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>
43 #include <linux/interrupt.h>
45 #include <linux/inet.h>
46 #include <linux/slab.h>
47 #include <linux/netdevice.h>
48 #ifdef CONFIG_NET_CLS_ACT
49 #include <net/pkt_sched.h>
51 #include <linux/string.h>
52 #include <linux/skbuff.h>
53 #include <linux/splice.h>
54 #include <linux/cache.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/init.h>
57 #include <linux/scatterlist.h>
58 #include <linux/errqueue.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
71 static struct kmem_cache *skbuff_head_cache __read_mostly;
72 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
74 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
75 struct pipe_buffer *buf)
80 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
81 struct pipe_buffer *buf)
86 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
87 struct pipe_buffer *buf)
93 /* Pipe buffer operations for a socket. */
94 static struct pipe_buf_operations sock_pipe_buf_ops = {
96 .map = generic_pipe_buf_map,
97 .unmap = generic_pipe_buf_unmap,
98 .confirm = generic_pipe_buf_confirm,
99 .release = sock_pipe_buf_release,
100 .steal = sock_pipe_buf_steal,
101 .get = sock_pipe_buf_get,
105 * Keep out-of-line to prevent kernel bloat.
106 * __builtin_return_address is not used because it is not always
111 * skb_over_panic - private function
116 * Out of line support code for skb_put(). Not user callable.
118 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
120 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
121 "data:%p tail:%#lx end:%#lx dev:%s\n",
122 here, skb->len, sz, skb->head, skb->data,
123 (unsigned long)skb->tail, (unsigned long)skb->end,
124 skb->dev ? skb->dev->name : "<NULL>");
127 EXPORT_SYMBOL(skb_over_panic);
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
140 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
147 EXPORT_SYMBOL(skb_under_panic);
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);
185 size = SKB_DATA_ALIGN(size);
186 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
192 * Only clear those fields we need to clear, not those that we will
193 * actually initialise below. Hence, don't put any more fields after
194 * the tail pointer in struct sk_buff!
196 memset(skb, 0, offsetof(struct sk_buff, tail));
197 skb->truesize = size + sizeof(struct sk_buff);
198 atomic_set(&skb->users, 1);
201 skb_reset_tail_pointer(skb);
202 skb->end = skb->tail + size;
203 /* make sure we initialize shinfo sequentially */
204 shinfo = skb_shinfo(skb);
205 atomic_set(&shinfo->dataref, 1);
206 shinfo->nr_frags = 0;
207 shinfo->gso_size = 0;
208 shinfo->gso_segs = 0;
209 shinfo->gso_type = 0;
210 shinfo->ip6_frag_id = 0;
211 shinfo->tx_flags.flags = 0;
212 shinfo->frag_list = NULL;
213 memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
216 struct sk_buff *child = skb + 1;
217 atomic_t *fclone_ref = (atomic_t *) (child + 1);
219 skb->fclone = SKB_FCLONE_ORIG;
220 atomic_set(fclone_ref, 1);
222 child->fclone = SKB_FCLONE_UNAVAILABLE;
227 kmem_cache_free(cache, skb);
231 EXPORT_SYMBOL(__alloc_skb);
234 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
235 * @dev: network device to receive on
236 * @length: length to allocate
237 * @gfp_mask: get_free_pages mask, passed to alloc_skb
239 * Allocate a new &sk_buff and assign it a usage count of one. The
240 * buffer has unspecified headroom built in. Users should allocate
241 * the headroom they think they need without accounting for the
242 * built in space. The built in space is used for optimisations.
244 * %NULL is returned if there is no free memory.
246 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
247 unsigned int length, gfp_t gfp_mask)
249 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
252 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
254 skb_reserve(skb, NET_SKB_PAD);
259 EXPORT_SYMBOL(__netdev_alloc_skb);
261 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
263 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
266 page = alloc_pages_node(node, gfp_mask, 0);
269 EXPORT_SYMBOL(__netdev_alloc_page);
271 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
274 skb_fill_page_desc(skb, i, page, off, size);
276 skb->data_len += size;
277 skb->truesize += size;
279 EXPORT_SYMBOL(skb_add_rx_frag);
282 * dev_alloc_skb - allocate an skbuff for receiving
283 * @length: length to allocate
285 * Allocate a new &sk_buff and assign it a usage count of one. The
286 * buffer has unspecified headroom built in. Users should allocate
287 * the headroom they think they need without accounting for the
288 * built in space. The built in space is used for optimisations.
290 * %NULL is returned if there is no free memory. Although this function
291 * allocates memory it can be called from an interrupt.
293 struct sk_buff *dev_alloc_skb(unsigned int length)
296 * There is more code here than it seems:
297 * __dev_alloc_skb is an inline
299 return __dev_alloc_skb(length, GFP_ATOMIC);
301 EXPORT_SYMBOL(dev_alloc_skb);
303 static void skb_drop_list(struct sk_buff **listp)
305 struct sk_buff *list = *listp;
310 struct sk_buff *this = list;
316 static inline void skb_drop_fraglist(struct sk_buff *skb)
318 skb_drop_list(&skb_shinfo(skb)->frag_list);
321 static void skb_clone_fraglist(struct sk_buff *skb)
323 struct sk_buff *list;
325 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
329 static void skb_release_data(struct sk_buff *skb)
332 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
333 &skb_shinfo(skb)->dataref)) {
334 if (skb_shinfo(skb)->nr_frags) {
336 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
337 put_page(skb_shinfo(skb)->frags[i].page);
340 if (skb_shinfo(skb)->frag_list)
341 skb_drop_fraglist(skb);
348 * Free an skbuff by memory without cleaning the state.
350 static void kfree_skbmem(struct sk_buff *skb)
352 struct sk_buff *other;
353 atomic_t *fclone_ref;
355 switch (skb->fclone) {
356 case SKB_FCLONE_UNAVAILABLE:
357 kmem_cache_free(skbuff_head_cache, skb);
360 case SKB_FCLONE_ORIG:
361 fclone_ref = (atomic_t *) (skb + 2);
362 if (atomic_dec_and_test(fclone_ref))
363 kmem_cache_free(skbuff_fclone_cache, skb);
366 case SKB_FCLONE_CLONE:
367 fclone_ref = (atomic_t *) (skb + 1);
370 /* The clone portion is available for
371 * fast-cloning again.
373 skb->fclone = SKB_FCLONE_UNAVAILABLE;
375 if (atomic_dec_and_test(fclone_ref))
376 kmem_cache_free(skbuff_fclone_cache, other);
381 static void skb_release_head_state(struct sk_buff *skb)
383 dst_release(skb->dst);
385 secpath_put(skb->sp);
387 if (skb->destructor) {
389 skb->destructor(skb);
391 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
392 nf_conntrack_put(skb->nfct);
393 nf_conntrack_put_reasm(skb->nfct_reasm);
395 #ifdef CONFIG_BRIDGE_NETFILTER
396 nf_bridge_put(skb->nf_bridge);
398 /* XXX: IS this still necessary? - JHS */
399 #ifdef CONFIG_NET_SCHED
401 #ifdef CONFIG_NET_CLS_ACT
407 /* Free everything but the sk_buff shell. */
408 static void skb_release_all(struct sk_buff *skb)
410 skb_release_head_state(skb);
411 skb_release_data(skb);
415 * __kfree_skb - private function
418 * Free an sk_buff. Release anything attached to the buffer.
419 * Clean the state. This is an internal helper function. Users should
420 * always call kfree_skb
423 void __kfree_skb(struct sk_buff *skb)
425 skb_release_all(skb);
428 EXPORT_SYMBOL(__kfree_skb);
431 * kfree_skb - free an sk_buff
432 * @skb: buffer to free
434 * Drop a reference to the buffer and free it if the usage count has
437 void kfree_skb(struct sk_buff *skb)
441 if (likely(atomic_read(&skb->users) == 1))
443 else if (likely(!atomic_dec_and_test(&skb->users)))
447 EXPORT_SYMBOL(kfree_skb);
450 * skb_recycle_check - check if skb can be reused for receive
452 * @skb_size: minimum receive buffer size
454 * Checks that the skb passed in is not shared or cloned, and
455 * that it is linear and its head portion at least as large as
456 * skb_size so that it can be recycled as a receive buffer.
