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>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
74 static struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here, skb->len, sz, skb->head, skb->data,
126 (unsigned long)skb->tail, (unsigned long)skb->end,
127 skb->dev ? skb->dev->name : "<NULL>");
132 * skb_under_panic - private function
137 * Out of line support code for skb_push(). Not user callable.
140 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here, skb->len, sz, skb->head, skb->data,
145 (unsigned long)skb->tail, (unsigned long)skb->end,
146 skb->dev ? skb->dev->name : "<NULL>");
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
187 size = SKB_DATA_ALIGN(size);
188 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
192 prefetchw(data + size);
195 * Only clear those fields we need to clear, not those that we will
196 * actually initialise below. Hence, don't put any more fields after
197 * the tail pointer in struct sk_buff!
199 memset(skb, 0, offsetof(struct sk_buff, tail));
200 skb->truesize = size + sizeof(struct sk_buff);
201 atomic_set(&skb->users, 1);
204 skb_reset_tail_pointer(skb);
205 skb->end = skb->tail + size;
206 #ifdef NET_SKBUFF_DATA_USES_OFFSET
207 skb->mac_header = ~0U;
210 /* make sure we initialize shinfo sequentially */
211 shinfo = skb_shinfo(skb);
212 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
213 atomic_set(&shinfo->dataref, 1);
214 kmemcheck_annotate_variable(shinfo->destructor_arg);
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)
254 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
256 skb_reserve(skb, NET_SKB_PAD);
261 EXPORT_SYMBOL(__netdev_alloc_skb);
263 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
266 skb_fill_page_desc(skb, i, page, off, size);
268 skb->data_len += size;
269 skb->truesize += size;
271 EXPORT_SYMBOL(skb_add_rx_frag);
274 * dev_alloc_skb - allocate an skbuff for receiving
275 * @length: length to allocate
277 * Allocate a new &sk_buff and assign it a usage count of one. The
278 * buffer has unspecified headroom built in. Users should allocate
279 * the headroom they think they need without accounting for the
280 * built in space. The built in space is used for optimisations.
282 * %NULL is returned if there is no free memory. Although this function
283 * allocates memory it can be called from an interrupt.
285 struct sk_buff *dev_alloc_skb(unsigned int length)
288 * There is more code here than it seems:
289 * __dev_alloc_skb is an inline
291 return __dev_alloc_skb(length, GFP_ATOMIC);
293 EXPORT_SYMBOL(dev_alloc_skb);
295 static void skb_drop_list(struct sk_buff **listp)
297 struct sk_buff *list = *listp;
302 struct sk_buff *this = list;
308 static inline void skb_drop_fraglist(struct sk_buff *skb)
310 skb_drop_list(&skb_shinfo(skb)->frag_list);
313 static void skb_clone_fraglist(struct sk_buff *skb)
315 struct sk_buff *list;
317 skb_walk_frags(skb, list)
321 static void skb_release_data(struct sk_buff *skb)
324 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
325 &skb_shinfo(skb)->dataref)) {
326 if (skb_shinfo(skb)->nr_frags) {
328 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
329 skb_frag_unref(skb, i);
333 * If skb buf is from userspace, we need to notify the caller
334 * the lower device DMA has done;
336 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
337 struct ubuf_info *uarg;
339 uarg = skb_shinfo(skb)->destructor_arg;
341 uarg->callback(uarg);
344 if (skb_has_frag_list(skb))
345 skb_drop_fraglist(skb);
352 * Free an skbuff by memory without cleaning the state.
354 static void kfree_skbmem(struct sk_buff *skb)
356 struct sk_buff *other;
357 atomic_t *fclone_ref;
359 switch (skb->fclone) {
360 case SKB_FCLONE_UNAVAILABLE:
361 kmem_cache_free(skbuff_head_cache, skb);
364 case SKB_FCLONE_ORIG:
365 fclone_ref = (atomic_t *) (skb + 2);
366 if (atomic_dec_and_test(fclone_ref))
367 kmem_cache_free(skbuff_fclone_cache, skb);
370 case SKB_FCLONE_CLONE:
371 fclone_ref = (atomic_t *) (skb + 1);
374 /* The clone portion is available for
375 * fast-cloning again.
377 skb->fclone = SKB_FCLONE_UNAVAILABLE;
379 if (atomic_dec_and_test(fclone_ref))
380 kmem_cache_free(skbuff_fclone_cache, other);
385 static void skb_release_head_state(struct sk_buff *skb)
389 secpath_put(skb->sp);
391 if (skb->destructor) {
393 skb->destructor(skb);
395 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
396 nf_conntrack_put(skb->nfct);
398 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
399 nf_conntrack_put_reasm(skb->nfct_reasm);
401 #ifdef CONFIG_BRIDGE_NETFILTER
402 nf_bridge_put(skb->nf_bridge);
404 /* XXX: IS this still necessary? - JHS */
405 #ifdef CONFIG_NET_SCHED
407 #ifdef CONFIG_NET_CLS_ACT
413 /* Free everything but the sk_buff shell. */
414 static void skb_release_all(struct sk_buff *skb)
416 skb_release_head_state(skb);
417 skb_release_data(skb);
421 * __kfree_skb - private function
424 * Free an sk_buff. Release anything attached to the buffer.
425 * Clean the state. This is an internal helper function. Users should
426 * always call kfree_skb
429 void __kfree_skb(struct sk_buff *skb)
431 skb_release_all(skb);
434 EXPORT_SYMBOL(__kfree_skb);
437 * kfree_skb - free an sk_buff
438 * @skb: buffer to free
440 * Drop a reference to the buffer and free it if the usage count has
443 void kfree_skb(struct sk_buff *skb)
447 if (likely(atomic_read(&skb->users) == 1))
449 else if (likely(!atomic_dec_and_test(&skb->users)))
451 trace_kfree_skb(skb, __builtin_return_address(0));
454 EXPORT_SYMBOL(kfree_skb);
457 * consume_skb - free an skbuff
458 * @skb: buffer to free
460 * Drop a ref to the buffer and free it if the usage count has hit zero
461 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
462 * is being dropped after a failure and notes that
464 void consume_skb(struct sk_buff *skb)
468 if (likely(atomic_read(&skb->users) == 1))
470 else if (likely(!atomic_dec_and_test(&skb->users)))
472 trace_consume_skb(skb);
475 EXPORT_SYMBOL(consume_skb);
478 * skb_recycle_check - check if skb can be reused for receive
480 * @skb_size: minimum receive buffer size
482 * Checks that the skb passed in is not shared or cloned, and
483 * that it is linear and its head portion at least as large as
484 * skb_size so that it can be recycled as a receive buffer.
485 * If these conditions are met, this function does any necessary
486 * reference count dropping and cleans up the skbuff as if it
487 * just came from __alloc_skb().
489 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
491 struct skb_shared_info *shinfo;
496 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
499 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
502 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
503 if (skb_end_pointer(skb) - skb->head < skb_size)
506 if (skb_shared(skb) || skb_cloned(skb))
509 skb_release_head_state(skb);
511 shinfo = skb_shinfo(skb);
512 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
513 atomic_set(&shinfo->dataref, 1);
515 memset(skb, 0, offsetof(struct sk_buff, tail));
516 skb->data = skb->head + NET_SKB_PAD;
517 skb_reset_tail_pointer(skb);
521 EXPORT_SYMBOL(skb_recycle_check);
523 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
525 new->tstamp = old->tstamp;
527 new->transport_header = old->transport_header;
528 new->network_header = old->network_header;
529 new->mac_header = old->mac_header;
530 skb_dst_copy(new, old);
531 new->rxhash = old->rxhash;
532 new->ooo_okay = old->ooo_okay;
533 new->l4_rxhash = old->l4_rxhash;
535 new->sp = secpath_get(old->sp);
537 memcpy(new->cb, old->cb, sizeof(old->cb));
538 new->csum = old->csum;
539 new->local_df = old->local_df;
540 new->pkt_type = old->pkt_type;
541 new->ip_summed = old->ip_summed;
542 skb_copy_queue_mapping(new, old);
543 new->priority = old->priority;
544 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
545 new->ipvs_property = old->ipvs_property;
547 new->protocol = old->protocol;
548 new->mark = old->mark;
549 new->skb_iif = old->skb_iif;
551 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
552 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
553 new->nf_trace = old->nf_trace;
555 #ifdef CONFIG_NET_SCHED
556 new->tc_index = old->tc_index;
557 #ifdef CONFIG_NET_CLS_ACT
558 new->tc_verd = old->tc_verd;
561 new->vlan_tci = old->vlan_tci;
563 skb_copy_secmark(new, old);
567 * You should not add any new code to this function. Add it to
568 * __copy_skb_header above instead.
