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 /* We do our best to align skb_shared_info on a separate cache
188 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
189 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
190 * Both skb->head and skb_shared_info are cache line aligned.
192 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
193 data = kmalloc_node_track_caller(size, gfp_mask, node);
196 /* kmalloc(size) might give us more room than requested.
197 * Put skb_shared_info exactly at the end of allocated zone,
198 * to allow max possible filling before reallocation.
200 size = SKB_WITH_OVERHEAD(ksize(data));
201 prefetchw(data + size);
204 * Only clear those fields we need to clear, not those that we will
205 * actually initialise below. Hence, don't put any more fields after
206 * the tail pointer in struct sk_buff!
208 memset(skb, 0, offsetof(struct sk_buff, tail));
209 /* Account for allocated memory : skb + skb->head */
210 skb->truesize = SKB_TRUESIZE(size);
211 atomic_set(&skb->users, 1);
214 skb_reset_tail_pointer(skb);
215 skb->end = skb->tail + size;
216 #ifdef NET_SKBUFF_DATA_USES_OFFSET
217 skb->mac_header = ~0U;
220 /* make sure we initialize shinfo sequentially */
221 shinfo = skb_shinfo(skb);
222 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
223 atomic_set(&shinfo->dataref, 1);
224 kmemcheck_annotate_variable(shinfo->destructor_arg);
227 struct sk_buff *child = skb + 1;
228 atomic_t *fclone_ref = (atomic_t *) (child + 1);
230 kmemcheck_annotate_bitfield(child, flags1);
231 kmemcheck_annotate_bitfield(child, flags2);
232 skb->fclone = SKB_FCLONE_ORIG;
233 atomic_set(fclone_ref, 1);
235 child->fclone = SKB_FCLONE_UNAVAILABLE;
240 kmem_cache_free(cache, skb);
244 EXPORT_SYMBOL(__alloc_skb);
247 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
248 * @dev: network device to receive on
249 * @length: length to allocate
250 * @gfp_mask: get_free_pages mask, passed to alloc_skb
252 * Allocate a new &sk_buff and assign it a usage count of one. The
253 * buffer has unspecified headroom built in. Users should allocate
254 * the headroom they think they need without accounting for the
255 * built in space. The built in space is used for optimisations.
257 * %NULL is returned if there is no free memory.
259 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
260 unsigned int length, gfp_t gfp_mask)
264 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
266 skb_reserve(skb, NET_SKB_PAD);
271 EXPORT_SYMBOL(__netdev_alloc_skb);
273 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
276 skb_fill_page_desc(skb, i, page, off, size);
278 skb->data_len += size;
279 skb->truesize += size;
281 EXPORT_SYMBOL(skb_add_rx_frag);
284 * dev_alloc_skb - allocate an skbuff for receiving
285 * @length: length to allocate
287 * Allocate a new &sk_buff and assign it a usage count of one. The
288 * buffer has unspecified headroom built in. Users should allocate
289 * the headroom they think they need without accounting for the
290 * built in space. The built in space is used for optimisations.
292 * %NULL is returned if there is no free memory. Although this function
293 * allocates memory it can be called from an interrupt.
295 struct sk_buff *dev_alloc_skb(unsigned int length)
298 * There is more code here than it seems:
299 * __dev_alloc_skb is an inline
301 return __dev_alloc_skb(length, GFP_ATOMIC);
303 EXPORT_SYMBOL(dev_alloc_skb);
305 static void skb_drop_list(struct sk_buff **listp)
307 struct sk_buff *list = *listp;
312 struct sk_buff *this = list;
318 static inline void skb_drop_fraglist(struct sk_buff *skb)
320 skb_drop_list(&skb_shinfo(skb)->frag_list);
323 static void skb_clone_fraglist(struct sk_buff *skb)
325 struct sk_buff *list;
327 skb_walk_frags(skb, list)
331 static void skb_release_data(struct sk_buff *skb)
334 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
335 &skb_shinfo(skb)->dataref)) {
336 if (skb_shinfo(skb)->nr_frags) {
338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
339 skb_frag_unref(skb, i);
343 * If skb buf is from userspace, we need to notify the caller
344 * the lower device DMA has done;
346 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
347 struct ubuf_info *uarg;
349 uarg = skb_shinfo(skb)->destructor_arg;
351 uarg->callback(uarg);
354 if (skb_has_frag_list(skb))
355 skb_drop_fraglist(skb);
362 * Free an skbuff by memory without cleaning the state.
364 static void kfree_skbmem(struct sk_buff *skb)
366 struct sk_buff *other;
367 atomic_t *fclone_ref;
369 switch (skb->fclone) {
370 case SKB_FCLONE_UNAVAILABLE:
371 kmem_cache_free(skbuff_head_cache, skb);
374 case SKB_FCLONE_ORIG:
375 fclone_ref = (atomic_t *) (skb + 2);
376 if (atomic_dec_and_test(fclone_ref))
377 kmem_cache_free(skbuff_fclone_cache, skb);
380 case SKB_FCLONE_CLONE:
381 fclone_ref = (atomic_t *) (skb + 1);
384 /* The clone portion is available for
385 * fast-cloning again.
