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_copy_ubufs - copy userspace skb frags buffers to kernel
615 * @skb: the skb to modify
616 * @gfp_mask: allocation priority
618 * This must be called on SKBTX_DEV_ZEROCOPY skb.
619 * It will copy all frags into kernel and drop the reference
620 * to userspace pages.
622 * If this function is called from an interrupt gfp_mask() must be
625 * Returns 0 on success or a negative error code on failure
626 * to allocate kernel memory to copy to.
628 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
631 int num_frags = skb_shinfo(skb)->nr_frags;
632 struct page *page, *head = NULL;
633 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
635 for (i = 0; i < num_frags; i++) {
637 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
639 page = alloc_page(GFP_ATOMIC);
642 struct page *next = (struct page *)head->private;
648 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
649 memcpy(page_address(page),
650 vaddr + f->page_offset, f->size);
651 kunmap_skb_frag(vaddr);
652 page->private = (unsigned long)head;
656 /* skb frags release userspace buffers */
657 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
658 put_page(skb_shinfo(skb)->frags[i].page);
660 uarg->callback(uarg);
662 /* skb frags point to kernel buffers */
663 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
664 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
665 skb_shinfo(skb)->frags[i - 1].page = head;
666 head = (struct page *)head->private;
669 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
675 * skb_clone - duplicate an sk_buff
676 * @skb: buffer to clone
677 * @gfp_mask: allocation priority
679 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
680 * copies share the same packet data but not structure. The new
681 * buffer has a reference count of 1. If the allocation fails the
682 * function returns %NULL otherwise the new buffer is returned.
684 * If this function is called from an interrupt gfp_mask() must be
688 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
692 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
693 if (skb_copy_ubufs(skb, gfp_mask))
698 if (skb->fclone == SKB_FCLONE_ORIG &&
699 n->fclone == SKB_FCLONE_UNAVAILABLE) {
700 atomic_t *fclone_ref = (atomic_t *) (n + 1);
701 n->fclone = SKB_FCLONE_CLONE;
702 atomic_inc(fclone_ref);
704 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
708 kmemcheck_annotate_bitfield(n, flags1);
709 kmemcheck_annotate_bitfield(n, flags2);
710 n->fclone = SKB_FCLONE_UNAVAILABLE;
713 return __skb_clone(n, skb);
715 EXPORT_SYMBOL(skb_clone);
717 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
719 #ifndef NET_SKBUFF_DATA_USES_OFFSET
721 * Shift between the two data areas in bytes
723 unsigned long offset = new->data - old->data;
726 __copy_skb_header(new, old);
728 #ifndef NET_SKBUFF_DATA_USES_OFFSET
729 /* {transport,network,mac}_header are relative to skb->head */
730 new->transport_header += offset;
731 new->network_header += offset;
732 if (skb_mac_header_was_set(new))
733 new->mac_header += offset;
735 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
736 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
737 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
741 * skb_copy - create private copy of an sk_buff
742 * @skb: buffer to copy
743 * @gfp_mask: allocation priority
745 * Make a copy of both an &sk_buff and its data. This is used when the
746 * caller wishes to modify the data and needs a private copy of the
747 * data to alter. Returns %NULL on failure or the pointer to the buffer
748 * on success. The returned buffer has a reference count of 1.
750 * As by-product this function converts non-linear &sk_buff to linear
751 * one, so that &sk_buff becomes completely private and caller is allowed
752 * to modify all the data of returned buffer. This means that this
753 * function is not recommended for use in circumstances when only
754 * header is going to be modified. Use pskb_copy() instead.
757 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
759 int headerlen = skb_headroom(skb);
760 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
761 struct sk_buff *n = alloc_skb(size, gfp_mask);
766 /* Set the data pointer */
767 skb_reserve(n, headerlen);
768 /* Set the tail pointer and length */
769 skb_put(n, skb->len);
771 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
774 copy_skb_header(n, skb);
777 EXPORT_SYMBOL(skb_copy);
780 * pskb_copy - create copy of an sk_buff with private head.
781 * @skb: buffer to copy
782 * @gfp_mask: allocation priority
784 * Make a copy of both an &sk_buff and part of its data, located
785 * in header. Fragmented data remain shared. This is used when
786 * the caller wishes to modify only header of &sk_buff and needs
787 * private copy of the header to alter. Returns %NULL on failure
788 * or the pointer to the buffer on success.
789 * The returned buffer has a reference count of 1.
792 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
794 unsigned int size = skb_end_pointer(skb) - skb->head;
795 struct sk_buff *n = alloc_skb(size, gfp_mask);
800 /* Set the data pointer */
801 skb_reserve(n, skb_headroom(skb));
802 /* Set the tail pointer and length */
803 skb_put(n, skb_headlen(skb));
805 skb_copy_from_linear_data(skb, n->data, n->len);
807 n->truesize += skb->data_len;
808 n->data_len = skb->data_len;
811 if (skb_shinfo(skb)->nr_frags) {
814 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
815 if (skb_copy_ubufs(skb, gfp_mask)) {
821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
822 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
823 get_page(skb_shinfo(n)->frags[i].page);
825 skb_shinfo(n)->nr_frags = i;
828 if (skb_has_frag_list(skb)) {
829 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
830 skb_clone_fraglist(n);
833 copy_skb_header(n, skb);
837 EXPORT_SYMBOL(pskb_copy);
840 * pskb_expand_head - reallocate header of &sk_buff
841 * @skb: buffer to reallocate
842 * @nhead: room to add at head
843 * @ntail: room to add at tail
844 * @gfp_mask: allocation priority
846 * Expands (or creates identical copy, if &nhead and &ntail are zero)
847 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
848 * reference count of 1. Returns zero in the case of success or error,
849 * if expansion failed. In the last case, &sk_buff is not changed.
851 * All the pointers pointing into skb header may change and must be
852 * reloaded after call to this function.
855 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
860 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
869 size = SKB_DATA_ALIGN(size);
871 /* Check if we can avoid taking references on fragments if we own
872 * the last reference on skb->head. (see skb_release_data())
877 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
878 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
882 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
883 memmove(skb->head + size, skb_shinfo(skb),
884 offsetof(struct skb_shared_info,
885 frags[skb_shinfo(skb)->nr_frags]));
886 memmove(skb->head + nhead, skb->head,
887 skb_tail_pointer(skb) - skb->head);
892 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
896 /* Copy only real data... and, alas, header. This should be
897 * optimized for the cases when header is void.