457 * If these conditions are met, this function does any necessary
458 * reference count dropping and cleans up the skbuff as if it
459 * just came from __alloc_skb().
461 int skb_recycle_check(struct sk_buff *skb, int skb_size)
463 struct skb_shared_info *shinfo;
465 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
468 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
469 if (skb_end_pointer(skb) - skb->head < skb_size)
472 if (skb_shared(skb) || skb_cloned(skb))
475 skb_release_head_state(skb);
476 shinfo = skb_shinfo(skb);
477 atomic_set(&shinfo->dataref, 1);
478 shinfo->nr_frags = 0;
479 shinfo->gso_size = 0;
480 shinfo->gso_segs = 0;
481 shinfo->gso_type = 0;
482 shinfo->ip6_frag_id = 0;
483 shinfo->frag_list = NULL;
485 memset(skb, 0, offsetof(struct sk_buff, tail));
486 skb->data = skb->head + NET_SKB_PAD;
487 skb_reset_tail_pointer(skb);
491 EXPORT_SYMBOL(skb_recycle_check);
493 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
495 new->tstamp = old->tstamp;
497 new->transport_header = old->transport_header;
498 new->network_header = old->network_header;
499 new->mac_header = old->mac_header;
500 new->dst = dst_clone(old->dst);
502 new->sp = secpath_get(old->sp);
504 memcpy(new->cb, old->cb, sizeof(old->cb));
505 new->csum_start = old->csum_start;
506 new->csum_offset = old->csum_offset;
507 new->local_df = old->local_df;
508 new->pkt_type = old->pkt_type;
509 new->ip_summed = old->ip_summed;
510 skb_copy_queue_mapping(new, old);
511 new->priority = old->priority;
512 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
513 new->ipvs_property = old->ipvs_property;
515 new->protocol = old->protocol;
516 new->mark = old->mark;
518 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
519 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
520 new->nf_trace = old->nf_trace;
522 #ifdef CONFIG_NET_SCHED
523 new->tc_index = old->tc_index;
524 #ifdef CONFIG_NET_CLS_ACT
525 new->tc_verd = old->tc_verd;
528 new->vlan_tci = old->vlan_tci;
530 skb_copy_secmark(new, old);
533 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
535 #define C(x) n->x = skb->x
537 n->next = n->prev = NULL;
539 __copy_skb_header(n, skb);
544 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
547 n->destructor = NULL;
554 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
558 atomic_set(&n->users, 1);
560 atomic_inc(&(skb_shinfo(skb)->dataref));
568 * skb_morph - morph one skb into another
569 * @dst: the skb to receive the contents
570 * @src: the skb to supply the contents
572 * This is identical to skb_clone except that the target skb is
573 * supplied by the user.
575 * The target skb is returned upon exit.
577 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
579 skb_release_all(dst);
580 return __skb_clone(dst, src);
582 EXPORT_SYMBOL_GPL(skb_morph);
585 * skb_clone - duplicate an sk_buff
586 * @skb: buffer to clone
587 * @gfp_mask: allocation priority
589 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
590 * copies share the same packet data but not structure. The new
591 * buffer has a reference count of 1. If the allocation fails the
592 * function returns %NULL otherwise the new buffer is returned.
594 * If this function is called from an interrupt gfp_mask() must be
598 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
603 if (skb->fclone == SKB_FCLONE_ORIG &&
604 n->fclone == SKB_FCLONE_UNAVAILABLE) {
605 atomic_t *fclone_ref = (atomic_t *) (n + 1);
606 n->fclone = SKB_FCLONE_CLONE;
607 atomic_inc(fclone_ref);
609 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
612 n->fclone = SKB_FCLONE_UNAVAILABLE;
615 return __skb_clone(n, skb);
617 EXPORT_SYMBOL(skb_clone);
619 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
621 #ifndef NET_SKBUFF_DATA_USES_OFFSET
623 * Shift between the two data areas in bytes
625 unsigned long offset = new->data - old->data;
628 __copy_skb_header(new, old);
630 #ifndef NET_SKBUFF_DATA_USES_OFFSET
631 /* {transport,network,mac}_header are relative to skb->head */
632 new->transport_header += offset;
633 new->network_header += offset;
634 new->mac_header += offset;
636 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
637 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
638 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
642 * skb_copy - create private copy of an sk_buff
643 * @skb: buffer to copy
644 * @gfp_mask: allocation priority
646 * Make a copy of both an &sk_buff and its data. This is used when the
647 * caller wishes to modify the data and needs a private copy of the
648 * data to alter. Returns %NULL on failure or the pointer to the buffer
649 * on success. The returned buffer has a reference count of 1.
651 * As by-product this function converts non-linear &sk_buff to linear
652 * one, so that &sk_buff becomes completely private and caller is allowed
653 * to modify all the data of returned buffer. This means that this
654 * function is not recommended for use in circumstances when only
655 * header is going to be modified. Use pskb_copy() instead.
658 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
660 int headerlen = skb->data - skb->head;
662 * Allocate the copy buffer
665 #ifdef NET_SKBUFF_DATA_USES_OFFSET
666 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
668 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
673 /* Set the data pointer */
674 skb_reserve(n, headerlen);
675 /* Set the tail pointer and length */
676 skb_put(n, skb->len);
678 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
681 copy_skb_header(n, skb);
684 EXPORT_SYMBOL(skb_copy);
687 * pskb_copy - create copy of an sk_buff with private head.
688 * @skb: buffer to copy
689 * @gfp_mask: allocation priority
691 * Make a copy of both an &sk_buff and part of its data, located
692 * in header. Fragmented data remain shared. This is used when
693 * the caller wishes to modify only header of &sk_buff and needs
694 * private copy of the header to alter. Returns %NULL on failure
695 * or the pointer to the buffer on success.
696 * The returned buffer has a reference count of 1.
699 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
702 * Allocate the copy buffer
705 #ifdef NET_SKBUFF_DATA_USES_OFFSET
706 n = alloc_skb(skb->end, gfp_mask);
708 n = alloc_skb(skb->end - skb->head, gfp_mask);
713 /* Set the data pointer */
714 skb_reserve(n, skb->data - skb->head);
715 /* Set the tail pointer and length */
716 skb_put(n, skb_headlen(skb));
718 skb_copy_from_linear_data(skb, n->data, n->len);
720 n->truesize += skb->data_len;
721 n->data_len = skb->data_len;
724 if (skb_shinfo(skb)->nr_frags) {
727 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
728 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
729 get_page(skb_shinfo(n)->frags[i].page);
731 skb_shinfo(n)->nr_frags = i;
734 if (skb_shinfo(skb)->frag_list) {
735 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
736 skb_clone_fraglist(n);
739 copy_skb_header(n, skb);
743 EXPORT_SYMBOL(pskb_copy);
746 * pskb_expand_head - reallocate header of &sk_buff
747 * @skb: buffer to reallocate
748 * @nhead: room to add at head
749 * @ntail: room to add at tail
750 * @gfp_mask: allocation priority
752 * Expands (or creates identical copy, if &nhead and &ntail are zero)
753 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
754 * reference count of 1. Returns zero in the case of success or error,
755 * if expansion failed. In the last case, &sk_buff is not changed.
757 * All the pointers pointing into skb header may change and must be
758 * reloaded after call to this function.