387 skb->fclone = SKB_FCLONE_UNAVAILABLE;
389 if (atomic_dec_and_test(fclone_ref))
390 kmem_cache_free(skbuff_fclone_cache, other);
395 static void skb_release_head_state(struct sk_buff *skb)
399 secpath_put(skb->sp);
401 if (skb->destructor) {
403 skb->destructor(skb);
405 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
406 nf_conntrack_put(skb->nfct);
408 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
409 nf_conntrack_put_reasm(skb->nfct_reasm);
411 #ifdef CONFIG_BRIDGE_NETFILTER
412 nf_bridge_put(skb->nf_bridge);
414 /* XXX: IS this still necessary? - JHS */
415 #ifdef CONFIG_NET_SCHED
417 #ifdef CONFIG_NET_CLS_ACT
423 /* Free everything but the sk_buff shell. */
424 static void skb_release_all(struct sk_buff *skb)
426 skb_release_head_state(skb);
427 skb_release_data(skb);
431 * __kfree_skb - private function
434 * Free an sk_buff. Release anything attached to the buffer.
435 * Clean the state. This is an internal helper function. Users should
436 * always call kfree_skb
439 void __kfree_skb(struct sk_buff *skb)
441 skb_release_all(skb);
444 EXPORT_SYMBOL(__kfree_skb);
447 * kfree_skb - free an sk_buff
448 * @skb: buffer to free
450 * Drop a reference to the buffer and free it if the usage count has
453 void kfree_skb(struct sk_buff *skb)
457 if (likely(atomic_read(&skb->users) == 1))
459 else if (likely(!atomic_dec_and_test(&skb->users)))
461 trace_kfree_skb(skb, __builtin_return_address(0));
464 EXPORT_SYMBOL(kfree_skb);
467 * consume_skb - free an skbuff
468 * @skb: buffer to free
470 * Drop a ref to the buffer and free it if the usage count has hit zero
471 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
472 * is being dropped after a failure and notes that
474 void consume_skb(struct sk_buff *skb)
478 if (likely(atomic_read(&skb->users) == 1))
480 else if (likely(!atomic_dec_and_test(&skb->users)))
482 trace_consume_skb(skb);
485 EXPORT_SYMBOL(consume_skb);
488 * skb_recycle_check - check if skb can be reused for receive
490 * @skb_size: minimum receive buffer size
492 * Checks that the skb passed in is not shared or cloned, and
493 * that it is linear and its head portion at least as large as
494 * skb_size so that it can be recycled as a receive buffer.
495 * If these conditions are met, this function does any necessary
496 * reference count dropping and cleans up the skbuff as if it
497 * just came from __alloc_skb().
499 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
501 struct skb_shared_info *shinfo;
506 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
509 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
512 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
513 if (skb_end_pointer(skb) - skb->head < skb_size)
516 if (skb_shared(skb) || skb_cloned(skb))
519 skb_release_head_state(skb);
521 shinfo = skb_shinfo(skb);
522 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
523 atomic_set(&shinfo->dataref, 1);
525 memset(skb, 0, offsetof(struct sk_buff, tail));
526 skb->data = skb->head + NET_SKB_PAD;
527 skb_reset_tail_pointer(skb);
531 EXPORT_SYMBOL(skb_recycle_check);
533 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
535 new->tstamp = old->tstamp;
537 new->transport_header = old->transport_header;
538 new->network_header = old->network_header;
539 new->mac_header = old->mac_header;
540 skb_dst_copy(new, old);
541 new->rxhash = old->rxhash;
542 new->ooo_okay = old->ooo_okay;
543 new->l4_rxhash = old->l4_rxhash;
545 new->sp = secpath_get(old->sp);
547 memcpy(new->cb, old->cb, sizeof(old->cb));
548 new->csum = old->csum;
549 new->local_df = old->local_df;
550 new->pkt_type = old->pkt_type;
551 new->ip_summed = old->ip_summed;
552 skb_copy_queue_mapping(new, old);
553 new->priority = old->priority;
554 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
555 new->ipvs_property = old->ipvs_property;
557 new->protocol = old->protocol;
558 new->mark = old->mark;
559 new->skb_iif = old->skb_iif;
561 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
562 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
563 new->nf_trace = old->nf_trace;
565 #ifdef CONFIG_NET_SCHED
566 new->tc_index = old->tc_index;
567 #ifdef CONFIG_NET_CLS_ACT
568 new->tc_verd = old->tc_verd;
571 new->vlan_tci = old->vlan_tci;
573 skb_copy_secmark(new, old);
577 * You should not add any new code to this function. Add it to
578 * __copy_skb_header above instead.
