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 put_page(skb_shinfo(skb)->frags[i].page);
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;
533 new->sp = secpath_get(old->sp);
535 memcpy(new->cb, old->cb, sizeof(old->cb));
536 new->csum = old->csum;
537 new->local_df = old->local_df;
538 new->pkt_type = old->pkt_type;
539 new->ip_summed = old->ip_summed;
540 skb_copy_queue_mapping(new, old);
541 new->priority = old->priority;
542 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
543 new->ipvs_property = old->ipvs_property;
545 new->protocol = old->protocol;
546 new->mark = old->mark;
547 new->skb_iif = old->skb_iif;
549 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
550 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
551 new->nf_trace = old->nf_trace;
553 #ifdef CONFIG_NET_SCHED
554 new->tc_index = old->tc_index;
555 #ifdef CONFIG_NET_CLS_ACT
556 new->tc_verd = old->tc_verd;
559 new->vlan_tci = old->vlan_tci;
561 skb_copy_secmark(new, old);
565 * You should not add any new code to this function. Add it to
566 * __copy_skb_header above instead.
568 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
570 #define C(x) n->x = skb->x
572 n->next = n->prev = NULL;
574 __copy_skb_header(n, skb);
579 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
582 n->destructor = NULL;
588 atomic_set(&n->users, 1);
590 atomic_inc(&(skb_shinfo(skb)->dataref));
598 * skb_morph - morph one skb into another
599 * @dst: the skb to receive the contents
600 * @src: the skb to supply the contents
602 * This is identical to skb_clone except that the target skb is
603 * supplied by the user.
605 * The target skb is returned upon exit.
607 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
609 skb_release_all(dst);
610 return __skb_clone(dst, src);
612 EXPORT_SYMBOL_GPL(skb_morph);
614 /* skb frags copy userspace buffers to kernel */
615 static int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
618 int num_frags = skb_shinfo(skb)->nr_frags;
619 struct page *page, *head = NULL;
620 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
622 for (i = 0; i < num_frags; i++) {
624 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
626 page = alloc_page(GFP_ATOMIC);
629 struct page *next = (struct page *)head->private;
635 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
636 memcpy(page_address(page),
637 vaddr + f->page_offset, f->size);
638 kunmap_skb_frag(vaddr);
639 page->private = (unsigned long)head;
643 /* skb frags release userspace buffers */
644 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
645 put_page(skb_shinfo(skb)->frags[i].page);
647 uarg->callback(uarg);
649 /* skb frags point to kernel buffers */
650 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
651 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
652 skb_shinfo(skb)->frags[i - 1].page = head;
653 head = (struct page *)head->private;
660 * skb_clone - duplicate an sk_buff
661 * @skb: buffer to clone
662 * @gfp_mask: allocation priority
664 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
665 * copies share the same packet data but not structure. The new
666 * buffer has a reference count of 1. If the allocation fails the
667 * function returns %NULL otherwise the new buffer is returned.
669 * If this function is called from an interrupt gfp_mask() must be
673 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
677 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
678 if (skb_copy_ubufs(skb, gfp_mask))
680 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
684 if (skb->fclone == SKB_FCLONE_ORIG &&
685 n->fclone == SKB_FCLONE_UNAVAILABLE) {
686 atomic_t *fclone_ref = (atomic_t *) (n + 1);
687 n->fclone = SKB_FCLONE_CLONE;
688 atomic_inc(fclone_ref);
690 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
694 kmemcheck_annotate_bitfield(n, flags1);
695 kmemcheck_annotate_bitfield(n, flags2);
696 n->fclone = SKB_FCLONE_UNAVAILABLE;
699 return __skb_clone(n, skb);
701 EXPORT_SYMBOL(skb_clone);
703 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
705 #ifndef NET_SKBUFF_DATA_USES_OFFSET
707 * Shift between the two data areas in bytes
709 unsigned long offset = new->data - old->data;
712 __copy_skb_header(new, old);
714 #ifndef NET_SKBUFF_DATA_USES_OFFSET
715 /* {transport,network,mac}_header are relative to skb->head */
716 new->transport_header += offset;
717 new->network_header += offset;
718 if (skb_mac_header_was_set(new))
719 new->mac_header += offset;
721 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
722 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
723 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
727 * skb_copy - create private copy of an sk_buff
728 * @skb: buffer to copy
729 * @gfp_mask: allocation priority
731 * Make a copy of both an &sk_buff and its data. This is used when the
732 * caller wishes to modify the data and needs a private copy of the
733 * data to alter. Returns %NULL on failure or the pointer to the buffer
734 * on success. The returned buffer has a reference count of 1.
736 * As by-product this function converts non-linear &sk_buff to linear
737 * one, so that &sk_buff becomes completely private and caller is allowed
738 * to modify all the data of returned buffer. This means that this
739 * function is not recommended for use in circumstances when only
740 * header is going to be modified. Use pskb_copy() instead.
743 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
745 int headerlen = skb_headroom(skb);
746 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
747 struct sk_buff *n = alloc_skb(size, gfp_mask);
752 /* Set the data pointer */
753 skb_reserve(n, headerlen);
754 /* Set the tail pointer and length */
755 skb_put(n, skb->len);
757 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
760 copy_skb_header(n, skb);
763 EXPORT_SYMBOL(skb_copy);
766 * pskb_copy - create copy of an sk_buff with private head.
767 * @skb: buffer to copy
768 * @gfp_mask: allocation priority
770 * Make a copy of both an &sk_buff and part of its data, located
771 * in header. Fragmented data remain shared. This is used when
772 * the caller wishes to modify only header of &sk_buff and needs
773 * private copy of the header to alter. Returns %NULL on failure
774 * or the pointer to the buffer on success.
775 * The returned buffer has a reference count of 1.
778 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
780 unsigned int size = skb_end_pointer(skb) - skb->head;
781 struct sk_buff *n = alloc_skb(size, gfp_mask);
786 /* Set the data pointer */
787 skb_reserve(n, skb_headroom(skb));
788 /* Set the tail pointer and length */
789 skb_put(n, skb_headlen(skb));
791 skb_copy_from_linear_data(skb, n->data, n->len);
793 n->truesize += skb->data_len;
794 n->data_len = skb->data_len;
797 if (skb_shinfo(skb)->nr_frags) {
800 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
801 if (skb_copy_ubufs(skb, gfp_mask)) {
806 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
808 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
809 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
810 get_page(skb_shinfo(n)->frags[i].page);
812 skb_shinfo(n)->nr_frags = i;
815 if (skb_has_frag_list(skb)) {
816 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
817 skb_clone_fraglist(n);
820 copy_skb_header(n, skb);
824 EXPORT_SYMBOL(pskb_copy);
827 * pskb_expand_head - reallocate header of &sk_buff
828 * @skb: buffer to reallocate
829 * @nhead: room to add at head
830 * @ntail: room to add at tail
831 * @gfp_mask: allocation priority
833 * Expands (or creates identical copy, if &nhead and &ntail are zero)
834 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
835 * reference count of 1. Returns zero in the case of success or error,
836 * if expansion failed. In the last case, &sk_buff is not changed.
838 * All the pointers pointing into skb header may change and must be
839 * reloaded after call to this function.