899 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
901 memcpy((struct skb_shared_info *)(data + size),
903 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
908 /* copy this zero copy skb frags */
909 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
910 if (skb_copy_ubufs(skb, gfp_mask))
913 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
914 get_page(skb_shinfo(skb)->frags[i].page);
916 if (skb_has_frag_list(skb))
917 skb_clone_fraglist(skb);
919 skb_release_data(skb);
921 off = (data + nhead) - skb->head;
926 #ifdef NET_SKBUFF_DATA_USES_OFFSET
930 skb->end = skb->head + size;
932 /* {transport,network,mac}_header and tail are relative to skb->head */
934 skb->transport_header += off;
935 skb->network_header += off;
936 if (skb_mac_header_was_set(skb))
937 skb->mac_header += off;
938 /* Only adjust this if it actually is csum_start rather than csum */
939 if (skb->ip_summed == CHECKSUM_PARTIAL)
940 skb->csum_start += nhead;
944 atomic_set(&skb_shinfo(skb)->dataref, 1);
952 EXPORT_SYMBOL(pskb_expand_head);
954 /* Make private copy of skb with writable head and some headroom */
956 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
958 struct sk_buff *skb2;
959 int delta = headroom - skb_headroom(skb);
962 skb2 = pskb_copy(skb, GFP_ATOMIC);
964 skb2 = skb_clone(skb, GFP_ATOMIC);
965 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
973 EXPORT_SYMBOL(skb_realloc_headroom);
976 * skb_copy_expand - copy and expand sk_buff
977 * @skb: buffer to copy
978 * @newheadroom: new free bytes at head
979 * @newtailroom: new free bytes at tail
980 * @gfp_mask: allocation priority
982 * Make a copy of both an &sk_buff and its data and while doing so
983 * allocate additional space.
985 * This is used when the caller wishes to modify the data and needs a
986 * private copy of the data to alter as well as more space for new fields.
987 * Returns %NULL on failure or the pointer to the buffer
988 * on success. The returned buffer has a reference count of 1.
990 * You must pass %GFP_ATOMIC as the allocation priority if this function
991 * is called from an interrupt.
993 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
994 int newheadroom, int newtailroom,
998 * Allocate the copy buffer
1000 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1002 int oldheadroom = skb_headroom(skb);
1003 int head_copy_len, head_copy_off;
1009 skb_reserve(n, newheadroom);
1011 /* Set the tail pointer and length */
1012 skb_put(n, skb->len);
1014 head_copy_len = oldheadroom;
1016 if (newheadroom <= head_copy_len)
1017 head_copy_len = newheadroom;
1019 head_copy_off = newheadroom - head_copy_len;
1021 /* Copy the linear header and data. */
1022 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1023 skb->len + head_copy_len))
1026 copy_skb_header(n, skb);
1028 off = newheadroom - oldheadroom;
1029 if (n->ip_summed == CHECKSUM_PARTIAL)
1030 n->csum_start += off;
1031 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1032 n->transport_header += off;
1033 n->network_header += off;
1034 if (skb_mac_header_was_set(skb))
1035 n->mac_header += off;
1040 EXPORT_SYMBOL(skb_copy_expand);
1043 * skb_pad - zero pad the tail of an skb
1044 * @skb: buffer to pad
1045 * @pad: space to pad
1047 * Ensure that a buffer is followed by a padding area that is zero
1048 * filled. Used by network drivers which may DMA or transfer data
1049 * beyond the buffer end onto the wire.
1051 * May return error in out of memory cases. The skb is freed on error.
1054 int skb_pad(struct sk_buff *skb, int pad)
1059 /* If the skbuff is non linear tailroom is always zero.. */
1060 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1061 memset(skb->data+skb->len, 0, pad);
1065 ntail = skb->data_len + pad - (skb->end - skb->tail);
1066 if (likely(skb_cloned(skb) || ntail > 0)) {
1067 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1072 /* FIXME: The use of this function with non-linear skb's really needs
1075 err = skb_linearize(skb);
1079 memset(skb->data + skb->len, 0, pad);
1086 EXPORT_SYMBOL(skb_pad);
1089 * skb_put - add data to a buffer
1090 * @skb: buffer to use
1091 * @len: amount of data to add
1093 * This function extends the used data area of the buffer. If this would
1094 * exceed the total buffer size the kernel will panic. A pointer to the
1095 * first byte of the extra data is returned.
1097 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1099 unsigned char *tmp = skb_tail_pointer(skb);
1100 SKB_LINEAR_ASSERT(skb);
1103 if (unlikely(skb->tail > skb->end))
1104 skb_over_panic(skb, len, __builtin_return_address(0));
1107 EXPORT_SYMBOL(skb_put);
1110 * skb_push - add data to the start of a buffer
1111 * @skb: buffer to use
1112 * @len: amount of data to add
1114 * This function extends the used data area of the buffer at the buffer
1115 * start. If this would exceed the total buffer headroom the kernel will
1116 * panic. A pointer to the first byte of the extra data is returned.
1118 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1122 if (unlikely(skb->data<skb->head))
1123 skb_under_panic(skb, len, __builtin_return_address(0));
1126 EXPORT_SYMBOL(skb_push);
1129 * skb_pull - remove data from the start of a buffer
1130 * @skb: buffer to use
1131 * @len: amount of data to remove
1133 * This function removes data from the start of a buffer, returning
1134 * the memory to the headroom. A pointer to the next data in the buffer
1135 * is returned. Once the data has been pulled future pushes will overwrite
1138 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1140 return skb_pull_inline(skb, len);
1142 EXPORT_SYMBOL(skb_pull);
1145 * skb_trim - remove end from a buffer
1146 * @skb: buffer to alter
1149 * Cut the length of a buffer down by removing data from the tail. If
1150 * the buffer is already under the length specified it is not modified.
1151 * The skb must be linear.
1153 void skb_trim(struct sk_buff *skb, unsigned int len)
1156 __skb_trim(skb, len);
1158 EXPORT_SYMBOL(skb_trim);
1160 /* Trims skb to length len. It can change skb pointers.
1163 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1165 struct sk_buff **fragp;
1166 struct sk_buff *frag;
1167 int offset = skb_headlen(skb);
1168 int nfrags = skb_shinfo(skb)->nr_frags;
1172 if (skb_cloned(skb) &&
1173 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1180 for (; i < nfrags; i++) {
1181 int end = offset + skb_shinfo(skb)->frags[i].size;
1188 skb_shinfo(skb)->frags[i++].size = len - offset;
1191 skb_shinfo(skb)->nr_frags = i;
1193 for (; i < nfrags; i++)
1194 put_page(skb_shinfo(skb)->frags[i].page);
1196 if (skb_has_frag_list(skb))
1197 skb_drop_fraglist(skb);
1201 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1202 fragp = &frag->next) {
1203 int end = offset + frag->len;
1205 if (skb_shared(frag)) {
1206 struct sk_buff *nfrag;
1208 nfrag = skb_clone(frag, GFP_ATOMIC);
1209 if (unlikely(!nfrag))
1212 nfrag->next = frag->next;
1224 unlikely((err = pskb_trim(frag, len - offset))))
1228 skb_drop_list(&frag->next);
1233 if (len > skb_headlen(skb)) {
1234 skb->data_len -= skb->len - len;
1239 skb_set_tail_pointer(skb, len);
1244 EXPORT_SYMBOL(___pskb_trim);
1247 * __pskb_pull_tail - advance tail of skb header
1248 * @skb: buffer to reallocate
1249 * @delta: number of bytes to advance tail
1251 * The function makes a sense only on a fragmented &sk_buff,
1252 * it expands header moving its tail forward and copying necessary
1253 * data from fragmented part.