580 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
582 #define C(x) n->x = skb->x
584 n->next = n->prev = NULL;
586 __copy_skb_header(n, skb);
591 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
594 n->destructor = NULL;
600 atomic_set(&n->users, 1);
602 atomic_inc(&(skb_shinfo(skb)->dataref));
610 * skb_morph - morph one skb into another
611 * @dst: the skb to receive the contents
612 * @src: the skb to supply the contents
614 * This is identical to skb_clone except that the target skb is
615 * supplied by the user.
617 * The target skb is returned upon exit.
619 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
621 skb_release_all(dst);
622 return __skb_clone(dst, src);
624 EXPORT_SYMBOL_GPL(skb_morph);
626 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
627 * @skb: the skb to modify
628 * @gfp_mask: allocation priority
630 * This must be called on SKBTX_DEV_ZEROCOPY skb.
631 * It will copy all frags into kernel and drop the reference
632 * to userspace pages.
634 * If this function is called from an interrupt gfp_mask() must be
637 * Returns 0 on success or a negative error code on failure
638 * to allocate kernel memory to copy to.
640 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
643 int num_frags = skb_shinfo(skb)->nr_frags;
644 struct page *page, *head = NULL;
645 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
647 for (i = 0; i < num_frags; i++) {
649 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
651 page = alloc_page(GFP_ATOMIC);
654 struct page *next = (struct page *)head->private;
660 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
661 memcpy(page_address(page),
662 vaddr + f->page_offset, f->size);
663 kunmap_skb_frag(vaddr);
664 page->private = (unsigned long)head;
668 /* skb frags release userspace buffers */
669 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
670 put_page(skb_shinfo(skb)->frags[i].page);
672 uarg->callback(uarg);
674 /* skb frags point to kernel buffers */
675 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
676 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
677 skb_shinfo(skb)->frags[i - 1].page = head;
678 head = (struct page *)head->private;
681 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
687 * skb_clone - duplicate an sk_buff
688 * @skb: buffer to clone
689 * @gfp_mask: allocation priority
691 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
692 * copies share the same packet data but not structure. The new
693 * buffer has a reference count of 1. If the allocation fails the
694 * function returns %NULL otherwise the new buffer is returned.
696 * If this function is called from an interrupt gfp_mask() must be
700 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
704 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
705 if (skb_copy_ubufs(skb, gfp_mask))
710 if (skb->fclone == SKB_FCLONE_ORIG &&
711 n->fclone == SKB_FCLONE_UNAVAILABLE) {
712 atomic_t *fclone_ref = (atomic_t *) (n + 1);
713 n->fclone = SKB_FCLONE_CLONE;
714 atomic_inc(fclone_ref);
716 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
720 kmemcheck_annotate_bitfield(n, flags1);
721 kmemcheck_annotate_bitfield(n, flags2);
722 n->fclone = SKB_FCLONE_UNAVAILABLE;
725 return __skb_clone(n, skb);
727 EXPORT_SYMBOL(skb_clone);
729 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
731 #ifndef NET_SKBUFF_DATA_USES_OFFSET
733 * Shift between the two data areas in bytes
735 unsigned long offset = new->data - old->data;
738 __copy_skb_header(new, old);
740 #ifndef NET_SKBUFF_DATA_USES_OFFSET
741 /* {transport,network,mac}_header are relative to skb->head */
742 new->transport_header += offset;
743 new->network_header += offset;
744 if (skb_mac_header_was_set(new))
745 new->mac_header += offset;
747 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
748 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
749 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
753 * skb_copy - create private copy of an sk_buff
754 * @skb: buffer to copy
755 * @gfp_mask: allocation priority
757 * Make a copy of both an &sk_buff and its data. This is used when the
758 * caller wishes to modify the data and needs a private copy of the
759 * data to alter. Returns %NULL on failure or the pointer to the buffer
760 * on success. The returned buffer has a reference count of 1.
762 * As by-product this function converts non-linear &sk_buff to linear
763 * one, so that &sk_buff becomes completely private and caller is allowed
764 * to modify all the data of returned buffer. This means that this
765 * function is not recommended for use in circumstances when only
766 * header is going to be modified. Use pskb_copy() instead.
769 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
771 int headerlen = skb_headroom(skb);
772 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
773 struct sk_buff *n = alloc_skb(size, gfp_mask);
778 /* Set the data pointer */
779 skb_reserve(n, headerlen);
780 /* Set the tail pointer and length */
781 skb_put(n, skb->len);
783 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
786 copy_skb_header(n, skb);
789 EXPORT_SYMBOL(skb_copy);
792 * pskb_copy - create copy of an sk_buff with private head.
793 * @skb: buffer to copy
794 * @gfp_mask: allocation priority
796 * Make a copy of both an &sk_buff and part of its data, located
797 * in header. Fragmented data remain shared. This is used when
798 * the caller wishes to modify only header of &sk_buff and needs
799 * private copy of the header to alter. Returns %NULL on failure
800 * or the pointer to the buffer on success.
