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 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
61 #include <net/protocol.h>
64 #include <net/checksum.h>
67 #include <asm/uaccess.h>
68 #include <asm/system.h>
69 #include <trace/events/skb.h>
73 static struct kmem_cache *skbuff_head_cache __read_mostly;
74 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
76 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
77 struct pipe_buffer *buf)
82 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
83 struct pipe_buffer *buf)
88 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
89 struct pipe_buffer *buf)
95 /* Pipe buffer operations for a socket. */
96 static const struct pipe_buf_operations sock_pipe_buf_ops = {
98 .map = generic_pipe_buf_map,
99 .unmap = generic_pipe_buf_unmap,
100 .confirm = generic_pipe_buf_confirm,
101 .release = sock_pipe_buf_release,
102 .steal = sock_pipe_buf_steal,
103 .get = sock_pipe_buf_get,
107 * Keep out-of-line to prevent kernel bloat.
108 * __builtin_return_address is not used because it is not always
113 * skb_over_panic - private function
118 * Out of line support code for skb_put(). Not user callable.
120 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
122 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
123 "data:%p tail:%#lx end:%#lx dev:%s\n",
124 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 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here, skb->len, sz, skb->head, skb->data,
144 (unsigned long)skb->tail, (unsigned long)skb->end,
145 skb->dev ? skb->dev->name : "<NULL>");
149 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
156 * __alloc_skb - allocate a network buffer
157 * @size: size to allocate
158 * @gfp_mask: allocation mask
159 * @fclone: allocate from fclone cache instead of head cache
160 * and allocate a cloned (child) skb
161 * @node: numa node to allocate memory on
163 * Allocate a new &sk_buff. The returned buffer has no headroom and a
164 * tail room of size bytes. The object has a reference count of one.
165 * The return is the buffer. On a failure the return is %NULL.
167 * Buffers may only be allocated from interrupts using a @gfp_mask of
170 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
171 int fclone, int node)
173 struct kmem_cache *cache;
174 struct skb_shared_info *shinfo;
178 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
181 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
186 size = SKB_DATA_ALIGN(size);
187 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
191 prefetchw(data + size);
194 * Only clear those fields we need to clear, not those that we will
195 * actually initialise below. Hence, don't put any more fields after
196 * the tail pointer in struct sk_buff!
198 memset(skb, 0, offsetof(struct sk_buff, tail));
199 skb->truesize = size + sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
203 skb_reset_tail_pointer(skb);
204 skb->end = skb->tail + size;
205 kmemcheck_annotate_bitfield(skb, flags1);
206 kmemcheck_annotate_bitfield(skb, flags2);
207 #ifdef NET_SKBUFF_DATA_USES_OFFSET
208 skb->mac_header = ~0U;
211 /* make sure we initialize shinfo sequentially */
212 shinfo = skb_shinfo(skb);
213 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
214 atomic_set(&shinfo->dataref, 1);
217 struct sk_buff *child = skb + 1;
218 atomic_t *fclone_ref = (atomic_t *) (child + 1);
220 kmemcheck_annotate_bitfield(child, flags1);
221 kmemcheck_annotate_bitfield(child, flags2);
222 skb->fclone = SKB_FCLONE_ORIG;
223 atomic_set(fclone_ref, 1);
225 child->fclone = SKB_FCLONE_UNAVAILABLE;
230 kmem_cache_free(cache, skb);
234 EXPORT_SYMBOL(__alloc_skb);
237 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
238 * @dev: network device to receive on
239 * @length: length to allocate
240 * @gfp_mask: get_free_pages mask, passed to alloc_skb
242 * Allocate a new &sk_buff and assign it a usage count of one. The
243 * buffer has unspecified headroom built in. Users should allocate
244 * the headroom they think they need without accounting for the
245 * built in space. The built in space is used for optimisations.
247 * %NULL is returned if there is no free memory.
249 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
250 unsigned int length, gfp_t gfp_mask)
252 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
255 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
257 skb_reserve(skb, NET_SKB_PAD);
262 EXPORT_SYMBOL(__netdev_alloc_skb);
264 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
266 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
269 page = alloc_pages_node(node, gfp_mask, 0);
272 EXPORT_SYMBOL(__netdev_alloc_page);
274 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
277 skb_fill_page_desc(skb, i, page, off, size);
279 skb->data_len += size;
280 skb->truesize += size;
282 EXPORT_SYMBOL(skb_add_rx_frag);
285 * dev_alloc_skb - allocate an skbuff for receiving
286 * @length: length to allocate
288 * Allocate a new &sk_buff and assign it a usage count of one. The
289 * buffer has unspecified headroom built in. Users should allocate
290 * the headroom they think they need without accounting for the
291 * built in space. The built in space is used for optimisations.
293 * %NULL is returned if there is no free memory. Although this function
294 * allocates memory it can be called from an interrupt.
296 struct sk_buff *dev_alloc_skb(unsigned int length)
299 * There is more code here than it seems:
300 * __dev_alloc_skb is an inline
302 return __dev_alloc_skb(length, GFP_ATOMIC);
304 EXPORT_SYMBOL(dev_alloc_skb);
306 static void skb_drop_list(struct sk_buff **listp)
308 struct sk_buff *list = *listp;
313 struct sk_buff *this = list;
319 static inline void skb_drop_fraglist(struct sk_buff *skb)
321 skb_drop_list(&skb_shinfo(skb)->frag_list);
324 static void skb_clone_fraglist(struct sk_buff *skb)
326 struct sk_buff *list;
328 skb_walk_frags(skb, list)
332 static void skb_release_data(struct sk_buff *skb)
335 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
336 &skb_shinfo(skb)->dataref)) {
337 if (skb_shinfo(skb)->nr_frags) {
339 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
340 put_page(skb_shinfo(skb)->frags[i].page);
343 if (skb_has_frags(skb))
344 skb_drop_fraglist(skb);
351 * Free an skbuff by memory without cleaning the state.
353 static void kfree_skbmem(struct sk_buff *skb)
355 struct sk_buff *other;
356 atomic_t *fclone_ref;
358 switch (skb->fclone) {
359 case SKB_FCLONE_UNAVAILABLE:
360 kmem_cache_free(skbuff_head_cache, skb);
363 case SKB_FCLONE_ORIG:
364 fclone_ref = (atomic_t *) (skb + 2);
365 if (atomic_dec_and_test(fclone_ref))
366 kmem_cache_free(skbuff_fclone_cache, skb);
369 case SKB_FCLONE_CLONE:
370 fclone_ref = (atomic_t *) (skb + 1);
373 /* The clone portion is available for
374 * fast-cloning again.
376 skb->fclone = SKB_FCLONE_UNAVAILABLE;
378 if (atomic_dec_and_test(fclone_ref))
379 kmem_cache_free(skbuff_fclone_cache, other);
384 static void skb_release_head_state(struct sk_buff *skb)
388 secpath_put(skb->sp);
390 if (skb->destructor) {
392 skb->destructor(skb);
394 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
395 nf_conntrack_put(skb->nfct);
396 nf_conntrack_put_reasm(skb->nfct_reasm);
398 #ifdef CONFIG_BRIDGE_NETFILTER
399 nf_bridge_put(skb->nf_bridge);
401 /* XXX: IS this still necessary? - JHS */
402 #ifdef CONFIG_NET_SCHED
404 #ifdef CONFIG_NET_CLS_ACT
410 /* Free everything but the sk_buff shell. */
411 static void skb_release_all(struct sk_buff *skb)
413 skb_release_head_state(skb);
414 skb_release_data(skb);
418 * __kfree_skb - private function
421 * Free an sk_buff. Release anything attached to the buffer.