842 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
847 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
856 size = SKB_DATA_ALIGN(size);
858 /* Check if we can avoid taking references on fragments if we own
859 * the last reference on skb->head. (see skb_release_data())
864 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
865 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
869 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
870 memmove(skb->head + size, skb_shinfo(skb),
871 offsetof(struct skb_shared_info,
872 frags[skb_shinfo(skb)->nr_frags]));
873 memmove(skb->head + nhead, skb->head,
874 skb_tail_pointer(skb) - skb->head);
879 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
883 /* Copy only real data... and, alas, header. This should be
884 * optimized for the cases when header is void.
886 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
888 memcpy((struct skb_shared_info *)(data + size),
890 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
895 /* copy this zero copy skb frags */
896 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
897 if (skb_copy_ubufs(skb, gfp_mask))
899 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
901 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
902 get_page(skb_shinfo(skb)->frags[i].page);
904 if (skb_has_frag_list(skb))
905 skb_clone_fraglist(skb);
907 skb_release_data(skb);
909 off = (data + nhead) - skb->head;
914 #ifdef NET_SKBUFF_DATA_USES_OFFSET
918 skb->end = skb->head + size;
920 /* {transport,network,mac}_header and tail are relative to skb->head */
922 skb->transport_header += off;
923 skb->network_header += off;
924 if (skb_mac_header_was_set(skb))
925 skb->mac_header += off;
926 /* Only adjust this if it actually is csum_start rather than csum */
927 if (skb->ip_summed == CHECKSUM_PARTIAL)
928 skb->csum_start += nhead;
932 atomic_set(&skb_shinfo(skb)->dataref, 1);
940 EXPORT_SYMBOL(pskb_expand_head);
942 /* Make private copy of skb with writable head and some headroom */
944 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
946 struct sk_buff *skb2;
947 int delta = headroom - skb_headroom(skb);
950 skb2 = pskb_copy(skb, GFP_ATOMIC);
952 skb2 = skb_clone(skb, GFP_ATOMIC);
953 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
961 EXPORT_SYMBOL(skb_realloc_headroom);
964 * skb_copy_expand - copy and expand sk_buff
965 * @skb: buffer to copy
966 * @newheadroom: new free bytes at head
967 * @newtailroom: new free bytes at tail
968 * @gfp_mask: allocation priority
970 * Make a copy of both an &sk_buff and its data and while doing so
971 * allocate additional space.
973 * This is used when the caller wishes to modify the data and needs a
974 * private copy of the data to alter as well as more space for new fields.
975 * Returns %NULL on failure or the pointer to the buffer
976 * on success. The returned buffer has a reference count of 1.
978 * You must pass %GFP_ATOMIC as the allocation priority if this function
979 * is called from an interrupt.
981 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
982 int newheadroom, int newtailroom,
986 * Allocate the copy buffer
988 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
990 int oldheadroom = skb_headroom(skb);
991 int head_copy_len, head_copy_off;
997 skb_reserve(n, newheadroom);
999 /* Set the tail pointer and length */
1000 skb_put(n, skb->len);
1002 head_copy_len = oldheadroom;
1004 if (newheadroom <= head_copy_len)
1005 head_copy_len = newheadroom;
1007 head_copy_off = newheadroom - head_copy_len;
1009 /* Copy the linear header and data. */
1010 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1011 skb->len + head_copy_len))
1014 copy_skb_header(n, skb);
1016 off = newheadroom - oldheadroom;
1017 if (n->ip_summed == CHECKSUM_PARTIAL)
1018 n->csum_start += off;
1019 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1020 n->transport_header += off;
1021 n->network_header += off;
1022 if (skb_mac_header_was_set(skb))
1023 n->mac_header += off;
1028 EXPORT_SYMBOL(skb_copy_expand);
1031 * skb_pad - zero pad the tail of an skb
1032 * @skb: buffer to pad
1033 * @pad: space to pad
1035 * Ensure that a buffer is followed by a padding area that is zero
1036 * filled. Used by network drivers which may DMA or transfer data
1037 * beyond the buffer end onto the wire.
1039 * May return error in out of memory cases. The skb is freed on error.
1042 int skb_pad(struct sk_buff *skb, int pad)
1047 /* If the skbuff is non linear tailroom is always zero.. */
1048 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1049 memset(skb->data+skb->len, 0, pad);
1053 ntail = skb->data_len + pad - (skb->end - skb->tail);
1054 if (likely(skb_cloned(skb) || ntail > 0)) {
1055 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1060 /* FIXME: The use of this function with non-linear skb's really needs
1063 err = skb_linearize(skb);
1067 memset(skb->data + skb->len, 0, pad);
1074 EXPORT_SYMBOL(skb_pad);
1077 * skb_put - add data to a buffer
1078 * @skb: buffer to use
1079 * @len: amount of data to add
1081 * This function extends the used data area of the buffer. If this would
1082 * exceed the total buffer size the kernel will panic. A pointer to the
1083 * first byte of the extra data is returned.
1085 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1087 unsigned char *tmp = skb_tail_pointer(skb);
1088 SKB_LINEAR_ASSERT(skb);
1091 if (unlikely(skb->tail > skb->end))
1092 skb_over_panic(skb, len, __builtin_return_address(0));
1095 EXPORT_SYMBOL(skb_put);
1098 * skb_push - add data to the start of a buffer
1099 * @skb: buffer to use
1100 * @len: amount of data to add
1102 * This function extends the used data area of the buffer at the buffer
1103 * start. If this would exceed the total buffer headroom the kernel will
1104 * panic. A pointer to the first byte of the extra data is returned.
1106 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1110 if (unlikely(skb->data<skb->head))
1111 skb_under_panic(skb, len, __builtin_return_address(0));
1114 EXPORT_SYMBOL(skb_push);
1117 * skb_pull - remove data from the start of a buffer
1118 * @skb: buffer to use
1119 * @len: amount of data to remove
1121 * This function removes data from the start of a buffer, returning
1122 * the memory to the headroom. A pointer to the next data in the buffer
1123 * is returned. Once the data has been pulled future pushes will overwrite
1126 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1128 return skb_pull_inline(skb, len);
1130 EXPORT_SYMBOL(skb_pull);
1133 * skb_trim - remove end from a buffer
1134 * @skb: buffer to alter
1137 * Cut the length of a buffer down by removing data from the tail. If
1138 * the buffer is already under the length specified it is not modified.
1139 * The skb must be linear.
1141 void skb_trim(struct sk_buff *skb, unsigned int len)
1144 __skb_trim(skb, len);
1146 EXPORT_SYMBOL(skb_trim);
1148 /* Trims skb to length len. It can change skb pointers.