570 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
572 #define C(x) n->x = skb->x
574 n->next = n->prev = NULL;
576 __copy_skb_header(n, skb);
581 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
584 n->destructor = NULL;
590 atomic_set(&n->users, 1);
592 atomic_inc(&(skb_shinfo(skb)->dataref));
600 * skb_morph - morph one skb into another
601 * @dst: the skb to receive the contents
602 * @src: the skb to supply the contents
604 * This is identical to skb_clone except that the target skb is
605 * supplied by the user.
607 * The target skb is returned upon exit.
609 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
611 skb_release_all(dst);
612 return __skb_clone(dst, src);
614 EXPORT_SYMBOL_GPL(skb_morph);
616 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
617 * @skb: the skb to modify
618 * @gfp_mask: allocation priority
620 * This must be called on SKBTX_DEV_ZEROCOPY skb.
621 * It will copy all frags into kernel and drop the reference
622 * to userspace pages.
624 * If this function is called from an interrupt gfp_mask() must be
627 * Returns 0 on success or a negative error code on failure
628 * to allocate kernel memory to copy to.
630 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
633 int num_frags = skb_shinfo(skb)->nr_frags;
634 struct page *page, *head = NULL;
635 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
637 for (i = 0; i < num_frags; i++) {
639 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
641 page = alloc_page(GFP_ATOMIC);
644 struct page *next = (struct page *)head->private;
650 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
651 memcpy(page_address(page),
652 vaddr + f->page_offset, f->size);
653 kunmap_skb_frag(vaddr);
654 page->private = (unsigned long)head;
658 /* skb frags release userspace buffers */
659 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
660 put_page(skb_shinfo(skb)->frags[i].page);
662 uarg->callback(uarg);
664 /* skb frags point to kernel buffers */
665 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
666 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
667 skb_shinfo(skb)->frags[i - 1].page = head;
668 head = (struct page *)head->private;
671 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
677 * skb_clone - duplicate an sk_buff
678 * @skb: buffer to clone
679 * @gfp_mask: allocation priority
681 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
682 * copies share the same packet data but not structure. The new
683 * buffer has a reference count of 1. If the allocation fails the
684 * function returns %NULL otherwise the new buffer is returned.
686 * If this function is called from an interrupt gfp_mask() must be
690 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
694 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
695 if (skb_copy_ubufs(skb, gfp_mask))
700 if (skb->fclone == SKB_FCLONE_ORIG &&
701 n->fclone == SKB_FCLONE_UNAVAILABLE) {
702 atomic_t *fclone_ref = (atomic_t *) (n + 1);
703 n->fclone = SKB_FCLONE_CLONE;
704 atomic_inc(fclone_ref);
706 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
710 kmemcheck_annotate_bitfield(n, flags1);
711 kmemcheck_annotate_bitfield(n, flags2);
712 n->fclone = SKB_FCLONE_UNAVAILABLE;
715 return __skb_clone(n, skb);
717 EXPORT_SYMBOL(skb_clone);
719 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
721 #ifndef NET_SKBUFF_DATA_USES_OFFSET
723 * Shift between the two data areas in bytes
725 unsigned long offset = new->data - old->data;
728 __copy_skb_header(new, old);
730 #ifndef NET_SKBUFF_DATA_USES_OFFSET
731 /* {transport,network,mac}_header are relative to skb->head */
732 new->transport_header += offset;
733 new->network_header += offset;
734 if (skb_mac_header_was_set(new))
735 new->mac_header += offset;
737 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
738 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
739 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
743 * skb_copy - create private copy of an sk_buff
744 * @skb: buffer to copy
745 * @gfp_mask: allocation priority
747 * Make a copy of both an &sk_buff and its data. This is used when the
748 * caller wishes to modify the data and needs a private copy of the
749 * data to alter. Returns %NULL on failure or the pointer to the buffer
750 * on success. The returned buffer has a reference count of 1.
752 * As by-product this function converts non-linear &sk_buff to linear
753 * one, so that &sk_buff becomes completely private and caller is allowed
754 * to modify all the data of returned buffer. This means that this
755 * function is not recommended for use in circumstances when only
756 * header is going to be modified. Use pskb_copy() instead.
759 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
761 int headerlen = skb_headroom(skb);
762 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
763 struct sk_buff *n = alloc_skb(size, gfp_mask);
768 /* Set the data pointer */
769 skb_reserve(n, headerlen);
770 /* Set the tail pointer and length */
771 skb_put(n, skb->len);
773 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
776 copy_skb_header(n, skb);
779 EXPORT_SYMBOL(skb_copy);
782 * pskb_copy - create copy of an sk_buff with private head.
783 * @skb: buffer to copy
784 * @gfp_mask: allocation priority
786 * Make a copy of both an &sk_buff and part of its data, located
787 * in header. Fragmented data remain shared. This is used when
788 * the caller wishes to modify only header of &sk_buff and needs
789 * private copy of the header to alter. Returns %NULL on failure
790 * or the pointer to the buffer on success.
791 * The returned buffer has a reference count of 1.
794 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
796 unsigned int size = skb_end_pointer(skb) - skb->head;
797 struct sk_buff *n = alloc_skb(size, gfp_mask);
802 /* Set the data pointer */
803 skb_reserve(n, skb_headroom(skb));
804 /* Set the tail pointer and length */
805 skb_put(n, skb_headlen(skb));
807 skb_copy_from_linear_data(skb, n->data, n->len);
809 n->truesize += skb->data_len;
810 n->data_len = skb->data_len;
813 if (skb_shinfo(skb)->nr_frags) {
816 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
817 if (skb_copy_ubufs(skb, gfp_mask)) {
823 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
824 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
825 skb_frag_ref(skb, i);
827 skb_shinfo(n)->nr_frags = i;
830 if (skb_has_frag_list(skb)) {
831 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
832 skb_clone_fraglist(n);
835 copy_skb_header(n, skb);
839 EXPORT_SYMBOL(pskb_copy);
842 * pskb_expand_head - reallocate header of &sk_buff
843 * @skb: buffer to reallocate
844 * @nhead: room to add at head
845 * @ntail: room to add at tail
846 * @gfp_mask: allocation priority
848 * Expands (or creates identical copy, if &nhead and &ntail are zero)
849 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
850 * reference count of 1. Returns zero in the case of success or error,
851 * if expansion failed. In the last case, &sk_buff is not changed.
853 * All the pointers pointing into skb header may change and must be
854 * reloaded after call to this function.
857 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
862 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
871 size = SKB_DATA_ALIGN(size);
873 /* Check if we can avoid taking references on fragments if we own
874 * the last reference on skb->head. (see skb_release_data())
879 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
880 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
884 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
885 memmove(skb->head + size, skb_shinfo(skb),
886 offsetof(struct skb_shared_info,
887 frags[skb_shinfo(skb)->nr_frags]));
888 memmove(skb->head + nhead, skb->head,
889 skb_tail_pointer(skb) - skb->head);
894 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
898 /* Copy only real data... and, alas, header. This should be
899 * optimized for the cases when header is void.