422 * Clean the state. This is an internal helper function. Users should
423 * always call kfree_skb
426 void __kfree_skb(struct sk_buff *skb)
428 skb_release_all(skb);
431 EXPORT_SYMBOL(__kfree_skb);
434 * kfree_skb - free an sk_buff
435 * @skb: buffer to free
437 * Drop a reference to the buffer and free it if the usage count has
440 void kfree_skb(struct sk_buff *skb)
444 if (likely(atomic_read(&skb->users) == 1))
446 else if (likely(!atomic_dec_and_test(&skb->users)))
448 trace_kfree_skb(skb, __builtin_return_address(0));
451 EXPORT_SYMBOL(kfree_skb);
454 * consume_skb - free an skbuff
455 * @skb: buffer to free
457 * Drop a ref to the buffer and free it if the usage count has hit zero
458 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
459 * is being dropped after a failure and notes that
461 void consume_skb(struct sk_buff *skb)
465 if (likely(atomic_read(&skb->users) == 1))
467 else if (likely(!atomic_dec_and_test(&skb->users)))
471 EXPORT_SYMBOL(consume_skb);
474 * skb_recycle_check - check if skb can be reused for receive
476 * @skb_size: minimum receive buffer size
478 * Checks that the skb passed in is not shared or cloned, and
479 * that it is linear and its head portion at least as large as
480 * skb_size so that it can be recycled as a receive buffer.
481 * If these conditions are met, this function does any necessary
482 * reference count dropping and cleans up the skbuff as if it
483 * just came from __alloc_skb().
485 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
487 struct skb_shared_info *shinfo;
492 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
495 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
496 if (skb_end_pointer(skb) - skb->head < skb_size)
499 if (skb_shared(skb) || skb_cloned(skb))
502 skb_release_head_state(skb);
504 shinfo = skb_shinfo(skb);
505 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
506 atomic_set(&shinfo->dataref, 1);
508 memset(skb, 0, offsetof(struct sk_buff, tail));
509 skb->data = skb->head + NET_SKB_PAD;
510 skb_reset_tail_pointer(skb);
514 EXPORT_SYMBOL(skb_recycle_check);
516 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
518 new->tstamp = old->tstamp;
520 new->transport_header = old->transport_header;
521 new->network_header = old->network_header;
522 new->mac_header = old->mac_header;
523 skb_dst_copy(new, old);
524 new->rxhash = old->rxhash;
526 new->sp = secpath_get(old->sp);
528 memcpy(new->cb, old->cb, sizeof(old->cb));
529 new->csum = old->csum;
530 new->local_df = old->local_df;
531 new->pkt_type = old->pkt_type;
532 new->ip_summed = old->ip_summed;
533 skb_copy_queue_mapping(new, old);
534 new->priority = old->priority;
535 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
536 new->ipvs_property = old->ipvs_property;
538 new->protocol = old->protocol;
539 new->mark = old->mark;
540 new->skb_iif = old->skb_iif;
542 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
543 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
544 new->nf_trace = old->nf_trace;
546 #ifdef CONFIG_NET_SCHED
547 new->tc_index = old->tc_index;
548 #ifdef CONFIG_NET_CLS_ACT
549 new->tc_verd = old->tc_verd;
552 new->vlan_tci = old->vlan_tci;
554 skb_copy_secmark(new, old);
558 * You should not add any new code to this function. Add it to
559 * __copy_skb_header above instead.
561 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
563 #define C(x) n->x = skb->x
565 n->next = n->prev = NULL;
567 __copy_skb_header(n, skb);
573 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
576 n->destructor = NULL;
582 atomic_set(&n->users, 1);
584 atomic_inc(&(skb_shinfo(skb)->dataref));
592 * skb_morph - morph one skb into another
593 * @dst: the skb to receive the contents
594 * @src: the skb to supply the contents
596 * This is identical to skb_clone except that the target skb is
597 * supplied by the user.
599 * The target skb is returned upon exit.
601 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
603 skb_release_all(dst);
604 return __skb_clone(dst, src);
606 EXPORT_SYMBOL_GPL(skb_morph);
609 * skb_clone - duplicate an sk_buff
610 * @skb: buffer to clone
611 * @gfp_mask: allocation priority
613 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
614 * copies share the same packet data but not structure. The new
615 * buffer has a reference count of 1. If the allocation fails the
616 * function returns %NULL otherwise the new buffer is returned.
618 * If this function is called from an interrupt gfp_mask() must be
622 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
627 if (skb->fclone == SKB_FCLONE_ORIG &&
628 n->fclone == SKB_FCLONE_UNAVAILABLE) {
629 atomic_t *fclone_ref = (atomic_t *) (n + 1);
630 n->fclone = SKB_FCLONE_CLONE;
631 atomic_inc(fclone_ref);
633 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
637 kmemcheck_annotate_bitfield(n, flags1);
638 kmemcheck_annotate_bitfield(n, flags2);
639 n->fclone = SKB_FCLONE_UNAVAILABLE;
642 return __skb_clone(n, skb);
644 EXPORT_SYMBOL(skb_clone);
646 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
648 #ifndef NET_SKBUFF_DATA_USES_OFFSET
650 * Shift between the two data areas in bytes
652 unsigned long offset = new->data - old->data;
655 __copy_skb_header(new, old);
657 #ifndef NET_SKBUFF_DATA_USES_OFFSET
658 /* {transport,network,mac}_header are relative to skb->head */
659 new->transport_header += offset;
660 new->network_header += offset;
661 if (skb_mac_header_was_set(new))
662 new->mac_header += offset;
664 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
665 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
666 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
670 * skb_copy - create private copy of an sk_buff
671 * @skb: buffer to copy
672 * @gfp_mask: allocation priority
674 * Make a copy of both an &sk_buff and its data. This is used when the
675 * caller wishes to modify the data and needs a private copy of the
676 * data to alter. Returns %NULL on failure or the pointer to the buffer
677 * on success. The returned buffer has a reference count of 1.
679 * As by-product this function converts non-linear &sk_buff to linear
680 * one, so that &sk_buff becomes completely private and caller is allowed
681 * to modify all the data of returned buffer. This means that this
682 * function is not recommended for use in circumstances when only
683 * header is going to be modified. Use pskb_copy() instead.
686 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
688 int headerlen = skb->data - skb->head;
690 * Allocate the copy buffer
693 #ifdef NET_SKBUFF_DATA_USES_OFFSET
694 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
696 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
701 /* Set the data pointer */
702 skb_reserve(n, headerlen);
703 /* Set the tail pointer and length */
704 skb_put(n, skb->len);
706 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
709 copy_skb_header(n, skb);
712 EXPORT_SYMBOL(skb_copy);
715 * pskb_copy - create copy of an sk_buff with private head.
716 * @skb: buffer to copy
717 * @gfp_mask: allocation priority
719 * Make a copy of both an &sk_buff and part of its data, located
720 * in header. Fragmented data remain shared. This is used when
721 * the caller wishes to modify only header of &sk_buff and needs
722 * private copy of the header to alter. Returns %NULL on failure
723 * or the pointer to the buffer on success.
724 * The returned buffer has a reference count of 1.
727 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
730 * Allocate the copy buffer
733 #ifdef NET_SKBUFF_DATA_USES_OFFSET
734 n = alloc_skb(skb->end, gfp_mask);
736 n = alloc_skb(skb->end - skb->head, gfp_mask);
741 /* Set the data pointer */
742 skb_reserve(n, skb->data - skb->head);
743 /* Set the tail pointer and length */
744 skb_put(n, skb_headlen(skb));
746 skb_copy_from_linear_data(skb, n->data, n->len);
748 n->truesize += skb->data_len;
749 n->data_len = skb->data_len;
752 if (skb_shinfo(skb)->nr_frags) {
755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
756 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
757 get_page(skb_shinfo(n)->frags[i].page);
759 skb_shinfo(n)->nr_frags = i;
762 if (skb_has_frags(skb)) {
763 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
764 skb_clone_fraglist(n);
767 copy_skb_header(n, skb);
771 EXPORT_SYMBOL(pskb_copy);
774 * pskb_expand_head - reallocate header of &sk_buff
775 * @skb: buffer to reallocate
776 * @nhead: room to add at head
777 * @ntail: room to add at tail
778 * @gfp_mask: allocation priority
780 * Expands (or creates identical copy, if &nhead and &ntail are zero)
781 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
782 * reference count of 1. Returns zero in the case of success or error,
783 * if expansion failed. In the last case, &sk_buff is not changed.
785 * All the pointers pointing into skb header may change and must be
786 * reloaded after call to this function.
