
625 if (!skb_is_recycleable(skb
, skb_size
))
632 EXPORT_SYMBOL(skb_recycle_check
);
634 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
636 new->tstamp
= old
->tstamp
;
638 new->transport_header
= old
->transport_header
;
639 new->network_header
= old
->network_header
;
640 new->mac_header
= old
->mac_header
;
641 skb_dst_copy(new, old
);
642 new->rxhash
= old
->rxhash
;
643 new->ooo_okay
= old
->ooo_okay
;
644 new->l4_rxhash
= old
->l4_rxhash
;
645 new->no_fcs
= old
->no_fcs
;
647 new->sp
= secpath_get(old
->sp
);
649 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
650 new->csum
= old
->csum
;
651 new->local_df
= old
->local_df
;
652 new->pkt_type
= old
->pkt_type
;
653 new->ip_summed
= old
->ip_summed
;
654 skb_copy_queue_mapping(new, old
);
655 new->priority
= old
->priority
;
656 #if IS_ENABLED(CONFIG_IP_VS)
657 new->ipvs_property
= old
->ipvs_property
;
659 new->protocol
= old
->protocol
;
660 new->mark
= old
->mark
;
661 new->skb_iif
= old
->skb_iif
;
663 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
664 new->nf_trace
= old
->nf_trace
;
666 #ifdef CONFIG_NET_SCHED
667 new->tc_index
= old
->tc_index
;
668 #ifdef CONFIG_NET_CLS_ACT
669 new->tc_verd
= old
->tc_verd
;
672 new->vlan_tci
= old
->vlan_tci
;
674 skb_copy_secmark(new, old
);
678 * You should not add any new code to this function. Add it to
679 * __copy_skb_header above instead.
681 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
683 #define C(x) n->x = skb->x
685 n
->next
= n
->prev
= NULL
;
687 __copy_skb_header(n
, skb
);
692 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
695 n
->destructor
= NULL
;
702 atomic_set(&n
->users
, 1);
704 atomic_inc(&(skb_shinfo(skb
)->dataref
));
712 * skb_morph - morph one skb into another
713 * @dst: the skb to receive the contents
714 * @src: the skb to supply the contents
716 * This is identical to skb_clone except that the target skb is
717 * supplied by the user.
719 * The target skb is returned upon exit.
721 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
723 skb_release_all(dst
);
724 return __skb_clone(dst
, src
);
726 EXPORT_SYMBOL_GPL(skb_morph
);
729 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
730 * @skb: the skb to modify
731 * @gfp_mask: allocation priority
733 * This must be called on SKBTX_DEV_ZEROCOPY skb.
734 * It will copy all frags into kernel and drop the reference
735 * to userspace pages.
737 * If this function is called from an interrupt gfp_mask() must be
740 * Returns 0 on success or a negative error code on failure
741 * to allocate kernel memory to copy to.
743 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
746 int num_frags
= skb_shinfo(skb
)->nr_frags
;
747 struct page
*page
, *head
= NULL
;
748 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
750 for (i
= 0; i
< num_frags
; i
++) {
752 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
754 page
= alloc_page(gfp_mask
);
757 struct page
*next
= (struct page
*)head
->private;
763 vaddr
= kmap_atomic(skb_frag_page(f
));
764 memcpy(page_address(page
),
765 vaddr
+ f
->page_offset
, skb_frag_size(f
));
766 kunmap_atomic(vaddr
);
767 page
->private = (unsigned long)head
;
771 /* skb frags release userspace buffers */
772 for (i
= 0; i
< num_frags
; i
++)
773 skb_frag_unref(skb
, i
);
775 uarg
->callback(uarg
);
777 /* skb frags point to kernel buffers */
778 for (i
= num_frags
- 1; i
>= 0; i
--) {
779 __skb_fill_page_desc(skb
, i
, head
, 0,
780 skb_shinfo(skb
)->frags
[i
].size
);
781 head
= (struct page
*)head
->private;
784 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
787 EXPORT_SYMBOL_GPL(skb_copy_ubufs
);
790 * skb_clone - duplicate an sk_buff
791 * @skb: buffer to clone
792 * @gfp_mask: allocation priority
794 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
795 * copies share the same packet data but not structure. The new
796 * buffer has a reference count of 1. If the allocation fails the
797 * function returns %NULL otherwise the new buffer is returned.
799 * If this function is called from an interrupt gfp_mask() must be
803 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
807 if (skb_orphan_frags(skb
, gfp_mask
))
811 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
812 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
813 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
814 n
->fclone
= SKB_FCLONE_CLONE
;
815 atomic_inc(fclone_ref
);
817 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
821 kmemcheck_annotate_bitfield(n
, flags1
);
822 kmemcheck_annotate_bitfield(n
, flags2
);
823 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
826 return __skb_clone(n
, skb
);
828 EXPORT_SYMBOL(skb_clone
);
830 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
832 #ifndef NET_SKBUFF_DATA_USES_OFFSET
834 * Shift between the two data areas in bytes
836 unsigned long offset
= new->data
- old
->data
;
839 __copy_skb_header(new, old
);
841 #ifndef NET_SKBUFF_DATA_USES_OFFSET
842 /* {transport,network,mac}_header are relative to skb->head */
843 new->transport_header
+= offset
;
844 new->network_header
+= offset
;
845 if (skb_mac_header_was_set(new))
846 new->mac_header
+= offset
;
848 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
849 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
850 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
854 * skb_copy - create private copy of an sk_buff
855 * @skb: buffer to copy
856 * @gfp_mask: allocation priority
858 * Make a copy of both an &sk_buff and its data. This is used when the
859 * caller wishes to modify the data and needs a private copy of the
860 * data to alter. Returns %NULL on failure or the pointer to the buffer
861 * on success. The returned buffer has a reference count of 1.
863 * As by-product this function converts non-linear &sk_buff to linear
864 * one, so that &sk_buff becomes completely private and caller is allowed
865 * to modify all the data of returned buffer. This means that this
866 * function is not recommended for use in circumstances when only
867 * header is going to be modified. Use pskb_copy() instead.
870 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
872 int headerlen
= skb_headroom(skb
);
873 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
874 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
879 /* Set the data pointer */
880 skb_reserve(n
, headerlen
);
881 /* Set the tail pointer and length */
882 skb_put(n
, skb
->len
);
884 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
887 copy_skb_header(n
, skb
);
890 EXPORT_SYMBOL(skb_copy
);
893 * __pskb_copy - create copy of an sk_buff with private head.
894 * @skb: buffer to copy
895 * @headroom: headroom of new skb
896 * @gfp_mask: allocation priority
898 * Make a copy of both an &sk_buff and part of its data, located
899 * in header. Fragmented data remain shared. This is used when
900 * the caller wishes to modify only header of &sk_buff and needs
901 * private copy of the header to alter. Returns %NULL on failure
902 * or the pointer to the buffer on success.
903 * The returned buffer has a reference count of 1.
906 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
908 unsigned int size
= skb_headlen(skb
) + headroom
;
909 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
914 /* Set the data pointer */
915 skb_reserve(n
, headroom
);
916 /* Set the tail pointer and length */
917 skb_put(n
, skb_headlen(skb
));
919 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
921 n
->truesize
+= skb
->data_len
;
922 n
->data_len
= skb
->data_len
;
925 if (skb_shinfo(skb
)->nr_frags
) {
928 if (skb_orphan_frags(skb
, gfp_mask
)) {
933 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
934 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
935 skb_frag_ref(skb
, i
);
937 skb_shinfo(n
)->nr_frags
= i
;
940 if (skb_has_frag_list(skb
)) {
941 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
942 skb_clone_fraglist(n
);
945 copy_skb_header(n
, skb
);
949 EXPORT_SYMBOL(__pskb_copy
);
952 * pskb_expand_head - reallocate header of &sk_buff
953 * @skb: buffer to reallocate
954 * @nhead: room to add at head
955 * @ntail: room to add at tail
956 * @gfp_mask: allocation priority
958 * Expands (or creates identical copy, if &nhead and &ntail are zero)
959 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
960 * reference count of 1. Returns zero in the case of success or error,
961 * if expansion failed. In the last case, &sk_buff is not changed.