801 * The returned buffer has a reference count of 1.
804 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
806 unsigned int size = skb_end_pointer(skb) - skb->head;
807 struct sk_buff *n = alloc_skb(size, gfp_mask);
812 /* Set the data pointer */
813 skb_reserve(n, skb_headroom(skb));
814 /* Set the tail pointer and length */
815 skb_put(n, skb_headlen(skb));
817 skb_copy_from_linear_data(skb, n->data, n->len);
819 n->truesize += skb->data_len;
820 n->data_len = skb->data_len;
823 if (skb_shinfo(skb)->nr_frags) {
826 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
827 if (skb_copy_ubufs(skb, gfp_mask)) {
833 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
834 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
835 skb_frag_ref(skb, i);
837 skb_shinfo(n)->nr_frags = i;
840 if (skb_has_frag_list(skb)) {
841 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
842 skb_clone_fraglist(n);
845 copy_skb_header(n, skb);
849 EXPORT_SYMBOL(pskb_copy);
852 * pskb_expand_head - reallocate header of &sk_buff
853 * @skb: buffer to reallocate
854 * @nhead: room to add at head
855 * @ntail: room to add at tail
856 * @gfp_mask: allocation priority
858 * Expands (or creates identical copy, if &nhead and &ntail are zero)
859 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
860 * reference count of 1. Returns zero in the case of success or error,
861 * if expansion failed. In the last case, &sk_buff is not changed.
863 * All the pointers pointing into skb header may change and must be
864 * reloaded after call to this function.
867 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
872 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
881 size = SKB_DATA_ALIGN(size);
883 /* Check if we can avoid taking references on fragments if we own
884 * the last reference on skb->head. (see skb_release_data())
889 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
890 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
894 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
895 memmove(skb->head + size, skb_shinfo(skb),
896 offsetof(struct skb_shared_info,
897 frags[skb_shinfo(skb)->nr_frags]));
898 memmove(skb->head + nhead, skb->head,
899 skb_tail_pointer(skb) - skb->head);
904 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
908 /* Copy only real data... and, alas, header. This should be
909 * optimized for the cases when header is void.
911 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
913 memcpy((struct skb_shared_info *)(data + size),
915 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
920 /* copy this zero copy skb frags */
921 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
922 if (skb_copy_ubufs(skb, gfp_mask))
925 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
926 skb_frag_ref(skb, i);
928 if (skb_has_frag_list(skb))
929 skb_clone_fraglist(skb);
931 skb_release_data(skb);
933 off = (data + nhead) - skb->head;
938 #ifdef NET_SKBUFF_DATA_USES_OFFSET
942 skb->end = skb->head + size;
944 /* {transport,network,mac}_header and tail are relative to skb->head */
946 skb->transport_header += off;
947 skb->network_header += off;
948 if (skb_mac_header_was_set(skb))
949 skb->mac_header += off;
950 /* Only adjust this if it actually is csum_start rather than csum */
951 if (skb->ip_summed == CHECKSUM_PARTIAL)
952 skb->csum_start += nhead;
956 atomic_set(&skb_shinfo(skb)->dataref, 1);
964 EXPORT_SYMBOL(pskb_expand_head);
966 /* Make private copy of skb with writable head and some headroom */
968 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
970 struct sk_buff *skb2;
971 int delta = headroom - skb_headroom(skb);
974 skb2 = pskb_copy(skb, GFP_ATOMIC);
976 skb2 = skb_clone(skb, GFP_ATOMIC);
977 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
985 EXPORT_SYMBOL(skb_realloc_headroom);
988 * skb_copy_expand - copy and expand sk_buff
989 * @skb: buffer to copy
990 * @newheadroom: new free bytes at head
991 * @newtailroom: new free bytes at tail
992 * @gfp_mask: allocation priority
994 * Make a copy of both an &sk_buff and its data and while doing so
995 * allocate additional space.
997 * This is used when the caller wishes to modify the data and needs a
998 * private copy of the data to alter as well as more space for new fields.
999 * Returns %NULL on failure or the pointer to the buffer
1000 * on success. The returned buffer has a reference count of 1.
1002 * You must pass %GFP_ATOMIC as the allocation priority if this function
1003 * is called from an interrupt.