761 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
766 #ifdef NET_SKBUFF_DATA_USES_OFFSET
767 int size = nhead + skb->end + ntail;
769 int size = nhead + (skb->end - skb->head) + ntail;
778 size = SKB_DATA_ALIGN(size);
780 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
784 /* Copy only real data... and, alas, header. This should be
785 * optimized for the cases when header is void. */
786 #ifdef NET_SKBUFF_DATA_USES_OFFSET
787 memcpy(data + nhead, skb->head, skb->tail);
789 memcpy(data + nhead, skb->head, skb->tail - skb->head);
791 memcpy(data + size, skb_end_pointer(skb),
792 sizeof(struct skb_shared_info));
794 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
795 get_page(skb_shinfo(skb)->frags[i].page);
797 if (skb_shinfo(skb)->frag_list)
798 skb_clone_fraglist(skb);
800 skb_release_data(skb);
802 off = (data + nhead) - skb->head;
806 #ifdef NET_SKBUFF_DATA_USES_OFFSET
810 skb->end = skb->head + size;
812 /* {transport,network,mac}_header and tail are relative to skb->head */
814 skb->transport_header += off;
815 skb->network_header += off;
816 skb->mac_header += off;
817 skb->csum_start += nhead;
821 atomic_set(&skb_shinfo(skb)->dataref, 1);
827 EXPORT_SYMBOL(pskb_expand_head);
829 /* Make private copy of skb with writable head and some headroom */
831 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
833 struct sk_buff *skb2;
834 int delta = headroom - skb_headroom(skb);
837 skb2 = pskb_copy(skb, GFP_ATOMIC);
839 skb2 = skb_clone(skb, GFP_ATOMIC);
840 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
848 EXPORT_SYMBOL(skb_realloc_headroom);
851 * skb_copy_expand - copy and expand sk_buff
852 * @skb: buffer to copy
853 * @newheadroom: new free bytes at head
854 * @newtailroom: new free bytes at tail
855 * @gfp_mask: allocation priority
857 * Make a copy of both an &sk_buff and its data and while doing so
858 * allocate additional space.
860 * This is used when the caller wishes to modify the data and needs a
861 * private copy of the data to alter as well as more space for new fields.
862 * Returns %NULL on failure or the pointer to the buffer
863 * on success. The returned buffer has a reference count of 1.
865 * You must pass %GFP_ATOMIC as the allocation priority if this function
866 * is called from an interrupt.
868 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
869 int newheadroom, int newtailroom,
873 * Allocate the copy buffer
875 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
877 int oldheadroom = skb_headroom(skb);
878 int head_copy_len, head_copy_off;
884 skb_reserve(n, newheadroom);
886 /* Set the tail pointer and length */
887 skb_put(n, skb->len);
889 head_copy_len = oldheadroom;
891 if (newheadroom <= head_copy_len)
892 head_copy_len = newheadroom;
894 head_copy_off = newheadroom - head_copy_len;
896 /* Copy the linear header and data. */
897 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
898 skb->len + head_copy_len))
901 copy_skb_header(n, skb);
903 off = newheadroom - oldheadroom;
904 n->csum_start += off;
905 #ifdef NET_SKBUFF_DATA_USES_OFFSET
906 n->transport_header += off;
907 n->network_header += off;
908 n->mac_header += off;
913 EXPORT_SYMBOL(skb_copy_expand);
916 * skb_pad - zero pad the tail of an skb
917 * @skb: buffer to pad
920 * Ensure that a buffer is followed by a padding area that is zero
921 * filled. Used by network drivers which may DMA or transfer data
922 * beyond the buffer end onto the wire.
924 * May return error in out of memory cases. The skb is freed on error.
927 int skb_pad(struct sk_buff *skb, int pad)
932 /* If the skbuff is non linear tailroom is always zero.. */
933 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
934 memset(skb->data+skb->len, 0, pad);
938 ntail = skb->data_len + pad - (skb->end - skb->tail);
939 if (likely(skb_cloned(skb) || ntail > 0)) {
940 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
945 /* FIXME: The use of this function with non-linear skb's really needs
948 err = skb_linearize(skb);
952 memset(skb->data + skb->len, 0, pad);
959 EXPORT_SYMBOL(skb_pad);
962 * skb_put - add data to a buffer
963 * @skb: buffer to use
964 * @len: amount of data to add
966 * This function extends the used data area of the buffer. If this would
967 * exceed the total buffer size the kernel will panic. A pointer to the
968 * first byte of the extra data is returned.
970 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
972 unsigned char *tmp = skb_tail_pointer(skb);
973 SKB_LINEAR_ASSERT(skb);
976 if (unlikely(skb->tail > skb->end))
977 skb_over_panic(skb, len, __builtin_return_address(0));
980 EXPORT_SYMBOL(skb_put);
983 * skb_push - add data to the start of a buffer
984 * @skb: buffer to use
985 * @len: amount of data to add
987 * This function extends the used data area of the buffer at the buffer
988 * start. If this would exceed the total buffer headroom the kernel will
989 * panic. A pointer to the first byte of the extra data is returned.
991 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
995 if (unlikely(skb->data<skb->head))
996 skb_under_panic(skb, len, __builtin_return_address(0));
999 EXPORT_SYMBOL(skb_push);
1002 * skb_pull - remove data from the start of a buffer
1003 * @skb: buffer to use
1004 * @len: amount of data to remove
1006 * This function removes data from the start of a buffer, returning
1007 * the memory to the headroom. A pointer to the next data in the buffer
1008 * is returned. Once the data has been pulled future pushes will overwrite
1011 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1013 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1015 EXPORT_SYMBOL(skb_pull);
1018 * skb_trim - remove end from a buffer
1019 * @skb: buffer to alter
1022 * Cut the length of a buffer down by removing data from the tail. If
1023 * the buffer is already under the length specified it is not modified.
1024 * The skb must be linear.
1026 void skb_trim(struct sk_buff *skb, unsigned int len)
1029 __skb_trim(skb, len);
1031 EXPORT_SYMBOL(skb_trim);
1033 /* Trims skb to length len. It can change skb pointers.
1036 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1038 struct sk_buff **fragp;
1039 struct sk_buff *frag;
1040 int offset = skb_headlen(skb);
1041 int nfrags = skb_shinfo(skb)->nr_frags;
1045 if (skb_cloned(skb) &&
1046 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1053 for (; i < nfrags; i++) {
1054 int end = offset + skb_shinfo(skb)->frags[i].size;
1061 skb_shinfo(skb)->frags[i++].size = len - offset;
1064 skb_shinfo(skb)->nr_frags = i;
1066 for (; i < nfrags; i++)
1067 put_page(skb_shinfo(skb)->frags[i].page);
1069 if (skb_shinfo(skb)->frag_list)
1070 skb_drop_fraglist(skb);
1074 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1075 fragp = &frag->next) {
1076 int end = offset + frag->len;
1078 if (skb_shared(frag)) {
1079 struct sk_buff *nfrag;
1081 nfrag = skb_clone(frag, GFP_ATOMIC);
1082 if (unlikely(!nfrag))
1085 nfrag->next = frag->next;
1097 unlikely((err = pskb_trim(frag, len - offset))))
1101 skb_drop_list(&frag->next);
1106 if (len > skb_headlen(skb)) {
1107 skb->data_len -= skb->len - len;
1112 skb_set_tail_pointer(skb, len);
1117 EXPORT_SYMBOL(___pskb_trim);
1120 * __pskb_pull_tail - advance tail of skb header
1121 * @skb: buffer to reallocate
1122 * @delta: number of bytes to advance tail
1124 * The function makes a sense only on a fragmented &sk_buff,
1125 * it expands header moving its tail forward and copying necessary
1126 * data from fragmented part.
1128 * &sk_buff MUST have reference count of 1.
1130 * Returns %NULL (and &sk_buff does not change) if pull failed
1131 * or value of new tail of skb in the case of success.
1133 * All the pointers pointing into skb header may change and must be
1134 * reloaded after call to this function.
1137 /* Moves tail of skb head forward, copying data from fragmented part,
1138 * when it is necessary.
1139 * 1. It may fail due to malloc failure.
1140 * 2. It may change skb pointers.
1142 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1144 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1146 /* If skb has not enough free space at tail, get new one
1147 * plus 128 bytes for future expansions. If we have enough
1148 * room at tail, reallocate without expansion only if skb is cloned.
1150 int i, k, eat = (skb->tail + delta) - skb->end;
1152 if (eat > 0 || skb_cloned(skb)) {
1153 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1158 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1161 /* Optimization: no fragments, no reasons to preestimate
1162 * size of pulled pages. Superb.