1151 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1153 struct sk_buff **fragp;
1154 struct sk_buff *frag;
1155 int offset = skb_headlen(skb);
1156 int nfrags = skb_shinfo(skb)->nr_frags;
1160 if (skb_cloned(skb) &&
1161 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1168 for (; i < nfrags; i++) {
1169 int end = offset + skb_shinfo(skb)->frags[i].size;
1176 skb_shinfo(skb)->frags[i++].size = len - offset;
1179 skb_shinfo(skb)->nr_frags = i;
1181 for (; i < nfrags; i++)
1182 put_page(skb_shinfo(skb)->frags[i].page);
1184 if (skb_has_frag_list(skb))
1185 skb_drop_fraglist(skb);
1189 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1190 fragp = &frag->next) {
1191 int end = offset + frag->len;
1193 if (skb_shared(frag)) {
1194 struct sk_buff *nfrag;
1196 nfrag = skb_clone(frag, GFP_ATOMIC);
1197 if (unlikely(!nfrag))
1200 nfrag->next = frag->next;
1212 unlikely((err = pskb_trim(frag, len - offset))))
1216 skb_drop_list(&frag->next);
1221 if (len > skb_headlen(skb)) {
1222 skb->data_len -= skb->len - len;
1227 skb_set_tail_pointer(skb, len);
1232 EXPORT_SYMBOL(___pskb_trim);
1235 * __pskb_pull_tail - advance tail of skb header
1236 * @skb: buffer to reallocate
1237 * @delta: number of bytes to advance tail
1239 * The function makes a sense only on a fragmented &sk_buff,
1240 * it expands header moving its tail forward and copying necessary
1241 * data from fragmented part.
1243 * &sk_buff MUST have reference count of 1.
1245 * Returns %NULL (and &sk_buff does not change) if pull failed
1246 * or value of new tail of skb in the case of success.
1248 * All the pointers pointing into skb header may change and must be
1249 * reloaded after call to this function.
1252 /* Moves tail of skb head forward, copying data from fragmented part,
1253 * when it is necessary.
1254 * 1. It may fail due to malloc failure.
1255 * 2. It may change skb pointers.
1257 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1259 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1261 /* If skb has not enough free space at tail, get new one
1262 * plus 128 bytes for future expansions. If we have enough
1263 * room at tail, reallocate without expansion only if skb is cloned.
1265 int i, k, eat = (skb->tail + delta) - skb->end;
1267 if (eat > 0 || skb_cloned(skb)) {
1268 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1273 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1276 /* Optimization: no fragments, no reasons to preestimate
1277 * size of pulled pages. Superb.
1279 if (!skb_has_frag_list(skb))
1282 /* Estimate size of pulled pages. */
1284 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1285 if (skb_shinfo(skb)->frags[i].size >= eat)
1287 eat -= skb_shinfo(skb)->frags[i].size;
1290 /* If we need update frag list, we are in troubles.
1291 * Certainly, it possible to add an offset to skb data,
1292 * but taking into account that pulling is expected to
1293 * be very rare operation, it is worth to fight against
1294 * further bloating skb head and crucify ourselves here instead.
1295 * Pure masohism, indeed. 8)8)
1298 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1299 struct sk_buff *clone = NULL;
1300 struct sk_buff *insp = NULL;
1305 if (list->len <= eat) {
1306 /* Eaten as whole. */
1311 /* Eaten partially. */
1313 if (skb_shared(list)) {
1314 /* Sucks! We need to fork list. :-( */
1315 clone = skb_clone(list, GFP_ATOMIC);
1321 /* This may be pulled without
1325 if (!pskb_pull(list, eat)) {
1333 /* Free pulled out fragments. */
1334 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1335 skb_shinfo(skb)->frag_list = list->next;
1338 /* And insert new clone at head. */
1341 skb_shinfo(skb)->frag_list = clone;
1344 /* Success! Now we may commit changes to skb data. */
1349 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1350 if (skb_shinfo(skb)->frags[i].size <= eat) {
1351 put_page(skb_shinfo(skb)->frags[i].page);
1352 eat -= skb_shinfo(skb)->frags[i].size;
1354 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1356 skb_shinfo(skb)->frags[k].page_offset += eat;
1357 skb_shinfo(skb)->frags[k].size -= eat;
1363 skb_shinfo(skb)->nr_frags = k;
1366 skb->data_len -= delta;
1368 return skb_tail_pointer(skb);
1370 EXPORT_SYMBOL(__pskb_pull_tail);
1373 * skb_copy_bits - copy bits from skb to kernel buffer
1375 * @offset: offset in source
1376 * @to: destination buffer
1377 * @len: number of bytes to copy
1379 * Copy the specified number of bytes from the source skb to the
1380 * destination buffer.
1383 * If its prototype is ever changed,
1384 * check arch/{*}/net/{*}.S files,
1385 * since it is called from BPF assembly code.
1387 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1389 int start = skb_headlen(skb);
1390 struct sk_buff *frag_iter;
1393 if (offset > (int)skb->len - len)
1397 if ((copy = start - offset) > 0) {
1400 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1401 if ((len -= copy) == 0)
1407 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1410 WARN_ON(start > offset + len);
1412 end = start + skb_shinfo(skb)->frags[i].size;
1413 if ((copy = end - offset) > 0) {
1419 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1421 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1422 offset - start, copy);
1423 kunmap_skb_frag(vaddr);
1425 if ((len -= copy) == 0)
1433 skb_walk_frags(skb, frag_iter) {
1436 WARN_ON(start > offset + len);
1438 end = start + frag_iter->len;
1439 if ((copy = end - offset) > 0) {
1442 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1444 if ((len -= copy) == 0)
1458 EXPORT_SYMBOL(skb_copy_bits);
1461 * Callback from splice_to_pipe(), if we need to release some pages
1462 * at the end of the spd in case we error'ed out in filling the pipe.
1464 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1466 put_page(spd->pages[i]);
1469 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1470 unsigned int *offset,
1471 struct sk_buff *skb, struct sock *sk)
1473 struct page *p = sk->sk_sndmsg_page;
1478 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1482 off = sk->sk_sndmsg_off = 0;
1483 /* hold one ref to this page until it's full */
1487 off = sk->sk_sndmsg_off;
1488 mlen = PAGE_SIZE - off;
1489 if (mlen < 64 && mlen < *len) {
1494 *len = min_t(unsigned int, *len, mlen);
1497 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1498 sk->sk_sndmsg_off += *len;
1506 * Fill page/offset/length into spd, if it can hold more pages.
1508 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1509 struct pipe_inode_info *pipe, struct page *page,
1510 unsigned int *len, unsigned int offset,
1511 struct sk_buff *skb, int linear,
1514 if (unlikely(spd->nr_pages == pipe->buffers))
1518 page = linear_to_page(page, len, &offset, skb, sk);
1524 spd->pages[spd->nr_pages] = page;
1525 spd->partial[spd->nr_pages].len = *len;
1526 spd->partial[spd->nr_pages].offset = offset;
1532 static inline void __segment_seek(struct page **page, unsigned int *poff,
1533 unsigned int *plen, unsigned int off)
1538 n = *poff / PAGE_SIZE;
1540 *page = nth_page(*page, n);
1542 *poff = *poff % PAGE_SIZE;
1546 static inline int __splice_segment(struct page *page, unsigned int poff,
1547 unsigned int plen, unsigned int *off,
1548 unsigned int *len, struct sk_buff *skb,
1549 struct splice_pipe_desc *spd, int linear,
1551 struct pipe_inode_info *pipe)
1556 /* skip this segment if already processed */
1562 /* ignore any bits we already processed */
1564 __segment_seek(&page, &poff, &plen, *off);
1569 unsigned int flen = min(*len, plen);
1571 /* the linear region may spread across several pages */
1572 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1574 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1577 __segment_seek(&page, &poff, &plen, flen);
1580 } while (*len && plen);
1586 * Map linear and fragment data from the skb to spd. It reports failure if the
1587 * pipe is full or if we already spliced the requested length.