963 * All the pointers pointing into skb header may change and must be
964 * reloaded after call to this function.
967 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
972 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
980 size
= SKB_DATA_ALIGN(size
);
982 data
= kmalloc(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
986 size
= SKB_WITH_OVERHEAD(ksize(data
));
988 /* Copy only real data... and, alas, header. This should be
989 * optimized for the cases when header is void.
991 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
993 memcpy((struct skb_shared_info
*)(data
+ size
),
995 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
998 * if shinfo is shared we must drop the old head gracefully, but if it
999 * is not we can just drop the old head and let the existing refcount
1000 * be since all we did is relocate the values
1002 if (skb_cloned(skb
)) {
1003 /* copy this zero copy skb frags */
1004 if (skb_orphan_frags(skb
, gfp_mask
))
1006 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1007 skb_frag_ref(skb
, i
);
1009 if (skb_has_frag_list(skb
))
1010 skb_clone_fraglist(skb
);
1012 skb_release_data(skb
);
1016 off
= (data
+ nhead
) - skb
->head
;
1021 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1025 skb
->end
= skb
->head
+ size
;
1027 /* {transport,network,mac}_header and tail are relative to skb->head */
1029 skb
->transport_header
+= off
;
1030 skb
->network_header
+= off
;
1031 if (skb_mac_header_was_set(skb
))
1032 skb
->mac_header
+= off
;
1033 /* Only adjust this if it actually is csum_start rather than csum */
1034 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1035 skb
->csum_start
+= nhead
;
1039 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1047 EXPORT_SYMBOL(pskb_expand_head
);
1049 /* Make private copy of skb with writable head and some headroom */
1051 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1053 struct sk_buff
*skb2
;
1054 int delta
= headroom
- skb_headroom(skb
);
1057 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1059 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1060 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1068 EXPORT_SYMBOL(skb_realloc_headroom
);
1071 * skb_copy_expand - copy and expand sk_buff
1072 * @skb: buffer to copy
1073 * @newheadroom: new free bytes at head
1074 * @newtailroom: new free bytes at tail
1075 * @gfp_mask: allocation priority
1077 * Make a copy of both an &sk_buff and its data and while doing so
1078 * allocate additional space.
1080 * This is used when the caller wishes to modify the data and needs a
1081 * private copy of the data to alter as well as more space for new fields.
1082 * Returns %NULL on failure or the pointer to the buffer
1083 * on success. The returned buffer has a reference count of 1.
1085 * You must pass %GFP_ATOMIC as the allocation priority if this function
1086 * is called from an interrupt.
1088 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1089 int newheadroom
, int newtailroom
,
1093 * Allocate the copy buffer
1095 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1097 int oldheadroom
= skb_headroom(skb
);
1098 int head_copy_len
, head_copy_off
;
1104 skb_reserve(n
, newheadroom
);
1106 /* Set the tail pointer and length */
1107 skb_put(n
, skb
->len
);
1109 head_copy_len
= oldheadroom
;
1111 if (newheadroom
<= head_copy_len
)
1112 head_copy_len
= newheadroom
;
1114 head_copy_off
= newheadroom
- head_copy_len
;
1116 /* Copy the linear header and data. */
1117 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1118 skb
->len
+ head_copy_len
))
1121 copy_skb_header(n
, skb
);
1123 off
= newheadroom
- oldheadroom
;
1124 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1125 n
->csum_start
+= off
;
1126 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1127 n
->transport_header
+= off
;
1128 n
->network_header
+= off
;
1129 if (skb_mac_header_was_set(skb
))
1130 n
->mac_header
+= off
;
1135 EXPORT_SYMBOL(skb_copy_expand
);
1138 * skb_pad - zero pad the tail of an skb
1139 * @skb: buffer to pad
1140 * @pad: space to pad
1142 * Ensure that a buffer is followed by a padding area that is zero
1143 * filled. Used by network drivers which may DMA or transfer data
1144 * beyond the buffer end onto the wire.
1146 * May return error in out of memory cases. The skb is freed on error.
1149 int skb_pad(struct sk_buff
*skb
, int pad
)
1154 /* If the skbuff is non linear tailroom is always zero.. */
1155 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1156 memset(skb
->data
+skb
->len
, 0, pad
);
1160 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1161 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1162 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1167 /* FIXME: The use of this function with non-linear skb's really needs
1170 err
= skb_linearize(skb
);
1174 memset(skb
->data
+ skb
->len
, 0, pad
);
1181 EXPORT_SYMBOL(skb_pad
);
1184 * skb_put - add data to a buffer
1185 * @skb: buffer to use
1186 * @len: amount of data to add
1188 * This function extends the used data area of the buffer. If this would
1189 * exceed the total buffer size the kernel will panic. A pointer to the
1190 * first byte of the extra data is returned.
1192 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1194 unsigned char *tmp
= skb_tail_pointer(skb
);
1195 SKB_LINEAR_ASSERT(skb
);
1198 if (unlikely(skb
->tail
> skb
->end
))
1199 skb_over_panic(skb
, len
, __builtin_return_address(0));
1202 EXPORT_SYMBOL(skb_put
);
1205 * skb_push - add data to the start of a buffer
1206 * @skb: buffer to use
1207 * @len: amount of data to add
1209 * This function extends the used data area of the buffer at the buffer
1210 * start. If this would exceed the total buffer headroom the kernel will
1211 * panic. A pointer to the first byte of the extra data is returned.
1213 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1217 if (unlikely(skb
->data
<skb
->head
))
1218 skb_under_panic(skb
, len
, __builtin_return_address(0));
1221 EXPORT_SYMBOL(skb_push
);
1224 * skb_pull - remove data from the start of a buffer
1225 * @skb: buffer to use
1226 * @len: amount of data to remove
1228 * This function removes data from the start of a buffer, returning
1229 * the memory to the headroom. A pointer to the next data in the buffer
1230 * is returned. Once the data has been pulled future pushes will overwrite
1233 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1235 return skb_pull_inline(skb
, len
);
1237 EXPORT_SYMBOL(skb_pull
);
1240 * skb_trim - remove end from a buffer
1241 * @skb: buffer to alter
1244 * Cut the length of a buffer down by removing data from the tail. If
1245 * the buffer is already under the length specified it is not modified.
1246 * The skb must be linear.
1248 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1251 __skb_trim(skb
, len
);
1253 EXPORT_SYMBOL(skb_trim
);
1255 /* Trims skb to length len. It can change skb pointers.
1258 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1260 struct sk_buff
**fragp
;
1261 struct sk_buff
*frag
;
1262 int offset
= skb_headlen(skb
);
1263 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1267 if (skb_cloned(skb
) &&
1268 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1275 for (; i
< nfrags
; i
++) {
1276 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1283 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1286 skb_shinfo(skb
)->nr_frags
= i
;
1288 for (; i
< nfrags
; i
++)
1289 skb_frag_unref(skb
, i
);
1291 if (skb_has_frag_list(skb
))
1292 skb_drop_fraglist(skb
);
1296 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1297 fragp
= &frag
->next
) {
1298 int end
= offset
+ frag
->len
;
1300 if (skb_shared(frag
)) {
1301 struct sk_buff
*nfrag
;
1303 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1304 if (unlikely(!nfrag
))
1307 nfrag
->next
= frag
->next
;
1319 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1323 skb_drop_list(&frag
->next
);
1328 if (len
> skb_headlen(skb
)) {
1329 skb
->data_len
-= skb
->len
- len
;
1334 skb_set_tail_pointer(skb
, len
);
1339 EXPORT_SYMBOL(___pskb_trim
);
1342 * __pskb_pull_tail - advance tail of skb header
1343 * @skb: buffer to reallocate
1344 * @delta: number of bytes to advance tail
1346 * The function makes a sense only on a fragmented &sk_buff,
1347 * it expands header moving its tail forward and copying necessary
1348 * data from fragmented part.