901 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
903 memcpy((struct skb_shared_info *)(data + size),
905 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
910 /* copy this zero copy skb frags */
911 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
912 if (skb_copy_ubufs(skb, gfp_mask))
915 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
916 skb_frag_ref(skb, i);
918 if (skb_has_frag_list(skb))
919 skb_clone_fraglist(skb);
921 skb_release_data(skb);
923 off = (data + nhead) - skb->head;
928 #ifdef NET_SKBUFF_DATA_USES_OFFSET
932 skb->end = skb->head + size;
934 /* {transport,network,mac}_header and tail are relative to skb->head */
936 skb->transport_header += off;
937 skb->network_header += off;
938 if (skb_mac_header_was_set(skb))
939 skb->mac_header += off;
940 /* Only adjust this if it actually is csum_start rather than csum */
941 if (skb->ip_summed == CHECKSUM_PARTIAL)
942 skb->csum_start += nhead;
946 atomic_set(&skb_shinfo(skb)->dataref, 1);
954 EXPORT_SYMBOL(pskb_expand_head);
956 /* Make private copy of skb with writable head and some headroom */
958 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
960 struct sk_buff *skb2;
961 int delta = headroom - skb_headroom(skb);
964 skb2 = pskb_copy(skb, GFP_ATOMIC);
966 skb2 = skb_clone(skb, GFP_ATOMIC);
967 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
975 EXPORT_SYMBOL(skb_realloc_headroom);
978 * skb_copy_expand - copy and expand sk_buff
979 * @skb: buffer to copy
980 * @newheadroom: new free bytes at head
981 * @newtailroom: new free bytes at tail
982 * @gfp_mask: allocation priority
984 * Make a copy of both an &sk_buff and its data and while doing so
985 * allocate additional space.
987 * This is used when the caller wishes to modify the data and needs a
988 * private copy of the data to alter as well as more space for new fields.
989 * Returns %NULL on failure or the pointer to the buffer
990 * on success. The returned buffer has a reference count of 1.
992 * You must pass %GFP_ATOMIC as the allocation priority if this function
993 * is called from an interrupt.
995 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
996 int newheadroom, int newtailroom,
1000 * Allocate the copy buffer
1002 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1004 int oldheadroom = skb_headroom(skb);
1005 int head_copy_len, head_copy_off;
1011 skb_reserve(n, newheadroom);
1013 /* Set the tail pointer and length */
1014 skb_put(n, skb->len);
1016 head_copy_len = oldheadroom;
1018 if (newheadroom <= head_copy_len)
1019 head_copy_len = newheadroom;
1021 head_copy_off = newheadroom - head_copy_len;
1023 /* Copy the linear header and data. */
1024 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1025 skb->len + head_copy_len))
1028 copy_skb_header(n, skb);
1030 off = newheadroom - oldheadroom;
1031 if (n->ip_summed == CHECKSUM_PARTIAL)
1032 n->csum_start += off;
1033 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1034 n->transport_header += off;
1035 n->network_header += off;
1036 if (skb_mac_header_was_set(skb))
1037 n->mac_header += off;
1042 EXPORT_SYMBOL(skb_copy_expand);
1045 * skb_pad - zero pad the tail of an skb
1046 * @skb: buffer to pad
1047 * @pad: space to pad
1049 * Ensure that a buffer is followed by a padding area that is zero
1050 * filled. Used by network drivers which may DMA or transfer data
1051 * beyond the buffer end onto the wire.
1053 * May return error in out of memory cases. The skb is freed on error.
1056 int skb_pad(struct sk_buff *skb, int pad)
1061 /* If the skbuff is non linear tailroom is always zero.. */
1062 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1063 memset(skb->data+skb->len, 0, pad);
1067 ntail = skb->data_len + pad - (skb->end - skb->tail);
1068 if (likely(skb_cloned(skb) || ntail > 0)) {
1069 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1074 /* FIXME: The use of this function with non-linear skb's really needs
1077 err = skb_linearize(skb);
1081 memset(skb->data + skb->len, 0, pad);
1088 EXPORT_SYMBOL(skb_pad);
1091 * skb_put - add data to a buffer
1092 * @skb: buffer to use
1093 * @len: amount of data to add
1095 * This function extends the used data area of the buffer. If this would
1096 * exceed the total buffer size the kernel will panic. A pointer to the
1097 * first byte of the extra data is returned.
1099 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1101 unsigned char *tmp = skb_tail_pointer(skb);
1102 SKB_LINEAR_ASSERT(skb);
1105 if (unlikely(skb->tail > skb->end))
1106 skb_over_panic(skb, len, __builtin_return_address(0));
1109 EXPORT_SYMBOL(skb_put);
1112 * skb_push - add data to the start of a buffer
1113 * @skb: buffer to use
1114 * @len: amount of data to add
1116 * This function extends the used data area of the buffer at the buffer
1117 * start. If this would exceed the total buffer headroom the kernel will
1118 * panic. A pointer to the first byte of the extra data is returned.
1120 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1124 if (unlikely(skb->data<skb->head))
1125 skb_under_panic(skb, len, __builtin_return_address(0));
1128 EXPORT_SYMBOL(skb_push);
1131 * skb_pull - remove data from the start of a buffer
1132 * @skb: buffer to use
1133 * @len: amount of data to remove
1135 * This function removes data from the start of a buffer, returning
1136 * the memory to the headroom. A pointer to the next data in the buffer
1137 * is returned. Once the data has been pulled future pushes will overwrite
1140 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1142 return skb_pull_inline(skb, len);
1144 EXPORT_SYMBOL(skb_pull);
1147 * skb_trim - remove end from a buffer
1148 * @skb: buffer to alter
1151 * Cut the length of a buffer down by removing data from the tail. If
1152 * the buffer is already under the length specified it is not modified.
1153 * The skb must be linear.
1155 void skb_trim(struct sk_buff *skb, unsigned int len)
1158 __skb_trim(skb, len);
1160 EXPORT_SYMBOL(skb_trim);
1162 /* Trims skb to length len. It can change skb pointers.
1165 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1167 struct sk_buff **fragp;
1168 struct sk_buff *frag;
1169 int offset = skb_headlen(skb);
1170 int nfrags = skb_shinfo(skb)->nr_frags;
1174 if (skb_cloned(skb) &&
1175 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1182 for (; i < nfrags; i++) {
1183 int end = offset + skb_shinfo(skb)->frags[i].size;
1190 skb_shinfo(skb)->frags[i++].size = len - offset;
1193 skb_shinfo(skb)->nr_frags = i;
1195 for (; i < nfrags; i++)
1196 skb_frag_unref(skb, i);
1198 if (skb_has_frag_list(skb))
1199 skb_drop_fraglist(skb);
1203 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1204 fragp = &frag->next) {
1205 int end = offset + frag->len;
1207 if (skb_shared(frag)) {
1208 struct sk_buff *nfrag;
1210 nfrag = skb_clone(frag, GFP_ATOMIC);
1211 if (unlikely(!nfrag))
1214 nfrag->next = frag->next;
1226 unlikely((err = pskb_trim(frag, len - offset))))
1230 skb_drop_list(&frag->next);
1235 if (len > skb_headlen(skb)) {
1236 skb->data_len -= skb->len - len;
1241 skb_set_tail_pointer(skb, len);
1246 EXPORT_SYMBOL(___pskb_trim);
1249 * __pskb_pull_tail - advance tail of skb header
1250 * @skb: buffer to reallocate
1251 * @delta: number of bytes to advance tail
1253 * The function makes a sense only on a fragmented &sk_buff,
1254 * it expands header moving its tail forward and copying necessary
1255 * data from fragmented part.
1257 * &sk_buff MUST have reference count of 1.
1259 * Returns %NULL (and &sk_buff does not change) if pull failed
1260 * or value of new tail of skb in the case of success.
1262 * All the pointers pointing into skb header may change and must be
1263 * reloaded after call to this function.
1266 /* Moves tail of skb head forward, copying data from fragmented part,
1267 * when it is necessary.
1268 * 1. It may fail due to malloc failure.
1269 * 2. It may change skb pointers.
1271 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1273 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1275 /* If skb has not enough free space at tail, get new one
1276 * plus 128 bytes for future expansions. If we have enough
1277 * room at tail, reallocate without expansion only if skb is cloned.
1279 int i, k, eat = (skb->tail + delta) - skb->end;
1281 if (eat > 0 || skb_cloned(skb)) {
1282 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1287 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1290 /* Optimization: no fragments, no reasons to preestimate
1291 * size of pulled pages. Superb.
1293 if (!skb_has_frag_list(skb))
1296 /* Estimate size of pulled pages. */
1298 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1299 if (skb_shinfo(skb)->frags[i].size >= eat)
1301 eat -= skb_shinfo(skb)->frags[i].size;
1304 /* If we need update frag list, we are in troubles.