1589 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1590 unsigned int *offset, unsigned int *len,
1591 struct splice_pipe_desc *spd, struct sock *sk)
1596 * map the linear part
1598 if (__splice_segment(virt_to_page(skb->data),
1599 (unsigned long) skb->data & (PAGE_SIZE - 1),
1601 offset, len, skb, spd, 1, sk, pipe))
1605 * then map the fragments
1607 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1608 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1610 if (__splice_segment(f->page, f->page_offset, f->size,
1611 offset, len, skb, spd, 0, sk, pipe))
1619 * Map data from the skb to a pipe. Should handle both the linear part,
1620 * the fragments, and the frag list. It does NOT handle frag lists within
1621 * the frag list, if such a thing exists. We'd probably need to recurse to
1622 * handle that cleanly.
1624 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1625 struct pipe_inode_info *pipe, unsigned int tlen,
1628 struct partial_page partial[PIPE_DEF_BUFFERS];
1629 struct page *pages[PIPE_DEF_BUFFERS];
1630 struct splice_pipe_desc spd = {
1634 .ops = &sock_pipe_buf_ops,
1635 .spd_release = sock_spd_release,
1637 struct sk_buff *frag_iter;
1638 struct sock *sk = skb->sk;
1641 if (splice_grow_spd(pipe, &spd))
1645 * __skb_splice_bits() only fails if the output has no room left,
1646 * so no point in going over the frag_list for the error case.
1648 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1654 * now see if we have a frag_list to map
1656 skb_walk_frags(skb, frag_iter) {
1659 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1666 * Drop the socket lock, otherwise we have reverse
1667 * locking dependencies between sk_lock and i_mutex
1668 * here as compared to sendfile(). We enter here
1669 * with the socket lock held, and splice_to_pipe() will
1670 * grab the pipe inode lock. For sendfile() emulation,
1671 * we call into ->sendpage() with the i_mutex lock held
1672 * and networking will grab the socket lock.
1675 ret = splice_to_pipe(pipe, &spd);
1679 splice_shrink_spd(pipe, &spd);
1684 * skb_store_bits - store bits from kernel buffer to skb
1685 * @skb: destination buffer
1686 * @offset: offset in destination
1687 * @from: source buffer
1688 * @len: number of bytes to copy
1690 * Copy the specified number of bytes from the source buffer to the
1691 * destination skb. This function handles all the messy bits of
1692 * traversing fragment lists and such.
1695 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1697 int start = skb_headlen(skb);
1698 struct sk_buff *frag_iter;
1701 if (offset > (int)skb->len - len)
1704 if ((copy = start - offset) > 0) {
1707 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1708 if ((len -= copy) == 0)
1714 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1715 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1718 WARN_ON(start > offset + len);
1720 end = start + frag->size;
1721 if ((copy = end - offset) > 0) {
1727 vaddr = kmap_skb_frag(frag);
1728 memcpy(vaddr + frag->page_offset + offset - start,
1730 kunmap_skb_frag(vaddr);
1732 if ((len -= copy) == 0)
1740 skb_walk_frags(skb, frag_iter) {
1743 WARN_ON(start > offset + len);
1745 end = start + frag_iter->len;
1746 if ((copy = end - offset) > 0) {
1749 if (skb_store_bits(frag_iter, offset - start,
1752 if ((len -= copy) == 0)
1765 EXPORT_SYMBOL(skb_store_bits);
1767 /* Checksum skb data. */
1769 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1770 int len, __wsum csum)
1772 int start = skb_headlen(skb);
1773 int i, copy = start - offset;
1774 struct sk_buff *frag_iter;
1777 /* Checksum header. */
1781 csum = csum_partial(skb->data + offset, copy, csum);
1782 if ((len -= copy) == 0)
1788 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1791 WARN_ON(start > offset + len);
1793 end = start + skb_shinfo(skb)->frags[i].size;
1794 if ((copy = end - offset) > 0) {
1797 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1801 vaddr = kmap_skb_frag(frag);
1802 csum2 = csum_partial(vaddr + frag->page_offset +
1803 offset - start, copy, 0);
1804 kunmap_skb_frag(vaddr);
1805 csum = csum_block_add(csum, csum2, pos);
1814 skb_walk_frags(skb, frag_iter) {
1817 WARN_ON(start > offset + len);
1819 end = start + frag_iter->len;
1820 if ((copy = end - offset) > 0) {
1824 csum2 = skb_checksum(frag_iter, offset - start,
1826 csum = csum_block_add(csum, csum2, pos);
1827 if ((len -= copy) == 0)
1838 EXPORT_SYMBOL(skb_checksum);
1840 /* Both of above in one bottle. */
1842 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1843 u8 *to, int len, __wsum csum)
1845 int start = skb_headlen(skb);
1846 int i, copy = start - offset;
1847 struct sk_buff *frag_iter;
1854 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1856 if ((len -= copy) == 0)
1863 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1866 WARN_ON(start > offset + len);
1868 end = start + skb_shinfo(skb)->frags[i].size;
1869 if ((copy = end - offset) > 0) {
1872 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1876 vaddr = kmap_skb_frag(frag);
1877 csum2 = csum_partial_copy_nocheck(vaddr +
1881 kunmap_skb_frag(vaddr);
1882 csum = csum_block_add(csum, csum2, pos);
1892 skb_walk_frags(skb, frag_iter) {
1896 WARN_ON(start > offset + len);
1898 end = start + frag_iter->len;
1899 if ((copy = end - offset) > 0) {
1902 csum2 = skb_copy_and_csum_bits(frag_iter,
1905 csum = csum_block_add(csum, csum2, pos);
1906 if ((len -= copy) == 0)
1917 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1919 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1924 if (skb->ip_summed == CHECKSUM_PARTIAL)
1925 csstart = skb_checksum_start_offset(skb);
1927 csstart = skb_headlen(skb);
1929 BUG_ON(csstart > skb_headlen(skb));
1931 skb_copy_from_linear_data(skb, to, csstart);
1934 if (csstart != skb->len)
1935 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1936 skb->len - csstart, 0);
1938 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1939 long csstuff = csstart + skb->csum_offset;
1941 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1944 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1947 * skb_dequeue - remove from the head of the queue
1948 * @list: list to dequeue from
1950 * Remove the head of the list. The list lock is taken so the function
1951 * may be used safely with other locking list functions. The head item is
1952 * returned or %NULL if the list is empty.
1955 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1957 unsigned long flags;
1958 struct sk_buff *result;
1960 spin_lock_irqsave(&list->lock, flags);
1961 result = __skb_dequeue(list);
1962 spin_unlock_irqrestore(&list->lock, flags);
1965 EXPORT_SYMBOL(skb_dequeue);
1968 * skb_dequeue_tail - remove from the tail of the queue
1969 * @list: list to dequeue from
1971 * Remove the tail of the list. The list lock is taken so the function
1972 * may be used safely with other locking list functions. The tail item is
1973 * returned or %NULL if the list is empty.