1255 * &sk_buff MUST have reference count of 1.
1257 * Returns %NULL (and &sk_buff does not change) if pull failed
1258 * or value of new tail of skb in the case of success.
1260 * All the pointers pointing into skb header may change and must be
1261 * reloaded after call to this function.
1264 /* Moves tail of skb head forward, copying data from fragmented part,
1265 * when it is necessary.
1266 * 1. It may fail due to malloc failure.
1267 * 2. It may change skb pointers.
1269 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1271 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1273 /* If skb has not enough free space at tail, get new one
1274 * plus 128 bytes for future expansions. If we have enough
1275 * room at tail, reallocate without expansion only if skb is cloned.
1277 int i, k, eat = (skb->tail + delta) - skb->end;
1279 if (eat > 0 || skb_cloned(skb)) {
1280 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1285 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1288 /* Optimization: no fragments, no reasons to preestimate
1289 * size of pulled pages. Superb.
1291 if (!skb_has_frag_list(skb))
1294 /* Estimate size of pulled pages. */
1296 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1297 if (skb_shinfo(skb)->frags[i].size >= eat)
1299 eat -= skb_shinfo(skb)->frags[i].size;
1302 /* If we need update frag list, we are in troubles.
1303 * Certainly, it possible to add an offset to skb data,
1304 * but taking into account that pulling is expected to
1305 * be very rare operation, it is worth to fight against
1306 * further bloating skb head and crucify ourselves here instead.
1307 * Pure masohism, indeed. 8)8)
1310 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311 struct sk_buff *clone = NULL;
1312 struct sk_buff *insp = NULL;
1317 if (list->len <= eat) {
1318 /* Eaten as whole. */
1323 /* Eaten partially. */
1325 if (skb_shared(list)) {
1326 /* Sucks! We need to fork list. :-( */
1327 clone = skb_clone(list, GFP_ATOMIC);
1333 /* This may be pulled without
1337 if (!pskb_pull(list, eat)) {
1345 /* Free pulled out fragments. */
1346 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1347 skb_shinfo(skb)->frag_list = list->next;
1350 /* And insert new clone at head. */
1353 skb_shinfo(skb)->frag_list = clone;
1356 /* Success! Now we may commit changes to skb data. */
1361 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362 if (skb_shinfo(skb)->frags[i].size <= eat) {
1363 put_page(skb_shinfo(skb)->frags[i].page);
1364 eat -= skb_shinfo(skb)->frags[i].size;
1366 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1368 skb_shinfo(skb)->frags[k].page_offset += eat;
1369 skb_shinfo(skb)->frags[k].size -= eat;
1375 skb_shinfo(skb)->nr_frags = k;
1378 skb->data_len -= delta;
1380 return skb_tail_pointer(skb);
1382 EXPORT_SYMBOL(__pskb_pull_tail);
1385 * skb_copy_bits - copy bits from skb to kernel buffer
1387 * @offset: offset in source
1388 * @to: destination buffer
1389 * @len: number of bytes to copy
1391 * Copy the specified number of bytes from the source skb to the
1392 * destination buffer.
1395 * If its prototype is ever changed,
1396 * check arch/{*}/net/{*}.S files,
1397 * since it is called from BPF assembly code.
1399 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1401 int start = skb_headlen(skb);
1402 struct sk_buff *frag_iter;
1405 if (offset > (int)skb->len - len)
1409 if ((copy = start - offset) > 0) {
1412 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1413 if ((len -= copy) == 0)
1419 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1422 WARN_ON(start > offset + len);
1424 end = start + skb_shinfo(skb)->frags[i].size;
1425 if ((copy = end - offset) > 0) {
1431 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1433 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1434 offset - start, copy);
1435 kunmap_skb_frag(vaddr);
1437 if ((len -= copy) == 0)
1445 skb_walk_frags(skb, frag_iter) {
1448 WARN_ON(start > offset + len);
1450 end = start + frag_iter->len;
1451 if ((copy = end - offset) > 0) {
1454 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1456 if ((len -= copy) == 0)
1470 EXPORT_SYMBOL(skb_copy_bits);
1473 * Callback from splice_to_pipe(), if we need to release some pages
1474 * at the end of the spd in case we error'ed out in filling the pipe.
1476 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1478 put_page(spd->pages[i]);
1481 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1482 unsigned int *offset,
1483 struct sk_buff *skb, struct sock *sk)
1485 struct page *p = sk->sk_sndmsg_page;
1490 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1494 off = sk->sk_sndmsg_off = 0;
1495 /* hold one ref to this page until it's full */
1499 off = sk->sk_sndmsg_off;
1500 mlen = PAGE_SIZE - off;
1501 if (mlen < 64 && mlen < *len) {
1506 *len = min_t(unsigned int, *len, mlen);
1509 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1510 sk->sk_sndmsg_off += *len;
1518 * Fill page/offset/length into spd, if it can hold more pages.
1520 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1521 struct pipe_inode_info *pipe, struct page *page,
1522 unsigned int *len, unsigned int offset,
1523 struct sk_buff *skb, int linear,
1526 if (unlikely(spd->nr_pages == pipe->buffers))
1530 page = linear_to_page(page, len, &offset, skb, sk);
1536 spd->pages[spd->nr_pages] = page;
1537 spd->partial[spd->nr_pages].len = *len;
1538 spd->partial[spd->nr_pages].offset = offset;
1544 static inline void __segment_seek(struct page **page, unsigned int *poff,
1545 unsigned int *plen, unsigned int off)
1550 n = *poff / PAGE_SIZE;
1552 *page = nth_page(*page, n);
1554 *poff = *poff % PAGE_SIZE;
1558 static inline int __splice_segment(struct page *page, unsigned int poff,
1559 unsigned int plen, unsigned int *off,
1560 unsigned int *len, struct sk_buff *skb,
1561 struct splice_pipe_desc *spd, int linear,
1563 struct pipe_inode_info *pipe)
1568 /* skip this segment if already processed */
1574 /* ignore any bits we already processed */
1576 __segment_seek(&page, &poff, &plen, *off);
1581 unsigned int flen = min(*len, plen);
1583 /* the linear region may spread across several pages */
1584 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1586 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1589 __segment_seek(&page, &poff, &plen, flen);
1592 } while (*len && plen);
1598 * Map linear and fragment data from the skb to spd. It reports failure if the
1599 * pipe is full or if we already spliced the requested length.
1601 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1602 unsigned int *offset, unsigned int *len,
1603 struct splice_pipe_desc *spd, struct sock *sk)
1608 * map the linear part
1610 if (__splice_segment(virt_to_page(skb->data),
1611 (unsigned long) skb->data & (PAGE_SIZE - 1),
1613 offset, len, skb, spd, 1, sk, pipe))
1617 * then map the fragments
1619 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1620 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1622 if (__splice_segment(f->page, f->page_offset, f->size,
1623 offset, len, skb, spd, 0, sk, pipe))
1631 * Map data from the skb to a pipe. Should handle both the linear part,
1632 * the fragments, and the frag list. It does NOT handle frag lists within
1633 * the frag list, if such a thing exists. We'd probably need to recurse to
1634 * handle that cleanly.