1005 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1006 int newheadroom, int newtailroom,
1010 * Allocate the copy buffer
1012 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1014 int oldheadroom = skb_headroom(skb);
1015 int head_copy_len, head_copy_off;
1021 skb_reserve(n, newheadroom);
1023 /* Set the tail pointer and length */
1024 skb_put(n, skb->len);
1026 head_copy_len = oldheadroom;
1028 if (newheadroom <= head_copy_len)
1029 head_copy_len = newheadroom;
1031 head_copy_off = newheadroom - head_copy_len;
1033 /* Copy the linear header and data. */
1034 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1035 skb->len + head_copy_len))
1038 copy_skb_header(n, skb);
1040 off = newheadroom - oldheadroom;
1041 if (n->ip_summed == CHECKSUM_PARTIAL)
1042 n->csum_start += off;
1043 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1044 n->transport_header += off;
1045 n->network_header += off;
1046 if (skb_mac_header_was_set(skb))
1047 n->mac_header += off;
1052 EXPORT_SYMBOL(skb_copy_expand);
1055 * skb_pad - zero pad the tail of an skb
1056 * @skb: buffer to pad
1057 * @pad: space to pad
1059 * Ensure that a buffer is followed by a padding area that is zero
1060 * filled. Used by network drivers which may DMA or transfer data
1061 * beyond the buffer end onto the wire.
1063 * May return error in out of memory cases. The skb is freed on error.
1066 int skb_pad(struct sk_buff *skb, int pad)
1071 /* If the skbuff is non linear tailroom is always zero.. */
1072 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1073 memset(skb->data+skb->len, 0, pad);
1077 ntail = skb->data_len + pad - (skb->end - skb->tail);
1078 if (likely(skb_cloned(skb) || ntail > 0)) {
1079 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1084 /* FIXME: The use of this function with non-linear skb's really needs
1087 err = skb_linearize(skb);
1091 memset(skb->data + skb->len, 0, pad);
1098 EXPORT_SYMBOL(skb_pad);
1101 * skb_put - add data to a buffer
1102 * @skb: buffer to use
1103 * @len: amount of data to add
1105 * This function extends the used data area of the buffer. If this would
1106 * exceed the total buffer size the kernel will panic. A pointer to the
1107 * first byte of the extra data is returned.
1109 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1111 unsigned char *tmp = skb_tail_pointer(skb);
1112 SKB_LINEAR_ASSERT(skb);
1115 if (unlikely(skb->tail > skb->end))
1116 skb_over_panic(skb, len, __builtin_return_address(0));
1119 EXPORT_SYMBOL(skb_put);
1122 * skb_push - add data to the start of a buffer
1123 * @skb: buffer to use
1124 * @len: amount of data to add
1126 * This function extends the used data area of the buffer at the buffer
1127 * start. If this would exceed the total buffer headroom the kernel will
1128 * panic. A pointer to the first byte of the extra data is returned.
1130 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1134 if (unlikely(skb->data<skb->head))
1135 skb_under_panic(skb, len, __builtin_return_address(0));
1138 EXPORT_SYMBOL(skb_push);
1141 * skb_pull - remove data from the start of a buffer
1142 * @skb: buffer to use
1143 * @len: amount of data to remove
1145 * This function removes data from the start of a buffer, returning
1146 * the memory to the headroom. A pointer to the next data in the buffer
1147 * is returned. Once the data has been pulled future pushes will overwrite
1150 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1152 return skb_pull_inline(skb, len);
1154 EXPORT_SYMBOL(skb_pull);
1157 * skb_trim - remove end from a buffer
1158 * @skb: buffer to alter
1161 * Cut the length of a buffer down by removing data from the tail. If
1162 * the buffer is already under the length specified it is not modified.
1163 * The skb must be linear.
1165 void skb_trim(struct sk_buff *skb, unsigned int len)
1168 __skb_trim(skb, len);
1170 EXPORT_SYMBOL(skb_trim);
1172 /* Trims skb to length len. It can change skb pointers.
1175 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1177 struct sk_buff **fragp;
1178 struct sk_buff *frag;
1179 int offset = skb_headlen(skb);
1180 int nfrags = skb_shinfo(skb)->nr_frags;
1184 if (skb_cloned(skb) &&
1185 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1192 for (; i < nfrags; i++) {
1193 int end = offset + skb_shinfo(skb)->frags[i].size;
1200 skb_shinfo(skb)->frags[i++].size = len - offset;
1203 skb_shinfo(skb)->nr_frags = i;
1205 for (; i < nfrags; i++)
1206 skb_frag_unref(skb, i);
1208 if (skb_has_frag_list(skb))
1209 skb_drop_fraglist(skb);
1213 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1214 fragp = &frag->next) {
1215 int end = offset + frag->len;
1217 if (skb_shared(frag)) {
1218 struct sk_buff *nfrag;
1220 nfrag = skb_clone(frag, GFP_ATOMIC);
1221 if (unlikely(!nfrag))
1224 nfrag->next = frag->next;
1236 unlikely((err = pskb_trim(frag, len - offset))))
1240 skb_drop_list(&frag->next);
1245 if (len > skb_headlen(skb)) {
1246 skb->data_len -= skb->len - len;
1251 skb_set_tail_pointer(skb, len);
1256 EXPORT_SYMBOL(___pskb_trim);
1259 * __pskb_pull_tail - advance tail of skb header
1260 * @skb: buffer to reallocate
1261 * @delta: number of bytes to advance tail
1263 * The function makes a sense only on a fragmented &sk_buff,
1264 * it expands header moving its tail forward and copying necessary
1265 * data from fragmented part.