1975 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1977 unsigned long flags;
1978 struct sk_buff *result;
1980 spin_lock_irqsave(&list->lock, flags);
1981 result = __skb_dequeue_tail(list);
1982 spin_unlock_irqrestore(&list->lock, flags);
1985 EXPORT_SYMBOL(skb_dequeue_tail);
1988 * skb_queue_purge - empty a list
1989 * @list: list to empty
1991 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1992 * the list and one reference dropped. This function takes the list
1993 * lock and is atomic with respect to other list locking functions.
1995 void skb_queue_purge(struct sk_buff_head *list)
1997 struct sk_buff *skb;
1998 while ((skb = skb_dequeue(list)) != NULL)
2001 EXPORT_SYMBOL(skb_queue_purge);
2004 * skb_queue_head - queue a buffer at the list head
2005 * @list: list to use
2006 * @newsk: buffer to queue
2008 * Queue a buffer at the start of the list. This function takes the
2009 * list lock and can be used safely with other locking &sk_buff functions
2012 * A buffer cannot be placed on two lists at the same time.
2014 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2016 unsigned long flags;
2018 spin_lock_irqsave(&list->lock, flags);
2019 __skb_queue_head(list, newsk);
2020 spin_unlock_irqrestore(&list->lock, flags);
2022 EXPORT_SYMBOL(skb_queue_head);
2025 * skb_queue_tail - queue a buffer at the list tail
2026 * @list: list to use
2027 * @newsk: buffer to queue
2029 * Queue a buffer at the tail of the list. This function takes the
2030 * list lock and can be used safely with other locking &sk_buff functions
2033 * A buffer cannot be placed on two lists at the same time.
2035 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2037 unsigned long flags;
2039 spin_lock_irqsave(&list->lock, flags);
2040 __skb_queue_tail(list, newsk);
2041 spin_unlock_irqrestore(&list->lock, flags);
2043 EXPORT_SYMBOL(skb_queue_tail);
2046 * skb_unlink - remove a buffer from a list
2047 * @skb: buffer to remove
2048 * @list: list to use
2050 * Remove a packet from a list. The list locks are taken and this
2051 * function is atomic with respect to other list locked calls
2053 * You must know what list the SKB is on.
2055 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2057 unsigned long flags;
2059 spin_lock_irqsave(&list->lock, flags);
2060 __skb_unlink(skb, list);
2061 spin_unlock_irqrestore(&list->lock, flags);
2063 EXPORT_SYMBOL(skb_unlink);
2066 * skb_append - append a buffer
2067 * @old: buffer to insert after
2068 * @newsk: buffer to insert
2069 * @list: list to use
2071 * Place a packet after a given packet in a list. The list locks are taken
2072 * and this function is atomic with respect to other list locked calls.
2073 * A buffer cannot be placed on two lists at the same time.
2075 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2077 unsigned long flags;
2079 spin_lock_irqsave(&list->lock, flags);
2080 __skb_queue_after(list, old, newsk);
2081 spin_unlock_irqrestore(&list->lock, flags);
2083 EXPORT_SYMBOL(skb_append);
2086 * skb_insert - insert a buffer
2087 * @old: buffer to insert before
2088 * @newsk: buffer to insert
2089 * @list: list to use
2091 * Place a packet before a given packet in a list. The list locks are
2092 * taken and this function is atomic with respect to other list locked
2095 * A buffer cannot be placed on two lists at the same time.
2097 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2099 unsigned long flags;
2101 spin_lock_irqsave(&list->lock, flags);
2102 __skb_insert(newsk, old->prev, old, list);
2103 spin_unlock_irqrestore(&list->lock, flags);
2105 EXPORT_SYMBOL(skb_insert);
2107 static inline void skb_split_inside_header(struct sk_buff *skb,
2108 struct sk_buff* skb1,
2109 const u32 len, const int pos)
2113 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2115 /* And move data appendix as is. */
2116 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2117 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2119 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2120 skb_shinfo(skb)->nr_frags = 0;
2121 skb1->data_len = skb->data_len;
2122 skb1->len += skb1->data_len;
2125 skb_set_tail_pointer(skb, len);
2128 static inline void skb_split_no_header(struct sk_buff *skb,
2129 struct sk_buff* skb1,
2130 const u32 len, int pos)
2133 const int nfrags = skb_shinfo(skb)->nr_frags;
2135 skb_shinfo(skb)->nr_frags = 0;
2136 skb1->len = skb1->data_len = skb->len - len;
2138 skb->data_len = len - pos;
2140 for (i = 0; i < nfrags; i++) {
2141 int size = skb_shinfo(skb)->frags[i].size;
2143 if (pos + size > len) {
2144 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2148 * We have two variants in this case:
2149 * 1. Move all the frag to the second
2150 * part, if it is possible. F.e.
2151 * this approach is mandatory for TUX,
2152 * where splitting is expensive.
2153 * 2. Split is accurately. We make this.
2155 get_page(skb_shinfo(skb)->frags[i].page);
2156 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2157 skb_shinfo(skb1)->frags[0].size -= len - pos;
2158 skb_shinfo(skb)->frags[i].size = len - pos;
2159 skb_shinfo(skb)->nr_frags++;
2163 skb_shinfo(skb)->nr_frags++;
2166 skb_shinfo(skb1)->nr_frags = k;
2170 * skb_split - Split fragmented skb to two parts at length len.
2171 * @skb: the buffer to split
2172 * @skb1: the buffer to receive the second part
2173 * @len: new length for skb
2175 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2177 int pos = skb_headlen(skb);
2179 if (len < pos) /* Split line is inside header. */
2180 skb_split_inside_header(skb, skb1, len, pos);
2181 else /* Second chunk has no header, nothing to copy. */
2182 skb_split_no_header(skb, skb1, len, pos);
2184 EXPORT_SYMBOL(skb_split);
2186 /* Shifting from/to a cloned skb is a no-go.
2188 * Caller cannot keep skb_shinfo related pointers past calling here!
2190 static int skb_prepare_for_shift(struct sk_buff *skb)
2192 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2196 * skb_shift - Shifts paged data partially from skb to another
2197 * @tgt: buffer into which tail data gets added
2198 * @skb: buffer from which the paged data comes from
2199 * @shiftlen: shift up to this many bytes
2201 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2202 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2203 * It's up to caller to free skb if everything was shifted.
2205 * If @tgt runs out of frags, the whole operation is aborted.
2207 * Skb cannot include anything else but paged data while tgt is allowed
2208 * to have non-paged data as well.
2210 * TODO: full sized shift could be optimized but that would need
2211 * specialized skb free'er to handle frags without up-to-date nr_frags.