1305 * Certainly, it possible to add an offset to skb data,
1306 * but taking into account that pulling is expected to
1307 * be very rare operation, it is worth to fight against
1308 * further bloating skb head and crucify ourselves here instead.
1309 * Pure masohism, indeed. 8)8)
1312 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1313 struct sk_buff *clone = NULL;
1314 struct sk_buff *insp = NULL;
1319 if (list->len <= eat) {
1320 /* Eaten as whole. */
1325 /* Eaten partially. */
1327 if (skb_shared(list)) {
1328 /* Sucks! We need to fork list. :-( */
1329 clone = skb_clone(list, GFP_ATOMIC);
1335 /* This may be pulled without
1339 if (!pskb_pull(list, eat)) {
1347 /* Free pulled out fragments. */
1348 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1349 skb_shinfo(skb)->frag_list = list->next;
1352 /* And insert new clone at head. */
1355 skb_shinfo(skb)->frag_list = clone;
1358 /* Success! Now we may commit changes to skb data. */
1363 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1364 if (skb_shinfo(skb)->frags[i].size <= eat) {
1365 skb_frag_unref(skb, i);
1366 eat -= skb_shinfo(skb)->frags[i].size;
1368 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1370 skb_shinfo(skb)->frags[k].page_offset += eat;
1371 skb_shinfo(skb)->frags[k].size -= eat;
1377 skb_shinfo(skb)->nr_frags = k;
1380 skb->data_len -= delta;
1382 return skb_tail_pointer(skb);
1384 EXPORT_SYMBOL(__pskb_pull_tail);
1387 * skb_copy_bits - copy bits from skb to kernel buffer
1389 * @offset: offset in source
1390 * @to: destination buffer
1391 * @len: number of bytes to copy
1393 * Copy the specified number of bytes from the source skb to the
1394 * destination buffer.
1397 * If its prototype is ever changed,
1398 * check arch/{*}/net/{*}.S files,
1399 * since it is called from BPF assembly code.
1401 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1403 int start = skb_headlen(skb);
1404 struct sk_buff *frag_iter;
1407 if (offset > (int)skb->len - len)
1411 if ((copy = start - offset) > 0) {
1414 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1415 if ((len -= copy) == 0)
1421 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1424 WARN_ON(start > offset + len);
1426 end = start + skb_shinfo(skb)->frags[i].size;
1427 if ((copy = end - offset) > 0) {
1433 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1435 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1436 offset - start, copy);
1437 kunmap_skb_frag(vaddr);
1439 if ((len -= copy) == 0)
1447 skb_walk_frags(skb, frag_iter) {
1450 WARN_ON(start > offset + len);
1452 end = start + frag_iter->len;
1453 if ((copy = end - offset) > 0) {
1456 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1458 if ((len -= copy) == 0)
1472 EXPORT_SYMBOL(skb_copy_bits);
1475 * Callback from splice_to_pipe(), if we need to release some pages
1476 * at the end of the spd in case we error'ed out in filling the pipe.
1478 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1480 put_page(spd->pages[i]);
1483 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1484 unsigned int *offset,
1485 struct sk_buff *skb, struct sock *sk)
1487 struct page *p = sk->sk_sndmsg_page;
1492 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1496 off = sk->sk_sndmsg_off = 0;
1497 /* hold one ref to this page until it's full */
1501 off = sk->sk_sndmsg_off;
1502 mlen = PAGE_SIZE - off;
1503 if (mlen < 64 && mlen < *len) {
1508 *len = min_t(unsigned int, *len, mlen);
1511 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1512 sk->sk_sndmsg_off += *len;
1520 * Fill page/offset/length into spd, if it can hold more pages.
1522 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1523 struct pipe_inode_info *pipe, struct page *page,
1524 unsigned int *len, unsigned int offset,
1525 struct sk_buff *skb, int linear,
1528 if (unlikely(spd->nr_pages == pipe->buffers))
1532 page = linear_to_page(page, len, &offset, skb, sk);
1538 spd->pages[spd->nr_pages] = page;
1539 spd->partial[spd->nr_pages].len = *len;
1540 spd->partial[spd->nr_pages].offset = offset;
1546 static inline void __segment_seek(struct page **page, unsigned int *poff,
1547 unsigned int *plen, unsigned int off)
1552 n = *poff / PAGE_SIZE;
1554 *page = nth_page(*page, n);
1556 *poff = *poff % PAGE_SIZE;
1560 static inline int __splice_segment(struct page *page, unsigned int poff,
1561 unsigned int plen, unsigned int *off,
1562 unsigned int *len, struct sk_buff *skb,
1563 struct splice_pipe_desc *spd, int linear,
1565 struct pipe_inode_info *pipe)
1570 /* skip this segment if already processed */
1576 /* ignore any bits we already processed */
1578 __segment_seek(&page, &poff, &plen, *off);
1583 unsigned int flen = min(*len, plen);
1585 /* the linear region may spread across several pages */
1586 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1588 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1591 __segment_seek(&page, &poff, &plen, flen);
1594 } while (*len && plen);
1600 * Map linear and fragment data from the skb to spd. It reports failure if the
1601 * pipe is full or if we already spliced the requested length.
1603 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1604 unsigned int *offset, unsigned int *len,
1605 struct splice_pipe_desc *spd, struct sock *sk)
1610 * map the linear part
1612 if (__splice_segment(virt_to_page(skb->data),
1613 (unsigned long) skb->data & (PAGE_SIZE - 1),
1615 offset, len, skb, spd, 1, sk, pipe))
1619 * then map the fragments
1621 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1622 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1624 if (__splice_segment(skb_frag_page(f),
1625 f->page_offset, f->size,
1626 offset, len, skb, spd, 0, sk, pipe))
1634 * Map data from the skb to a pipe. Should handle both the linear part,
1635 * the fragments, and the frag list. It does NOT handle frag lists within
1636 * the frag list, if such a thing exists. We'd probably need to recurse to
1637 * handle that cleanly.
1639 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1640 struct pipe_inode_info *pipe, unsigned int tlen,
1643 struct partial_page partial[PIPE_DEF_BUFFERS];
1644 struct page *pages[PIPE_DEF_BUFFERS];
1645 struct splice_pipe_desc spd = {
1649 .ops = &sock_pipe_buf_ops,
1650 .spd_release = sock_spd_release,
1652 struct sk_buff *frag_iter;
1653 struct sock *sk = skb->sk;
1656 if (splice_grow_spd(pipe, &spd))
1660 * __skb_splice_bits() only fails if the output has no room left,
1661 * so no point in going over the frag_list for the error case.
1663 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1669 * now see if we have a frag_list to map
1671 skb_walk_frags(skb, frag_iter) {
1674 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1681 * Drop the socket lock, otherwise we have reverse
1682 * locking dependencies between sk_lock and i_mutex
1683 * here as compared to sendfile(). We enter here
1684 * with the socket lock held, and splice_to_pipe() will
1685 * grab the pipe inode lock. For sendfile() emulation,
1686 * we call into ->sendpage() with the i_mutex lock held
1687 * and networking will grab the socket lock.
1690 ret = splice_to_pipe(pipe, &spd);
1694 splice_shrink_spd(pipe, &spd);
1699 * skb_store_bits - store bits from kernel buffer to skb
1700 * @skb: destination buffer
1701 * @offset: offset in destination
1702 * @from: source buffer
1703 * @len: number of bytes to copy
1705 * Copy the specified number of bytes from the source buffer to the
1706 * destination skb. This function handles all the messy bits of
1707 * traversing fragment lists and such.