1350 * &sk_buff MUST have reference count of 1.
1352 * Returns %NULL (and &sk_buff does not change) if pull failed
1353 * or value of new tail of skb in the case of success.
1355 * All the pointers pointing into skb header may change and must be
1356 * reloaded after call to this function.
1359 /* Moves tail of skb head forward, copying data from fragmented part,
1360 * when it is necessary.
1361 * 1. It may fail due to malloc failure.
1362 * 2. It may change skb pointers.
1364 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1366 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1368 /* If skb has not enough free space at tail, get new one
1369 * plus 128 bytes for future expansions. If we have enough
1370 * room at tail, reallocate without expansion only if skb is cloned.
1372 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1374 if (eat
> 0 || skb_cloned(skb
)) {
1375 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1380 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1383 /* Optimization: no fragments, no reasons to preestimate
1384 * size of pulled pages. Superb.
1386 if (!skb_has_frag_list(skb
))
1389 /* Estimate size of pulled pages. */
1391 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1392 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1399 /* If we need update frag list, we are in troubles.
1400 * Certainly, it possible to add an offset to skb data,
1401 * but taking into account that pulling is expected to
1402 * be very rare operation, it is worth to fight against
1403 * further bloating skb head and crucify ourselves here instead.
1404 * Pure masohism, indeed. 8)8)
1407 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1408 struct sk_buff
*clone
= NULL
;
1409 struct sk_buff
*insp
= NULL
;
1414 if (list
->len
<= eat
) {
1415 /* Eaten as whole. */
1420 /* Eaten partially. */
1422 if (skb_shared(list
)) {
1423 /* Sucks! We need to fork list. :-( */
1424 clone
= skb_clone(list
, GFP_ATOMIC
);
1430 /* This may be pulled without
1434 if (!pskb_pull(list
, eat
)) {
1442 /* Free pulled out fragments. */
1443 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1444 skb_shinfo(skb
)->frag_list
= list
->next
;
1447 /* And insert new clone at head. */
1450 skb_shinfo(skb
)->frag_list
= clone
;
1453 /* Success! Now we may commit changes to skb data. */
1458 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1459 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1462 skb_frag_unref(skb
, i
);
1465 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1467 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1468 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1474 skb_shinfo(skb
)->nr_frags
= k
;
1477 skb
->data_len
-= delta
;
1479 return skb_tail_pointer(skb
);
1481 EXPORT_SYMBOL(__pskb_pull_tail
);
1484 * skb_copy_bits - copy bits from skb to kernel buffer
1486 * @offset: offset in source
1487 * @to: destination buffer
1488 * @len: number of bytes to copy
1490 * Copy the specified number of bytes from the source skb to the
1491 * destination buffer.
1494 * If its prototype is ever changed,
1495 * check arch/{*}/net/{*}.S files,
1496 * since it is called from BPF assembly code.
1498 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1500 int start
= skb_headlen(skb
);
1501 struct sk_buff
*frag_iter
;
1504 if (offset
> (int)skb
->len
- len
)
1508 if ((copy
= start
- offset
) > 0) {
1511 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1512 if ((len
-= copy
) == 0)
1518 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1520 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1522 WARN_ON(start
> offset
+ len
);
1524 end
= start
+ skb_frag_size(f
);
1525 if ((copy
= end
- offset
) > 0) {
1531 vaddr
= kmap_atomic(skb_frag_page(f
));
1533 vaddr
+ f
->page_offset
+ offset
- start
,
1535 kunmap_atomic(vaddr
);
1537 if ((len
-= copy
) == 0)
1545 skb_walk_frags(skb
, frag_iter
) {
1548 WARN_ON(start
> offset
+ len
);
1550 end
= start
+ frag_iter
->len
;
1551 if ((copy
= end
- offset
) > 0) {
1554 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1556 if ((len
-= copy
) == 0)
1570 EXPORT_SYMBOL(skb_copy_bits
);
1573 * Callback from splice_to_pipe(), if we need to release some pages
1574 * at the end of the spd in case we error'ed out in filling the pipe.
1576 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1578 put_page(spd
->pages
[i
]);
1581 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1582 unsigned int *offset
,
1583 struct sk_buff
*skb
, struct sock
*sk
)
1585 struct page
*p
= sk
->sk_sndmsg_page
;
1590 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1594 off
= sk
->sk_sndmsg_off
= 0;
1595 /* hold one ref to this page until it's full */
1599 /* If we are the only user of the page, we can reset offset */
1600 if (page_count(p
) == 1)
1601 sk
->sk_sndmsg_off
= 0;
1602 off
= sk
->sk_sndmsg_off
;
1603 mlen
= PAGE_SIZE
- off
;
1604 if (mlen
< 64 && mlen
< *len
) {
1609 *len
= min_t(unsigned int, *len
, mlen
);
1612 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1613 sk
->sk_sndmsg_off
+= *len
;
1619 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1621 unsigned int offset
)
1623 return spd
->nr_pages
&&
1624 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1625 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1626 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1630 * Fill page/offset/length into spd, if it can hold more pages.
1632 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1633 struct pipe_inode_info
*pipe
, struct page
*page
,
1634 unsigned int *len
, unsigned int offset
,
1635 struct sk_buff
*skb
, bool linear
,
1638 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1642 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1646 if (spd_can_coalesce(spd
, page
, offset
)) {
1647 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1651 spd
->pages
[spd
->nr_pages
] = page
;
1652 spd
->partial
[spd
->nr_pages
].len
= *len
;
1653 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1659 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1660 unsigned int *plen
, unsigned int off
)
1665 n
= *poff
/ PAGE_SIZE
;
1667 *page
= nth_page(*page
, n
);
1669 *poff
= *poff
% PAGE_SIZE
;
1673 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1674 unsigned int plen
, unsigned int *off
,
1675 unsigned int *len
, struct sk_buff
*skb
,
1676 struct splice_pipe_desc
*spd
, bool linear
,
1678 struct pipe_inode_info
*pipe
)
1683 /* skip this segment if already processed */
1689 /* ignore any bits we already processed */
1691 __segment_seek(&page
, &poff
, &plen
, *off
);
1696 unsigned int flen
= min(*len
, plen
);
1698 /* the linear region may spread across several pages */
1699 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1701 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1704 __segment_seek(&page
, &poff
, &plen
, flen
);
1707 } while (*len
&& plen
);
1713 * Map linear and fragment data from the skb to spd. It reports true if the
1714 * pipe is full or if we already spliced the requested length.
1716 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1717 unsigned int *offset
, unsigned int *len
,
1718 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1722 /* map the linear part :
1723 * If skb->head_frag is set, this 'linear' part is backed by a
1724 * fragment, and if the head is not shared with any clones then
1725 * we can avoid a copy since we own the head portion of this page.