1267 * &sk_buff MUST have reference count of 1.
1269 * Returns %NULL (and &sk_buff does not change) if pull failed
1270 * or value of new tail of skb in the case of success.
1272 * All the pointers pointing into skb header may change and must be
1273 * reloaded after call to this function.
1276 /* Moves tail of skb head forward, copying data from fragmented part,
1277 * when it is necessary.
1278 * 1. It may fail due to malloc failure.
1279 * 2. It may change skb pointers.
1281 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1283 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1285 /* If skb has not enough free space at tail, get new one
1286 * plus 128 bytes for future expansions. If we have enough
1287 * room at tail, reallocate without expansion only if skb is cloned.
1289 int i, k, eat = (skb->tail + delta) - skb->end;
1291 if (eat > 0 || skb_cloned(skb)) {
1292 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1297 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1300 /* Optimization: no fragments, no reasons to preestimate
1301 * size of pulled pages. Superb.
1303 if (!skb_has_frag_list(skb))
1306 /* Estimate size of pulled pages. */
1308 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1309 if (skb_shinfo(skb)->frags[i].size >= eat)
1311 eat -= skb_shinfo(skb)->frags[i].size;
1314 /* If we need update frag list, we are in troubles.
1315 * Certainly, it possible to add an offset to skb data,
1316 * but taking into account that pulling is expected to
1317 * be very rare operation, it is worth to fight against
1318 * further bloating skb head and crucify ourselves here instead.
1319 * Pure masohism, indeed. 8)8)
1322 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1323 struct sk_buff *clone = NULL;
1324 struct sk_buff *insp = NULL;
1329 if (list->len <= eat) {
1330 /* Eaten as whole. */
1335 /* Eaten partially. */
1337 if (skb_shared(list)) {
1338 /* Sucks! We need to fork list. :-( */
1339 clone = skb_clone(list, GFP_ATOMIC);
1345 /* This may be pulled without
1349 if (!pskb_pull(list, eat)) {
1357 /* Free pulled out fragments. */
1358 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1359 skb_shinfo(skb)->frag_list = list->next;
1362 /* And insert new clone at head. */
1365 skb_shinfo(skb)->frag_list = clone;
1368 /* Success! Now we may commit changes to skb data. */
1373 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1374 if (skb_shinfo(skb)->frags[i].size <= eat) {
1375 skb_frag_unref(skb, i);
1376 eat -= skb_shinfo(skb)->frags[i].size;
1378 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1380 skb_shinfo(skb)->frags[k].page_offset += eat;
1381 skb_shinfo(skb)->frags[k].size -= eat;
1387 skb_shinfo(skb)->nr_frags = k;
1390 skb->data_len -= delta;
1392 return skb_tail_pointer(skb);
1394 EXPORT_SYMBOL(__pskb_pull_tail);
1397 * skb_copy_bits - copy bits from skb to kernel buffer
1399 * @offset: offset in source
1400 * @to: destination buffer
1401 * @len: number of bytes to copy
1403 * Copy the specified number of bytes from the source skb to the
1404 * destination buffer.
1407 * If its prototype is ever changed,
1408 * check arch/{*}/net/{*}.S files,
1409 * since it is called from BPF assembly code.
1411 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1413 int start = skb_headlen(skb);
1414 struct sk_buff *frag_iter;
1417 if (offset > (int)skb->len - len)
1421 if ((copy = start - offset) > 0) {
1424 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1425 if ((len -= copy) == 0)
1431 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1434 WARN_ON(start > offset + len);
1436 end = start + skb_shinfo(skb)->frags[i].size;
1437 if ((copy = end - offset) > 0) {
1443 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1445 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1446 offset - start, copy);
1447 kunmap_skb_frag(vaddr);
1449 if ((len -= copy) == 0)
1457 skb_walk_frags(skb, frag_iter) {
1460 WARN_ON(start > offset + len);
1462 end = start + frag_iter->len;
1463 if ((copy = end - offset) > 0) {
1466 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1468 if ((len -= copy) == 0)
1482 EXPORT_SYMBOL(skb_copy_bits);
1485 * Callback from splice_to_pipe(), if we need to release some pages
1486 * at the end of the spd in case we error'ed out in filling the pipe.
1488 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1490 put_page(spd->pages[i]);
1493 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1494 unsigned int *offset,
1495 struct sk_buff *skb, struct sock *sk)
1497 struct page *p = sk->sk_sndmsg_page;
1502 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1506 off = sk->sk_sndmsg_off = 0;
1507 /* hold one ref to this page until it's full */
1511 off = sk->sk_sndmsg_off;
1512 mlen = PAGE_SIZE - off;
1513 if (mlen < 64 && mlen < *len) {
1518 *len = min_t(unsigned int, *len, mlen);
1521 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1522 sk->sk_sndmsg_off += *len;
1530 * Fill page/offset/length into spd, if it can hold more pages.