1710 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1712 int start = skb_headlen(skb);
1713 struct sk_buff *frag_iter;
1716 if (offset > (int)skb->len - len)
1719 if ((copy = start - offset) > 0) {
1722 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1723 if ((len -= copy) == 0)
1729 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1730 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1733 WARN_ON(start > offset + len);
1735 end = start + frag->size;
1736 if ((copy = end - offset) > 0) {
1742 vaddr = kmap_skb_frag(frag);
1743 memcpy(vaddr + frag->page_offset + offset - start,
1745 kunmap_skb_frag(vaddr);
1747 if ((len -= copy) == 0)
1755 skb_walk_frags(skb, frag_iter) {
1758 WARN_ON(start > offset + len);
1760 end = start + frag_iter->len;
1761 if ((copy = end - offset) > 0) {
1764 if (skb_store_bits(frag_iter, offset - start,
1767 if ((len -= copy) == 0)
1780 EXPORT_SYMBOL(skb_store_bits);
1782 /* Checksum skb data. */
1784 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1785 int len, __wsum csum)
1787 int start = skb_headlen(skb);
1788 int i, copy = start - offset;
1789 struct sk_buff *frag_iter;
1792 /* Checksum header. */
1796 csum = csum_partial(skb->data + offset, copy, csum);
1797 if ((len -= copy) == 0)
1803 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1806 WARN_ON(start > offset + len);
1808 end = start + skb_shinfo(skb)->frags[i].size;
1809 if ((copy = end - offset) > 0) {
1812 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1816 vaddr = kmap_skb_frag(frag);
1817 csum2 = csum_partial(vaddr + frag->page_offset +
1818 offset - start, copy, 0);
1819 kunmap_skb_frag(vaddr);
1820 csum = csum_block_add(csum, csum2, pos);
1829 skb_walk_frags(skb, frag_iter) {
1832 WARN_ON(start > offset + len);
1834 end = start + frag_iter->len;
1835 if ((copy = end - offset) > 0) {
1839 csum2 = skb_checksum(frag_iter, offset - start,
1841 csum = csum_block_add(csum, csum2, pos);
1842 if ((len -= copy) == 0)
1853 EXPORT_SYMBOL(skb_checksum);
1855 /* Both of above in one bottle. */
1857 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1858 u8 *to, int len, __wsum csum)
1860 int start = skb_headlen(skb);
1861 int i, copy = start - offset;
1862 struct sk_buff *frag_iter;
1869 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1871 if ((len -= copy) == 0)
1878 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1881 WARN_ON(start > offset + len);
1883 end = start + skb_shinfo(skb)->frags[i].size;
1884 if ((copy = end - offset) > 0) {
1887 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1891 vaddr = kmap_skb_frag(frag);
1892 csum2 = csum_partial_copy_nocheck(vaddr +
1896 kunmap_skb_frag(vaddr);
1897 csum = csum_block_add(csum, csum2, pos);
1907 skb_walk_frags(skb, frag_iter) {
1911 WARN_ON(start > offset + len);
1913 end = start + frag_iter->len;
1914 if ((copy = end - offset) > 0) {
1917 csum2 = skb_copy_and_csum_bits(frag_iter,
1920 csum = csum_block_add(csum, csum2, pos);
1921 if ((len -= copy) == 0)
1932 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1934 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1939 if (skb->ip_summed == CHECKSUM_PARTIAL)
1940 csstart = skb_checksum_start_offset(skb);
1942 csstart = skb_headlen(skb);
1944 BUG_ON(csstart > skb_headlen(skb));
1946 skb_copy_from_linear_data(skb, to, csstart);
1949 if (csstart != skb->len)
1950 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1951 skb->len - csstart, 0);
1953 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1954 long csstuff = csstart + skb->csum_offset;
1956 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1959 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1962 * skb_dequeue - remove from the head of the queue
1963 * @list: list to dequeue from
1965 * Remove the head of the list. The list lock is taken so the function
1966 * may be used safely with other locking list functions. The head item is
1967 * returned or %NULL if the list is empty.
1970 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1972 unsigned long flags;
1973 struct sk_buff *result;
1975 spin_lock_irqsave(&list->lock, flags);
1976 result = __skb_dequeue(list);
1977 spin_unlock_irqrestore(&list->lock, flags);
1980 EXPORT_SYMBOL(skb_dequeue);
1983 * skb_dequeue_tail - remove from the tail of the queue
1984 * @list: list to dequeue from
1986 * Remove the tail of the list. The list lock is taken so the function
1987 * may be used safely with other locking list functions. The tail item is
1988 * returned or %NULL if the list is empty.
1990 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1992 unsigned long flags;
1993 struct sk_buff *result;
1995 spin_lock_irqsave(&list->lock, flags);
1996 result = __skb_dequeue_tail(list);
1997 spin_unlock_irqrestore(&list->lock, flags);
2000 EXPORT_SYMBOL(skb_dequeue_tail);
2003 * skb_queue_purge - empty a list
2004 * @list: list to empty
2006 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2007 * the list and one reference dropped. This function takes the list
2008 * lock and is atomic with respect to other list locking functions.
2010 void skb_queue_purge(struct sk_buff_head *list)
2012 struct sk_buff *skb;
2013 while ((skb = skb_dequeue(list)) != NULL)
2016 EXPORT_SYMBOL(skb_queue_purge);
2019 * skb_queue_head - queue a buffer at the list head
2020 * @list: list to use
2021 * @newsk: buffer to queue
2023 * Queue a buffer at the start of the list. This function takes the
2024 * list lock and can be used safely with other locking &sk_buff functions
2027 * A buffer cannot be placed on two lists at the same time.
2029 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2031 unsigned long flags;
2033 spin_lock_irqsave(&list->lock, flags);
2034 __skb_queue_head(list, newsk);
2035 spin_unlock_irqrestore(&list->lock, flags);
2037 EXPORT_SYMBOL(skb_queue_head);
2040 * skb_queue_tail - queue a buffer at the list tail
2041 * @list: list to use
2042 * @newsk: buffer to queue
2044 * Queue a buffer at the tail of the list. This function takes the
2045 * list lock and can be used safely with other locking &sk_buff functions
2048 * A buffer cannot be placed on two lists at the same time.
2050 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2052 unsigned long flags;
2054 spin_lock_irqsave(&list->lock, flags);
2055 __skb_queue_tail(list, newsk);
2056 spin_unlock_irqrestore(&list->lock, flags);
2058 EXPORT_SYMBOL(skb_queue_tail);
2061 * skb_unlink - remove a buffer from a list
2062 * @skb: buffer to remove
2063 * @list: list to use
2065 * Remove a packet from a list. The list locks are taken and this
2066 * function is atomic with respect to other list locked calls
2068 * You must know what list the SKB is on.
2070 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2072 unsigned long flags;
2074 spin_lock_irqsave(&list->lock, flags);
2075 __skb_unlink(skb, list);
2076 spin_unlock_irqrestore(&list->lock, flags);
2078 EXPORT_SYMBOL(skb_unlink);
2081 * skb_append - append a buffer
2082 * @old: buffer to insert after
2083 * @newsk: buffer to insert
2084 * @list: list to use
2086 * Place a packet after a given packet in a list. The list locks are taken
2087 * and this function is atomic with respect to other list locked calls.
2088 * A buffer cannot be placed on two lists at the same time.
2090 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2092 unsigned long flags;
2094 spin_lock_irqsave(&list->lock, flags);
2095 __skb_queue_after(list, old, newsk);
2096 spin_unlock_irqrestore(&list->lock, flags);
2098 EXPORT_SYMBOL(skb_append);
2101 * skb_insert - insert a buffer
2102 * @old: buffer to insert before
2103 * @newsk: buffer to insert
2104 * @list: list to use
2106 * Place a packet before a given packet in a list. The list locks are
2107 * taken and this function is atomic with respect to other list locked
2110 * A buffer cannot be placed on two lists at the same time.
2112 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2114 unsigned long flags;
2116 spin_lock_irqsave(&list->lock, flags);
2117 __skb_insert(newsk, old->prev, old, list);
2118 spin_unlock_irqrestore(&list->lock, flags);
2120 EXPORT_SYMBOL(skb_insert);
2122 static inline void skb_split_inside_header(struct sk_buff *skb,
2123 struct sk_buff* skb1,
2124 const u32 len, const int pos)
2128 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2130 /* And move data appendix as is. */
2131 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2132 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2134 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2135 skb_shinfo(skb)->nr_frags = 0;
2136 skb1->data_len = skb->data_len;
2137 skb1->len += skb1->data_len;
2140 skb_set_tail_pointer(skb, len);
2143 static inline void skb_split_no_header(struct sk_buff *skb,
2144 struct sk_buff* skb1,
2145 const u32 len, int pos)
2148 const int nfrags = skb_shinfo(skb)->nr_frags;
2150 skb_shinfo(skb)->nr_frags = 0;
2151 skb1->len = skb1->data_len = skb->len - len;
2153 skb->data_len = len - pos;
2155 for (i = 0; i < nfrags; i++) {
2156 int size = skb_shinfo(skb)->frags[i].size;
2158 if (pos + size > len) {
2159 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2163 * We have two variants in this case:
2164 * 1. Move all the frag to the second
2165 * part, if it is possible. F.e.