1164 if (!skb_shinfo(skb)->frag_list)
1167 /* Estimate size of pulled pages. */
1169 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1170 if (skb_shinfo(skb)->frags[i].size >= eat)
1172 eat -= skb_shinfo(skb)->frags[i].size;
1175 /* If we need update frag list, we are in troubles.
1176 * Certainly, it possible to add an offset to skb data,
1177 * but taking into account that pulling is expected to
1178 * be very rare operation, it is worth to fight against
1179 * further bloating skb head and crucify ourselves here instead.
1180 * Pure masohism, indeed. 8)8)
1183 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1184 struct sk_buff *clone = NULL;
1185 struct sk_buff *insp = NULL;
1190 if (list->len <= eat) {
1191 /* Eaten as whole. */
1196 /* Eaten partially. */
1198 if (skb_shared(list)) {
1199 /* Sucks! We need to fork list. :-( */
1200 clone = skb_clone(list, GFP_ATOMIC);
1206 /* This may be pulled without
1210 if (!pskb_pull(list, eat)) {
1218 /* Free pulled out fragments. */
1219 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1220 skb_shinfo(skb)->frag_list = list->next;
1223 /* And insert new clone at head. */
1226 skb_shinfo(skb)->frag_list = clone;
1229 /* Success! Now we may commit changes to skb data. */
1234 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1235 if (skb_shinfo(skb)->frags[i].size <= eat) {
1236 put_page(skb_shinfo(skb)->frags[i].page);
1237 eat -= skb_shinfo(skb)->frags[i].size;
1239 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1241 skb_shinfo(skb)->frags[k].page_offset += eat;
1242 skb_shinfo(skb)->frags[k].size -= eat;
1248 skb_shinfo(skb)->nr_frags = k;
1251 skb->data_len -= delta;
1253 return skb_tail_pointer(skb);
1255 EXPORT_SYMBOL(__pskb_pull_tail);
1257 /* Copy some data bits from skb to kernel buffer. */
1259 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1262 int start = skb_headlen(skb);
1264 if (offset > (int)skb->len - len)
1268 if ((copy = start - offset) > 0) {
1271 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1272 if ((len -= copy) == 0)
1278 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1281 WARN_ON(start > offset + len);
1283 end = start + skb_shinfo(skb)->frags[i].size;
1284 if ((copy = end - offset) > 0) {
1290 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1292 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1293 offset - start, copy);
1294 kunmap_skb_frag(vaddr);
1296 if ((len -= copy) == 0)
1304 if (skb_shinfo(skb)->frag_list) {
1305 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1307 for (; list; list = list->next) {
1310 WARN_ON(start > offset + len);
1312 end = start + list->len;
1313 if ((copy = end - offset) > 0) {
1316 if (skb_copy_bits(list, offset - start,
1319 if ((len -= copy) == 0)
1333 EXPORT_SYMBOL(skb_copy_bits);
1336 * Callback from splice_to_pipe(), if we need to release some pages
1337 * at the end of the spd in case we error'ed out in filling the pipe.
1339 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1341 put_page(spd->pages[i]);
1344 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1345 unsigned int *offset,
1346 struct sk_buff *skb)
1348 struct sock *sk = skb->sk;
1349 struct page *p = sk->sk_sndmsg_page;
1354 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1358 off = sk->sk_sndmsg_off = 0;
1359 /* hold one ref to this page until it's full */
1363 off = sk->sk_sndmsg_off;
1364 mlen = PAGE_SIZE - off;
1365 if (mlen < 64 && mlen < *len) {
1370 *len = min_t(unsigned int, *len, mlen);
1373 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1374 sk->sk_sndmsg_off += *len;
1382 * Fill page/offset/length into spd, if it can hold more pages.
1384 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1385 unsigned int *len, unsigned int offset,
1386 struct sk_buff *skb, int linear)
1388 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1392 page = linear_to_page(page, len, &offset, skb);
1398 spd->pages[spd->nr_pages] = page;
1399 spd->partial[spd->nr_pages].len = *len;
1400 spd->partial[spd->nr_pages].offset = offset;
1406 static inline void __segment_seek(struct page **page, unsigned int *poff,
1407 unsigned int *plen, unsigned int off)
1412 n = *poff / PAGE_SIZE;
1414 *page = nth_page(*page, n);
1416 *poff = *poff % PAGE_SIZE;
1420 static inline int __splice_segment(struct page *page, unsigned int poff,
1421 unsigned int plen, unsigned int *off,
1422 unsigned int *len, struct sk_buff *skb,
1423 struct splice_pipe_desc *spd, int linear)
1428 /* skip this segment if already processed */
1434 /* ignore any bits we already processed */
1436 __segment_seek(&page, &poff, &plen, *off);
1441 unsigned int flen = min(*len, plen);
1443 /* the linear region may spread across several pages */
1444 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1446 if (spd_fill_page(spd, page, &flen, poff, skb, linear))
1449 __segment_seek(&page, &poff, &plen, flen);
1452 } while (*len && plen);
1458 * Map linear and fragment data from the skb to spd. It reports failure if the
1459 * pipe is full or if we already spliced the requested length.
1461 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1463 struct splice_pipe_desc *spd)
1468 * map the linear part
1470 if (__splice_segment(virt_to_page(skb->data),
1471 (unsigned long) skb->data & (PAGE_SIZE - 1),
1473 offset, len, skb, spd, 1))
1477 * then map the fragments
1479 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1480 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1482 if (__splice_segment(f->page, f->page_offset, f->size,
1483 offset, len, skb, spd, 0))
1491 * Map data from the skb to a pipe. Should handle both the linear part,
1492 * the fragments, and the frag list. It does NOT handle frag lists within
1493 * the frag list, if such a thing exists. We'd probably need to recurse to
1494 * handle that cleanly.
1496 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1497 struct pipe_inode_info *pipe, unsigned int tlen,
1500 struct partial_page partial[PIPE_BUFFERS];
1501 struct page *pages[PIPE_BUFFERS];
1502 struct splice_pipe_desc spd = {
1506 .ops = &sock_pipe_buf_ops,
1507 .spd_release = sock_spd_release,
1511 * __skb_splice_bits() only fails if the output has no room left,
1512 * so no point in going over the frag_list for the error case.
1514 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1520 * now see if we have a frag_list to map
1522 if (skb_shinfo(skb)->frag_list) {
1523 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1525 for (; list && tlen; list = list->next) {
1526 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1533 struct sock *sk = skb->sk;
1537 * Drop the socket lock, otherwise we have reverse
1538 * locking dependencies between sk_lock and i_mutex
1539 * here as compared to sendfile(). We enter here
1540 * with the socket lock held, and splice_to_pipe() will
1541 * grab the pipe inode lock. For sendfile() emulation,
1542 * we call into ->sendpage() with the i_mutex lock held
1543 * and networking will grab the socket lock.
1546 ret = splice_to_pipe(pipe, &spd);
1555 * skb_store_bits - store bits from kernel buffer to skb
1556 * @skb: destination buffer
1557 * @offset: offset in destination
1558 * @from: source buffer
1559 * @len: number of bytes to copy
1561 * Copy the specified number of bytes from the source buffer to the
1562 * destination skb. This function handles all the messy bits of
1563 * traversing fragment lists and such.