2213 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2215 int from, to, merge, todo;
2216 struct skb_frag_struct *fragfrom, *fragto;
2218 BUG_ON(shiftlen > skb->len);
2219 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2223 to = skb_shinfo(tgt)->nr_frags;
2224 fragfrom = &skb_shinfo(skb)->frags[from];
2226 /* Actual merge is delayed until the point when we know we can
2227 * commit all, so that we don't have to undo partial changes
2230 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2235 todo -= fragfrom->size;
2237 if (skb_prepare_for_shift(skb) ||
2238 skb_prepare_for_shift(tgt))
2241 /* All previous frag pointers might be stale! */
2242 fragfrom = &skb_shinfo(skb)->frags[from];
2243 fragto = &skb_shinfo(tgt)->frags[merge];
2245 fragto->size += shiftlen;
2246 fragfrom->size -= shiftlen;
2247 fragfrom->page_offset += shiftlen;
2255 /* Skip full, not-fitting skb to avoid expensive operations */
2256 if ((shiftlen == skb->len) &&
2257 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2260 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2263 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2264 if (to == MAX_SKB_FRAGS)
2267 fragfrom = &skb_shinfo(skb)->frags[from];
2268 fragto = &skb_shinfo(tgt)->frags[to];
2270 if (todo >= fragfrom->size) {
2271 *fragto = *fragfrom;
2272 todo -= fragfrom->size;
2277 get_page(fragfrom->page);
2278 fragto->page = fragfrom->page;
2279 fragto->page_offset = fragfrom->page_offset;
2280 fragto->size = todo;
2282 fragfrom->page_offset += todo;
2283 fragfrom->size -= todo;
2291 /* Ready to "commit" this state change to tgt */
2292 skb_shinfo(tgt)->nr_frags = to;
2295 fragfrom = &skb_shinfo(skb)->frags[0];
2296 fragto = &skb_shinfo(tgt)->frags[merge];
2298 fragto->size += fragfrom->size;
2299 put_page(fragfrom->page);
2302 /* Reposition in the original skb */
2304 while (from < skb_shinfo(skb)->nr_frags)
2305 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2306 skb_shinfo(skb)->nr_frags = to;
2308 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2311 /* Most likely the tgt won't ever need its checksum anymore, skb on
2312 * the other hand might need it if it needs to be resent
2314 tgt->ip_summed = CHECKSUM_PARTIAL;
2315 skb->ip_summed = CHECKSUM_PARTIAL;
2317 /* Yak, is it really working this way? Some helper please? */
2318 skb->len -= shiftlen;
2319 skb->data_len -= shiftlen;
2320 skb->truesize -= shiftlen;
2321 tgt->len += shiftlen;
2322 tgt->data_len += shiftlen;
2323 tgt->truesize += shiftlen;
2329 * skb_prepare_seq_read - Prepare a sequential read of skb data
2330 * @skb: the buffer to read
2331 * @from: lower offset of data to be read
2332 * @to: upper offset of data to be read
2333 * @st: state variable
2335 * Initializes the specified state variable. Must be called before
2336 * invoking skb_seq_read() for the first time.
2338 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2339 unsigned int to, struct skb_seq_state *st)
2341 st->lower_offset = from;
2342 st->upper_offset = to;
2343 st->root_skb = st->cur_skb = skb;
2344 st->frag_idx = st->stepped_offset = 0;
2345 st->frag_data = NULL;
2347 EXPORT_SYMBOL(skb_prepare_seq_read);
2350 * skb_seq_read - Sequentially read skb data
2351 * @consumed: number of bytes consumed by the caller so far
2352 * @data: destination pointer for data to be returned
2353 * @st: state variable
2355 * Reads a block of skb data at &consumed relative to the
2356 * lower offset specified to skb_prepare_seq_read(). Assigns
2357 * the head of the data block to &data and returns the length
2358 * of the block or 0 if the end of the skb data or the upper
2359 * offset has been reached.
2361 * The caller is not required to consume all of the data
2362 * returned, i.e. &consumed is typically set to the number
2363 * of bytes already consumed and the next call to
2364 * skb_seq_read() will return the remaining part of the block.
2366 * Note 1: The size of each block of data returned can be arbitrary,
2367 * this limitation is the cost for zerocopy seqeuental
2368 * reads of potentially non linear data.
2370 * Note 2: Fragment lists within fragments are not implemented
2371 * at the moment, state->root_skb could be replaced with
2372 * a stack for this purpose.
2374 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2375 struct skb_seq_state *st)
2377 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2380 if (unlikely(abs_offset >= st->upper_offset))
2384 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2386 if (abs_offset < block_limit && !st->frag_data) {
2387 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2388 return block_limit - abs_offset;
2391 if (st->frag_idx == 0 && !st->frag_data)
2392 st->stepped_offset += skb_headlen(st->cur_skb);
2394 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2395 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2396 block_limit = frag->size + st->stepped_offset;
2398 if (abs_offset < block_limit) {
2400 st->frag_data = kmap_skb_frag(frag);
2402 *data = (u8 *) st->frag_data + frag->page_offset +
2403 (abs_offset - st->stepped_offset);
2405 return block_limit - abs_offset;
2408 if (st->frag_data) {
2409 kunmap_skb_frag(st->frag_data);
2410 st->frag_data = NULL;
2414 st->stepped_offset += frag->size;
2417 if (st->frag_data) {
2418 kunmap_skb_frag(st->frag_data);
2419 st->frag_data = NULL;
2422 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2423 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2426 } else if (st->cur_skb->next) {
2427 st->cur_skb = st->cur_skb->next;
2434 EXPORT_SYMBOL(skb_seq_read);
2437 * skb_abort_seq_read - Abort a sequential read of skb data
2438 * @st: state variable
2440 * Must be called if skb_seq_read() was not called until it
2443 void skb_abort_seq_read(struct skb_seq_state *st)
2446 kunmap_skb_frag(st->frag_data);
2448 EXPORT_SYMBOL(skb_abort_seq_read);
2450 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2452 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2453 struct ts_config *conf,
2454 struct ts_state *state)
2456 return skb_seq_read(offset, text, TS_SKB_CB(state));
2459 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2461 skb_abort_seq_read(TS_SKB_CB(state));
2465 * skb_find_text - Find a text pattern in skb data
2466 * @skb: the buffer to look in
2467 * @from: search offset
2469 * @config: textsearch configuration
2470 * @state: uninitialized textsearch state variable
2472 * Finds a pattern in the skb data according to the specified
2473 * textsearch configuration. Use textsearch_next() to retrieve
2474 * subsequent occurrences of the pattern. Returns the offset
2475 * to the first occurrence or UINT_MAX if no match was found.
2477 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2478 unsigned int to, struct ts_config *config,
2479 struct ts_state *state)
2483 config->get_next_block = skb_ts_get_next_block;
2484 config->finish = skb_ts_finish;
2486 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2488 ret = textsearch_find(config, state);
2489 return (ret <= to - from ? ret : UINT_MAX);
2491 EXPORT_SYMBOL(skb_find_text);
2494 * skb_append_datato_frags: - append the user data to a skb
2495 * @sk: sock structure
2496 * @skb: skb structure to be appened with user data.