1636 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1637 struct pipe_inode_info *pipe, unsigned int tlen,
1640 struct partial_page partial[PIPE_DEF_BUFFERS];
1641 struct page *pages[PIPE_DEF_BUFFERS];
1642 struct splice_pipe_desc spd = {
1646 .ops = &sock_pipe_buf_ops,
1647 .spd_release = sock_spd_release,
1649 struct sk_buff *frag_iter;
1650 struct sock *sk = skb->sk;
1653 if (splice_grow_spd(pipe, &spd))
1657 * __skb_splice_bits() only fails if the output has no room left,
1658 * so no point in going over the frag_list for the error case.
1660 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1666 * now see if we have a frag_list to map
1668 skb_walk_frags(skb, frag_iter) {
1671 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1678 * Drop the socket lock, otherwise we have reverse
1679 * locking dependencies between sk_lock and i_mutex
1680 * here as compared to sendfile(). We enter here
1681 * with the socket lock held, and splice_to_pipe() will
1682 * grab the pipe inode lock. For sendfile() emulation,
1683 * we call into ->sendpage() with the i_mutex lock held
1684 * and networking will grab the socket lock.
1687 ret = splice_to_pipe(pipe, &spd);
1691 splice_shrink_spd(pipe, &spd);
1696 * skb_store_bits - store bits from kernel buffer to skb
1697 * @skb: destination buffer
1698 * @offset: offset in destination
1699 * @from: source buffer
1700 * @len: number of bytes to copy
1702 * Copy the specified number of bytes from the source buffer to the
1703 * destination skb. This function handles all the messy bits of
1704 * traversing fragment lists and such.
1707 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1709 int start = skb_headlen(skb);
1710 struct sk_buff *frag_iter;
1713 if (offset > (int)skb->len - len)
1716 if ((copy = start - offset) > 0) {
1719 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1720 if ((len -= copy) == 0)
1726 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1730 WARN_ON(start > offset + len);
1732 end = start + frag->size;
1733 if ((copy = end - offset) > 0) {
1739 vaddr = kmap_skb_frag(frag);
1740 memcpy(vaddr + frag->page_offset + offset - start,
1742 kunmap_skb_frag(vaddr);
1744 if ((len -= copy) == 0)
1752 skb_walk_frags(skb, frag_iter) {
1755 WARN_ON(start > offset + len);
1757 end = start + frag_iter->len;
1758 if ((copy = end - offset) > 0) {
1761 if (skb_store_bits(frag_iter, offset - start,
1764 if ((len -= copy) == 0)
1777 EXPORT_SYMBOL(skb_store_bits);
1779 /* Checksum skb data. */
1781 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1782 int len, __wsum csum)
1784 int start = skb_headlen(skb);
1785 int i, copy = start - offset;
1786 struct sk_buff *frag_iter;
1789 /* Checksum header. */
1793 csum = csum_partial(skb->data + offset, copy, csum);
1794 if ((len -= copy) == 0)
1800 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1803 WARN_ON(start > offset + len);
1805 end = start + skb_shinfo(skb)->frags[i].size;
1806 if ((copy = end - offset) > 0) {
1809 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1813 vaddr = kmap_skb_frag(frag);
1814 csum2 = csum_partial(vaddr + frag->page_offset +
1815 offset - start, copy, 0);
1816 kunmap_skb_frag(vaddr);
1817 csum = csum_block_add(csum, csum2, pos);
1826 skb_walk_frags(skb, frag_iter) {
1829 WARN_ON(start > offset + len);
1831 end = start + frag_iter->len;
1832 if ((copy = end - offset) > 0) {
1836 csum2 = skb_checksum(frag_iter, offset - start,
1838 csum = csum_block_add(csum, csum2, pos);
1839 if ((len -= copy) == 0)
1850 EXPORT_SYMBOL(skb_checksum);
1852 /* Both of above in one bottle. */
1854 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1855 u8 *to, int len, __wsum csum)
1857 int start = skb_headlen(skb);
1858 int i, copy = start - offset;
1859 struct sk_buff *frag_iter;
1866 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1868 if ((len -= copy) == 0)
1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1878 WARN_ON(start > offset + len);
1880 end = start + skb_shinfo(skb)->frags[i].size;
1881 if ((copy = end - offset) > 0) {
1884 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1888 vaddr = kmap_skb_frag(frag);
1889 csum2 = csum_partial_copy_nocheck(vaddr +
1893 kunmap_skb_frag(vaddr);
1894 csum = csum_block_add(csum, csum2, pos);
1904 skb_walk_frags(skb, frag_iter) {
1908 WARN_ON(start > offset + len);
1910 end = start + frag_iter->len;
1911 if ((copy = end - offset) > 0) {
1914 csum2 = skb_copy_and_csum_bits(frag_iter,
1917 csum = csum_block_add(csum, csum2, pos);
1918 if ((len -= copy) == 0)
1929 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1931 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1936 if (skb->ip_summed == CHECKSUM_PARTIAL)
1937 csstart = skb_checksum_start_offset(skb);
1939 csstart = skb_headlen(skb);
1941 BUG_ON(csstart > skb_headlen(skb));
1943 skb_copy_from_linear_data(skb, to, csstart);
1946 if (csstart != skb->len)
1947 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1948 skb->len - csstart, 0);
1950 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1951 long csstuff = csstart + skb->csum_offset;
1953 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1956 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1959 * skb_dequeue - remove from the head of the queue
1960 * @list: list to dequeue from
1962 * Remove the head of the list. The list lock is taken so the function
1963 * may be used safely with other locking list functions. The head item is
1964 * returned or %NULL if the list is empty.
1967 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1969 unsigned long flags;
1970 struct sk_buff *result;
1972 spin_lock_irqsave(&list->lock, flags);
1973 result = __skb_dequeue(list);
1974 spin_unlock_irqrestore(&list->lock, flags);
1977 EXPORT_SYMBOL(skb_dequeue);
1980 * skb_dequeue_tail - remove from the tail of the queue
1981 * @list: list to dequeue from
1983 * Remove the tail of the list. The list lock is taken so the function
1984 * may be used safely with other locking list functions. The tail item is
1985 * returned or %NULL if the list is empty.
1987 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1989 unsigned long flags;
1990 struct sk_buff *result;
1992 spin_lock_irqsave(&list->lock, flags);
1993 result = __skb_dequeue_tail(list);
1994 spin_unlock_irqrestore(&list->lock, flags);
1997 EXPORT_SYMBOL(skb_dequeue_tail);
2000 * skb_queue_purge - empty a list
2001 * @list: list to empty
2003 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 * the list and one reference dropped. This function takes the list
2005 * lock and is atomic with respect to other list locking functions.