1566 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1569 int start = skb_headlen(skb);
1571 if (offset > (int)skb->len - len)
1574 if ((copy = start - offset) > 0) {
1577 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1578 if ((len -= copy) == 0)
1584 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1585 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1588 WARN_ON(start > offset + len);
1590 end = start + frag->size;
1591 if ((copy = end - offset) > 0) {
1597 vaddr = kmap_skb_frag(frag);
1598 memcpy(vaddr + frag->page_offset + offset - start,
1600 kunmap_skb_frag(vaddr);
1602 if ((len -= copy) == 0)
1610 if (skb_shinfo(skb)->frag_list) {
1611 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1613 for (; list; list = list->next) {
1616 WARN_ON(start > offset + len);
1618 end = start + list->len;
1619 if ((copy = end - offset) > 0) {
1622 if (skb_store_bits(list, offset - start,
1625 if ((len -= copy) == 0)
1639 EXPORT_SYMBOL(skb_store_bits);
1641 /* Checksum skb data. */
1643 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1644 int len, __wsum csum)
1646 int start = skb_headlen(skb);
1647 int i, copy = start - offset;
1650 /* Checksum header. */
1654 csum = csum_partial(skb->data + offset, copy, csum);
1655 if ((len -= copy) == 0)
1661 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1664 WARN_ON(start > offset + len);
1666 end = start + skb_shinfo(skb)->frags[i].size;
1667 if ((copy = end - offset) > 0) {
1670 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1674 vaddr = kmap_skb_frag(frag);
1675 csum2 = csum_partial(vaddr + frag->page_offset +
1676 offset - start, copy, 0);
1677 kunmap_skb_frag(vaddr);
1678 csum = csum_block_add(csum, csum2, pos);
1687 if (skb_shinfo(skb)->frag_list) {
1688 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1690 for (; list; list = list->next) {
1693 WARN_ON(start > offset + len);
1695 end = start + list->len;
1696 if ((copy = end - offset) > 0) {
1700 csum2 = skb_checksum(list, offset - start,
1702 csum = csum_block_add(csum, csum2, pos);
1703 if ((len -= copy) == 0)
1715 EXPORT_SYMBOL(skb_checksum);
1717 /* Both of above in one bottle. */
1719 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1720 u8 *to, int len, __wsum csum)
1722 int start = skb_headlen(skb);
1723 int i, copy = start - offset;
1730 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1732 if ((len -= copy) == 0)
1739 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1742 WARN_ON(start > offset + len);
1744 end = start + skb_shinfo(skb)->frags[i].size;
1745 if ((copy = end - offset) > 0) {
1748 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1752 vaddr = kmap_skb_frag(frag);
1753 csum2 = csum_partial_copy_nocheck(vaddr +
1757 kunmap_skb_frag(vaddr);
1758 csum = csum_block_add(csum, csum2, pos);
1768 if (skb_shinfo(skb)->frag_list) {
1769 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1771 for (; list; list = list->next) {
1775 WARN_ON(start > offset + len);
1777 end = start + list->len;
1778 if ((copy = end - offset) > 0) {
1781 csum2 = skb_copy_and_csum_bits(list,
1784 csum = csum_block_add(csum, csum2, pos);
1785 if ((len -= copy) == 0)
1797 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1799 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1804 if (skb->ip_summed == CHECKSUM_PARTIAL)
1805 csstart = skb->csum_start - skb_headroom(skb);
1807 csstart = skb_headlen(skb);
1809 BUG_ON(csstart > skb_headlen(skb));
1811 skb_copy_from_linear_data(skb, to, csstart);
1814 if (csstart != skb->len)
1815 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1816 skb->len - csstart, 0);
1818 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1819 long csstuff = csstart + skb->csum_offset;
1821 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1824 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1827 * skb_dequeue - remove from the head of the queue
1828 * @list: list to dequeue from
1830 * Remove the head of the list. The list lock is taken so the function
1831 * may be used safely with other locking list functions. The head item is
1832 * returned or %NULL if the list is empty.
1835 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1837 unsigned long flags;
1838 struct sk_buff *result;
1840 spin_lock_irqsave(&list->lock, flags);
1841 result = __skb_dequeue(list);
1842 spin_unlock_irqrestore(&list->lock, flags);
1845 EXPORT_SYMBOL(skb_dequeue);
1848 * skb_dequeue_tail - remove from the tail of the queue
1849 * @list: list to dequeue from
1851 * Remove the tail of the list. The list lock is taken so the function
1852 * may be used safely with other locking list functions. The tail item is
1853 * returned or %NULL if the list is empty.
1855 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1857 unsigned long flags;
1858 struct sk_buff *result;
1860 spin_lock_irqsave(&list->lock, flags);
1861 result = __skb_dequeue_tail(list);
1862 spin_unlock_irqrestore(&list->lock, flags);
1865 EXPORT_SYMBOL(skb_dequeue_tail);
1868 * skb_queue_purge - empty a list
1869 * @list: list to empty
1871 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1872 * the list and one reference dropped. This function takes the list
1873 * lock and is atomic with respect to other list locking functions.
1875 void skb_queue_purge(struct sk_buff_head *list)
1877 struct sk_buff *skb;
1878 while ((skb = skb_dequeue(list)) != NULL)
1881 EXPORT_SYMBOL(skb_queue_purge);
1884 * skb_queue_head - queue a buffer at the list head
1885 * @list: list to use
1886 * @newsk: buffer to queue
1888 * Queue a buffer at the start of the list. This function takes the
1889 * list lock and can be used safely with other locking &sk_buff functions
1892 * A buffer cannot be placed on two lists at the same time.
1894 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1896 unsigned long flags;
1898 spin_lock_irqsave(&list->lock, flags);
1899 __skb_queue_head(list, newsk);
1900 spin_unlock_irqrestore(&list->lock, flags);
1902 EXPORT_SYMBOL(skb_queue_head);
1905 * skb_queue_tail - queue a buffer at the list tail
1906 * @list: list to use
1907 * @newsk: buffer to queue
1909 * Queue a buffer at the tail 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_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1917 unsigned long flags;
1919 spin_lock_irqsave(&list->lock, flags);
1920 __skb_queue_tail(list, newsk);
1921 spin_unlock_irqrestore(&list->lock, flags);
1923 EXPORT_SYMBOL(skb_queue_tail);
1926 * skb_unlink - remove a buffer from a list
1927 * @skb: buffer to remove
1928 * @list: list to use
1930 * Remove a packet from a list. The list locks are taken and this
1931 * function is atomic with respect to other list locked calls
1933 * You must know what list the SKB is on.
1935 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1937 unsigned long flags;
1939 spin_lock_irqsave(&list->lock, flags);
1940 __skb_unlink(skb, list);
1941 spin_unlock_irqrestore(&list->lock, flags);
1943 EXPORT_SYMBOL(skb_unlink);
1946 * skb_append - append a buffer
1947 * @old: buffer to insert after
1948 * @newsk: buffer to insert
1949 * @list: list to use
1951 * Place a packet after a given packet in a list. The list locks are taken
1952 * and this function is atomic with respect to other list locked calls.
1953 * A buffer cannot be placed on two lists at the same time.
1955 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1957 unsigned long flags;
1959 spin_lock_irqsave(&list->lock, flags);
1960 __skb_queue_after(list, old, newsk);
1961 spin_unlock_irqrestore(&list->lock, flags);
1963 EXPORT_SYMBOL(skb_append);
1966 * skb_insert - insert a buffer
1967 * @old: buffer to insert before
1968 * @newsk: buffer to insert
1969 * @list: list to use
1971 * Place a packet before a given packet in a list. The list locks are
1972 * taken and this function is atomic with respect to other list locked
1975 * A buffer cannot be placed on two lists at the same time.
1977 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1979 unsigned long flags;
1981 spin_lock_irqsave(&list->lock, flags);
1982 __skb_insert(newsk, old->prev, old, list);
1983 spin_unlock_irqrestore(&list->lock, flags);
1985 EXPORT_SYMBOL(skb_insert);
1987 static inline void skb_split_inside_header(struct sk_buff *skb,
1988 struct sk_buff* skb1,
1989 const u32 len, const int pos)
1993 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1995 /* And move data appendix as is. */
1996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1997 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1999 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2000 skb_shinfo(skb)->nr_frags = 0;
2001 skb1->data_len = skb->data_len;
2002 skb1->len += skb1->data_len;
2005 skb_set_tail_pointer(skb, len);
2008 static inline void skb_split_no_header(struct sk_buff *skb,
2009 struct sk_buff* skb1,
2010 const u32 len, int pos)
2013 const int nfrags = skb_shinfo(skb)->nr_frags;
2015 skb_shinfo(skb)->nr_frags = 0;
2016 skb1->len = skb1->data_len = skb->len - len;
2018 skb->data_len = len - pos;
2020 for (i = 0; i < nfrags; i++) {
2021 int size = skb_shinfo(skb)->frags[i].size;
2023 if (pos + size > len) {
2024 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2028 * We have two variants in this case:
2029 * 1. Move all the frag to the second
2030 * part, if it is possible. F.e.