789 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
794 #ifdef NET_SKBUFF_DATA_USES_OFFSET
795 int size = nhead + skb->end + ntail;
797 int size = nhead + (skb->end - skb->head) + ntail;
806 size = SKB_DATA_ALIGN(size);
808 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
812 /* Copy only real data... and, alas, header. This should be
813 * optimized for the cases when header is void. */
814 #ifdef NET_SKBUFF_DATA_USES_OFFSET
815 memcpy(data + nhead, skb->head, skb->tail);
817 memcpy(data + nhead, skb->head, skb->tail - skb->head);
819 memcpy(data + size, skb_end_pointer(skb),
820 sizeof(struct skb_shared_info));
822 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
823 get_page(skb_shinfo(skb)->frags[i].page);
825 if (skb_has_frags(skb))
826 skb_clone_fraglist(skb);
828 skb_release_data(skb);
830 off = (data + nhead) - skb->head;
834 #ifdef NET_SKBUFF_DATA_USES_OFFSET
838 skb->end = skb->head + size;
840 /* {transport,network,mac}_header and tail are relative to skb->head */
842 skb->transport_header += off;
843 skb->network_header += off;
844 if (skb_mac_header_was_set(skb))
845 skb->mac_header += off;
846 skb->csum_start += nhead;
850 atomic_set(&skb_shinfo(skb)->dataref, 1);
856 EXPORT_SYMBOL(pskb_expand_head);
858 /* Make private copy of skb with writable head and some headroom */
860 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
862 struct sk_buff *skb2;
863 int delta = headroom - skb_headroom(skb);
866 skb2 = pskb_copy(skb, GFP_ATOMIC);
868 skb2 = skb_clone(skb, GFP_ATOMIC);
869 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
877 EXPORT_SYMBOL(skb_realloc_headroom);
880 * skb_copy_expand - copy and expand sk_buff
881 * @skb: buffer to copy
882 * @newheadroom: new free bytes at head
883 * @newtailroom: new free bytes at tail
884 * @gfp_mask: allocation priority
886 * Make a copy of both an &sk_buff and its data and while doing so
887 * allocate additional space.
889 * This is used when the caller wishes to modify the data and needs a
890 * private copy of the data to alter as well as more space for new fields.
891 * Returns %NULL on failure or the pointer to the buffer
892 * on success. The returned buffer has a reference count of 1.
894 * You must pass %GFP_ATOMIC as the allocation priority if this function
895 * is called from an interrupt.
897 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
898 int newheadroom, int newtailroom,
902 * Allocate the copy buffer
904 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
906 int oldheadroom = skb_headroom(skb);
907 int head_copy_len, head_copy_off;
913 skb_reserve(n, newheadroom);
915 /* Set the tail pointer and length */
916 skb_put(n, skb->len);
918 head_copy_len = oldheadroom;
920 if (newheadroom <= head_copy_len)
921 head_copy_len = newheadroom;
923 head_copy_off = newheadroom - head_copy_len;
925 /* Copy the linear header and data. */
926 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
927 skb->len + head_copy_len))
930 copy_skb_header(n, skb);
932 off = newheadroom - oldheadroom;
933 n->csum_start += off;
934 #ifdef NET_SKBUFF_DATA_USES_OFFSET
935 n->transport_header += off;
936 n->network_header += off;
937 if (skb_mac_header_was_set(skb))
938 n->mac_header += off;
943 EXPORT_SYMBOL(skb_copy_expand);
946 * skb_pad - zero pad the tail of an skb
947 * @skb: buffer to pad
950 * Ensure that a buffer is followed by a padding area that is zero
951 * filled. Used by network drivers which may DMA or transfer data
952 * beyond the buffer end onto the wire.
954 * May return error in out of memory cases. The skb is freed on error.
957 int skb_pad(struct sk_buff *skb, int pad)
962 /* If the skbuff is non linear tailroom is always zero.. */
963 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
964 memset(skb->data+skb->len, 0, pad);
968 ntail = skb->data_len + pad - (skb->end - skb->tail);
969 if (likely(skb_cloned(skb) || ntail > 0)) {
970 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
975 /* FIXME: The use of this function with non-linear skb's really needs
978 err = skb_linearize(skb);
982 memset(skb->data + skb->len, 0, pad);
989 EXPORT_SYMBOL(skb_pad);
992 * skb_put - add data to a buffer
993 * @skb: buffer to use
994 * @len: amount of data to add
996 * This function extends the used data area of the buffer. If this would
997 * exceed the total buffer size the kernel will panic. A pointer to the
998 * first byte of the extra data is returned.
1000 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1002 unsigned char *tmp = skb_tail_pointer(skb);
1003 SKB_LINEAR_ASSERT(skb);
1006 if (unlikely(skb->tail > skb->end))
1007 skb_over_panic(skb, len, __builtin_return_address(0));
1010 EXPORT_SYMBOL(skb_put);
1013 * skb_push - add data to the start of a buffer
1014 * @skb: buffer to use
1015 * @len: amount of data to add
1017 * This function extends the used data area of the buffer at the buffer
1018 * start. If this would exceed the total buffer headroom the kernel will
1019 * panic. A pointer to the first byte of the extra data is returned.
1021 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1025 if (unlikely(skb->data<skb->head))
1026 skb_under_panic(skb, len, __builtin_return_address(0));
1029 EXPORT_SYMBOL(skb_push);
1032 * skb_pull - remove data from the start of a buffer
1033 * @skb: buffer to use
1034 * @len: amount of data to remove
1036 * This function removes data from the start of a buffer, returning
1037 * the memory to the headroom. A pointer to the next data in the buffer
1038 * is returned. Once the data has been pulled future pushes will overwrite
1041 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1043 return skb_pull_inline(skb, len);
1045 EXPORT_SYMBOL(skb_pull);
1048 * skb_trim - remove end from a buffer
1049 * @skb: buffer to alter
1052 * Cut the length of a buffer down by removing data from the tail. If
1053 * the buffer is already under the length specified it is not modified.
1054 * The skb must be linear.
1056 void skb_trim(struct sk_buff *skb, unsigned int len)
1059 __skb_trim(skb, len);
1061 EXPORT_SYMBOL(skb_trim);
1063 /* Trims skb to length len. It can change skb pointers.
1066 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1068 struct sk_buff **fragp;
1069 struct sk_buff *frag;
1070 int offset = skb_headlen(skb);
1071 int nfrags = skb_shinfo(skb)->nr_frags;
1075 if (skb_cloned(skb) &&
1076 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1083 for (; i < nfrags; i++) {
1084 int end = offset + skb_shinfo(skb)->frags[i].size;
1091 skb_shinfo(skb)->frags[i++].size = len - offset;
1094 skb_shinfo(skb)->nr_frags = i;
1096 for (; i < nfrags; i++)
1097 put_page(skb_shinfo(skb)->frags[i].page);
1099 if (skb_has_frags(skb))
1100 skb_drop_fraglist(skb);
1104 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1105 fragp = &frag->next) {
1106 int end = offset + frag->len;
1108 if (skb_shared(frag)) {
1109 struct sk_buff *nfrag;
1111 nfrag = skb_clone(frag, GFP_ATOMIC);
1112 if (unlikely(!nfrag))
1115 nfrag->next = frag->next;
1127 unlikely((err = pskb_trim(frag, len - offset))))
1131 skb_drop_list(&frag->next);
1136 if (len > skb_headlen(skb)) {
1137 skb->data_len -= skb->len - len;
1142 skb_set_tail_pointer(skb, len);
1147 EXPORT_SYMBOL(___pskb_trim);
1150 * __pskb_pull_tail - advance tail of skb header
1151 * @skb: buffer to reallocate
1152 * @delta: number of bytes to advance tail
1154 * The function makes a sense only on a fragmented &sk_buff,
1155 * it expands header moving its tail forward and copying necessary
1156 * data from fragmented part.
1158 * &sk_buff MUST have reference count of 1.
1160 * Returns %NULL (and &sk_buff does not change) if pull failed
1161 * or value of new tail of skb in the case of success.
1163 * All the pointers pointing into skb header may change and must be
1164 * reloaded after call to this function.
1167 /* Moves tail of skb head forward, copying data from fragmented part,
1168 * when it is necessary.
1169 * 1. It may fail due to malloc failure.
1170 * 2. It may change skb pointers.