2166 * this approach is mandatory for TUX,
2167 * where splitting is expensive.
2168 * 2. Split is accurately. We make this.
2170 skb_frag_ref(skb, i);
2171 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2172 skb_shinfo(skb1)->frags[0].size -= len - pos;
2173 skb_shinfo(skb)->frags[i].size = len - pos;
2174 skb_shinfo(skb)->nr_frags++;
2178 skb_shinfo(skb)->nr_frags++;
2181 skb_shinfo(skb1)->nr_frags = k;
2185 * skb_split - Split fragmented skb to two parts at length len.
2186 * @skb: the buffer to split
2187 * @skb1: the buffer to receive the second part
2188 * @len: new length for skb
2190 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2192 int pos = skb_headlen(skb);
2194 if (len < pos) /* Split line is inside header. */
2195 skb_split_inside_header(skb, skb1, len, pos);
2196 else /* Second chunk has no header, nothing to copy. */
2197 skb_split_no_header(skb, skb1, len, pos);
2199 EXPORT_SYMBOL(skb_split);
2201 /* Shifting from/to a cloned skb is a no-go.
2203 * Caller cannot keep skb_shinfo related pointers past calling here!
2205 static int skb_prepare_for_shift(struct sk_buff *skb)
2207 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2211 * skb_shift - Shifts paged data partially from skb to another
2212 * @tgt: buffer into which tail data gets added
2213 * @skb: buffer from which the paged data comes from
2214 * @shiftlen: shift up to this many bytes
2216 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2217 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2218 * It's up to caller to free skb if everything was shifted.
2220 * If @tgt runs out of frags, the whole operation is aborted.
2222 * Skb cannot include anything else but paged data while tgt is allowed
2223 * to have non-paged data as well.
2225 * TODO: full sized shift could be optimized but that would need
2226 * specialized skb free'er to handle frags without up-to-date nr_frags.
2228 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2230 int from, to, merge, todo;
2231 struct skb_frag_struct *fragfrom, *fragto;
2233 BUG_ON(shiftlen > skb->len);
2234 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2238 to = skb_shinfo(tgt)->nr_frags;
2239 fragfrom = &skb_shinfo(skb)->frags[from];
2241 /* Actual merge is delayed until the point when we know we can
2242 * commit all, so that we don't have to undo partial changes
2245 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2246 fragfrom->page_offset)) {
2251 todo -= fragfrom->size;
2253 if (skb_prepare_for_shift(skb) ||
2254 skb_prepare_for_shift(tgt))
2257 /* All previous frag pointers might be stale! */
2258 fragfrom = &skb_shinfo(skb)->frags[from];
2259 fragto = &skb_shinfo(tgt)->frags[merge];
2261 fragto->size += shiftlen;
2262 fragfrom->size -= shiftlen;
2263 fragfrom->page_offset += shiftlen;
2271 /* Skip full, not-fitting skb to avoid expensive operations */
2272 if ((shiftlen == skb->len) &&
2273 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2276 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2279 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2280 if (to == MAX_SKB_FRAGS)
2283 fragfrom = &skb_shinfo(skb)->frags[from];
2284 fragto = &skb_shinfo(tgt)->frags[to];
2286 if (todo >= fragfrom->size) {
2287 *fragto = *fragfrom;
2288 todo -= fragfrom->size;
2293 __skb_frag_ref(fragfrom);
2294 fragto->page = fragfrom->page;
2295 fragto->page_offset = fragfrom->page_offset;
2296 fragto->size = todo;
2298 fragfrom->page_offset += todo;
2299 fragfrom->size -= todo;
2307 /* Ready to "commit" this state change to tgt */
2308 skb_shinfo(tgt)->nr_frags = to;
2311 fragfrom = &skb_shinfo(skb)->frags[0];
2312 fragto = &skb_shinfo(tgt)->frags[merge];
2314 fragto->size += fragfrom->size;
2315 __skb_frag_unref(fragfrom);
2318 /* Reposition in the original skb */
2320 while (from < skb_shinfo(skb)->nr_frags)
2321 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2322 skb_shinfo(skb)->nr_frags = to;
2324 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2327 /* Most likely the tgt won't ever need its checksum anymore, skb on
2328 * the other hand might need it if it needs to be resent
2330 tgt->ip_summed = CHECKSUM_PARTIAL;
2331 skb->ip_summed = CHECKSUM_PARTIAL;
2333 /* Yak, is it really working this way? Some helper please? */
2334 skb->len -= shiftlen;
2335 skb->data_len -= shiftlen;
2336 skb->truesize -= shiftlen;
2337 tgt->len += shiftlen;
2338 tgt->data_len += shiftlen;
2339 tgt->truesize += shiftlen;
2345 * skb_prepare_seq_read - Prepare a sequential read of skb data
2346 * @skb: the buffer to read
2347 * @from: lower offset of data to be read
2348 * @to: upper offset of data to be read
2349 * @st: state variable
2351 * Initializes the specified state variable. Must be called before
2352 * invoking skb_seq_read() for the first time.
2354 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2355 unsigned int to, struct skb_seq_state *st)
2357 st->lower_offset = from;
2358 st->upper_offset = to;
2359 st->root_skb = st->cur_skb = skb;
2360 st->frag_idx = st->stepped_offset = 0;
2361 st->frag_data = NULL;
2363 EXPORT_SYMBOL(skb_prepare_seq_read);
2366 * skb_seq_read - Sequentially read skb data
2367 * @consumed: number of bytes consumed by the caller so far
2368 * @data: destination pointer for data to be returned
2369 * @st: state variable
2371 * Reads a block of skb data at &consumed relative to the
2372 * lower offset specified to skb_prepare_seq_read(). Assigns
2373 * the head of the data block to &data and returns the length
2374 * of the block or 0 if the end of the skb data or the upper
2375 * offset has been reached.
2377 * The caller is not required to consume all of the data
2378 * returned, i.e. &consumed is typically set to the number
2379 * of bytes already consumed and the next call to
2380 * skb_seq_read() will return the remaining part of the block.
2382 * Note 1: The size of each block of data returned can be arbitrary,
2383 * this limitation is the cost for zerocopy seqeuental
2384 * reads of potentially non linear data.
2386 * Note 2: Fragment lists within fragments are not implemented
2387 * at the moment, state->root_skb could be replaced with
2388 * a stack for this purpose.
2390 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2391 struct skb_seq_state *st)
2393 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2396 if (unlikely(abs_offset >= st->upper_offset))
2400 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2402 if (abs_offset < block_limit && !st->frag_data) {
2403 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2404 return block_limit - abs_offset;
2407 if (st->frag_idx == 0 && !st->frag_data)
2408 st->stepped_offset += skb_headlen(st->cur_skb);
2410 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2411 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2412 block_limit = frag->size + st->stepped_offset;
2414 if (abs_offset < block_limit) {
2416 st->frag_data = kmap_skb_frag(frag);
2418 *data = (u8 *) st->frag_data + frag->page_offset +
2419 (abs_offset - st->stepped_offset);
2421 return block_limit - abs_offset;
2424 if (st->frag_data) {
2425 kunmap_skb_frag(st->frag_data);
2426 st->frag_data = NULL;
2430 st->stepped_offset += frag->size;
2433 if (st->frag_data) {
2434 kunmap_skb_frag(st->frag_data);
2435 st->frag_data = NULL;
2438 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2439 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2442 } else if (st->cur_skb->next) {
2443 st->cur_skb = st->cur_skb->next;
2450 EXPORT_SYMBOL(skb_seq_read);
2453 * skb_abort_seq_read - Abort a sequential read of skb data
2454 * @st: state variable
2456 * Must be called if skb_seq_read() was not called until it
2459 void skb_abort_seq_read(struct skb_seq_state *st)
2462 kunmap_skb_frag(st->frag_data);
2464 EXPORT_SYMBOL(skb_abort_seq_read);
2466 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2468 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2469 struct ts_config *conf,
2470 struct ts_state *state)
2472 return skb_seq_read(offset, text, TS_SKB_CB(state));
2475 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2477 skb_abort_seq_read(TS_SKB_CB(state));
2481 * skb_find_text - Find a text pattern in skb data
2482 * @skb: the buffer to look in
2483 * @from: search offset
2485 * @config: textsearch configuration
2486 * @state: uninitialized textsearch state variable
2488 * Finds a pattern in the skb data according to the specified
2489 * textsearch configuration. Use textsearch_next() to retrieve
2490 * subsequent occurrences of the pattern. Returns the offset
2491 * to the first occurrence or UINT_MAX if no match was found.