2007 void skb_queue_purge(struct sk_buff_head *list)
2009 struct sk_buff *skb;
2010 while ((skb = skb_dequeue(list)) != NULL)
2013 EXPORT_SYMBOL(skb_queue_purge);
2016 * skb_queue_head - queue a buffer at the list head
2017 * @list: list to use
2018 * @newsk: buffer to queue
2020 * Queue a buffer at the start of the list. This function takes the
2021 * list lock and can be used safely with other locking &sk_buff functions
2024 * A buffer cannot be placed on two lists at the same time.
2026 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2028 unsigned long flags;
2030 spin_lock_irqsave(&list->lock, flags);
2031 __skb_queue_head(list, newsk);
2032 spin_unlock_irqrestore(&list->lock, flags);
2034 EXPORT_SYMBOL(skb_queue_head);
2037 * skb_queue_tail - queue a buffer at the list tail
2038 * @list: list to use
2039 * @newsk: buffer to queue
2041 * Queue a buffer at the tail of the list. This function takes the
2042 * list lock and can be used safely with other locking &sk_buff functions
2045 * A buffer cannot be placed on two lists at the same time.
2047 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2049 unsigned long flags;
2051 spin_lock_irqsave(&list->lock, flags);
2052 __skb_queue_tail(list, newsk);
2053 spin_unlock_irqrestore(&list->lock, flags);
2055 EXPORT_SYMBOL(skb_queue_tail);
2058 * skb_unlink - remove a buffer from a list
2059 * @skb: buffer to remove
2060 * @list: list to use
2062 * Remove a packet from a list. The list locks are taken and this
2063 * function is atomic with respect to other list locked calls
2065 * You must know what list the SKB is on.
2067 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2069 unsigned long flags;
2071 spin_lock_irqsave(&list->lock, flags);
2072 __skb_unlink(skb, list);
2073 spin_unlock_irqrestore(&list->lock, flags);
2075 EXPORT_SYMBOL(skb_unlink);
2078 * skb_append - append a buffer
2079 * @old: buffer to insert after
2080 * @newsk: buffer to insert
2081 * @list: list to use
2083 * Place a packet after a given packet in a list. The list locks are taken
2084 * and this function is atomic with respect to other list locked calls.
2085 * A buffer cannot be placed on two lists at the same time.
2087 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2089 unsigned long flags;
2091 spin_lock_irqsave(&list->lock, flags);
2092 __skb_queue_after(list, old, newsk);
2093 spin_unlock_irqrestore(&list->lock, flags);
2095 EXPORT_SYMBOL(skb_append);
2098 * skb_insert - insert a buffer
2099 * @old: buffer to insert before
2100 * @newsk: buffer to insert
2101 * @list: list to use
2103 * Place a packet before a given packet in a list. The list locks are
2104 * taken and this function is atomic with respect to other list locked
2107 * A buffer cannot be placed on two lists at the same time.
2109 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2111 unsigned long flags;
2113 spin_lock_irqsave(&list->lock, flags);
2114 __skb_insert(newsk, old->prev, old, list);
2115 spin_unlock_irqrestore(&list->lock, flags);
2117 EXPORT_SYMBOL(skb_insert);
2119 static inline void skb_split_inside_header(struct sk_buff *skb,
2120 struct sk_buff* skb1,
2121 const u32 len, const int pos)
2125 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2127 /* And move data appendix as is. */
2128 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2129 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2131 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2132 skb_shinfo(skb)->nr_frags = 0;
2133 skb1->data_len = skb->data_len;
2134 skb1->len += skb1->data_len;
2137 skb_set_tail_pointer(skb, len);
2140 static inline void skb_split_no_header(struct sk_buff *skb,
2141 struct sk_buff* skb1,
2142 const u32 len, int pos)
2145 const int nfrags = skb_shinfo(skb)->nr_frags;
2147 skb_shinfo(skb)->nr_frags = 0;
2148 skb1->len = skb1->data_len = skb->len - len;
2150 skb->data_len = len - pos;
2152 for (i = 0; i < nfrags; i++) {
2153 int size = skb_shinfo(skb)->frags[i].size;
2155 if (pos + size > len) {
2156 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2160 * We have two variants in this case:
2161 * 1. Move all the frag to the second
2162 * part, if it is possible. F.e.
2163 * this approach is mandatory for TUX,
2164 * where splitting is expensive.
2165 * 2. Split is accurately. We make this.
2167 get_page(skb_shinfo(skb)->frags[i].page);
2168 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2169 skb_shinfo(skb1)->frags[0].size -= len - pos;
2170 skb_shinfo(skb)->frags[i].size = len - pos;
2171 skb_shinfo(skb)->nr_frags++;
2175 skb_shinfo(skb)->nr_frags++;
2178 skb_shinfo(skb1)->nr_frags = k;
2182 * skb_split - Split fragmented skb to two parts at length len.
2183 * @skb: the buffer to split
2184 * @skb1: the buffer to receive the second part
2185 * @len: new length for skb
2187 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2189 int pos = skb_headlen(skb);
2191 if (len < pos) /* Split line is inside header. */
2192 skb_split_inside_header(skb, skb1, len, pos);
2193 else /* Second chunk has no header, nothing to copy. */
2194 skb_split_no_header(skb, skb1, len, pos);
2196 EXPORT_SYMBOL(skb_split);
2198 /* Shifting from/to a cloned skb is a no-go.
2200 * Caller cannot keep skb_shinfo related pointers past calling here!
2202 static int skb_prepare_for_shift(struct sk_buff *skb)
2204 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2208 * skb_shift - Shifts paged data partially from skb to another
2209 * @tgt: buffer into which tail data gets added
2210 * @skb: buffer from which the paged data comes from
2211 * @shiftlen: shift up to this many bytes
2213 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2214 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2215 * It's up to caller to free skb if everything was shifted.
2217 * If @tgt runs out of frags, the whole operation is aborted.
2219 * Skb cannot include anything else but paged data while tgt is allowed
2220 * to have non-paged data as well.
2222 * TODO: full sized shift could be optimized but that would need
2223 * specialized skb free'er to handle frags without up-to-date nr_frags.