1532 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1533 struct pipe_inode_info *pipe, struct page *page,
1534 unsigned int *len, unsigned int offset,
1535 struct sk_buff *skb, int linear,
1538 if (unlikely(spd->nr_pages == pipe->buffers))
1542 page = linear_to_page(page, len, &offset, skb, sk);
1548 spd->pages[spd->nr_pages] = page;
1549 spd->partial[spd->nr_pages].len = *len;
1550 spd->partial[spd->nr_pages].offset = offset;
1556 static inline void __segment_seek(struct page **page, unsigned int *poff,
1557 unsigned int *plen, unsigned int off)
1562 n = *poff / PAGE_SIZE;
1564 *page = nth_page(*page, n);
1566 *poff = *poff % PAGE_SIZE;
1570 static inline int __splice_segment(struct page *page, unsigned int poff,
1571 unsigned int plen, unsigned int *off,
1572 unsigned int *len, struct sk_buff *skb,
1573 struct splice_pipe_desc *spd, int linear,
1575 struct pipe_inode_info *pipe)
1580 /* skip this segment if already processed */
1586 /* ignore any bits we already processed */
1588 __segment_seek(&page, &poff, &plen, *off);
1593 unsigned int flen = min(*len, plen);
1595 /* the linear region may spread across several pages */
1596 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1598 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1601 __segment_seek(&page, &poff, &plen, flen);
1604 } while (*len && plen);
1610 * Map linear and fragment data from the skb to spd. It reports failure if the
1611 * pipe is full or if we already spliced the requested length.
1613 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1614 unsigned int *offset, unsigned int *len,
1615 struct splice_pipe_desc *spd, struct sock *sk)
1620 * map the linear part
1622 if (__splice_segment(virt_to_page(skb->data),
1623 (unsigned long) skb->data & (PAGE_SIZE - 1),
1625 offset, len, skb, spd, 1, sk, pipe))
1629 * then map the fragments
1631 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1632 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1634 if (__splice_segment(skb_frag_page(f),
1635 f->page_offset, f->size,
1636 offset, len, skb, spd, 0, sk, pipe))
1644 * Map data from the skb to a pipe. Should handle both the linear part,
1645 * the fragments, and the frag list. It does NOT handle frag lists within
1646 * the frag list, if such a thing exists. We'd probably need to recurse to
1647 * handle that cleanly.
1649 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1650 struct pipe_inode_info *pipe, unsigned int tlen,
1653 struct partial_page partial[PIPE_DEF_BUFFERS];
1654 struct page *pages[PIPE_DEF_BUFFERS];
1655 struct splice_pipe_desc spd = {
1659 .ops = &sock_pipe_buf_ops,
1660 .spd_release = sock_spd_release,
1662 struct sk_buff *frag_iter;
1663 struct sock *sk = skb->sk;
1666 if (splice_grow_spd(pipe, &spd))
1670 * __skb_splice_bits() only fails if the output has no room left,
1671 * so no point in going over the frag_list for the error case.
1673 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1679 * now see if we have a frag_list to map
1681 skb_walk_frags(skb, frag_iter) {
1684 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1691 * Drop the socket lock, otherwise we have reverse
1692 * locking dependencies between sk_lock and i_mutex
1693 * here as compared to sendfile(). We enter here
1694 * with the socket lock held, and splice_to_pipe() will
1695 * grab the pipe inode lock. For sendfile() emulation,
1696 * we call into ->sendpage() with the i_mutex lock held
1697 * and networking will grab the socket lock.
1700 ret = splice_to_pipe(pipe, &spd);
1704 splice_shrink_spd(pipe, &spd);
1709 * skb_store_bits - store bits from kernel buffer to skb
1710 * @skb: destination buffer
1711 * @offset: offset in destination
1712 * @from: source buffer
1713 * @len: number of bytes to copy
1715 * Copy the specified number of bytes from the source buffer to the
1716 * destination skb. This function handles all the messy bits of
1717 * traversing fragment lists and such.