1727 if (__splice_segment(virt_to_page(skb
->data
),
1728 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1730 offset
, len
, skb
, spd
,
1731 skb_head_is_locked(skb
),
1736 * then map the fragments
1738 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1739 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1741 if (__splice_segment(skb_frag_page(f
),
1742 f
->page_offset
, skb_frag_size(f
),
1743 offset
, len
, skb
, spd
, false, sk
, pipe
))
1751 * Map data from the skb to a pipe. Should handle both the linear part,
1752 * the fragments, and the frag list. It does NOT handle frag lists within
1753 * the frag list, if such a thing exists. We'd probably need to recurse to
1754 * handle that cleanly.
1756 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1757 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1760 struct partial_page partial
[MAX_SKB_FRAGS
];
1761 struct page
*pages
[MAX_SKB_FRAGS
];
1762 struct splice_pipe_desc spd
= {
1765 .nr_pages_max
= MAX_SKB_FRAGS
,
1767 .ops
= &sock_pipe_buf_ops
,
1768 .spd_release
= sock_spd_release
,
1770 struct sk_buff
*frag_iter
;
1771 struct sock
*sk
= skb
->sk
;
1775 * __skb_splice_bits() only fails if the output has no room left,
1776 * so no point in going over the frag_list for the error case.
1778 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1784 * now see if we have a frag_list to map
1786 skb_walk_frags(skb
, frag_iter
) {
1789 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1796 * Drop the socket lock, otherwise we have reverse
1797 * locking dependencies between sk_lock and i_mutex
1798 * here as compared to sendfile(). We enter here
1799 * with the socket lock held, and splice_to_pipe() will
1800 * grab the pipe inode lock. For sendfile() emulation,
1801 * we call into ->sendpage() with the i_mutex lock held
1802 * and networking will grab the socket lock.
1805 ret
= splice_to_pipe(pipe
, &spd
);
1813 * skb_store_bits - store bits from kernel buffer to skb
1814 * @skb: destination buffer
1815 * @offset: offset in destination
1816 * @from: source buffer
1817 * @len: number of bytes to copy
1819 * Copy the specified number of bytes from the source buffer to the
1820 * destination skb. This function handles all the messy bits of
1821 * traversing fragment lists and such.
1824 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1826 int start
= skb_headlen(skb
);
1827 struct sk_buff
*frag_iter
;
1830 if (offset
> (int)skb
->len
- len
)
1833 if ((copy
= start
- offset
) > 0) {
1836 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1837 if ((len
-= copy
) == 0)
1843 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1844 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1847 WARN_ON(start
> offset
+ len
);
1849 end
= start
+ skb_frag_size(frag
);
1850 if ((copy
= end
- offset
) > 0) {
1856 vaddr
= kmap_atomic(skb_frag_page(frag
));
1857 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1859 kunmap_atomic(vaddr
);
1861 if ((len
-= copy
) == 0)
1869 skb_walk_frags(skb
, frag_iter
) {
1872 WARN_ON(start
> offset
+ len
);
1874 end
= start
+ frag_iter
->len
;
1875 if ((copy
= end
- offset
) > 0) {
1878 if (skb_store_bits(frag_iter
, offset
- start
,
1881 if ((len
-= copy
) == 0)
1894 EXPORT_SYMBOL(skb_store_bits
);
1896 /* Checksum skb data. */
1898 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1899 int len
, __wsum csum
)
1901 int start
= skb_headlen(skb
);
1902 int i
, copy
= start
- offset
;
1903 struct sk_buff
*frag_iter
;
1906 /* Checksum header. */
1910 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1911 if ((len
-= copy
) == 0)
1917 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1919 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1921 WARN_ON(start
> offset
+ len
);
1923 end
= start
+ skb_frag_size(frag
);
1924 if ((copy
= end
- offset
) > 0) {
1930 vaddr
= kmap_atomic(skb_frag_page(frag
));
1931 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1932 offset
- start
, copy
, 0);
1933 kunmap_atomic(vaddr
);
1934 csum
= csum_block_add(csum
, csum2
, pos
);
1943 skb_walk_frags(skb
, frag_iter
) {
1946 WARN_ON(start
> offset
+ len
);
1948 end
= start
+ frag_iter
->len
;
1949 if ((copy
= end
- offset
) > 0) {
1953 csum2
= skb_checksum(frag_iter
, offset
- start
,
1955 csum
= csum_block_add(csum
, csum2
, pos
);
1956 if ((len
-= copy
) == 0)
1967 EXPORT_SYMBOL(skb_checksum
);
1969 /* Both of above in one bottle. */
1971 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1972 u8
*to
, int len
, __wsum csum
)
1974 int start
= skb_headlen(skb
);
1975 int i
, copy
= start
- offset
;
1976 struct sk_buff
*frag_iter
;
1983 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1985 if ((len
-= copy
) == 0)
1992 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1995 WARN_ON(start
> offset
+ len
);
1997 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1998 if ((copy
= end
- offset
) > 0) {
2001 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2005 vaddr
= kmap_atomic(skb_frag_page(frag
));
2006 csum2
= csum_partial_copy_nocheck(vaddr
+
2010 kunmap_atomic(vaddr
);
2011 csum
= csum_block_add(csum
, csum2
, pos
);
2021 skb_walk_frags(skb
, frag_iter
) {
2025 WARN_ON(start
> offset
+ len
);
2027 end
= start
+ frag_iter
->len
;
2028 if ((copy
= end
- offset
) > 0) {
2031 csum2
= skb_copy_and_csum_bits(frag_iter
,
2034 csum
= csum_block_add(csum
, csum2
, pos
);
2035 if ((len
-= copy
) == 0)
2046 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2048 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2053 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2054 csstart
= skb_checksum_start_offset(skb
);
2056 csstart
= skb_headlen(skb
);
2058 BUG_ON(csstart
> skb_headlen(skb
));
2060 skb_copy_from_linear_data(skb
, to
, csstart
);
2063 if (csstart
!= skb
->len
)
2064 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2065 skb
->len
- csstart
, 0);
2067 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2068 long csstuff
= csstart
+ skb
->csum_offset
;
2070 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2073 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2076 * skb_dequeue - remove from the head of the queue
2077 * @list: list to dequeue from
2079 * Remove the head of the list. The list lock is taken so the function
2080 * may be used safely with other locking list functions. The head item is
2081 * returned or %NULL if the list is empty.
2084 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2086 unsigned long flags
;
2087 struct sk_buff
*result
;
2089 spin_lock_irqsave(&list
->lock
, flags
);
2090 result
= __skb_dequeue(list
);
2091 spin_unlock_irqrestore(&list
->lock
, flags
);
2094 EXPORT_SYMBOL(skb_dequeue
);
2097 * skb_dequeue_tail - remove from the tail of the queue
2098 * @list: list to dequeue from
2100 * Remove the tail of the list. The list lock is taken so the function
2101 * may be used safely with other locking list functions. The tail item is
2102 * returned or %NULL if the list is empty.
2104 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2106 unsigned long flags
;
2107 struct sk_buff
*result
;
2109 spin_lock_irqsave(&list
->lock
, flags
);
2110 result
= __skb_dequeue_tail(list
);
2111 spin_unlock_irqrestore(&list
->lock
, flags
);
2114 EXPORT_SYMBOL(skb_dequeue_tail
);
2117 * skb_queue_purge - empty a list
2118 * @list: list to empty
2120 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2121 * the list and one reference dropped. This function takes the list
2122 * lock and is atomic with respect to other list locking functions.
2124 void skb_queue_purge(struct sk_buff_head
*list
)
2126 struct sk_buff
*skb
;
2127 while ((skb
= skb_dequeue(list
)) != NULL
)
2130 EXPORT_SYMBOL(skb_queue_purge
);
2133 * skb_queue_head - queue a buffer at the list head
2134 * @list: list to use
2135 * @newsk: buffer to queue
2137 * Queue a buffer at the start of the list. This function takes the
2138 * list lock and can be used safely with other locking &sk_buff functions
2141 * A buffer cannot be placed on two lists at the same time.