2031 * this approach is mandatory for TUX,
2032 * where splitting is expensive.
2033 * 2. Split is accurately. We make this.
2035 get_page(skb_shinfo(skb)->frags[i].page);
2036 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2037 skb_shinfo(skb1)->frags[0].size -= len - pos;
2038 skb_shinfo(skb)->frags[i].size = len - pos;
2039 skb_shinfo(skb)->nr_frags++;
2043 skb_shinfo(skb)->nr_frags++;
2046 skb_shinfo(skb1)->nr_frags = k;
2050 * skb_split - Split fragmented skb to two parts at length len.
2051 * @skb: the buffer to split
2052 * @skb1: the buffer to receive the second part
2053 * @len: new length for skb
2055 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2057 int pos = skb_headlen(skb);
2059 if (len < pos) /* Split line is inside header. */
2060 skb_split_inside_header(skb, skb1, len, pos);
2061 else /* Second chunk has no header, nothing to copy. */
2062 skb_split_no_header(skb, skb1, len, pos);
2064 EXPORT_SYMBOL(skb_split);
2066 /* Shifting from/to a cloned skb is a no-go.
2068 * Caller cannot keep skb_shinfo related pointers past calling here!
2070 static int skb_prepare_for_shift(struct sk_buff *skb)
2072 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2076 * skb_shift - Shifts paged data partially from skb to another
2077 * @tgt: buffer into which tail data gets added
2078 * @skb: buffer from which the paged data comes from
2079 * @shiftlen: shift up to this many bytes
2081 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2082 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2083 * It's up to caller to free skb if everything was shifted.
2085 * If @tgt runs out of frags, the whole operation is aborted.
2087 * Skb cannot include anything else but paged data while tgt is allowed
2088 * to have non-paged data as well.
2090 * TODO: full sized shift could be optimized but that would need
2091 * specialized skb free'er to handle frags without up-to-date nr_frags.
2093 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2095 int from, to, merge, todo;
2096 struct skb_frag_struct *fragfrom, *fragto;
2098 BUG_ON(shiftlen > skb->len);
2099 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2103 to = skb_shinfo(tgt)->nr_frags;
2104 fragfrom = &skb_shinfo(skb)->frags[from];
2106 /* Actual merge is delayed until the point when we know we can
2107 * commit all, so that we don't have to undo partial changes
2110 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2115 todo -= fragfrom->size;
2117 if (skb_prepare_for_shift(skb) ||
2118 skb_prepare_for_shift(tgt))
2121 /* All previous frag pointers might be stale! */
2122 fragfrom = &skb_shinfo(skb)->frags[from];
2123 fragto = &skb_shinfo(tgt)->frags[merge];
2125 fragto->size += shiftlen;
2126 fragfrom->size -= shiftlen;
2127 fragfrom->page_offset += shiftlen;
2135 /* Skip full, not-fitting skb to avoid expensive operations */
2136 if ((shiftlen == skb->len) &&
2137 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2140 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2143 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2144 if (to == MAX_SKB_FRAGS)
2147 fragfrom = &skb_shinfo(skb)->frags[from];
2148 fragto = &skb_shinfo(tgt)->frags[to];
2150 if (todo >= fragfrom->size) {
2151 *fragto = *fragfrom;
2152 todo -= fragfrom->size;
2157 get_page(fragfrom->page);
2158 fragto->page = fragfrom->page;
2159 fragto->page_offset = fragfrom->page_offset;
2160 fragto->size = todo;
2162 fragfrom->page_offset += todo;
2163 fragfrom->size -= todo;
2171 /* Ready to "commit" this state change to tgt */
2172 skb_shinfo(tgt)->nr_frags = to;
2175 fragfrom = &skb_shinfo(skb)->frags[0];
2176 fragto = &skb_shinfo(tgt)->frags[merge];
2178 fragto->size += fragfrom->size;
2179 put_page(fragfrom->page);
2182 /* Reposition in the original skb */
2184 while (from < skb_shinfo(skb)->nr_frags)
2185 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2186 skb_shinfo(skb)->nr_frags = to;
2188 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2191 /* Most likely the tgt won't ever need its checksum anymore, skb on
2192 * the other hand might need it if it needs to be resent
2194 tgt->ip_summed = CHECKSUM_PARTIAL;
2195 skb->ip_summed = CHECKSUM_PARTIAL;
2197 /* Yak, is it really working this way? Some helper please? */
2198 skb->len -= shiftlen;
2199 skb->data_len -= shiftlen;
2200 skb->truesize -= shiftlen;
2201 tgt->len += shiftlen;
2202 tgt->data_len += shiftlen;
2203 tgt->truesize += shiftlen;
2209 * skb_prepare_seq_read - Prepare a sequential read of skb data
2210 * @skb: the buffer to read
2211 * @from: lower offset of data to be read
2212 * @to: upper offset of data to be read
2213 * @st: state variable
2215 * Initializes the specified state variable. Must be called before
2216 * invoking skb_seq_read() for the first time.
2218 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2219 unsigned int to, struct skb_seq_state *st)
2221 st->lower_offset = from;
2222 st->upper_offset = to;
2223 st->root_skb = st->cur_skb = skb;
2224 st->frag_idx = st->stepped_offset = 0;
2225 st->frag_data = NULL;
2227 EXPORT_SYMBOL(skb_prepare_seq_read);
2230 * skb_seq_read - Sequentially read skb data
2231 * @consumed: number of bytes consumed by the caller so far
2232 * @data: destination pointer for data to be returned
2233 * @st: state variable
2235 * Reads a block of skb data at &consumed relative to the
2236 * lower offset specified to skb_prepare_seq_read(). Assigns
2237 * the head of the data block to &data and returns the length
2238 * of the block or 0 if the end of the skb data or the upper
2239 * offset has been reached.
2241 * The caller is not required to consume all of the data
2242 * returned, i.e. &consumed is typically set to the number
2243 * of bytes already consumed and the next call to
2244 * skb_seq_read() will return the remaining part of the block.
2246 * Note 1: The size of each block of data returned can be arbitary,
2247 * this limitation is the cost for zerocopy seqeuental
2248 * reads of potentially non linear data.
2250 * Note 2: Fragment lists within fragments are not implemented
2251 * at the moment, state->root_skb could be replaced with
2252 * a stack for this purpose.
2254 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2255 struct skb_seq_state *st)
2257 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2260 if (unlikely(abs_offset >= st->upper_offset))
2264 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2266 if (abs_offset < block_limit) {
2267 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2268 return block_limit - abs_offset;
2271 if (st->frag_idx == 0 && !st->frag_data)
2272 st->stepped_offset += skb_headlen(st->cur_skb);
2274 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2275 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2276 block_limit = frag->size + st->stepped_offset;
2278 if (abs_offset < block_limit) {
2280 st->frag_data = kmap_skb_frag(frag);
2282 *data = (u8 *) st->frag_data + frag->page_offset +
2283 (abs_offset - st->stepped_offset);
2285 return block_limit - abs_offset;
2288 if (st->frag_data) {
2289 kunmap_skb_frag(st->frag_data);
2290 st->frag_data = NULL;
2294 st->stepped_offset += frag->size;
2297 if (st->frag_data) {
2298 kunmap_skb_frag(st->frag_data);
2299 st->frag_data = NULL;
2302 if (st->root_skb == st->cur_skb &&
2303 skb_shinfo(st->root_skb)->frag_list) {
2304 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2307 } else if (st->cur_skb->next) {
2308 st->cur_skb = st->cur_skb->next;
2315 EXPORT_SYMBOL(skb_seq_read);
2318 * skb_abort_seq_read - Abort a sequential read of skb data
2319 * @st: state variable
2321 * Must be called if skb_seq_read() was not called until it
2324 void skb_abort_seq_read(struct skb_seq_state *st)
2327 kunmap_skb_frag(st->frag_data);
2329 EXPORT_SYMBOL(skb_abort_seq_read);
2331 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2333 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2334 struct ts_config *conf,
2335 struct ts_state *state)
2337 return skb_seq_read(offset, text, TS_SKB_CB(state));
2340 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2342 skb_abort_seq_read(TS_SKB_CB(state));
2346 * skb_find_text - Find a text pattern in skb data
2347 * @skb: the buffer to look in
2348 * @from: search offset
2350 * @config: textsearch configuration
2351 * @state: uninitialized textsearch state variable
2353 * Finds a pattern in the skb data according to the specified
2354 * textsearch configuration. Use textsearch_next() to retrieve
2355 * subsequent occurrences of the pattern. Returns the offset
2356 * to the first occurrence or UINT_MAX if no match was found.