1172 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1174 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1176 /* If skb has not enough free space at tail, get new one
1177 * plus 128 bytes for future expansions. If we have enough
1178 * room at tail, reallocate without expansion only if skb is cloned.
1180 int i, k, eat = (skb->tail + delta) - skb->end;
1182 if (eat > 0 || skb_cloned(skb)) {
1183 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1188 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1191 /* Optimization: no fragments, no reasons to preestimate
1192 * size of pulled pages. Superb.
1194 if (!skb_has_frags(skb))
1197 /* Estimate size of pulled pages. */
1199 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1200 if (skb_shinfo(skb)->frags[i].size >= eat)
1202 eat -= skb_shinfo(skb)->frags[i].size;
1205 /* If we need update frag list, we are in troubles.
1206 * Certainly, it possible to add an offset to skb data,
1207 * but taking into account that pulling is expected to
1208 * be very rare operation, it is worth to fight against
1209 * further bloating skb head and crucify ourselves here instead.
1210 * Pure masohism, indeed. 8)8)
1213 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1214 struct sk_buff *clone = NULL;
1215 struct sk_buff *insp = NULL;
1220 if (list->len <= eat) {
1221 /* Eaten as whole. */
1226 /* Eaten partially. */
1228 if (skb_shared(list)) {
1229 /* Sucks! We need to fork list. :-( */
1230 clone = skb_clone(list, GFP_ATOMIC);
1236 /* This may be pulled without
1240 if (!pskb_pull(list, eat)) {
1248 /* Free pulled out fragments. */
1249 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1250 skb_shinfo(skb)->frag_list = list->next;
1253 /* And insert new clone at head. */
1256 skb_shinfo(skb)->frag_list = clone;
1259 /* Success! Now we may commit changes to skb data. */
1264 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1265 if (skb_shinfo(skb)->frags[i].size <= eat) {
1266 put_page(skb_shinfo(skb)->frags[i].page);
1267 eat -= skb_shinfo(skb)->frags[i].size;
1269 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1271 skb_shinfo(skb)->frags[k].page_offset += eat;
1272 skb_shinfo(skb)->frags[k].size -= eat;
1278 skb_shinfo(skb)->nr_frags = k;
1281 skb->data_len -= delta;
1283 return skb_tail_pointer(skb);
1285 EXPORT_SYMBOL(__pskb_pull_tail);
1287 /* Copy some data bits from skb to kernel buffer. */
1289 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1291 int start = skb_headlen(skb);
1292 struct sk_buff *frag_iter;
1295 if (offset > (int)skb->len - len)
1299 if ((copy = start - offset) > 0) {
1302 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1303 if ((len -= copy) == 0)
1309 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1312 WARN_ON(start > offset + len);
1314 end = start + skb_shinfo(skb)->frags[i].size;
1315 if ((copy = end - offset) > 0) {
1321 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1323 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1324 offset - start, copy);
1325 kunmap_skb_frag(vaddr);
1327 if ((len -= copy) == 0)
1335 skb_walk_frags(skb, frag_iter) {
1338 WARN_ON(start > offset + len);
1340 end = start + frag_iter->len;
1341 if ((copy = end - offset) > 0) {
1344 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1346 if ((len -= copy) == 0)
1359 EXPORT_SYMBOL(skb_copy_bits);
1362 * Callback from splice_to_pipe(), if we need to release some pages
1363 * at the end of the spd in case we error'ed out in filling the pipe.
1365 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1367 put_page(spd->pages[i]);
1370 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1371 unsigned int *offset,
1372 struct sk_buff *skb, struct sock *sk)
1374 struct page *p = sk->sk_sndmsg_page;
1379 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1383 off = sk->sk_sndmsg_off = 0;
1384 /* hold one ref to this page until it's full */
1388 off = sk->sk_sndmsg_off;
1389 mlen = PAGE_SIZE - off;
1390 if (mlen < 64 && mlen < *len) {
1395 *len = min_t(unsigned int, *len, mlen);
1398 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1399 sk->sk_sndmsg_off += *len;
1407 * Fill page/offset/length into spd, if it can hold more pages.
1409 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1410 struct pipe_inode_info *pipe, struct page *page,
1411 unsigned int *len, unsigned int offset,
1412 struct sk_buff *skb, int linear,
1415 if (unlikely(spd->nr_pages == pipe->buffers))
1419 page = linear_to_page(page, len, &offset, skb, sk);
1425 spd->pages[spd->nr_pages] = page;
1426 spd->partial[spd->nr_pages].len = *len;
1427 spd->partial[spd->nr_pages].offset = offset;
1433 static inline void __segment_seek(struct page **page, unsigned int *poff,
1434 unsigned int *plen, unsigned int off)
1439 n = *poff / PAGE_SIZE;
1441 *page = nth_page(*page, n);
1443 *poff = *poff % PAGE_SIZE;
1447 static inline int __splice_segment(struct page *page, unsigned int poff,
1448 unsigned int plen, unsigned int *off,
1449 unsigned int *len, struct sk_buff *skb,
1450 struct splice_pipe_desc *spd, int linear,
1452 struct pipe_inode_info *pipe)
1457 /* skip this segment if already processed */
1463 /* ignore any bits we already processed */
1465 __segment_seek(&page, &poff, &plen, *off);
1470 unsigned int flen = min(*len, plen);
1472 /* the linear region may spread across several pages */
1473 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1475 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1478 __segment_seek(&page, &poff, &plen, flen);
1481 } while (*len && plen);
1487 * Map linear and fragment data from the skb to spd. It reports failure if the
1488 * pipe is full or if we already spliced the requested length.
1490 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1491 unsigned int *offset, unsigned int *len,
1492 struct splice_pipe_desc *spd, struct sock *sk)
1497 * map the linear part
1499 if (__splice_segment(virt_to_page(skb->data),
1500 (unsigned long) skb->data & (PAGE_SIZE - 1),
1502 offset, len, skb, spd, 1, sk, pipe))
1506 * then map the fragments
1508 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1509 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1511 if (__splice_segment(f->page, f->page_offset, f->size,
1512 offset, len, skb, spd, 0, sk, pipe))
1520 * Map data from the skb to a pipe. Should handle both the linear part,
1521 * the fragments, and the frag list. It does NOT handle frag lists within
1522 * the frag list, if such a thing exists. We'd probably need to recurse to
1523 * handle that cleanly.
1525 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1526 struct pipe_inode_info *pipe, unsigned int tlen,
1529 struct partial_page partial[PIPE_DEF_BUFFERS];
1530 struct page *pages[PIPE_DEF_BUFFERS];
1531 struct splice_pipe_desc spd = {
1535 .ops = &sock_pipe_buf_ops,
1536 .spd_release = sock_spd_release,
1538 struct sk_buff *frag_iter;
1539 struct sock *sk = skb->sk;
1542 if (splice_grow_spd(pipe, &spd))
1546 * __skb_splice_bits() only fails if the output has no room left,
1547 * so no point in going over the frag_list for the error case.
1549 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1555 * now see if we have a frag_list to map
1557 skb_walk_frags(skb, frag_iter) {
1560 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1567 * Drop the socket lock, otherwise we have reverse
1568 * locking dependencies between sk_lock and i_mutex
1569 * here as compared to sendfile(). We enter here
1570 * with the socket lock held, and splice_to_pipe() will
1571 * grab the pipe inode lock. For sendfile() emulation,
1572 * we call into ->sendpage() with the i_mutex lock held
1573 * and networking will grab the socket lock.
1576 ret = splice_to_pipe(pipe, &spd);
1580 splice_shrink_spd(pipe, &spd);
1585 * skb_store_bits - store bits from kernel buffer to skb
1586 * @skb: destination buffer
1587 * @offset: offset in destination
1588 * @from: source buffer
1589 * @len: number of bytes to copy
1591 * Copy the specified number of bytes from the source buffer to the
1592 * destination skb. This function handles all the messy bits of
1593 * traversing fragment lists and such.