2143 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2145 unsigned long flags
;
2147 spin_lock_irqsave(&list
->lock
, flags
);
2148 __skb_queue_head(list
, newsk
);
2149 spin_unlock_irqrestore(&list
->lock
, flags
);
2151 EXPORT_SYMBOL(skb_queue_head
);
2154 * skb_queue_tail - queue a buffer at the list tail
2155 * @list: list to use
2156 * @newsk: buffer to queue
2158 * Queue a buffer at the tail of the list. This function takes the
2159 * list lock and can be used safely with other locking &sk_buff functions
2162 * A buffer cannot be placed on two lists at the same time.
2164 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2166 unsigned long flags
;
2168 spin_lock_irqsave(&list
->lock
, flags
);
2169 __skb_queue_tail(list
, newsk
);
2170 spin_unlock_irqrestore(&list
->lock
, flags
);
2172 EXPORT_SYMBOL(skb_queue_tail
);
2175 * skb_unlink - remove a buffer from a list
2176 * @skb: buffer to remove
2177 * @list: list to use
2179 * Remove a packet from a list. The list locks are taken and this
2180 * function is atomic with respect to other list locked calls
2182 * You must know what list the SKB is on.
2184 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2186 unsigned long flags
;
2188 spin_lock_irqsave(&list
->lock
, flags
);
2189 __skb_unlink(skb
, list
);
2190 spin_unlock_irqrestore(&list
->lock
, flags
);
2192 EXPORT_SYMBOL(skb_unlink
);
2195 * skb_append - append a buffer
2196 * @old: buffer to insert after
2197 * @newsk: buffer to insert
2198 * @list: list to use
2200 * Place a packet after a given packet in a list. The list locks are taken
2201 * and this function is atomic with respect to other list locked calls.
2202 * A buffer cannot be placed on two lists at the same time.
2204 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2206 unsigned long flags
;
2208 spin_lock_irqsave(&list
->lock
, flags
);
2209 __skb_queue_after(list
, old
, newsk
);
2210 spin_unlock_irqrestore(&list
->lock
, flags
);
2212 EXPORT_SYMBOL(skb_append
);
2215 * skb_insert - insert a buffer
2216 * @old: buffer to insert before
2217 * @newsk: buffer to insert
2218 * @list: list to use
2220 * Place a packet before a given packet in a list. The list locks are
2221 * taken and this function is atomic with respect to other list locked
2224 * A buffer cannot be placed on two lists at the same time.
2226 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2228 unsigned long flags
;
2230 spin_lock_irqsave(&list
->lock
, flags
);
2231 __skb_insert(newsk
, old
->prev
, old
, list
);
2232 spin_unlock_irqrestore(&list
->lock
, flags
);
2234 EXPORT_SYMBOL(skb_insert
);
2236 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2237 struct sk_buff
* skb1
,
2238 const u32 len
, const int pos
)
2242 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2244 /* And move data appendix as is. */
2245 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2246 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2248 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2249 skb_shinfo(skb
)->nr_frags
= 0;
2250 skb1
->data_len
= skb
->data_len
;
2251 skb1
->len
+= skb1
->data_len
;
2254 skb_set_tail_pointer(skb
, len
);
2257 static inline void skb_split_no_header(struct sk_buff
*skb
,
2258 struct sk_buff
* skb1
,
2259 const u32 len
, int pos
)
2262 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2264 skb_shinfo(skb
)->nr_frags
= 0;
2265 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2267 skb
->data_len
= len
- pos
;
2269 for (i
= 0; i
< nfrags
; i
++) {
2270 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2272 if (pos
+ size
> len
) {
2273 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2277 * We have two variants in this case:
2278 * 1. Move all the frag to the second
2279 * part, if it is possible. F.e.
2280 * this approach is mandatory for TUX,
2281 * where splitting is expensive.
2282 * 2. Split is accurately. We make this.
2284 skb_frag_ref(skb
, i
);
2285 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2286 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2287 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2288 skb_shinfo(skb
)->nr_frags
++;
2292 skb_shinfo(skb
)->nr_frags
++;
2295 skb_shinfo(skb1
)->nr_frags
= k
;
2299 * skb_split - Split fragmented skb to two parts at length len.
2300 * @skb: the buffer to split
2301 * @skb1: the buffer to receive the second part
2302 * @len: new length for skb
2304 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2306 int pos
= skb_headlen(skb
);
2308 if (len
< pos
) /* Split line is inside header. */
2309 skb_split_inside_header(skb
, skb1
, len
, pos
);
2310 else /* Second chunk has no header, nothing to copy. */
2311 skb_split_no_header(skb
, skb1
, len
, pos
);
2313 EXPORT_SYMBOL(skb_split
);
2315 /* Shifting from/to a cloned skb is a no-go.
2317 * Caller cannot keep skb_shinfo related pointers past calling here!
2319 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2321 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2325 * skb_shift - Shifts paged data partially from skb to another
2326 * @tgt: buffer into which tail data gets added
2327 * @skb: buffer from which the paged data comes from
2328 * @shiftlen: shift up to this many bytes
2330 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2331 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2332 * It's up to caller to free skb if everything was shifted.
2334 * If @tgt runs out of frags, the whole operation is aborted.
2336 * Skb cannot include anything else but paged data while tgt is allowed
2337 * to have non-paged data as well.
2339 * TODO: full sized shift could be optimized but that would need
2340 * specialized skb free'er to handle frags without up-to-date nr_frags.
2342 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2344 int from
, to
, merge
, todo
;
2345 struct skb_frag_struct
*fragfrom
, *fragto
;
2347 BUG_ON(shiftlen
> skb
->len
);
2348 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2352 to
= skb_shinfo(tgt
)->nr_frags
;
2353 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2355 /* Actual merge is delayed until the point when we know we can
2356 * commit all, so that we don't have to undo partial changes
2359 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2360 fragfrom
->page_offset
)) {
2365 todo
-= skb_frag_size(fragfrom
);
2367 if (skb_prepare_for_shift(skb
) ||
2368 skb_prepare_for_shift(tgt
))
2371 /* All previous frag pointers might be stale! */
2372 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2373 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2375 skb_frag_size_add(fragto
, shiftlen
);
2376 skb_frag_size_sub(fragfrom
, shiftlen
);
2377 fragfrom
->page_offset
+= shiftlen
;
2385 /* Skip full, not-fitting skb to avoid expensive operations */
2386 if ((shiftlen
== skb
->len
) &&
2387 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2390 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2393 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2394 if (to
== MAX_SKB_FRAGS
)
2397 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2398 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2400 if (todo
>= skb_frag_size(fragfrom
)) {
2401 *fragto
= *fragfrom
;
2402 todo
-= skb_frag_size(fragfrom
);
2407 __skb_frag_ref(fragfrom
);
2408 fragto
->page
= fragfrom
->page
;
2409 fragto
->page_offset
= fragfrom
->page_offset
;
2410 skb_frag_size_set(fragto
, todo
);
2412 fragfrom
->page_offset
+= todo
;
2413 skb_frag_size_sub(fragfrom
, todo
);
2421 /* Ready to "commit" this state change to tgt */
2422 skb_shinfo(tgt
)->nr_frags
= to
;
2425 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2426 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2428 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2429 __skb_frag_unref(fragfrom
);
2432 /* Reposition in the original skb */
2434 while (from
< skb_shinfo(skb
)->nr_frags
)
2435 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2436 skb_shinfo(skb
)->nr_frags
= to
;
2438 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2441 /* Most likely the tgt won't ever need its checksum anymore, skb on
2442 * the other hand might need it if it needs to be resent
2444 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2445 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2447 /* Yak, is it really working this way? Some helper please? */
2448 skb
->len
-= shiftlen
;
2449 skb
->data_len
-= shiftlen
;
2450 skb
->truesize
-= shiftlen
;
2451 tgt
->len
+= shiftlen
;
2452 tgt
->data_len
+= shiftlen
;
2453 tgt
->truesize
+= shiftlen
;
2459 * skb_prepare_seq_read - Prepare a sequential read of skb data
2460 * @skb: the buffer to read
2461 * @from: lower offset of data to be read
2462 * @to: upper offset of data to be read
2463 * @st: state variable
2465 * Initializes the specified state variable. Must be called before
2466 * invoking skb_seq_read() for the first time.