2497 * @getfrag: call back function to be used for getting the user data
2498 * @from: pointer to user message iov
2499 * @length: length of the iov message
2501 * Description: This procedure append the user data in the fragment part
2502 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2504 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2505 int (*getfrag)(void *from, char *to, int offset,
2506 int len, int odd, struct sk_buff *skb),
2507 void *from, int length)
2510 skb_frag_t *frag = NULL;
2511 struct page *page = NULL;
2517 /* Return error if we don't have space for new frag */
2518 frg_cnt = skb_shinfo(skb)->nr_frags;
2519 if (frg_cnt >= MAX_SKB_FRAGS)
2522 /* allocate a new page for next frag */
2523 page = alloc_pages(sk->sk_allocation, 0);
2525 /* If alloc_page fails just return failure and caller will
2526 * free previous allocated pages by doing kfree_skb()
2531 /* initialize the next frag */
2532 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2533 skb->truesize += PAGE_SIZE;
2534 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2536 /* get the new initialized frag */
2537 frg_cnt = skb_shinfo(skb)->nr_frags;
2538 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2540 /* copy the user data to page */
2541 left = PAGE_SIZE - frag->page_offset;
2542 copy = (length > left)? left : length;
2544 ret = getfrag(from, (page_address(frag->page) +
2545 frag->page_offset + frag->size),
2546 offset, copy, 0, skb);
2550 /* copy was successful so update the size parameters */
2553 skb->data_len += copy;
2557 } while (length > 0);
2561 EXPORT_SYMBOL(skb_append_datato_frags);
2564 * skb_pull_rcsum - pull skb and update receive checksum
2565 * @skb: buffer to update
2566 * @len: length of data pulled
2568 * This function performs an skb_pull on the packet and updates
2569 * the CHECKSUM_COMPLETE checksum. It should be used on
2570 * receive path processing instead of skb_pull unless you know
2571 * that the checksum difference is zero (e.g., a valid IP header)
2572 * or you are setting ip_summed to CHECKSUM_NONE.
2574 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2576 BUG_ON(len > skb->len);
2578 BUG_ON(skb->len < skb->data_len);
2579 skb_postpull_rcsum(skb, skb->data, len);
2580 return skb->data += len;
2582 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2585 * skb_segment - Perform protocol segmentation on skb.
2586 * @skb: buffer to segment
2587 * @features: features for the output path (see dev->features)
2589 * This function performs segmentation on the given skb. It returns
2590 * a pointer to the first in a list of new skbs for the segments.
2591 * In case of error it returns ERR_PTR(err).
2593 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2595 struct sk_buff *segs = NULL;
2596 struct sk_buff *tail = NULL;
2597 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2598 unsigned int mss = skb_shinfo(skb)->gso_size;
2599 unsigned int doffset = skb->data - skb_mac_header(skb);
2600 unsigned int offset = doffset;
2601 unsigned int headroom;
2603 int sg = !!(features & NETIF_F_SG);
2604 int nfrags = skb_shinfo(skb)->nr_frags;
2609 __skb_push(skb, doffset);
2610 headroom = skb_headroom(skb);
2611 pos = skb_headlen(skb);
2614 struct sk_buff *nskb;
2619 len = skb->len - offset;
2623 hsize = skb_headlen(skb) - offset;
2626 if (hsize > len || !sg)
2629 if (!hsize && i >= nfrags) {
2630 BUG_ON(fskb->len != len);
2633 nskb = skb_clone(fskb, GFP_ATOMIC);
2636 if (unlikely(!nskb))
2639 hsize = skb_end_pointer(nskb) - nskb->head;
2640 if (skb_cow_head(nskb, doffset + headroom)) {
2645 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2647 skb_release_head_state(nskb);
2648 __skb_push(nskb, doffset);
2650 nskb = alloc_skb(hsize + doffset + headroom,
2653 if (unlikely(!nskb))
2656 skb_reserve(nskb, headroom);
2657 __skb_put(nskb, doffset);
2666 __copy_skb_header(nskb, skb);
2667 nskb->mac_len = skb->mac_len;
2669 /* nskb and skb might have different headroom */
2670 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2671 nskb->csum_start += skb_headroom(nskb) - headroom;
2673 skb_reset_mac_header(nskb);
2674 skb_set_network_header(nskb, skb->mac_len);
2675 nskb->transport_header = (nskb->network_header +
2676 skb_network_header_len(skb));
2677 skb_copy_from_linear_data(skb, nskb->data, doffset);
2679 if (fskb != skb_shinfo(skb)->frag_list)
2683 nskb->ip_summed = CHECKSUM_NONE;
2684 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2690 frag = skb_shinfo(nskb)->frags;
2692 skb_copy_from_linear_data_offset(skb, offset,
2693 skb_put(nskb, hsize), hsize);
2695 while (pos < offset + len && i < nfrags) {
2696 *frag = skb_shinfo(skb)->frags[i];
2697 get_page(frag->page);
2701 frag->page_offset += offset - pos;
2702 frag->size -= offset - pos;
2705 skb_shinfo(nskb)->nr_frags++;
2707 if (pos + size <= offset + len) {
2711 frag->size -= pos + size - (offset + len);
2718 if (pos < offset + len) {
2719 struct sk_buff *fskb2 = fskb;
2721 BUG_ON(pos + fskb->len != offset + len);
2727 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2733 SKB_FRAG_ASSERT(nskb);
2734 skb_shinfo(nskb)->frag_list = fskb2;
2738 nskb->data_len = len - hsize;
2739 nskb->len += nskb->data_len;
2740 nskb->truesize += nskb->data_len;
2741 } while ((offset += len) < skb->len);
2746 while ((skb = segs)) {
2750 return ERR_PTR(err);
2752 EXPORT_SYMBOL_GPL(skb_segment);
2754 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2756 struct sk_buff *p = *head;
2757 struct sk_buff *nskb;
2758 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2759 struct skb_shared_info *pinfo = skb_shinfo(p);
2760 unsigned int headroom;
2761 unsigned int len = skb_gro_len(skb);
2762 unsigned int offset = skb_gro_offset(skb);
2763 unsigned int headlen = skb_headlen(skb);
2765 if (p->len + len >= 65536)
2768 if (pinfo->frag_list)
2770 else if (headlen <= offset) {
2773 int i = skbinfo->nr_frags;
2774 int nr_frags = pinfo->nr_frags + i;
2778 if (nr_frags > MAX_SKB_FRAGS)
2781 pinfo->nr_frags = nr_frags;
2782 skbinfo->nr_frags = 0;
2784 frag = pinfo->frags + nr_frags;
2785 frag2 = skbinfo->frags + i;
2790 frag->page_offset += offset;
2791 frag->size -= offset;
2793 skb->truesize -= skb->data_len;
2794 skb->len -= skb->data_len;
2797 NAPI_GRO_CB(skb)->free = 1;
2799 } else if (skb_gro_len(p) != pinfo->gso_size)
2802 headroom = skb_headroom(p);
2803 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2804 if (unlikely(!nskb))
2807 __copy_skb_header(nskb, p);
2808 nskb->mac_len = p->mac_len;
2810 skb_reserve(nskb, headroom);
2811 __skb_put(nskb, skb_gro_offset(p));
2813 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2814 skb_set_network_header(nskb, skb_network_offset(p));
2815 skb_set_transport_header(nskb, skb_transport_offset(p));
2817 __skb_pull(p, skb_gro_offset(p));
2818 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2819 p->data - skb_mac_header(p));
2821 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2822 skb_shinfo(nskb)->frag_list = p;
2823 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2824 pinfo->gso_size = 0;
2825 skb_header_release(p);
2828 nskb->data_len += p->len;
2829 nskb->truesize += p->len;
2830 nskb->len += p->len;
2833 nskb->next = p->next;
2839 if (offset > headlen) {
2840 unsigned int eat = offset - headlen;
2842 skbinfo->frags[0].page_offset += eat;
2843 skbinfo->frags[0].size -= eat;
2844 skb->data_len -= eat;
2849 __skb_pull(skb, offset);
2851 p->prev->next = skb;
2853 skb_header_release(skb);
2856 NAPI_GRO_CB(p)->count++;
2861 NAPI_GRO_CB(skb)->same_flow = 1;
2864 EXPORT_SYMBOL_GPL(skb_gro_receive);
2866 void __init skb_init(void)
2868 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2869 sizeof(struct sk_buff),
2871 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2873 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2874 (2*sizeof(struct sk_buff)) +
2877 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2882 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2883 * @skb: Socket buffer containing the buffers to be mapped
2884 * @sg: The scatter-gather list to map into
2885 * @offset: The offset into the buffer's contents to start mapping
2886 * @len: Length of buffer space to be mapped
2888 * Fill the specified scatter-gather list with mappings/pointers into a
2889 * region of the buffer space attached to a socket buffer.