1596 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1598 int start = skb_headlen(skb);
1599 struct sk_buff *frag_iter;
1602 if (offset > (int)skb->len - len)
1605 if ((copy = start - offset) > 0) {
1608 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1609 if ((len -= copy) == 0)
1615 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1616 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1619 WARN_ON(start > offset + len);
1621 end = start + frag->size;
1622 if ((copy = end - offset) > 0) {
1628 vaddr = kmap_skb_frag(frag);
1629 memcpy(vaddr + frag->page_offset + offset - start,
1631 kunmap_skb_frag(vaddr);
1633 if ((len -= copy) == 0)
1641 skb_walk_frags(skb, frag_iter) {
1644 WARN_ON(start > offset + len);
1646 end = start + frag_iter->len;
1647 if ((copy = end - offset) > 0) {
1650 if (skb_store_bits(frag_iter, offset - start,
1653 if ((len -= copy) == 0)
1666 EXPORT_SYMBOL(skb_store_bits);
1668 /* Checksum skb data. */
1670 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1671 int len, __wsum csum)
1673 int start = skb_headlen(skb);
1674 int i, copy = start - offset;
1675 struct sk_buff *frag_iter;
1678 /* Checksum header. */
1682 csum = csum_partial(skb->data + offset, copy, csum);
1683 if ((len -= copy) == 0)
1689 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1692 WARN_ON(start > offset + len);
1694 end = start + skb_shinfo(skb)->frags[i].size;
1695 if ((copy = end - offset) > 0) {
1698 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1702 vaddr = kmap_skb_frag(frag);
1703 csum2 = csum_partial(vaddr + frag->page_offset +
1704 offset - start, copy, 0);
1705 kunmap_skb_frag(vaddr);
1706 csum = csum_block_add(csum, csum2, pos);
1715 skb_walk_frags(skb, frag_iter) {
1718 WARN_ON(start > offset + len);
1720 end = start + frag_iter->len;
1721 if ((copy = end - offset) > 0) {
1725 csum2 = skb_checksum(frag_iter, offset - start,
1727 csum = csum_block_add(csum, csum2, pos);
1728 if ((len -= copy) == 0)
1739 EXPORT_SYMBOL(skb_checksum);
1741 /* Both of above in one bottle. */
1743 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1744 u8 *to, int len, __wsum csum)
1746 int start = skb_headlen(skb);
1747 int i, copy = start - offset;
1748 struct sk_buff *frag_iter;
1755 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1757 if ((len -= copy) == 0)
1764 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1767 WARN_ON(start > offset + len);
1769 end = start + skb_shinfo(skb)->frags[i].size;
1770 if ((copy = end - offset) > 0) {
1773 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1777 vaddr = kmap_skb_frag(frag);
1778 csum2 = csum_partial_copy_nocheck(vaddr +
1782 kunmap_skb_frag(vaddr);
1783 csum = csum_block_add(csum, csum2, pos);
1793 skb_walk_frags(skb, frag_iter) {
1797 WARN_ON(start > offset + len);
1799 end = start + frag_iter->len;
1800 if ((copy = end - offset) > 0) {
1803 csum2 = skb_copy_and_csum_bits(frag_iter,
1806 csum = csum_block_add(csum, csum2, pos);
1807 if ((len -= copy) == 0)
1818 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1820 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1825 if (skb->ip_summed == CHECKSUM_PARTIAL)
1826 csstart = skb->csum_start - skb_headroom(skb);
1828 csstart = skb_headlen(skb);
1830 BUG_ON(csstart > skb_headlen(skb));
1832 skb_copy_from_linear_data(skb, to, csstart);
1835 if (csstart != skb->len)
1836 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1837 skb->len - csstart, 0);
1839 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1840 long csstuff = csstart + skb->csum_offset;
1842 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1845 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1848 * skb_dequeue - remove from the head of the queue
1849 * @list: list to dequeue from
1851 * Remove the head of the list. The list lock is taken so the function
1852 * may be used safely with other locking list functions. The head item is
1853 * returned or %NULL if the list is empty.
1856 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1858 unsigned long flags;
1859 struct sk_buff *result;
1861 spin_lock_irqsave(&list->lock, flags);
1862 result = __skb_dequeue(list);
1863 spin_unlock_irqrestore(&list->lock, flags);
1866 EXPORT_SYMBOL(skb_dequeue);
1869 * skb_dequeue_tail - remove from the tail of the queue
1870 * @list: list to dequeue from
1872 * Remove the tail of the list. The list lock is taken so the function
1873 * may be used safely with other locking list functions. The tail item is
1874 * returned or %NULL if the list is empty.
1876 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1878 unsigned long flags;
1879 struct sk_buff *result;
1881 spin_lock_irqsave(&list->lock, flags);
1882 result = __skb_dequeue_tail(list);
1883 spin_unlock_irqrestore(&list->lock, flags);
1886 EXPORT_SYMBOL(skb_dequeue_tail);
1889 * skb_queue_purge - empty a list
1890 * @list: list to empty
1892 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1893 * the list and one reference dropped. This function takes the list
1894 * lock and is atomic with respect to other list locking functions.
1896 void skb_queue_purge(struct sk_buff_head *list)
1898 struct sk_buff *skb;
1899 while ((skb = skb_dequeue(list)) != NULL)
1902 EXPORT_SYMBOL(skb_queue_purge);
1905 * skb_queue_head - queue a buffer at the list head
1906 * @list: list to use
1907 * @newsk: buffer to queue
1909 * Queue a buffer at the start of the list. This function takes the
1910 * list lock and can be used safely with other locking &sk_buff functions
1913 * A buffer cannot be placed on two lists at the same time.
1915 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1917 unsigned long flags;
1919 spin_lock_irqsave(&list->lock, flags);
1920 __skb_queue_head(list, newsk);
1921 spin_unlock_irqrestore(&list->lock, flags);
1923 EXPORT_SYMBOL(skb_queue_head);
1926 * skb_queue_tail - queue a buffer at the list tail
1927 * @list: list to use
1928 * @newsk: buffer to queue
1930 * Queue a buffer at the tail of the list. This function takes the
1931 * list lock and can be used safely with other locking &sk_buff functions
1934 * A buffer cannot be placed on two lists at the same time.
1936 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1938 unsigned long flags;
1940 spin_lock_irqsave(&list->lock, flags);
1941 __skb_queue_tail(list, newsk);
1942 spin_unlock_irqrestore(&list->lock, flags);
1944 EXPORT_SYMBOL(skb_queue_tail);
1947 * skb_unlink - remove a buffer from a list
1948 * @skb: buffer to remove
1949 * @list: list to use
1951 * Remove a packet from a list. The list locks are taken and this
1952 * function is atomic with respect to other list locked calls
1954 * You must know what list the SKB is on.
1956 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1958 unsigned long flags;
1960 spin_lock_irqsave(&list->lock, flags);
1961 __skb_unlink(skb, list);
1962 spin_unlock_irqrestore(&list->lock, flags);
1964 EXPORT_SYMBOL(skb_unlink);
1967 * skb_append - append a buffer
1968 * @old: buffer to insert after
1969 * @newsk: buffer to insert
1970 * @list: list to use
1972 * Place a packet after a given packet in a list. The list locks are taken
1973 * and this function is atomic with respect to other list locked calls.
1974 * A buffer cannot be placed on two lists at the same time.
1976 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1978 unsigned long flags;
1980 spin_lock_irqsave(&list->lock, flags);
1981 __skb_queue_after(list, old, newsk);
1982 spin_unlock_irqrestore(&list->lock, flags);
1984 EXPORT_SYMBOL(skb_append);
1987 * skb_insert - insert a buffer
1988 * @old: buffer to insert before
1989 * @newsk: buffer to insert
1990 * @list: list to use
1992 * Place a packet before a given packet in a list. The list locks are
1993 * taken and this function is atomic with respect to other list locked
1996 * A buffer cannot be placed on two lists at the same time.