2493 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2494 unsigned int to, struct ts_config *config,
2495 struct ts_state *state)
2499 config->get_next_block = skb_ts_get_next_block;
2500 config->finish = skb_ts_finish;
2502 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2504 ret = textsearch_find(config, state);
2505 return (ret <= to - from ? ret : UINT_MAX);
2507 EXPORT_SYMBOL(skb_find_text);
2510 * skb_append_datato_frags: - append the user data to a skb
2511 * @sk: sock structure
2512 * @skb: skb structure to be appened with user data.
2513 * @getfrag: call back function to be used for getting the user data
2514 * @from: pointer to user message iov
2515 * @length: length of the iov message
2517 * Description: This procedure append the user data in the fragment part
2518 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2520 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2521 int (*getfrag)(void *from, char *to, int offset,
2522 int len, int odd, struct sk_buff *skb),
2523 void *from, int length)
2526 skb_frag_t *frag = NULL;
2527 struct page *page = NULL;
2533 /* Return error if we don't have space for new frag */
2534 frg_cnt = skb_shinfo(skb)->nr_frags;
2535 if (frg_cnt >= MAX_SKB_FRAGS)
2538 /* allocate a new page for next frag */
2539 page = alloc_pages(sk->sk_allocation, 0);
2541 /* If alloc_page fails just return failure and caller will
2542 * free previous allocated pages by doing kfree_skb()
2547 /* initialize the next frag */
2548 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2549 skb->truesize += PAGE_SIZE;
2550 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2552 /* get the new initialized frag */
2553 frg_cnt = skb_shinfo(skb)->nr_frags;
2554 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2556 /* copy the user data to page */
2557 left = PAGE_SIZE - frag->page_offset;
2558 copy = (length > left)? left : length;
2560 ret = getfrag(from, skb_frag_address(frag) + frag->size,
2561 offset, copy, 0, skb);
2565 /* copy was successful so update the size parameters */
2568 skb->data_len += copy;
2572 } while (length > 0);
2576 EXPORT_SYMBOL(skb_append_datato_frags);
2579 * skb_pull_rcsum - pull skb and update receive checksum
2580 * @skb: buffer to update
2581 * @len: length of data pulled
2583 * This function performs an skb_pull on the packet and updates
2584 * the CHECKSUM_COMPLETE checksum. It should be used on
2585 * receive path processing instead of skb_pull unless you know
2586 * that the checksum difference is zero (e.g., a valid IP header)
2587 * or you are setting ip_summed to CHECKSUM_NONE.
2589 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2591 BUG_ON(len > skb->len);
2593 BUG_ON(skb->len < skb->data_len);
2594 skb_postpull_rcsum(skb, skb->data, len);
2595 return skb->data += len;
2597 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2600 * skb_segment - Perform protocol segmentation on skb.
2601 * @skb: buffer to segment
2602 * @features: features for the output path (see dev->features)
2604 * This function performs segmentation on the given skb. It returns
2605 * a pointer to the first in a list of new skbs for the segments.
2606 * In case of error it returns ERR_PTR(err).
2608 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2610 struct sk_buff *segs = NULL;
2611 struct sk_buff *tail = NULL;
2612 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2613 unsigned int mss = skb_shinfo(skb)->gso_size;
2614 unsigned int doffset = skb->data - skb_mac_header(skb);
2615 unsigned int offset = doffset;
2616 unsigned int headroom;
2618 int sg = !!(features & NETIF_F_SG);
2619 int nfrags = skb_shinfo(skb)->nr_frags;
2624 __skb_push(skb, doffset);
2625 headroom = skb_headroom(skb);
2626 pos = skb_headlen(skb);
2629 struct sk_buff *nskb;
2634 len = skb->len - offset;
2638 hsize = skb_headlen(skb) - offset;
2641 if (hsize > len || !sg)
2644 if (!hsize && i >= nfrags) {
2645 BUG_ON(fskb->len != len);
2648 nskb = skb_clone(fskb, GFP_ATOMIC);
2651 if (unlikely(!nskb))
2654 hsize = skb_end_pointer(nskb) - nskb->head;
2655 if (skb_cow_head(nskb, doffset + headroom)) {
2660 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2662 skb_release_head_state(nskb);
2663 __skb_push(nskb, doffset);
2665 nskb = alloc_skb(hsize + doffset + headroom,
2668 if (unlikely(!nskb))
2671 skb_reserve(nskb, headroom);
2672 __skb_put(nskb, doffset);
2681 __copy_skb_header(nskb, skb);
2682 nskb->mac_len = skb->mac_len;
2684 /* nskb and skb might have different headroom */
2685 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2686 nskb->csum_start += skb_headroom(nskb) - headroom;
2688 skb_reset_mac_header(nskb);
2689 skb_set_network_header(nskb, skb->mac_len);
2690 nskb->transport_header = (nskb->network_header +
2691 skb_network_header_len(skb));
2692 skb_copy_from_linear_data(skb, nskb->data, doffset);
2694 if (fskb != skb_shinfo(skb)->frag_list)
2698 nskb->ip_summed = CHECKSUM_NONE;
2699 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2705 frag = skb_shinfo(nskb)->frags;
2707 skb_copy_from_linear_data_offset(skb, offset,
2708 skb_put(nskb, hsize), hsize);
2710 while (pos < offset + len && i < nfrags) {
2711 *frag = skb_shinfo(skb)->frags[i];
2712 __skb_frag_ref(frag);
2716 frag->page_offset += offset - pos;
2717 frag->size -= offset - pos;
2720 skb_shinfo(nskb)->nr_frags++;
2722 if (pos + size <= offset + len) {
2726 frag->size -= pos + size - (offset + len);
2733 if (pos < offset + len) {
2734 struct sk_buff *fskb2 = fskb;
2736 BUG_ON(pos + fskb->len != offset + len);
2742 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2748 SKB_FRAG_ASSERT(nskb);
2749 skb_shinfo(nskb)->frag_list = fskb2;
2753 nskb->data_len = len - hsize;
2754 nskb->len += nskb->data_len;
2755 nskb->truesize += nskb->data_len;
2756 } while ((offset += len) < skb->len);
2761 while ((skb = segs)) {
2765 return ERR_PTR(err);
2767 EXPORT_SYMBOL_GPL(skb_segment);
2769 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2771 struct sk_buff *p = *head;
2772 struct sk_buff *nskb;
2773 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2774 struct skb_shared_info *pinfo = skb_shinfo(p);
2775 unsigned int headroom;
2776 unsigned int len = skb_gro_len(skb);
2777 unsigned int offset = skb_gro_offset(skb);
2778 unsigned int headlen = skb_headlen(skb);
2780 if (p->len + len >= 65536)
2783 if (pinfo->frag_list)
2785 else if (headlen <= offset) {
2788 int i = skbinfo->nr_frags;
2789 int nr_frags = pinfo->nr_frags + i;
2793 if (nr_frags > MAX_SKB_FRAGS)
2796 pinfo->nr_frags = nr_frags;
2797 skbinfo->nr_frags = 0;
2799 frag = pinfo->frags + nr_frags;
2800 frag2 = skbinfo->frags + i;
2805 frag->page_offset += offset;
2806 frag->size -= offset;
2808 skb->truesize -= skb->data_len;
2809 skb->len -= skb->data_len;
2812 NAPI_GRO_CB(skb)->free = 1;
2814 } else if (skb_gro_len(p) != pinfo->gso_size)
2817 headroom = skb_headroom(p);
2818 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2819 if (unlikely(!nskb))
2822 __copy_skb_header(nskb, p);
2823 nskb->mac_len = p->mac_len;
2825 skb_reserve(nskb, headroom);
2826 __skb_put(nskb, skb_gro_offset(p));
2828 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2829 skb_set_network_header(nskb, skb_network_offset(p));
2830 skb_set_transport_header(nskb, skb_transport_offset(p));
2832 __skb_pull(p, skb_gro_offset(p));
2833 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2834 p->data - skb_mac_header(p));
2836 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2837 skb_shinfo(nskb)->frag_list = p;
2838 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2839 pinfo->gso_size = 0;
2840 skb_header_release(p);
2843 nskb->data_len += p->len;
2844 nskb->truesize += p->len;
2845 nskb->len += p->len;
2848 nskb->next = p->next;
2854 if (offset > headlen) {
2855 unsigned int eat = offset - headlen;
2857 skbinfo->frags[0].page_offset += eat;
2858 skbinfo->frags[0].size -= eat;
2859 skb->data_len -= eat;
2864 __skb_pull(skb, offset);
2866 p->prev->next = skb;
2868 skb_header_release(skb);
2871 NAPI_GRO_CB(p)->count++;
2876 NAPI_GRO_CB(skb)->same_flow = 1;
2879 EXPORT_SYMBOL_GPL(skb_gro_receive);
2881 void __init skb_init(void)
2883 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2884 sizeof(struct sk_buff),
2886 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2888 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2889 (2*sizeof(struct sk_buff)) +
2892 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2897 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2898 * @skb: Socket buffer containing the buffers to be mapped
2899 * @sg: The scatter-gather list to map into
2900 * @offset: The offset into the buffer's contents to start mapping
2901 * @len: Length of buffer space to be mapped
2903 * Fill the specified scatter-gather list with mappings/pointers into a
2904 * region of the buffer space attached to a socket buffer.