2468 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2469 unsigned int to
, struct skb_seq_state
*st
)
2471 st
->lower_offset
= from
;
2472 st
->upper_offset
= to
;
2473 st
->root_skb
= st
->cur_skb
= skb
;
2474 st
->frag_idx
= st
->stepped_offset
= 0;
2475 st
->frag_data
= NULL
;
2477 EXPORT_SYMBOL(skb_prepare_seq_read
);
2480 * skb_seq_read - Sequentially read skb data
2481 * @consumed: number of bytes consumed by the caller so far
2482 * @data: destination pointer for data to be returned
2483 * @st: state variable
2485 * Reads a block of skb data at &consumed relative to the
2486 * lower offset specified to skb_prepare_seq_read(). Assigns
2487 * the head of the data block to &data and returns the length
2488 * of the block or 0 if the end of the skb data or the upper
2489 * offset has been reached.
2491 * The caller is not required to consume all of the data
2492 * returned, i.e. &consumed is typically set to the number
2493 * of bytes already consumed and the next call to
2494 * skb_seq_read() will return the remaining part of the block.
2496 * Note 1: The size of each block of data returned can be arbitrary,
2497 * this limitation is the cost for zerocopy seqeuental
2498 * reads of potentially non linear data.
2500 * Note 2: Fragment lists within fragments are not implemented
2501 * at the moment, state->root_skb could be replaced with
2502 * a stack for this purpose.
2504 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2505 struct skb_seq_state
*st
)
2507 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2510 if (unlikely(abs_offset
>= st
->upper_offset
))
2514 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2516 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2517 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2518 return block_limit
- abs_offset
;
2521 if (st
->frag_idx
== 0 && !st
->frag_data
)
2522 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2524 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2525 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2526 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2528 if (abs_offset
< block_limit
) {
2530 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2532 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2533 (abs_offset
- st
->stepped_offset
);
2535 return block_limit
- abs_offset
;
2538 if (st
->frag_data
) {
2539 kunmap_atomic(st
->frag_data
);
2540 st
->frag_data
= NULL
;
2544 st
->stepped_offset
+= skb_frag_size(frag
);
2547 if (st
->frag_data
) {
2548 kunmap_atomic(st
->frag_data
);
2549 st
->frag_data
= NULL
;
2552 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2553 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2556 } else if (st
->cur_skb
->next
) {
2557 st
->cur_skb
= st
->cur_skb
->next
;
2564 EXPORT_SYMBOL(skb_seq_read
);
2567 * skb_abort_seq_read - Abort a sequential read of skb data
2568 * @st: state variable
2570 * Must be called if skb_seq_read() was not called until it
2573 void skb_abort_seq_read(struct skb_seq_state
*st
)
2576 kunmap_atomic(st
->frag_data
);
2578 EXPORT_SYMBOL(skb_abort_seq_read
);
2580 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2582 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2583 struct ts_config
*conf
,
2584 struct ts_state
*state
)
2586 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2589 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2591 skb_abort_seq_read(TS_SKB_CB(state
));
2595 * skb_find_text - Find a text pattern in skb data
2596 * @skb: the buffer to look in
2597 * @from: search offset
2599 * @config: textsearch configuration
2600 * @state: uninitialized textsearch state variable
2602 * Finds a pattern in the skb data according to the specified
2603 * textsearch configuration. Use textsearch_next() to retrieve
2604 * subsequent occurrences of the pattern. Returns the offset
2605 * to the first occurrence or UINT_MAX if no match was found.
2607 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2608 unsigned int to
, struct ts_config
*config
,
2609 struct ts_state
*state
)
2613 config
->get_next_block
= skb_ts_get_next_block
;
2614 config
->finish
= skb_ts_finish
;
2616 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2618 ret
= textsearch_find(config
, state
);
2619 return (ret
<= to
- from
? ret
: UINT_MAX
);
2621 EXPORT_SYMBOL(skb_find_text
);
2624 * skb_append_datato_frags - append the user data to a skb
2625 * @sk: sock structure
2626 * @skb: skb structure to be appened with user data.
2627 * @getfrag: call back function to be used for getting the user data
2628 * @from: pointer to user message iov
2629 * @length: length of the iov message
2631 * Description: This procedure append the user data in the fragment part
2632 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2634 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2635 int (*getfrag
)(void *from
, char *to
, int offset
,
2636 int len
, int odd
, struct sk_buff
*skb
),
2637 void *from
, int length
)
2640 skb_frag_t
*frag
= NULL
;
2641 struct page
*page
= NULL
;
2647 /* Return error if we don't have space for new frag */
2648 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2649 if (frg_cnt
>= MAX_SKB_FRAGS
)
2652 /* allocate a new page for next frag */
2653 page
= alloc_pages(sk
->sk_allocation
, 0);
2655 /* If alloc_page fails just return failure and caller will
2656 * free previous allocated pages by doing kfree_skb()
2661 /* initialize the next frag */
2662 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2663 skb
->truesize
+= PAGE_SIZE
;
2664 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2666 /* get the new initialized frag */
2667 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2668 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2670 /* copy the user data to page */
2671 left
= PAGE_SIZE
- frag
->page_offset
;
2672 copy
= (length
> left
)? left
: length
;
2674 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2675 offset
, copy
, 0, skb
);
2679 /* copy was successful so update the size parameters */
2680 skb_frag_size_add(frag
, copy
);
2682 skb
->data_len
+= copy
;
2686 } while (length
> 0);
2690 EXPORT_SYMBOL(skb_append_datato_frags
);
2693 * skb_pull_rcsum - pull skb and update receive checksum
2694 * @skb: buffer to update
2695 * @len: length of data pulled
2697 * This function performs an skb_pull on the packet and updates
2698 * the CHECKSUM_COMPLETE checksum. It should be used on
2699 * receive path processing instead of skb_pull unless you know
2700 * that the checksum difference is zero (e.g., a valid IP header)
2701 * or you are setting ip_summed to CHECKSUM_NONE.
2703 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2705 BUG_ON(len
> skb
->len
);
2707 BUG_ON(skb
->len
< skb
->data_len
);
2708 skb_postpull_rcsum(skb
, skb
->data
, len
);
2709 return skb
->data
+= len
;
2711 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2714 * skb_segment - Perform protocol segmentation on skb.
2715 * @skb: buffer to segment
2716 * @features: features for the output path (see dev->features)
2718 * This function performs segmentation on the given skb. It returns
2719 * a pointer to the first in a list of new skbs for the segments.
2720 * In case of error it returns ERR_PTR(err).