1998 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2000 unsigned long flags;
2002 spin_lock_irqsave(&list->lock, flags);
2003 __skb_insert(newsk, old->prev, old, list);
2004 spin_unlock_irqrestore(&list->lock, flags);
2006 EXPORT_SYMBOL(skb_insert);
2008 static inline void skb_split_inside_header(struct sk_buff *skb,
2009 struct sk_buff* skb1,
2010 const u32 len, const int pos)
2014 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2016 /* And move data appendix as is. */
2017 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2018 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2020 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2021 skb_shinfo(skb)->nr_frags = 0;
2022 skb1->data_len = skb->data_len;
2023 skb1->len += skb1->data_len;
2026 skb_set_tail_pointer(skb, len);
2029 static inline void skb_split_no_header(struct sk_buff *skb,
2030 struct sk_buff* skb1,
2031 const u32 len, int pos)
2034 const int nfrags = skb_shinfo(skb)->nr_frags;
2036 skb_shinfo(skb)->nr_frags = 0;
2037 skb1->len = skb1->data_len = skb->len - len;
2039 skb->data_len = len - pos;
2041 for (i = 0; i < nfrags; i++) {
2042 int size = skb_shinfo(skb)->frags[i].size;
2044 if (pos + size > len) {
2045 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2049 * We have two variants in this case:
2050 * 1. Move all the frag to the second
2051 * part, if it is possible. F.e.
2052 * this approach is mandatory for TUX,
2053 * where splitting is expensive.
2054 * 2. Split is accurately. We make this.
2056 get_page(skb_shinfo(skb)->frags[i].page);
2057 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2058 skb_shinfo(skb1)->frags[0].size -= len - pos;
2059 skb_shinfo(skb)->frags[i].size = len - pos;
2060 skb_shinfo(skb)->nr_frags++;
2064 skb_shinfo(skb)->nr_frags++;
2067 skb_shinfo(skb1)->nr_frags = k;
2071 * skb_split - Split fragmented skb to two parts at length len.
2072 * @skb: the buffer to split
2073 * @skb1: the buffer to receive the second part
2074 * @len: new length for skb
2076 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2078 int pos = skb_headlen(skb);
2080 if (len < pos) /* Split line is inside header. */
2081 skb_split_inside_header(skb, skb1, len, pos);
2082 else /* Second chunk has no header, nothing to copy. */
2083 skb_split_no_header(skb, skb1, len, pos);
2085 EXPORT_SYMBOL(skb_split);
2087 /* Shifting from/to a cloned skb is a no-go.
2089 * Caller cannot keep skb_shinfo related pointers past calling here!
2091 static int skb_prepare_for_shift(struct sk_buff *skb)
2093 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2097 * skb_shift - Shifts paged data partially from skb to another
2098 * @tgt: buffer into which tail data gets added
2099 * @skb: buffer from which the paged data comes from
2100 * @shiftlen: shift up to this many bytes
2102 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2103 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2104 * It's up to caller to free skb if everything was shifted.
2106 * If @tgt runs out of frags, the whole operation is aborted.
2108 * Skb cannot include anything else but paged data while tgt is allowed
2109 * to have non-paged data as well.
2111 * TODO: full sized shift could be optimized but that would need
2112 * specialized skb free'er to handle frags without up-to-date nr_frags.
2114 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2116 int from, to, merge, todo;
2117 struct skb_frag_struct *fragfrom, *fragto;
2119 BUG_ON(shiftlen > skb->len);
2120 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2124 to = skb_shinfo(tgt)->nr_frags;
2125 fragfrom = &skb_shinfo(skb)->frags[from];
2127 /* Actual merge is delayed until the point when we know we can
2128 * commit all, so that we don't have to undo partial changes
2131 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2136 todo -= fragfrom->size;
2138 if (skb_prepare_for_shift(skb) ||
2139 skb_prepare_for_shift(tgt))
2142 /* All previous frag pointers might be stale! */
2143 fragfrom = &skb_shinfo(skb)->frags[from];
2144 fragto = &skb_shinfo(tgt)->frags[merge];
2146 fragto->size += shiftlen;
2147 fragfrom->size -= shiftlen;
2148 fragfrom->page_offset += shiftlen;
2156 /* Skip full, not-fitting skb to avoid expensive operations */
2157 if ((shiftlen == skb->len) &&
2158 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2161 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2164 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2165 if (to == MAX_SKB_FRAGS)
2168 fragfrom = &skb_shinfo(skb)->frags[from];
2169 fragto = &skb_shinfo(tgt)->frags[to];
2171 if (todo >= fragfrom->size) {
2172 *fragto = *fragfrom;
2173 todo -= fragfrom->size;
2178 get_page(fragfrom->page);
2179 fragto->page = fragfrom->page;
2180 fragto->page_offset = fragfrom->page_offset;
2181 fragto->size = todo;
2183 fragfrom->page_offset += todo;
2184 fragfrom->size -= todo;
2192 /* Ready to "commit" this state change to tgt */
2193 skb_shinfo(tgt)->nr_frags = to;
2196 fragfrom = &skb_shinfo(skb)->frags[0];
2197 fragto = &skb_shinfo(tgt)->frags[merge];
2199 fragto->size += fragfrom->size;
2200 put_page(fragfrom->page);
2203 /* Reposition in the original skb */
2205 while (from < skb_shinfo(skb)->nr_frags)
2206 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2207 skb_shinfo(skb)->nr_frags = to;
2209 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2212 /* Most likely the tgt won't ever need its checksum anymore, skb on
2213 * the other hand might need it if it needs to be resent
2215 tgt->ip_summed = CHECKSUM_PARTIAL;
2216 skb->ip_summed = CHECKSUM_PARTIAL;
2218 /* Yak, is it really working this way? Some helper please? */
2219 skb->len -= shiftlen;
2220 skb->data_len -= shiftlen;
2221 skb->truesize -= shiftlen;
2222 tgt->len += shiftlen;
2223 tgt->data_len += shiftlen;
2224 tgt->truesize += shiftlen;
2230 * skb_prepare_seq_read - Prepare a sequential read of skb data
2231 * @skb: the buffer to read
2232 * @from: lower offset of data to be read
2233 * @to: upper offset of data to be read
2234 * @st: state variable
2236 * Initializes the specified state variable. Must be called before
2237 * invoking skb_seq_read() for the first time.
2239 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2240 unsigned int to, struct skb_seq_state *st)
2242 st->lower_offset = from;
2243 st->upper_offset = to;
2244 st->root_skb = st->cur_skb = skb;
2245 st->frag_idx = st->stepped_offset = 0;
2246 st->frag_data = NULL;
2248 EXPORT_SYMBOL(skb_prepare_seq_read);
2251 * skb_seq_read - Sequentially read skb data
2252 * @consumed: number of bytes consumed by the caller so far
2253 * @data: destination pointer for data to be returned
2254 * @st: state variable
2256 * Reads a block of skb data at &consumed relative to the
2257 * lower offset specified to skb_prepare_seq_read(). Assigns
2258 * the head of the data block to &data and returns the length
2259 * of the block or 0 if the end of the skb data or the upper
2260 * offset has been reached.
2262 * The caller is not required to consume all of the data
2263 * returned, i.e. &consumed is typically set to the number
2264 * of bytes already consumed and the next call to
2265 * skb_seq_read() will return the remaining part of the block.
2267 * Note 1: The size of each block of data returned can be arbitary,
2268 * this limitation is the cost for zerocopy seqeuental
2269 * reads of potentially non linear data.
2271 * Note 2: Fragment lists within fragments are not implemented
2272 * at the moment, state->root_skb could be replaced with
2273 * a stack for this purpose.
2275 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2276 struct skb_seq_state *st)
2278 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2281 if (unlikely(abs_offset >= st->upper_offset))
2285 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2287 if (abs_offset < block_limit && !st->frag_data) {
2288 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2289 return block_limit - abs_offset;
2292 if (st->frag_idx == 0 && !st->frag_data)
2293 st->stepped_offset += skb_headlen(st->cur_skb);
2295 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2296 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2297 block_limit = frag->size + st->stepped_offset;
2299 if (abs_offset < block_limit) {
2301 st->frag_data = kmap_skb_frag(frag);
2303 *data = (u8 *) st->frag_data + frag->page_offset +
2304 (abs_offset - st->stepped_offset);
2306 return block_limit - abs_offset;
2309 if (st->frag_data) {
2310 kunmap_skb_frag(st->frag_data);
2311 st->frag_data = NULL;
2315 st->stepped_offset += frag->size;
2318 if (st->frag_data) {
2319 kunmap_skb_frag(st->frag_data);
2320 st->frag_data = NULL;
2323 if (st->root_skb == st->cur_skb && skb_has_frags(st->root_skb)) {
2324 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2327 } else if (st->cur_skb->next) {
2328 st->cur_skb = st->cur_skb->next;
2335 EXPORT_SYMBOL(skb_seq_read);
2338 * skb_abort_seq_read - Abort a sequential read of skb data
2339 * @st: state variable
2341 * Must be called if skb_seq_read() was not called until it
2344 void skb_abort_seq_read(struct skb_seq_state *st)
2347 kunmap_skb_frag(st->frag_data);
2349 EXPORT_SYMBOL(skb_abort_seq_read);
2351 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2353 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2354 struct ts_config *conf,
2355 struct ts_state *state)
2357 return skb_seq_read(offset, text, TS_SKB_CB(state));
2360 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2362 skb_abort_seq_read(TS_SKB_CB(state));
2366 * skb_find_text - Find a text pattern in skb data
2367 * @skb: the buffer to look in
2368 * @from: search offset
2370 * @config: textsearch configuration
2371 * @state: uninitialized textsearch state variable
2373 * Finds a pattern in the skb data according to the specified
2374 * textsearch configuration. Use textsearch_next() to retrieve
2375 * subsequent occurrences of the pattern. Returns the offset
2376 * to the first occurrence or UINT_MAX if no match was found.