2225 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2227 int from, to, merge, todo;
2228 struct skb_frag_struct *fragfrom, *fragto;
2230 BUG_ON(shiftlen > skb->len);
2231 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2235 to = skb_shinfo(tgt)->nr_frags;
2236 fragfrom = &skb_shinfo(skb)->frags[from];
2238 /* Actual merge is delayed until the point when we know we can
2239 * commit all, so that we don't have to undo partial changes
2242 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2247 todo -= fragfrom->size;
2249 if (skb_prepare_for_shift(skb) ||
2250 skb_prepare_for_shift(tgt))
2253 /* All previous frag pointers might be stale! */
2254 fragfrom = &skb_shinfo(skb)->frags[from];
2255 fragto = &skb_shinfo(tgt)->frags[merge];
2257 fragto->size += shiftlen;
2258 fragfrom->size -= shiftlen;
2259 fragfrom->page_offset += shiftlen;
2267 /* Skip full, not-fitting skb to avoid expensive operations */
2268 if ((shiftlen == skb->len) &&
2269 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2272 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2275 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2276 if (to == MAX_SKB_FRAGS)
2279 fragfrom = &skb_shinfo(skb)->frags[from];
2280 fragto = &skb_shinfo(tgt)->frags[to];
2282 if (todo >= fragfrom->size) {
2283 *fragto = *fragfrom;
2284 todo -= fragfrom->size;
2289 get_page(fragfrom->page);
2290 fragto->page = fragfrom->page;
2291 fragto->page_offset = fragfrom->page_offset;
2292 fragto->size = todo;
2294 fragfrom->page_offset += todo;
2295 fragfrom->size -= todo;
2303 /* Ready to "commit" this state change to tgt */
2304 skb_shinfo(tgt)->nr_frags = to;
2307 fragfrom = &skb_shinfo(skb)->frags[0];
2308 fragto = &skb_shinfo(tgt)->frags[merge];
2310 fragto->size += fragfrom->size;
2311 put_page(fragfrom->page);
2314 /* Reposition in the original skb */
2316 while (from < skb_shinfo(skb)->nr_frags)
2317 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2318 skb_shinfo(skb)->nr_frags = to;
2320 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2323 /* Most likely the tgt won't ever need its checksum anymore, skb on
2324 * the other hand might need it if it needs to be resent
2326 tgt->ip_summed = CHECKSUM_PARTIAL;
2327 skb->ip_summed = CHECKSUM_PARTIAL;
2329 /* Yak, is it really working this way? Some helper please? */
2330 skb->len -= shiftlen;
2331 skb->data_len -= shiftlen;
2332 skb->truesize -= shiftlen;
2333 tgt->len += shiftlen;
2334 tgt->data_len += shiftlen;
2335 tgt->truesize += shiftlen;
2341 * skb_prepare_seq_read - Prepare a sequential read of skb data
2342 * @skb: the buffer to read
2343 * @from: lower offset of data to be read
2344 * @to: upper offset of data to be read
2345 * @st: state variable
2347 * Initializes the specified state variable. Must be called before
2348 * invoking skb_seq_read() for the first time.
2350 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2351 unsigned int to, struct skb_seq_state *st)
2353 st->lower_offset = from;
2354 st->upper_offset = to;
2355 st->root_skb = st->cur_skb = skb;
2356 st->frag_idx = st->stepped_offset = 0;
2357 st->frag_data = NULL;
2359 EXPORT_SYMBOL(skb_prepare_seq_read);
2362 * skb_seq_read - Sequentially read skb data
2363 * @consumed: number of bytes consumed by the caller so far
2364 * @data: destination pointer for data to be returned
2365 * @st: state variable
2367 * Reads a block of skb data at &consumed relative to the
2368 * lower offset specified to skb_prepare_seq_read(). Assigns
2369 * the head of the data block to &data and returns the length
2370 * of the block or 0 if the end of the skb data or the upper
2371 * offset has been reached.
2373 * The caller is not required to consume all of the data
2374 * returned, i.e. &consumed is typically set to the number
2375 * of bytes already consumed and the next call to
2376 * skb_seq_read() will return the remaining part of the block.
2378 * Note 1: The size of each block of data returned can be arbitrary,
2379 * this limitation is the cost for zerocopy seqeuental
2380 * reads of potentially non linear data.
2382 * Note 2: Fragment lists within fragments are not implemented
2383 * at the moment, state->root_skb could be replaced with
2384 * a stack for this purpose.
2386 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2387 struct skb_seq_state *st)
2389 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2392 if (unlikely(abs_offset >= st->upper_offset))
2396 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2398 if (abs_offset < block_limit && !st->frag_data) {
2399 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2400 return block_limit - abs_offset;
2403 if (st->frag_idx == 0 && !st->frag_data)
2404 st->stepped_offset += skb_headlen(st->cur_skb);
2406 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2407 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2408 block_limit = frag->size + st->stepped_offset;
2410 if (abs_offset < block_limit) {
2412 st->frag_data = kmap_skb_frag(frag);
2414 *data = (u8 *) st->frag_data + frag->page_offset +
2415 (abs_offset - st->stepped_offset);
2417 return block_limit - abs_offset;
2420 if (st->frag_data) {
2421 kunmap_skb_frag(st->frag_data);
2422 st->frag_data = NULL;
2426 st->stepped_offset += frag->size;
2429 if (st->frag_data) {
2430 kunmap_skb_frag(st->frag_data);
2431 st->frag_data = NULL;
2434 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2435 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2438 } else if (st->cur_skb->next) {
2439 st->cur_skb = st->cur_skb->next;
2446 EXPORT_SYMBOL(skb_seq_read);
2449 * skb_abort_seq_read - Abort a sequential read of skb data
2450 * @st: state variable
2452 * Must be called if skb_seq_read() was not called until it
2455 void skb_abort_seq_read(struct skb_seq_state *st)
2458 kunmap_skb_frag(st->frag_data);
2460 EXPORT_SYMBOL(skb_abort_seq_read);
2462 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2464 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2465 struct ts_config *conf,
2466 struct ts_state *state)
2468 return skb_seq_read(offset, text, TS_SKB_CB(state));
2471 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2473 skb_abort_seq_read(TS_SKB_CB(state));
2477 * skb_find_text - Find a text pattern in skb data
2478 * @skb: the buffer to look in
2479 * @from: search offset
2481 * @config: textsearch configuration
2482 * @state: uninitialized textsearch state variable
2484 * Finds a pattern in the skb data according to the specified
2485 * textsearch configuration. Use textsearch_next() to retrieve
2486 * subsequent occurrences of the pattern. Returns the offset
2487 * to the first occurrence or UINT_MAX if no match was found.
2489 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2490 unsigned int to, struct ts_config *config,
2491 struct ts_state *state)
2495 config->get_next_block = skb_ts_get_next_block;
2496 config->finish = skb_ts_finish;
2498 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2500 ret = textsearch_find(config, state);
2501 return (ret <= to - from ? ret : UINT_MAX);
2503 EXPORT_SYMBOL(skb_find_text);
2506 * skb_append_datato_frags: - append the user data to a skb
2507 * @sk: sock structure
2508 * @skb: skb structure to be appened with user data.