2722 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2724 struct sk_buff
*segs
= NULL
;
2725 struct sk_buff
*tail
= NULL
;
2726 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2727 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2728 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2729 unsigned int offset
= doffset
;
2730 unsigned int headroom
;
2732 int sg
= !!(features
& NETIF_F_SG
);
2733 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2738 __skb_push(skb
, doffset
);
2739 headroom
= skb_headroom(skb
);
2740 pos
= skb_headlen(skb
);
2743 struct sk_buff
*nskb
;
2748 len
= skb
->len
- offset
;
2752 hsize
= skb_headlen(skb
) - offset
;
2755 if (hsize
> len
|| !sg
)
2758 if (!hsize
&& i
>= nfrags
) {
2759 BUG_ON(fskb
->len
!= len
);
2762 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2765 if (unlikely(!nskb
))
2768 hsize
= skb_end_offset(nskb
);
2769 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2774 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2775 skb_release_head_state(nskb
);
2776 __skb_push(nskb
, doffset
);
2778 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2781 if (unlikely(!nskb
))
2784 skb_reserve(nskb
, headroom
);
2785 __skb_put(nskb
, doffset
);
2794 __copy_skb_header(nskb
, skb
);
2795 nskb
->mac_len
= skb
->mac_len
;
2797 /* nskb and skb might have different headroom */
2798 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2799 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2801 skb_reset_mac_header(nskb
);
2802 skb_set_network_header(nskb
, skb
->mac_len
);
2803 nskb
->transport_header
= (nskb
->network_header
+
2804 skb_network_header_len(skb
));
2805 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2807 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2811 nskb
->ip_summed
= CHECKSUM_NONE
;
2812 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2818 frag
= skb_shinfo(nskb
)->frags
;
2820 skb_copy_from_linear_data_offset(skb
, offset
,
2821 skb_put(nskb
, hsize
), hsize
);
2823 while (pos
< offset
+ len
&& i
< nfrags
) {
2824 *frag
= skb_shinfo(skb
)->frags
[i
];
2825 __skb_frag_ref(frag
);
2826 size
= skb_frag_size(frag
);
2829 frag
->page_offset
+= offset
- pos
;
2830 skb_frag_size_sub(frag
, offset
- pos
);
2833 skb_shinfo(nskb
)->nr_frags
++;
2835 if (pos
+ size
<= offset
+ len
) {
2839 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2846 if (pos
< offset
+ len
) {
2847 struct sk_buff
*fskb2
= fskb
;
2849 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2855 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2861 SKB_FRAG_ASSERT(nskb
);
2862 skb_shinfo(nskb
)->frag_list
= fskb2
;
2866 nskb
->data_len
= len
- hsize
;
2867 nskb
->len
+= nskb
->data_len
;
2868 nskb
->truesize
+= nskb
->data_len
;
2869 } while ((offset
+= len
) < skb
->len
);
2874 while ((skb
= segs
)) {
2878 return ERR_PTR(err
);
2880 EXPORT_SYMBOL_GPL(skb_segment
);
2882 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2884 struct sk_buff
*p
= *head
;
2885 struct sk_buff
*nskb
;
2886 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2887 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2888 unsigned int headroom
;
2889 unsigned int len
= skb_gro_len(skb
);
2890 unsigned int offset
= skb_gro_offset(skb
);
2891 unsigned int headlen
= skb_headlen(skb
);
2892 unsigned int delta_truesize
;
2894 if (p
->len
+ len
>= 65536)
2897 if (pinfo
->frag_list
)
2899 else if (headlen
<= offset
) {
2902 int i
= skbinfo
->nr_frags
;
2903 int nr_frags
= pinfo
->nr_frags
+ i
;
2907 if (nr_frags
> MAX_SKB_FRAGS
)
2910 pinfo
->nr_frags
= nr_frags
;
2911 skbinfo
->nr_frags
= 0;
2913 frag
= pinfo
->frags
+ nr_frags
;
2914 frag2
= skbinfo
->frags
+ i
;
2919 frag
->page_offset
+= offset
;
2920 skb_frag_size_sub(frag
, offset
);
2922 /* all fragments truesize : remove (head size + sk_buff) */
2923 delta_truesize
= skb
->truesize
-
2924 SKB_TRUESIZE(skb_end_offset(skb
));
2926 skb
->truesize
-= skb
->data_len
;
2927 skb
->len
-= skb
->data_len
;
2930 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2932 } else if (skb
->head_frag
) {
2933 int nr_frags
= pinfo
->nr_frags
;
2934 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2935 struct page
*page
= virt_to_head_page(skb
->head
);
2936 unsigned int first_size
= headlen
- offset
;
2937 unsigned int first_offset
;
2939 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2942 first_offset
= skb
->data
-
2943 (unsigned char *)page_address(page
) +
2946 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
2948 frag
->page
.p
= page
;
2949 frag
->page_offset
= first_offset
;
2950 skb_frag_size_set(frag
, first_size
);
2952 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
2953 /* We dont need to clear skbinfo->nr_frags here */
2955 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
2956 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
2958 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2961 headroom
= skb_headroom(p
);
2962 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2963 if (unlikely(!nskb
))
2966 __copy_skb_header(nskb
, p
);
2967 nskb
->mac_len
= p
->mac_len
;
2969 skb_reserve(nskb
, headroom
);
2970 __skb_put(nskb
, skb_gro_offset(p
));
2972 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2973 skb_set_network_header(nskb
, skb_network_offset(p
));
2974 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2976 __skb_pull(p
, skb_gro_offset(p
));
2977 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2978 p
->data
- skb_mac_header(p
));
2980 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2981 skb_shinfo(nskb
)->frag_list
= p
;
2982 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2983 pinfo
->gso_size
= 0;
2984 skb_header_release(p
);
2987 nskb
->data_len
+= p
->len
;
2988 nskb
->truesize
+= p
->truesize
;
2989 nskb
->len
+= p
->len
;
2992 nskb
->next
= p
->next
;
2998 delta_truesize
= skb
->truesize
;
2999 if (offset
> headlen
) {
3000 unsigned int eat
= offset
- headlen
;
3002 skbinfo
->frags
[0].page_offset
+= eat
;
3003 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
3004 skb
->data_len
-= eat
;
3009 __skb_pull(skb
, offset
);
3011 p
->prev
->next
= skb
;
3013 skb_header_release(skb
);
3016 NAPI_GRO_CB(p
)->count
++;
3018 p
->truesize
+= delta_truesize
;
3021 NAPI_GRO_CB(skb
)->same_flow
= 1;
3024 EXPORT_SYMBOL_GPL(skb_gro_receive
);
3026 void __init
skb_init(void)
3028 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
3029 sizeof(struct sk_buff
),
3031 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3033 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
3034 (2*sizeof(struct sk_buff
)) +
3037 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3042 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3043 * @skb: Socket buffer containing the buffers to be mapped
3044 * @sg: The scatter-gather list to map into
3045 * @offset: The offset into the buffer's contents to start mapping
3046 * @len: Length of buffer space to be mapped
3048 * Fill the specified scatter-gather list with mappings/pointers into a
3049 * region of the buffer space attached to a socket buffer.
3052 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3054 int start
= skb_headlen(skb
);
3055 int i
, copy
= start
- offset
;
3056 struct sk_buff
*frag_iter
;
3062 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3064 if ((len
-= copy
) == 0)
3069 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3072 WARN_ON(start
> offset
+ len
);
3074 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3075 if ((copy
= end
- offset
) > 0) {
3076 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3080 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3081 frag
->page_offset
+offset
-start
);
3090 skb_walk_frags(skb
, frag_iter
) {
3093 WARN_ON(start
> offset
+ len
);
3095 end
= start
+ frag_iter
->len
;
3096 if ((copy
= end
- offset
) > 0) {
3099 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3101 if ((len
-= copy
) == 0)
3111 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3113 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3115 sg_mark_end(&sg
[nsg
- 1]);
3119 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3122 * skb_cow_data - Check that a socket buffer's data buffers are writable
3123 * @skb: The socket buffer to check.
3124 * @tailbits: Amount of trailing space to be added
3125 * @trailer: Returned pointer to the skb where the @tailbits space begins
3127 * Make sure that the data buffers attached to a socket buffer are
3128 * writable. If they are not, private copies are made of the data buffers
3129 * and the socket buffer is set to use these instead.