2378 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2379 unsigned int to, struct ts_config *config,
2380 struct ts_state *state)
2384 config->get_next_block = skb_ts_get_next_block;
2385 config->finish = skb_ts_finish;
2387 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2389 ret = textsearch_find(config, state);
2390 return (ret <= to - from ? ret : UINT_MAX);
2392 EXPORT_SYMBOL(skb_find_text);
2395 * skb_append_datato_frags: - append the user data to a skb
2396 * @sk: sock structure
2397 * @skb: skb structure to be appened with user data.
2398 * @getfrag: call back function to be used for getting the user data
2399 * @from: pointer to user message iov
2400 * @length: length of the iov message
2402 * Description: This procedure append the user data in the fragment part
2403 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2405 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2406 int (*getfrag)(void *from, char *to, int offset,
2407 int len, int odd, struct sk_buff *skb),
2408 void *from, int length)
2411 skb_frag_t *frag = NULL;
2412 struct page *page = NULL;
2418 /* Return error if we don't have space for new frag */
2419 frg_cnt = skb_shinfo(skb)->nr_frags;
2420 if (frg_cnt >= MAX_SKB_FRAGS)
2423 /* allocate a new page for next frag */
2424 page = alloc_pages(sk->sk_allocation, 0);
2426 /* If alloc_page fails just return failure and caller will
2427 * free previous allocated pages by doing kfree_skb()
2432 /* initialize the next frag */
2433 sk->sk_sndmsg_page = page;
2434 sk->sk_sndmsg_off = 0;
2435 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2436 skb->truesize += PAGE_SIZE;
2437 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2439 /* get the new initialized frag */
2440 frg_cnt = skb_shinfo(skb)->nr_frags;
2441 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2443 /* copy the user data to page */
2444 left = PAGE_SIZE - frag->page_offset;
2445 copy = (length > left)? left : length;
2447 ret = getfrag(from, (page_address(frag->page) +
2448 frag->page_offset + frag->size),
2449 offset, copy, 0, skb);
2453 /* copy was successful so update the size parameters */
2454 sk->sk_sndmsg_off += copy;
2457 skb->data_len += copy;
2461 } while (length > 0);
2465 EXPORT_SYMBOL(skb_append_datato_frags);
2468 * skb_pull_rcsum - pull skb and update receive checksum
2469 * @skb: buffer to update
2470 * @len: length of data pulled
2472 * This function performs an skb_pull on the packet and updates
2473 * the CHECKSUM_COMPLETE checksum. It should be used on
2474 * receive path processing instead of skb_pull unless you know
2475 * that the checksum difference is zero (e.g., a valid IP header)
2476 * or you are setting ip_summed to CHECKSUM_NONE.
2478 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2480 BUG_ON(len > skb->len);
2482 BUG_ON(skb->len < skb->data_len);
2483 skb_postpull_rcsum(skb, skb->data, len);
2484 return skb->data += len;
2487 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2490 * skb_segment - Perform protocol segmentation on skb.
2491 * @skb: buffer to segment
2492 * @features: features for the output path (see dev->features)
2494 * This function performs segmentation on the given skb. It returns
2495 * a pointer to the first in a list of new skbs for the segments.
2496 * In case of error it returns ERR_PTR(err).
2498 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2500 struct sk_buff *segs = NULL;
2501 struct sk_buff *tail = NULL;
2502 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2503 unsigned int mss = skb_shinfo(skb)->gso_size;
2504 unsigned int doffset = skb->data - skb_mac_header(skb);
2505 unsigned int offset = doffset;
2506 unsigned int headroom;
2508 int sg = features & NETIF_F_SG;
2509 int nfrags = skb_shinfo(skb)->nr_frags;
2514 __skb_push(skb, doffset);
2515 headroom = skb_headroom(skb);
2516 pos = skb_headlen(skb);
2519 struct sk_buff *nskb;
2524 len = skb->len - offset;
2528 hsize = skb_headlen(skb) - offset;
2531 if (hsize > len || !sg)
2534 if (!hsize && i >= nfrags) {
2535 BUG_ON(fskb->len != len);
2538 nskb = skb_clone(fskb, GFP_ATOMIC);
2541 if (unlikely(!nskb))
2544 hsize = skb_end_pointer(nskb) - nskb->head;
2545 if (skb_cow_head(nskb, doffset + headroom)) {
2550 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2552 skb_release_head_state(nskb);
2553 __skb_push(nskb, doffset);
2555 nskb = alloc_skb(hsize + doffset + headroom,
2558 if (unlikely(!nskb))
2561 skb_reserve(nskb, headroom);
2562 __skb_put(nskb, doffset);
2571 __copy_skb_header(nskb, skb);
2572 nskb->mac_len = skb->mac_len;
2574 skb_reset_mac_header(nskb);
2575 skb_set_network_header(nskb, skb->mac_len);
2576 nskb->transport_header = (nskb->network_header +
2577 skb_network_header_len(skb));
2578 skb_copy_from_linear_data(skb, nskb->data, doffset);
2580 if (fskb != skb_shinfo(skb)->frag_list)
2584 nskb->ip_summed = CHECKSUM_NONE;
2585 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2591 frag = skb_shinfo(nskb)->frags;
2593 skb_copy_from_linear_data_offset(skb, offset,
2594 skb_put(nskb, hsize), hsize);
2596 while (pos < offset + len && i < nfrags) {
2597 *frag = skb_shinfo(skb)->frags[i];
2598 get_page(frag->page);
2602 frag->page_offset += offset - pos;
2603 frag->size -= offset - pos;
2606 skb_shinfo(nskb)->nr_frags++;
2608 if (pos + size <= offset + len) {
2612 frag->size -= pos + size - (offset + len);
2619 if (pos < offset + len) {
2620 struct sk_buff *fskb2 = fskb;
2622 BUG_ON(pos + fskb->len != offset + len);
2628 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2634 SKB_FRAG_ASSERT(nskb);
2635 skb_shinfo(nskb)->frag_list = fskb2;
2639 nskb->data_len = len - hsize;
2640 nskb->len += nskb->data_len;
2641 nskb->truesize += nskb->data_len;
2642 } while ((offset += len) < skb->len);
2647 while ((skb = segs)) {
2651 return ERR_PTR(err);
2653 EXPORT_SYMBOL_GPL(skb_segment);
2655 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2657 struct sk_buff *p = *head;
2658 struct sk_buff *nskb;
2659 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2660 struct skb_shared_info *pinfo = skb_shinfo(p);
2661 unsigned int headroom;
2662 unsigned int len = skb_gro_len(skb);
2663 unsigned int offset = skb_gro_offset(skb);
2664 unsigned int headlen = skb_headlen(skb);
2666 if (p->len + len >= 65536)
2669 if (pinfo->frag_list)
2671 else if (headlen <= offset) {
2674 int i = skbinfo->nr_frags;
2675 int nr_frags = pinfo->nr_frags + i;
2679 if (nr_frags > MAX_SKB_FRAGS)
2682 pinfo->nr_frags = nr_frags;
2683 skbinfo->nr_frags = 0;
2685 frag = pinfo->frags + nr_frags;
2686 frag2 = skbinfo->frags + i;
2691 frag->page_offset += offset;
2692 frag->size -= offset;
2694 skb->truesize -= skb->data_len;
2695 skb->len -= skb->data_len;
2698 NAPI_GRO_CB(skb)->free = 1;
2700 } else if (skb_gro_len(p) != pinfo->gso_size)
2703 headroom = skb_headroom(p);
2704 nskb = netdev_alloc_skb(p->dev, headroom + skb_gro_offset(p));
2705 if (unlikely(!nskb))
2708 __copy_skb_header(nskb, p);
2709 nskb->mac_len = p->mac_len;
2711 skb_reserve(nskb, headroom);
2712 __skb_put(nskb, skb_gro_offset(p));
2714 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2715 skb_set_network_header(nskb, skb_network_offset(p));
2716 skb_set_transport_header(nskb, skb_transport_offset(p));
2718 __skb_pull(p, skb_gro_offset(p));
2719 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2720 p->data - skb_mac_header(p));
2722 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2723 skb_shinfo(nskb)->frag_list = p;
2724 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2725 pinfo->gso_size = 0;
2726 skb_header_release(p);
2729 nskb->data_len += p->len;
2730 nskb->truesize += p->len;
2731 nskb->len += p->len;
2734 nskb->next = p->next;
2740 if (offset > headlen) {
2741 skbinfo->frags[0].page_offset += offset - headlen;
2742 skbinfo->frags[0].size -= offset - headlen;
2746 __skb_pull(skb, offset);
2748 p->prev->next = skb;
2750 skb_header_release(skb);
2753 NAPI_GRO_CB(p)->count++;
2758 NAPI_GRO_CB(skb)->same_flow = 1;
2761 EXPORT_SYMBOL_GPL(skb_gro_receive);
2763 void __init skb_init(void)
2765 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2766 sizeof(struct sk_buff),
2768 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2770 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2771 (2*sizeof(struct sk_buff)) +
2774 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2779 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2780 * @skb: Socket buffer containing the buffers to be mapped
2781 * @sg: The scatter-gather list to map into
2782 * @offset: The offset into the buffer's contents to start mapping
2783 * @len: Length of buffer space to be mapped
2785 * Fill the specified scatter-gather list with mappings/pointers into a
2786 * region of the buffer space attached to a socket buffer.