2907 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2909 int start = skb_headlen(skb);
2910 int i, copy = start - offset;
2911 struct sk_buff *frag_iter;
2917 sg_set_buf(sg, skb->data + offset, copy);
2919 if ((len -= copy) == 0)
2924 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2927 WARN_ON(start > offset + len);
2929 end = start + skb_shinfo(skb)->frags[i].size;
2930 if ((copy = end - offset) > 0) {
2931 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2935 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
2936 frag->page_offset+offset-start);
2945 skb_walk_frags(skb, frag_iter) {
2948 WARN_ON(start > offset + len);
2950 end = start + frag_iter->len;
2951 if ((copy = end - offset) > 0) {
2954 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2956 if ((len -= copy) == 0)
2966 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2968 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2970 sg_mark_end(&sg[nsg - 1]);
2974 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2977 * skb_cow_data - Check that a socket buffer's data buffers are writable
2978 * @skb: The socket buffer to check.
2979 * @tailbits: Amount of trailing space to be added
2980 * @trailer: Returned pointer to the skb where the @tailbits space begins
2982 * Make sure that the data buffers attached to a socket buffer are
2983 * writable. If they are not, private copies are made of the data buffers
2984 * and the socket buffer is set to use these instead.
2986 * If @tailbits is given, make sure that there is space to write @tailbits
2987 * bytes of data beyond current end of socket buffer. @trailer will be
2988 * set to point to the skb in which this space begins.
2990 * The number of scatterlist elements required to completely map the
2991 * COW'd and extended socket buffer will be returned.
2993 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2997 struct sk_buff *skb1, **skb_p;
2999 /* If skb is cloned or its head is paged, reallocate
3000 * head pulling out all the pages (pages are considered not writable
3001 * at the moment even if they are anonymous).
3003 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3004 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3007 /* Easy case. Most of packets will go this way. */
3008 if (!skb_has_frag_list(skb)) {
3009 /* A little of trouble, not enough of space for trailer.
3010 * This should not happen, when stack is tuned to generate
3011 * good frames. OK, on miss we reallocate and reserve even more
3012 * space, 128 bytes is fair. */
3014 if (skb_tailroom(skb) < tailbits &&
3015 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3023 /* Misery. We are in troubles, going to mincer fragments... */
3026 skb_p = &skb_shinfo(skb)->frag_list;
3029 while ((skb1 = *skb_p) != NULL) {
3032 /* The fragment is partially pulled by someone,
3033 * this can happen on input. Copy it and everything
3036 if (skb_shared(skb1))
3039 /* If the skb is the last, worry about trailer. */
3041 if (skb1->next == NULL && tailbits) {
3042 if (skb_shinfo(skb1)->nr_frags ||
3043 skb_has_frag_list(skb1) ||
3044 skb_tailroom(skb1) < tailbits)
3045 ntail = tailbits + 128;
3051 skb_shinfo(skb1)->nr_frags ||
3052 skb_has_frag_list(skb1)) {
3053 struct sk_buff *skb2;
3055 /* Fuck, we are miserable poor guys... */
3057 skb2 = skb_copy(skb1, GFP_ATOMIC);
3059 skb2 = skb_copy_expand(skb1,
3063 if (unlikely(skb2 == NULL))
3067 skb_set_owner_w(skb2, skb1->sk);
3069 /* Looking around. Are we still alive?
3070 * OK, link new skb, drop old one */
3072 skb2->next = skb1->next;
3079 skb_p = &skb1->next;
3084 EXPORT_SYMBOL_GPL(skb_cow_data);
3086 static void sock_rmem_free(struct sk_buff *skb)
3088 struct sock *sk = skb->sk;
3090 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3094 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3096 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3098 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3099 (unsigned)sk->sk_rcvbuf)
3104 skb->destructor = sock_rmem_free;
3105 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3107 /* before exiting rcu section, make sure dst is refcounted */
3110 skb_queue_tail(&sk->sk_error_queue, skb);
3111 if (!sock_flag(sk, SOCK_DEAD))
3112 sk->sk_data_ready(sk, skb->len);
3115 EXPORT_SYMBOL(sock_queue_err_skb);
3117 void skb_tstamp_tx(struct sk_buff *orig_skb,
3118 struct skb_shared_hwtstamps *hwtstamps)
3120 struct sock *sk = orig_skb->sk;
3121 struct sock_exterr_skb *serr;
3122 struct sk_buff *skb;
3128 skb = skb_clone(orig_skb, GFP_ATOMIC);
3133 *skb_hwtstamps(skb) =
3137 * no hardware time stamps available,
3138 * so keep the shared tx_flags and only
3139 * store software time stamp
3141 skb->tstamp = ktime_get_real();
3144 serr = SKB_EXT_ERR(skb);
3145 memset(serr, 0, sizeof(*serr));
3146 serr->ee.ee_errno = ENOMSG;
3147 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3149 err = sock_queue_err_skb(sk, skb);
3154 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3158 * skb_partial_csum_set - set up and verify partial csum values for packet
3159 * @skb: the skb to set
3160 * @start: the number of bytes after skb->data to start checksumming.
3161 * @off: the offset from start to place the checksum.
3163 * For untrusted partially-checksummed packets, we need to make sure the values
3164 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3166 * This function checks and sets those values and skb->ip_summed: if this
3167 * returns false you should drop the packet.
3169 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3171 if (unlikely(start > skb_headlen(skb)) ||
3172 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3173 if (net_ratelimit())
3175 "bad partial csum: csum=%u/%u len=%u\n",
3176 start, off, skb_headlen(skb));
3179 skb->ip_summed = CHECKSUM_PARTIAL;
3180 skb->csum_start = skb_headroom(skb) + start;
3181 skb->csum_offset = off;
3184 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3186 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3188 if (net_ratelimit())
3189 pr_warning("%s: received packets cannot be forwarded"
3190 " while LRO is enabled\n", skb->dev->name);
3192 EXPORT_SYMBOL(__skb_warn_lro_forwarding);