3131 * If @tailbits is given, make sure that there is space to write @tailbits
3132 * bytes of data beyond current end of socket buffer. @trailer will be
3133 * set to point to the skb in which this space begins.
3135 * The number of scatterlist elements required to completely map the
3136 * COW'd and extended socket buffer will be returned.
3138 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3142 struct sk_buff
*skb1
, **skb_p
;
3144 /* If skb is cloned or its head is paged, reallocate
3145 * head pulling out all the pages (pages are considered not writable
3146 * at the moment even if they are anonymous).
3148 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3149 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3152 /* Easy case. Most of packets will go this way. */
3153 if (!skb_has_frag_list(skb
)) {
3154 /* A little of trouble, not enough of space for trailer.
3155 * This should not happen, when stack is tuned to generate
3156 * good frames. OK, on miss we reallocate and reserve even more
3157 * space, 128 bytes is fair. */
3159 if (skb_tailroom(skb
) < tailbits
&&
3160 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3168 /* Misery. We are in troubles, going to mincer fragments... */
3171 skb_p
= &skb_shinfo(skb
)->frag_list
;
3174 while ((skb1
= *skb_p
) != NULL
) {
3177 /* The fragment is partially pulled by someone,
3178 * this can happen on input. Copy it and everything
3181 if (skb_shared(skb1
))
3184 /* If the skb is the last, worry about trailer. */
3186 if (skb1
->next
== NULL
&& tailbits
) {
3187 if (skb_shinfo(skb1
)->nr_frags
||
3188 skb_has_frag_list(skb1
) ||
3189 skb_tailroom(skb1
) < tailbits
)
3190 ntail
= tailbits
+ 128;
3196 skb_shinfo(skb1
)->nr_frags
||
3197 skb_has_frag_list(skb1
)) {
3198 struct sk_buff
*skb2
;
3200 /* Fuck, we are miserable poor guys... */
3202 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3204 skb2
= skb_copy_expand(skb1
,
3208 if (unlikely(skb2
== NULL
))
3212 skb_set_owner_w(skb2
, skb1
->sk
);
3214 /* Looking around. Are we still alive?
3215 * OK, link new skb, drop old one */
3217 skb2
->next
= skb1
->next
;
3224 skb_p
= &skb1
->next
;
3229 EXPORT_SYMBOL_GPL(skb_cow_data
);
3231 static void sock_rmem_free(struct sk_buff
*skb
)
3233 struct sock
*sk
= skb
->sk
;
3235 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3239 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3241 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3245 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3246 (unsigned int)sk
->sk_rcvbuf
)
3251 skb
->destructor
= sock_rmem_free
;
3252 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3254 /* before exiting rcu section, make sure dst is refcounted */
3257 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3258 if (!sock_flag(sk
, SOCK_DEAD
))
3259 sk
->sk_data_ready(sk
, len
);
3262 EXPORT_SYMBOL(sock_queue_err_skb
);
3264 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3265 struct skb_shared_hwtstamps
*hwtstamps
)
3267 struct sock
*sk
= orig_skb
->sk
;
3268 struct sock_exterr_skb
*serr
;
3269 struct sk_buff
*skb
;
3275 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3280 *skb_hwtstamps(skb
) =
3284 * no hardware time stamps available,
3285 * so keep the shared tx_flags and only
3286 * store software time stamp
3288 skb
->tstamp
= ktime_get_real();
3291 serr
= SKB_EXT_ERR(skb
);
3292 memset(serr
, 0, sizeof(*serr
));
3293 serr
->ee
.ee_errno
= ENOMSG
;
3294 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3296 err
= sock_queue_err_skb(sk
, skb
);
3301 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3303 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3305 struct sock
*sk
= skb
->sk
;
3306 struct sock_exterr_skb
*serr
;
3309 skb
->wifi_acked_valid
= 1;
3310 skb
->wifi_acked
= acked
;
3312 serr
= SKB_EXT_ERR(skb
);
3313 memset(serr
, 0, sizeof(*serr
));
3314 serr
->ee
.ee_errno
= ENOMSG
;
3315 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3317 err
= sock_queue_err_skb(sk
, skb
);
3321 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3325 * skb_partial_csum_set - set up and verify partial csum values for packet
3326 * @skb: the skb to set
3327 * @start: the number of bytes after skb->data to start checksumming.
3328 * @off: the offset from start to place the checksum.
3330 * For untrusted partially-checksummed packets, we need to make sure the values
3331 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3333 * This function checks and sets those values and skb->ip_summed: if this
3334 * returns false you should drop the packet.
3336 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3338 if (unlikely(start
> skb_headlen(skb
)) ||
3339 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3340 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3341 start
, off
, skb_headlen(skb
));
3344 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3345 skb
->csum_start
= skb_headroom(skb
) + start
;
3346 skb
->csum_offset
= off
;
3349 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3351 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3353 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3356 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);
3358 void kfree_skb_partial(struct sk_buff
*skb
, bool head_stolen
)
3361 kmem_cache_free(skbuff_head_cache
, skb
);
3365 EXPORT_SYMBOL(kfree_skb_partial
);
3368 * skb_try_coalesce - try to merge skb to prior one
3370 * @from: buffer to add
3371 * @fragstolen: pointer to boolean
3372 * @delta_truesize: how much more was allocated than was requested
3374 bool skb_try_coalesce(struct sk_buff
*to
, struct sk_buff
*from
,
3375 bool *fragstolen
, int *delta_truesize
)
3377 int i
, delta
, len
= from
->len
;
3379 *fragstolen
= false;
3384 if (len
<= skb_tailroom(to
)) {
3385 BUG_ON(skb_copy_bits(from
, 0, skb_put(to
, len
), len
));
3386 *delta_truesize
= 0;
3390 if (skb_has_frag_list(to
) || skb_has_frag_list(from
))
3393 if (skb_headlen(from
) != 0) {
3395 unsigned int offset
;
3397 if (skb_shinfo(to
)->nr_frags
+
3398 skb_shinfo(from
)->nr_frags
>= MAX_SKB_FRAGS
)
3401 if (skb_head_is_locked(from
))
3404 delta
= from
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
3406 page
= virt_to_head_page(from
->head
);
3407 offset
= from
->data
- (unsigned char *)page_address(page
);
3409 skb_fill_page_desc(to
, skb_shinfo(to
)->nr_frags
,
3410 page
, offset
, skb_headlen(from
));
3413 if (skb_shinfo(to
)->nr_frags
+
3414 skb_shinfo(from
)->nr_frags
> MAX_SKB_FRAGS
)
3417 delta
= from
->truesize
-
3418 SKB_TRUESIZE(skb_end_pointer(from
) - from
->head
);
3421 WARN_ON_ONCE(delta
< len
);
3423 memcpy(skb_shinfo(to
)->frags
+ skb_shinfo(to
)->nr_frags
,
3424 skb_shinfo(from
)->frags
,
3425 skb_shinfo(from
)->nr_frags
* sizeof(skb_frag_t
));
3426 skb_shinfo(to
)->nr_frags
+= skb_shinfo(from
)->nr_frags
;
3428 if (!skb_cloned(from
))
3429 skb_shinfo(from
)->nr_frags
= 0;
3431 /* if the skb is cloned this does nothing since we set nr_frags to 0 */
3432 for (i
= 0; i
< skb_shinfo(from
)->nr_frags
; i
++)
3433 skb_frag_ref(from
, i
);
3435 to
->truesize
+= delta
;
3437 to
->data_len
+= len
;
3439 *delta_truesize
= delta
;
3442 EXPORT_SYMBOL(skb_try_coalesce
);