2789 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2791 int start = skb_headlen(skb);
2792 int i, copy = start - offset;
2793 struct sk_buff *frag_iter;
2799 sg_set_buf(sg, skb->data + offset, copy);
2801 if ((len -= copy) == 0)
2806 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2809 WARN_ON(start > offset + len);
2811 end = start + skb_shinfo(skb)->frags[i].size;
2812 if ((copy = end - offset) > 0) {
2813 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2817 sg_set_page(&sg[elt], frag->page, copy,
2818 frag->page_offset+offset-start);
2827 skb_walk_frags(skb, frag_iter) {
2830 WARN_ON(start > offset + len);
2832 end = start + frag_iter->len;
2833 if ((copy = end - offset) > 0) {
2836 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2838 if ((len -= copy) == 0)
2848 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2850 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2852 sg_mark_end(&sg[nsg - 1]);
2856 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2859 * skb_cow_data - Check that a socket buffer's data buffers are writable
2860 * @skb: The socket buffer to check.
2861 * @tailbits: Amount of trailing space to be added
2862 * @trailer: Returned pointer to the skb where the @tailbits space begins
2864 * Make sure that the data buffers attached to a socket buffer are
2865 * writable. If they are not, private copies are made of the data buffers
2866 * and the socket buffer is set to use these instead.
2868 * If @tailbits is given, make sure that there is space to write @tailbits
2869 * bytes of data beyond current end of socket buffer. @trailer will be
2870 * set to point to the skb in which this space begins.
2872 * The number of scatterlist elements required to completely map the
2873 * COW'd and extended socket buffer will be returned.
2875 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2879 struct sk_buff *skb1, **skb_p;
2881 /* If skb is cloned or its head is paged, reallocate
2882 * head pulling out all the pages (pages are considered not writable
2883 * at the moment even if they are anonymous).
2885 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2886 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2889 /* Easy case. Most of packets will go this way. */
2890 if (!skb_has_frags(skb)) {
2891 /* A little of trouble, not enough of space for trailer.
2892 * This should not happen, when stack is tuned to generate
2893 * good frames. OK, on miss we reallocate and reserve even more
2894 * space, 128 bytes is fair. */
2896 if (skb_tailroom(skb) < tailbits &&
2897 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2905 /* Misery. We are in troubles, going to mincer fragments... */
2908 skb_p = &skb_shinfo(skb)->frag_list;
2911 while ((skb1 = *skb_p) != NULL) {
2914 /* The fragment is partially pulled by someone,
2915 * this can happen on input. Copy it and everything
2918 if (skb_shared(skb1))
2921 /* If the skb is the last, worry about trailer. */
2923 if (skb1->next == NULL && tailbits) {
2924 if (skb_shinfo(skb1)->nr_frags ||
2925 skb_has_frags(skb1) ||
2926 skb_tailroom(skb1) < tailbits)
2927 ntail = tailbits + 128;
2933 skb_shinfo(skb1)->nr_frags ||
2934 skb_has_frags(skb1)) {
2935 struct sk_buff *skb2;
2937 /* Fuck, we are miserable poor guys... */
2939 skb2 = skb_copy(skb1, GFP_ATOMIC);
2941 skb2 = skb_copy_expand(skb1,
2945 if (unlikely(skb2 == NULL))
2949 skb_set_owner_w(skb2, skb1->sk);
2951 /* Looking around. Are we still alive?
2952 * OK, link new skb, drop old one */
2954 skb2->next = skb1->next;
2961 skb_p = &skb1->next;
2966 EXPORT_SYMBOL_GPL(skb_cow_data);
2968 static void sock_rmem_free(struct sk_buff *skb)
2970 struct sock *sk = skb->sk;
2972 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2976 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2978 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
2980 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
2981 (unsigned)sk->sk_rcvbuf)
2986 skb->destructor = sock_rmem_free;
2987 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2989 skb_queue_tail(&sk->sk_error_queue, skb);
2990 if (!sock_flag(sk, SOCK_DEAD))
2991 sk->sk_data_ready(sk, skb->len);
2994 EXPORT_SYMBOL(sock_queue_err_skb);
2996 void skb_tstamp_tx(struct sk_buff *orig_skb,
2997 struct skb_shared_hwtstamps *hwtstamps)
2999 struct sock *sk = orig_skb->sk;
3000 struct sock_exterr_skb *serr;
3001 struct sk_buff *skb;
3007 skb = skb_clone(orig_skb, GFP_ATOMIC);
3012 *skb_hwtstamps(skb) =
3016 * no hardware time stamps available,
3017 * so keep the skb_shared_tx and only
3018 * store software time stamp
3020 skb->tstamp = ktime_get_real();
3023 serr = SKB_EXT_ERR(skb);
3024 memset(serr, 0, sizeof(*serr));
3025 serr->ee.ee_errno = ENOMSG;
3026 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3028 err = sock_queue_err_skb(sk, skb);
3033 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3037 * skb_partial_csum_set - set up and verify partial csum values for packet
3038 * @skb: the skb to set
3039 * @start: the number of bytes after skb->data to start checksumming.
3040 * @off: the offset from start to place the checksum.
3042 * For untrusted partially-checksummed packets, we need to make sure the values
3043 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3045 * This function checks and sets those values and skb->ip_summed: if this
3046 * returns false you should drop the packet.
3048 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3050 if (unlikely(start > skb_headlen(skb)) ||
3051 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3052 if (net_ratelimit())
3054 "bad partial csum: csum=%u/%u len=%u\n",
3055 start, off, skb_headlen(skb));
3058 skb->ip_summed = CHECKSUM_PARTIAL;
3059 skb->csum_start = skb_headroom(skb) + start;
3060 skb->csum_offset = off;
3063 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3065 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3067 if (net_ratelimit())
3068 pr_warning("%s: received packets cannot be forwarded"
3069 " while LRO is enabled\n", skb->dev->name);
3071 EXPORT_SYMBOL(__skb_warn_lro_forwarding);