2892 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2894 int start = skb_headlen(skb);
2895 int i, copy = start - offset;
2896 struct sk_buff *frag_iter;
2902 sg_set_buf(sg, skb->data + offset, copy);
2904 if ((len -= copy) == 0)
2909 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2912 WARN_ON(start > offset + len);
2914 end = start + skb_shinfo(skb)->frags[i].size;
2915 if ((copy = end - offset) > 0) {
2916 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2920 sg_set_page(&sg[elt], frag->page, copy,
2921 frag->page_offset+offset-start);
2930 skb_walk_frags(skb, frag_iter) {
2933 WARN_ON(start > offset + len);
2935 end = start + frag_iter->len;
2936 if ((copy = end - offset) > 0) {
2939 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2941 if ((len -= copy) == 0)
2951 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2953 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2955 sg_mark_end(&sg[nsg - 1]);
2959 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2962 * skb_cow_data - Check that a socket buffer's data buffers are writable
2963 * @skb: The socket buffer to check.
2964 * @tailbits: Amount of trailing space to be added
2965 * @trailer: Returned pointer to the skb where the @tailbits space begins
2967 * Make sure that the data buffers attached to a socket buffer are
2968 * writable. If they are not, private copies are made of the data buffers
2969 * and the socket buffer is set to use these instead.
2971 * If @tailbits is given, make sure that there is space to write @tailbits
2972 * bytes of data beyond current end of socket buffer. @trailer will be
2973 * set to point to the skb in which this space begins.
2975 * The number of scatterlist elements required to completely map the
2976 * COW'd and extended socket buffer will be returned.
2978 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2982 struct sk_buff *skb1, **skb_p;
2984 /* If skb is cloned or its head is paged, reallocate
2985 * head pulling out all the pages (pages are considered not writable
2986 * at the moment even if they are anonymous).
2988 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2989 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2992 /* Easy case. Most of packets will go this way. */
2993 if (!skb_has_frag_list(skb)) {
2994 /* A little of trouble, not enough of space for trailer.
2995 * This should not happen, when stack is tuned to generate
2996 * good frames. OK, on miss we reallocate and reserve even more
2997 * space, 128 bytes is fair. */
2999 if (skb_tailroom(skb) < tailbits &&
3000 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3008 /* Misery. We are in troubles, going to mincer fragments... */
3011 skb_p = &skb_shinfo(skb)->frag_list;
3014 while ((skb1 = *skb_p) != NULL) {
3017 /* The fragment is partially pulled by someone,
3018 * this can happen on input. Copy it and everything
3021 if (skb_shared(skb1))
3024 /* If the skb is the last, worry about trailer. */
3026 if (skb1->next == NULL && tailbits) {
3027 if (skb_shinfo(skb1)->nr_frags ||
3028 skb_has_frag_list(skb1) ||
3029 skb_tailroom(skb1) < tailbits)
3030 ntail = tailbits + 128;
3036 skb_shinfo(skb1)->nr_frags ||
3037 skb_has_frag_list(skb1)) {
3038 struct sk_buff *skb2;
3040 /* Fuck, we are miserable poor guys... */
3042 skb2 = skb_copy(skb1, GFP_ATOMIC);
3044 skb2 = skb_copy_expand(skb1,
3048 if (unlikely(skb2 == NULL))
3052 skb_set_owner_w(skb2, skb1->sk);
3054 /* Looking around. Are we still alive?
3055 * OK, link new skb, drop old one */
3057 skb2->next = skb1->next;
3064 skb_p = &skb1->next;
3069 EXPORT_SYMBOL_GPL(skb_cow_data);
3071 static void sock_rmem_free(struct sk_buff *skb)
3073 struct sock *sk = skb->sk;
3075 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3079 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3081 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3083 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3084 (unsigned)sk->sk_rcvbuf)
3089 skb->destructor = sock_rmem_free;
3090 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3092 /* before exiting rcu section, make sure dst is refcounted */
3095 skb_queue_tail(&sk->sk_error_queue, skb);
3096 if (!sock_flag(sk, SOCK_DEAD))
3097 sk->sk_data_ready(sk, skb->len);
3100 EXPORT_SYMBOL(sock_queue_err_skb);
3102 void skb_tstamp_tx(struct sk_buff *orig_skb,
3103 struct skb_shared_hwtstamps *hwtstamps)
3105 struct sock *sk = orig_skb->sk;
3106 struct sock_exterr_skb *serr;
3107 struct sk_buff *skb;
3113 skb = skb_clone(orig_skb, GFP_ATOMIC);
3118 *skb_hwtstamps(skb) =
3122 * no hardware time stamps available,
3123 * so keep the shared tx_flags and only
3124 * store software time stamp
3126 skb->tstamp = ktime_get_real();
3129 serr = SKB_EXT_ERR(skb);
3130 memset(serr, 0, sizeof(*serr));
3131 serr->ee.ee_errno = ENOMSG;
3132 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3134 err = sock_queue_err_skb(sk, skb);
3139 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3143 * skb_partial_csum_set - set up and verify partial csum values for packet
3144 * @skb: the skb to set
3145 * @start: the number of bytes after skb->data to start checksumming.
3146 * @off: the offset from start to place the checksum.
3148 * For untrusted partially-checksummed packets, we need to make sure the values
3149 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3151 * This function checks and sets those values and skb->ip_summed: if this
3152 * returns false you should drop the packet.
3154 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3156 if (unlikely(start > skb_headlen(skb)) ||
3157 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3158 if (net_ratelimit())
3160 "bad partial csum: csum=%u/%u len=%u\n",
3161 start, off, skb_headlen(skb));
3164 skb->ip_summed = CHECKSUM_PARTIAL;
3165 skb->csum_start = skb_headroom(skb) + start;
3166 skb->csum_offset = off;
3169 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3171 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3173 if (net_ratelimit())
3174 pr_warning("%s: received packets cannot be forwarded"
3175 " while LRO is enabled\n", skb->dev->name);
3177 EXPORT_SYMBOL(__skb_warn_lro_forwarding);