2509 * @getfrag: call back function to be used for getting the user data
2510 * @from: pointer to user message iov
2511 * @length: length of the iov message
2513 * Description: This procedure append the user data in the fragment part
2514 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2516 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2517 int (*getfrag)(void *from, char *to, int offset,
2518 int len, int odd, struct sk_buff *skb),
2519 void *from, int length)
2522 skb_frag_t *frag = NULL;
2523 struct page *page = NULL;
2529 /* Return error if we don't have space for new frag */
2530 frg_cnt = skb_shinfo(skb)->nr_frags;
2531 if (frg_cnt >= MAX_SKB_FRAGS)
2534 /* allocate a new page for next frag */
2535 page = alloc_pages(sk->sk_allocation, 0);
2537 /* If alloc_page fails just return failure and caller will
2538 * free previous allocated pages by doing kfree_skb()
2543 /* initialize the next frag */
2544 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2545 skb->truesize += PAGE_SIZE;
2546 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2548 /* get the new initialized frag */
2549 frg_cnt = skb_shinfo(skb)->nr_frags;
2550 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2552 /* copy the user data to page */
2553 left = PAGE_SIZE - frag->page_offset;
2554 copy = (length > left)? left : length;
2556 ret = getfrag(from, (page_address(frag->page) +
2557 frag->page_offset + frag->size),
2558 offset, copy, 0, skb);
2562 /* copy was successful so update the size parameters */
2565 skb->data_len += copy;
2569 } while (length > 0);
2573 EXPORT_SYMBOL(skb_append_datato_frags);
2576 * skb_pull_rcsum - pull skb and update receive checksum
2577 * @skb: buffer to update
2578 * @len: length of data pulled
2580 * This function performs an skb_pull on the packet and updates
2581 * the CHECKSUM_COMPLETE checksum. It should be used on
2582 * receive path processing instead of skb_pull unless you know
2583 * that the checksum difference is zero (e.g., a valid IP header)
2584 * or you are setting ip_summed to CHECKSUM_NONE.
2586 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2588 BUG_ON(len > skb->len);
2590 BUG_ON(skb->len < skb->data_len);
2591 skb_postpull_rcsum(skb, skb->data, len);
2592 return skb->data += len;
2594 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2597 * skb_segment - Perform protocol segmentation on skb.
2598 * @skb: buffer to segment
2599 * @features: features for the output path (see dev->features)
2601 * This function performs segmentation on the given skb. It returns
2602 * a pointer to the first in a list of new skbs for the segments.
2603 * In case of error it returns ERR_PTR(err).
2605 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2607 struct sk_buff *segs = NULL;
2608 struct sk_buff *tail = NULL;
2609 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2610 unsigned int mss = skb_shinfo(skb)->gso_size;
2611 unsigned int doffset = skb->data - skb_mac_header(skb);
2612 unsigned int offset = doffset;
2613 unsigned int headroom;
2615 int sg = !!(features & NETIF_F_SG);
2616 int nfrags = skb_shinfo(skb)->nr_frags;
2621 __skb_push(skb, doffset);
2622 headroom = skb_headroom(skb);
2623 pos = skb_headlen(skb);
2626 struct sk_buff *nskb;
2631 len = skb->len - offset;
2635 hsize = skb_headlen(skb) - offset;
2638 if (hsize > len || !sg)
2641 if (!hsize && i >= nfrags) {
2642 BUG_ON(fskb->len != len);
2645 nskb = skb_clone(fskb, GFP_ATOMIC);
2648 if (unlikely(!nskb))
2651 hsize = skb_end_pointer(nskb) - nskb->head;
2652 if (skb_cow_head(nskb, doffset + headroom)) {
2657 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2659 skb_release_head_state(nskb);
2660 __skb_push(nskb, doffset);
2662 nskb = alloc_skb(hsize + doffset + headroom,
2665 if (unlikely(!nskb))
2668 skb_reserve(nskb, headroom);
2669 __skb_put(nskb, doffset);
2678 __copy_skb_header(nskb, skb);
2679 nskb->mac_len = skb->mac_len;
2681 /* nskb and skb might have different headroom */
2682 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2683 nskb->csum_start += skb_headroom(nskb) - headroom;
2685 skb_reset_mac_header(nskb);
2686 skb_set_network_header(nskb, skb->mac_len);
2687 nskb->transport_header = (nskb->network_header +
2688 skb_network_header_len(skb));
2689 skb_copy_from_linear_data(skb, nskb->data, doffset);
2691 if (fskb != skb_shinfo(skb)->frag_list)
2695 nskb->ip_summed = CHECKSUM_NONE;
2696 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2702 frag = skb_shinfo(nskb)->frags;
2704 skb_copy_from_linear_data_offset(skb, offset,
2705 skb_put(nskb, hsize), hsize);
2707 while (pos < offset + len && i < nfrags) {
2708 *frag = skb_shinfo(skb)->frags[i];
2709 get_page(frag->page);
2713 frag->page_offset += offset - pos;
2714 frag->size -= offset - pos;
2717 skb_shinfo(nskb)->nr_frags++;
2719 if (pos + size <= offset + len) {
2723 frag->size -= pos + size - (offset + len);
2730 if (pos < offset + len) {
2731 struct sk_buff *fskb2 = fskb;
2733 BUG_ON(pos + fskb->len != offset + len);
2739 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2745 SKB_FRAG_ASSERT(nskb);
2746 skb_shinfo(nskb)->frag_list = fskb2;
2750 nskb->data_len = len - hsize;
2751 nskb->len += nskb->data_len;
2752 nskb->truesize += nskb->data_len;
2753 } while ((offset += len) < skb->len);
2758 while ((skb = segs)) {
2762 return ERR_PTR(err);
2764 EXPORT_SYMBOL_GPL(skb_segment);
2766 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2768 struct sk_buff *p = *head;
2769 struct sk_buff *nskb;
2770 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2771 struct skb_shared_info *pinfo = skb_shinfo(p);
2772 unsigned int headroom;
2773 unsigned int len = skb_gro_len(skb);
2774 unsigned int offset = skb_gro_offset(skb);
2775 unsigned int headlen = skb_headlen(skb);
2777 if (p->len + len >= 65536)
2780 if (pinfo->frag_list)
2782 else if (headlen <= offset) {
2785 int i = skbinfo->nr_frags;
2786 int nr_frags = pinfo->nr_frags + i;
2790 if (nr_frags > MAX_SKB_FRAGS)
2793 pinfo->nr_frags = nr_frags;
2794 skbinfo->nr_frags = 0;
2796 frag = pinfo->frags + nr_frags;
2797 frag2 = skbinfo->frags + i;
2802 frag->page_offset += offset;
2803 frag->size -= offset;
2805 skb->truesize -= skb->data_len;
2806 skb->len -= skb->data_len;
2809 NAPI_GRO_CB(skb)->free = 1;
2811 } else if (skb_gro_len(p) != pinfo->gso_size)
2814 headroom = skb_headroom(p);
2815 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2816 if (unlikely(!nskb))
2819 __copy_skb_header(nskb, p);
2820 nskb->mac_len = p->mac_len;
2822 skb_reserve(nskb, headroom);
2823 __skb_put(nskb, skb_gro_offset(p));
2825 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2826 skb_set_network_header(nskb, skb_network_offset(p));
2827 skb_set_transport_header(nskb, skb_transport_offset(p));
2829 __skb_pull(p, skb_gro_offset(p));
2830 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2831 p->data - skb_mac_header(p));
2833 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2834 skb_shinfo(nskb)->frag_list = p;
2835 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2836 pinfo->gso_size = 0;
2837 skb_header_release(p);
2840 nskb->data_len += p->len;
2841 nskb->truesize += p->len;
2842 nskb->len += p->len;
2845 nskb->next = p->next;
2851 if (offset > headlen) {
2852 unsigned int eat = offset - headlen;
2854 skbinfo->frags[0].page_offset += eat;
2855 skbinfo->frags[0].size -= eat;
2856 skb->data_len -= eat;
2861 __skb_pull(skb, offset);
2863 p->prev->next = skb;
2865 skb_header_release(skb);
2868 NAPI_GRO_CB(p)->count++;
2873 NAPI_GRO_CB(skb)->same_flow = 1;
2876 EXPORT_SYMBOL_GPL(skb_gro_receive);
2878 void __init skb_init(void)
2880 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2881 sizeof(struct sk_buff),
2883 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2885 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2886 (2*sizeof(struct sk_buff)) +
2889 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2894 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2895 * @skb: Socket buffer containing the buffers to be mapped
2896 * @sg: The scatter-gather list to map into
2897 * @offset: The offset into the buffer's contents to start mapping
2898 * @len: Length of buffer space to be mapped
2900 * Fill the specified scatter-gather list with mappings/pointers into a
2901 * region of the buffer space attached to a socket buffer.