2358 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2359 unsigned int to, struct ts_config *config,
2360 struct ts_state *state)
2364 config->get_next_block = skb_ts_get_next_block;
2365 config->finish = skb_ts_finish;
2367 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2369 ret = textsearch_find(config, state);
2370 return (ret <= to - from ? ret : UINT_MAX);
2372 EXPORT_SYMBOL(skb_find_text);
2375 * skb_append_datato_frags: - append the user data to a skb
2376 * @sk: sock structure
2377 * @skb: skb structure to be appened with user data.
2378 * @getfrag: call back function to be used for getting the user data
2379 * @from: pointer to user message iov
2380 * @length: length of the iov message
2382 * Description: This procedure append the user data in the fragment part
2383 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2385 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2386 int (*getfrag)(void *from, char *to, int offset,
2387 int len, int odd, struct sk_buff *skb),
2388 void *from, int length)
2391 skb_frag_t *frag = NULL;
2392 struct page *page = NULL;
2398 /* Return error if we don't have space for new frag */
2399 frg_cnt = skb_shinfo(skb)->nr_frags;
2400 if (frg_cnt >= MAX_SKB_FRAGS)
2403 /* allocate a new page for next frag */
2404 page = alloc_pages(sk->sk_allocation, 0);
2406 /* If alloc_page fails just return failure and caller will
2407 * free previous allocated pages by doing kfree_skb()
2412 /* initialize the next frag */
2413 sk->sk_sndmsg_page = page;
2414 sk->sk_sndmsg_off = 0;
2415 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2416 skb->truesize += PAGE_SIZE;
2417 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2419 /* get the new initialized frag */
2420 frg_cnt = skb_shinfo(skb)->nr_frags;
2421 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2423 /* copy the user data to page */
2424 left = PAGE_SIZE - frag->page_offset;
2425 copy = (length > left)? left : length;
2427 ret = getfrag(from, (page_address(frag->page) +
2428 frag->page_offset + frag->size),
2429 offset, copy, 0, skb);
2433 /* copy was successful so update the size parameters */
2434 sk->sk_sndmsg_off += copy;
2437 skb->data_len += copy;
2441 } while (length > 0);
2445 EXPORT_SYMBOL(skb_append_datato_frags);
2448 * skb_pull_rcsum - pull skb and update receive checksum
2449 * @skb: buffer to update
2450 * @len: length of data pulled
2452 * This function performs an skb_pull on the packet and updates
2453 * the CHECKSUM_COMPLETE checksum. It should be used on
2454 * receive path processing instead of skb_pull unless you know
2455 * that the checksum difference is zero (e.g., a valid IP header)
2456 * or you are setting ip_summed to CHECKSUM_NONE.
2458 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2460 BUG_ON(len > skb->len);
2462 BUG_ON(skb->len < skb->data_len);
2463 skb_postpull_rcsum(skb, skb->data, len);
2464 return skb->data += len;
2467 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2470 * skb_segment - Perform protocol segmentation on skb.
2471 * @skb: buffer to segment
2472 * @features: features for the output path (see dev->features)
2474 * This function performs segmentation on the given skb. It returns
2475 * a pointer to the first in a list of new skbs for the segments.
2476 * In case of error it returns ERR_PTR(err).
2478 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2480 struct sk_buff *segs = NULL;
2481 struct sk_buff *tail = NULL;
2482 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2483 unsigned int mss = skb_shinfo(skb)->gso_size;
2484 unsigned int doffset = skb->data - skb_mac_header(skb);
2485 unsigned int offset = doffset;
2486 unsigned int headroom;
2488 int sg = features & NETIF_F_SG;
2489 int nfrags = skb_shinfo(skb)->nr_frags;
2494 __skb_push(skb, doffset);
2495 headroom = skb_headroom(skb);
2496 pos = skb_headlen(skb);
2499 struct sk_buff *nskb;
2504 len = skb->len - offset;
2508 hsize = skb_headlen(skb) - offset;
2511 if (hsize > len || !sg)
2514 if (!hsize && i >= nfrags) {
2515 BUG_ON(fskb->len != len);
2518 nskb = skb_clone(fskb, GFP_ATOMIC);
2521 if (unlikely(!nskb))
2524 hsize = skb_end_pointer(nskb) - nskb->head;
2525 if (skb_cow_head(nskb, doffset + headroom)) {
2530 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2532 skb_release_head_state(nskb);
2533 __skb_push(nskb, doffset);
2535 nskb = alloc_skb(hsize + doffset + headroom,
2538 if (unlikely(!nskb))
2541 skb_reserve(nskb, headroom);
2542 __skb_put(nskb, doffset);
2551 __copy_skb_header(nskb, skb);
2552 nskb->mac_len = skb->mac_len;
2554 skb_reset_mac_header(nskb);
2555 skb_set_network_header(nskb, skb->mac_len);
2556 nskb->transport_header = (nskb->network_header +
2557 skb_network_header_len(skb));
2558 skb_copy_from_linear_data(skb, nskb->data, doffset);
2560 if (pos >= offset + len)
2564 nskb->ip_summed = CHECKSUM_NONE;
2565 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2571 frag = skb_shinfo(nskb)->frags;
2573 skb_copy_from_linear_data_offset(skb, offset,
2574 skb_put(nskb, hsize), hsize);
2576 while (pos < offset + len && i < nfrags) {
2577 *frag = skb_shinfo(skb)->frags[i];
2578 get_page(frag->page);
2582 frag->page_offset += offset - pos;
2583 frag->size -= offset - pos;
2586 skb_shinfo(nskb)->nr_frags++;
2588 if (pos + size <= offset + len) {
2592 frag->size -= pos + size - (offset + len);
2599 if (pos < offset + len) {
2600 struct sk_buff *fskb2 = fskb;
2602 BUG_ON(pos + fskb->len != offset + len);
2608 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2614 BUG_ON(skb_shinfo(nskb)->frag_list);
2615 skb_shinfo(nskb)->frag_list = fskb2;
2619 nskb->data_len = len - hsize;
2620 nskb->len += nskb->data_len;
2621 nskb->truesize += nskb->data_len;
2622 } while ((offset += len) < skb->len);
2627 while ((skb = segs)) {
2631 return ERR_PTR(err);
2633 EXPORT_SYMBOL_GPL(skb_segment);
2635 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2637 struct sk_buff *p = *head;
2638 struct sk_buff *nskb;
2639 unsigned int headroom;
2640 unsigned int len = skb_gro_len(skb);
2642 if (p->len + len >= 65536)
2645 if (skb_shinfo(p)->frag_list)
2647 else if (skb_headlen(skb) <= skb_gro_offset(skb)) {
2648 if (skb_shinfo(p)->nr_frags + skb_shinfo(skb)->nr_frags >
2652 skb_shinfo(skb)->frags[0].page_offset +=
2653 skb_gro_offset(skb) - skb_headlen(skb);
2654 skb_shinfo(skb)->frags[0].size -=
2655 skb_gro_offset(skb) - skb_headlen(skb);
2657 memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags,
2658 skb_shinfo(skb)->frags,
2659 skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
2661 skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags;
2662 skb_shinfo(skb)->nr_frags = 0;
2664 skb->truesize -= skb->data_len;
2665 skb->len -= skb->data_len;
2668 NAPI_GRO_CB(skb)->free = 1;
2672 headroom = skb_headroom(p);
2673 nskb = netdev_alloc_skb(p->dev, headroom + skb_gro_offset(p));
2674 if (unlikely(!nskb))
2677 __copy_skb_header(nskb, p);
2678 nskb->mac_len = p->mac_len;
2680 skb_reserve(nskb, headroom);
2681 __skb_put(nskb, skb_gro_offset(p));
2683 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2684 skb_set_network_header(nskb, skb_network_offset(p));
2685 skb_set_transport_header(nskb, skb_transport_offset(p));
2687 __skb_pull(p, skb_gro_offset(p));
2688 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2689 p->data - skb_mac_header(p));
2691 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2692 skb_shinfo(nskb)->frag_list = p;
2693 skb_shinfo(nskb)->gso_size = skb_shinfo(p)->gso_size;
2694 skb_header_release(p);
2697 nskb->data_len += p->len;
2698 nskb->truesize += p->len;
2699 nskb->len += p->len;
2702 nskb->next = p->next;
2708 if (skb_gro_offset(skb) > skb_headlen(skb)) {
2709 skb_shinfo(skb)->frags[0].page_offset +=
2710 skb_gro_offset(skb) - skb_headlen(skb);
2711 skb_shinfo(skb)->frags[0].size -=
2712 skb_gro_offset(skb) - skb_headlen(skb);
2713 skb_gro_reset_offset(skb);
2714 skb_gro_pull(skb, skb_headlen(skb));
2717 __skb_pull(skb, skb_gro_offset(skb));
2719 p->prev->next = skb;
2721 skb_header_release(skb);
2724 NAPI_GRO_CB(p)->count++;
2729 NAPI_GRO_CB(skb)->same_flow = 1;
2732 EXPORT_SYMBOL_GPL(skb_gro_receive);
2734 void __init skb_init(void)
2736 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2737 sizeof(struct sk_buff),
2739 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2741 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2742 (2*sizeof(struct sk_buff)) +
2745 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2750 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2751 * @skb: Socket buffer containing the buffers to be mapped
2752 * @sg: The scatter-gather list to map into
2753 * @offset: The offset into the buffer's contents to start mapping
2754 * @len: Length of buffer space to be mapped
2756 * Fill the specified scatter-gather list with mappings/pointers into a
2757 * region of the buffer space attached to a socket buffer.
2760 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2762 int start = skb_headlen(skb);
2763 int i, copy = start - offset;
2769 sg_set_buf(sg, skb->data + offset, copy);
2771 if ((len -= copy) == 0)
2776 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2779 WARN_ON(start > offset + len);
2781 end = start + skb_shinfo(skb)->frags[i].size;
2782 if ((copy = end - offset) > 0) {
2783 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2787 sg_set_page(&sg[elt], frag->page, copy,
2788 frag->page_offset+offset-start);
2797 if (skb_shinfo(skb)->frag_list) {
2798 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2800 for (; list; list = list->next) {
2803 WARN_ON(start > offset + len);
2805 end = start + list->len;
2806 if ((copy = end - offset) > 0) {
2809 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2811 if ((len -= copy) == 0)
2822 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2824 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2826 sg_mark_end(&sg[nsg - 1]);
2830 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2833 * skb_cow_data - Check that a socket buffer's data buffers are writable
2834 * @skb: The socket buffer to check.
2835 * @tailbits: Amount of trailing space to be added
2836 * @trailer: Returned pointer to the skb where the @tailbits space begins
2838 * Make sure that the data buffers attached to a socket buffer are
2839 * writable. If they are not, private copies are made of the data buffers
2840 * and the socket buffer is set to use these instead.
2842 * If @tailbits is given, make sure that there is space to write @tailbits
2843 * bytes of data beyond current end of socket buffer. @trailer will be
2844 * set to point to the skb in which this space begins.
2846 * The number of scatterlist elements required to completely map the
2847 * COW'd and extended socket buffer will be returned.
2849 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2853 struct sk_buff *skb1, **skb_p;
2855 /* If skb is cloned or its head is paged, reallocate
2856 * head pulling out all the pages (pages are considered not writable
2857 * at the moment even if they are anonymous).
2859 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2860 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2863 /* Easy case. Most of packets will go this way. */
2864 if (!skb_shinfo(skb)->frag_list) {
2865 /* A little of trouble, not enough of space for trailer.
2866 * This should not happen, when stack is tuned to generate
2867 * good frames. OK, on miss we reallocate and reserve even more
2868 * space, 128 bytes is fair. */
2870 if (skb_tailroom(skb) < tailbits &&
2871 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2879 /* Misery. We are in troubles, going to mincer fragments... */
2882 skb_p = &skb_shinfo(skb)->frag_list;
2885 while ((skb1 = *skb_p) != NULL) {
2888 /* The fragment is partially pulled by someone,
2889 * this can happen on input. Copy it and everything
2892 if (skb_shared(skb1))
2895 /* If the skb is the last, worry about trailer. */
2897 if (skb1->next == NULL && tailbits) {
2898 if (skb_shinfo(skb1)->nr_frags ||
2899 skb_shinfo(skb1)->frag_list ||
2900 skb_tailroom(skb1) < tailbits)
2901 ntail = tailbits + 128;
2907 skb_shinfo(skb1)->nr_frags ||
2908 skb_shinfo(skb1)->frag_list) {
2909 struct sk_buff *skb2;
2911 /* Fuck, we are miserable poor guys... */
2913 skb2 = skb_copy(skb1, GFP_ATOMIC);
2915 skb2 = skb_copy_expand(skb1,
2919 if (unlikely(skb2 == NULL))
2923 skb_set_owner_w(skb2, skb1->sk);
2925 /* Looking around. Are we still alive?
2926 * OK, link new skb, drop old one */
2928 skb2->next = skb1->next;
2935 skb_p = &skb1->next;
2940 EXPORT_SYMBOL_GPL(skb_cow_data);
2942 void skb_tstamp_tx(struct sk_buff *orig_skb,
2943 struct skb_shared_hwtstamps *hwtstamps)
2945 struct sock *sk = orig_skb->sk;
2946 struct sock_exterr_skb *serr;
2947 struct sk_buff *skb;
2953 skb = skb_clone(orig_skb, GFP_ATOMIC);
2958 *skb_hwtstamps(skb) =
2962 * no hardware time stamps available,
2963 * so keep the skb_shared_tx and only
2964 * store software time stamp
2966 skb->tstamp = ktime_get_real();
2969 serr = SKB_EXT_ERR(skb);
2970 memset(serr, 0, sizeof(*serr));
2971 serr->ee.ee_errno = ENOMSG;
2972 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
2973 err = sock_queue_err_skb(sk, skb);
2977 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
2981 * skb_partial_csum_set - set up and verify partial csum values for packet
2982 * @skb: the skb to set
2983 * @start: the number of bytes after skb->data to start checksumming.
2984 * @off: the offset from start to place the checksum.
2986 * For untrusted partially-checksummed packets, we need to make sure the values
2987 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2989 * This function checks and sets those values and skb->ip_summed: if this
2990 * returns false you should drop the packet.
2992 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2994 if (unlikely(start > skb->len - 2) ||
2995 unlikely((int)start + off > skb->len - 2)) {
2996 if (net_ratelimit())
2998 "bad partial csum: csum=%u/%u len=%u\n",
2999 start, off, skb->len);
3002 skb->ip_summed = CHECKSUM_PARTIAL;
3003 skb->csum_start = skb_headroom(skb) + start;
3004 skb->csum_offset = off;
3007 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3009 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3011 if (net_ratelimit())
3012 pr_warning("%s: received packets cannot be forwarded"
3013 " while LRO is enabled\n", skb->dev->name);
3015 EXPORT_SYMBOL(__skb_warn_lro_forwarding);