2904 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2906 int start = skb_headlen(skb);
2907 int i, copy = start - offset;
2908 struct sk_buff *frag_iter;
2914 sg_set_buf(sg, skb->data + offset, copy);
2916 if ((len -= copy) == 0)
2921 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2924 WARN_ON(start > offset + len);
2926 end = start + skb_shinfo(skb)->frags[i].size;
2927 if ((copy = end - offset) > 0) {
2928 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2932 sg_set_page(&sg[elt], frag->page, copy,
2933 frag->page_offset+offset-start);
2942 skb_walk_frags(skb, frag_iter) {
2945 WARN_ON(start > offset + len);
2947 end = start + frag_iter->len;
2948 if ((copy = end - offset) > 0) {
2951 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2953 if ((len -= copy) == 0)
2963 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2965 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2967 sg_mark_end(&sg[nsg - 1]);
2971 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2974 * skb_cow_data - Check that a socket buffer's data buffers are writable
2975 * @skb: The socket buffer to check.
2976 * @tailbits: Amount of trailing space to be added
2977 * @trailer: Returned pointer to the skb where the @tailbits space begins
2979 * Make sure that the data buffers attached to a socket buffer are
2980 * writable. If they are not, private copies are made of the data buffers
2981 * and the socket buffer is set to use these instead.
2983 * If @tailbits is given, make sure that there is space to write @tailbits
2984 * bytes of data beyond current end of socket buffer. @trailer will be
2985 * set to point to the skb in which this space begins.
2987 * The number of scatterlist elements required to completely map the
2988 * COW'd and extended socket buffer will be returned.
2990 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2994 struct sk_buff *skb1, **skb_p;
2996 /* If skb is cloned or its head is paged, reallocate
2997 * head pulling out all the pages (pages are considered not writable
2998 * at the moment even if they are anonymous).
3000 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3001 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3004 /* Easy case. Most of packets will go this way. */
3005 if (!skb_has_frag_list(skb)) {
3006 /* A little of trouble, not enough of space for trailer.
3007 * This should not happen, when stack is tuned to generate
3008 * good frames. OK, on miss we reallocate and reserve even more
3009 * space, 128 bytes is fair. */
3011 if (skb_tailroom(skb) < tailbits &&
3012 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3020 /* Misery. We are in troubles, going to mincer fragments... */
3023 skb_p = &skb_shinfo(skb)->frag_list;
3026 while ((skb1 = *skb_p) != NULL) {
3029 /* The fragment is partially pulled by someone,
3030 * this can happen on input. Copy it and everything
3033 if (skb_shared(skb1))
3036 /* If the skb is the last, worry about trailer. */
3038 if (skb1->next == NULL && tailbits) {
3039 if (skb_shinfo(skb1)->nr_frags ||
3040 skb_has_frag_list(skb1) ||
3041 skb_tailroom(skb1) < tailbits)
3042 ntail = tailbits + 128;
3048 skb_shinfo(skb1)->nr_frags ||
3049 skb_has_frag_list(skb1)) {
3050 struct sk_buff *skb2;
3052 /* Fuck, we are miserable poor guys... */
3054 skb2 = skb_copy(skb1, GFP_ATOMIC);
3056 skb2 = skb_copy_expand(skb1,
3060 if (unlikely(skb2 == NULL))
3064 skb_set_owner_w(skb2, skb1->sk);
3066 /* Looking around. Are we still alive?
3067 * OK, link new skb, drop old one */
3069 skb2->next = skb1->next;
3076 skb_p = &skb1->next;
3081 EXPORT_SYMBOL_GPL(skb_cow_data);
3083 static void sock_rmem_free(struct sk_buff *skb)
3085 struct sock *sk = skb->sk;
3087 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3091 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3093 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3095 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3096 (unsigned)sk->sk_rcvbuf)
3101 skb->destructor = sock_rmem_free;
3102 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3104 /* before exiting rcu section, make sure dst is refcounted */
3107 skb_queue_tail(&sk->sk_error_queue, skb);
3108 if (!sock_flag(sk, SOCK_DEAD))
3109 sk->sk_data_ready(sk, skb->len);
3112 EXPORT_SYMBOL(sock_queue_err_skb);
3114 void skb_tstamp_tx(struct sk_buff *orig_skb,
3115 struct skb_shared_hwtstamps *hwtstamps)
3117 struct sock *sk = orig_skb->sk;
3118 struct sock_exterr_skb *serr;
3119 struct sk_buff *skb;
3125 skb = skb_clone(orig_skb, GFP_ATOMIC);
3130 *skb_hwtstamps(skb) =
3134 * no hardware time stamps available,
3135 * so keep the shared tx_flags and only
3136 * store software time stamp
3138 skb->tstamp = ktime_get_real();
3141 serr = SKB_EXT_ERR(skb);
3142 memset(serr, 0, sizeof(*serr));
3143 serr->ee.ee_errno = ENOMSG;
3144 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3146 err = sock_queue_err_skb(sk, skb);
3151 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3155 * skb_partial_csum_set - set up and verify partial csum values for packet
3156 * @skb: the skb to set
3157 * @start: the number of bytes after skb->data to start checksumming.
3158 * @off: the offset from start to place the checksum.
3160 * For untrusted partially-checksummed packets, we need to make sure the values
3161 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3163 * This function checks and sets those values and skb->ip_summed: if this
3164 * returns false you should drop the packet.
3166 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3168 if (unlikely(start > skb_headlen(skb)) ||
3169 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3170 if (net_ratelimit())
3172 "bad partial csum: csum=%u/%u len=%u\n",
3173 start, off, skb_headlen(skb));
3176 skb->ip_summed = CHECKSUM_PARTIAL;
3177 skb->csum_start = skb_headroom(skb) + start;
3178 skb->csum_offset = off;
3181 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3183 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3185 if (net_ratelimit())
3186 pr_warning("%s: received packets cannot be forwarded"
3187 " while LRO is enabled\n", skb->dev->name);
3189 EXPORT_SYMBOL(__skb_warn_lro_forwarding);