1720 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1722 int start = skb_headlen(skb);
1723 struct sk_buff *frag_iter;
1726 if (offset > (int)skb->len - len)
1729 if ((copy = start - offset) > 0) {
1732 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1733 if ((len -= copy) == 0)
1739 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1740 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1743 WARN_ON(start > offset + len);
1745 end = start + frag->size;
1746 if ((copy = end - offset) > 0) {
1752 vaddr = kmap_skb_frag(frag);
1753 memcpy(vaddr + frag->page_offset + offset - start,
1755 kunmap_skb_frag(vaddr);
1757 if ((len -= copy) == 0)
1765 skb_walk_frags(skb, frag_iter) {
1768 WARN_ON(start > offset + len);
1770 end = start + frag_iter->len;
1771 if ((copy = end - offset) > 0) {
1774 if (skb_store_bits(frag_iter, offset - start,
1777 if ((len -= copy) == 0)
1790 EXPORT_SYMBOL(skb_store_bits);
1792 /* Checksum skb data. */
1794 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1795 int len, __wsum csum)
1797 int start = skb_headlen(skb);
1798 int i, copy = start - offset;
1799 struct sk_buff *frag_iter;
1802 /* Checksum header. */
1806 csum = csum_partial(skb->data + offset, copy, csum);
1807 if ((len -= copy) == 0)
1813 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1816 WARN_ON(start > offset + len);
1818 end = start + skb_shinfo(skb)->frags[i].size;
1819 if ((copy = end - offset) > 0) {
1822 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1826 vaddr = kmap_skb_frag(frag);
1827 csum2 = csum_partial(vaddr + frag->page_offset +
1828 offset - start, copy, 0);
1829 kunmap_skb_frag(vaddr);
1830 csum = csum_block_add(csum, csum2, pos);
1839 skb_walk_frags(skb, frag_iter) {
1842 WARN_ON(start > offset + len);
1844 end = start + frag_iter->len;
1845 if ((copy = end - offset) > 0) {
1849 csum2 = skb_checksum(frag_iter, offset - start,
1851 csum = csum_block_add(csum, csum2, pos);
1852 if ((len -= copy) == 0)
1863 EXPORT_SYMBOL(skb_checksum);
1865 /* Both of above in one bottle. */
1867 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1868 u8 *to, int len, __wsum csum)
1870 int start = skb_headlen(skb);
1871 int i, copy = start - offset;
1872 struct sk_buff *frag_iter;
1879 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1881 if ((len -= copy) == 0)
1888 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1891 WARN_ON(start > offset + len);
1893 end = start + skb_shinfo(skb)->frags[i].size;
1894 if ((copy = end - offset) > 0) {
1897 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1901 vaddr = kmap_skb_frag(frag);
1902 csum2 = csum_partial_copy_nocheck(vaddr +
1906 kunmap_skb_frag(vaddr);
1907 csum = csum_block_add(csum, csum2, pos);
1917 skb_walk_frags(skb, frag_iter) {
1921 WARN_ON(start > offset + len);
1923 end = start + frag_iter->len;
1924 if ((copy = end - offset) > 0) {
1927 csum2 = skb_copy_and_csum_bits(frag_iter,
1930 csum = csum_block_add(csum, csum2, pos);
1931 if ((len -= copy) == 0)
1942 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1944 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1949 if (skb->ip_summed == CHECKSUM_PARTIAL)
1950 csstart = skb_checksum_start_offset(skb);
1952 csstart = skb_headlen(skb);
1954 BUG_ON(csstart > skb_headlen(skb));
1956 skb_copy_from_linear_data(skb, to, csstart);
1959 if (csstart != skb->len)
1960 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1961 skb->len - csstart, 0);
1963 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1964 long csstuff = csstart + skb->csum_offset;
1966 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1969 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1972 * skb_dequeue - remove from the head of the queue
1973 * @list: list to dequeue from
1975 * Remove the head of the list. The list lock is taken so the function
1976 * may be used safely with other locking list functions. The head item is
1977 * returned or %NULL if the list is empty.
1980 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1982 unsigned long flags;
1983 struct sk_buff *result;
1985 spin_lock_irqsave(&list->lock, flags);
1986 result = __skb_dequeue(list);
1987 spin_unlock_irqrestore(&list->lock, flags);
1990 EXPORT_SYMBOL(skb_dequeue);
1993 * skb_dequeue_tail - remove from the tail of the queue
1994 * @list: list to dequeue from
1996 * Remove the tail of the list. The list lock is taken so the function
1997 * may be used safely with other locking list functions. The tail item is
1998 * returned or %NULL if the list is empty.
2000 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2002 unsigned long flags;
2003 struct sk_buff *result;
2005 spin_lock_irqsave(&list->lock, flags);
2006 result = __skb_dequeue_tail(list);
2007 spin_unlock_irqrestore(&list->lock, flags);
2010 EXPORT_SYMBOL(skb_dequeue_tail);
2013 * skb_queue_purge - empty a list
2014 * @list: list to empty
2016 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2017 * the list and one reference dropped. This function takes the list
2018 * lock and is atomic with respect to other list locking functions.
2020 void skb_queue_purge(struct sk_buff_head *list)
2022 struct sk_buff *skb;
2023 while ((skb = skb_dequeue(list)) != NULL)
2026 EXPORT_SYMBOL(skb_queue_purge);
2029 * skb_queue_head - queue a buffer at the list head
2030 * @list: list to use
2031 * @newsk: buffer to queue
2033 * Queue a buffer at the start of the list. This function takes the
2034 * list lock and can be used safely with other locking &sk_buff functions
2037 * A buffer cannot be placed on two lists at the same time.
2039 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2041 unsigned long flags;
2043 spin_lock_irqsave(&list->lock, flags);
2044 __skb_queue_head(list, newsk);
2045 spin_unlock_irqrestore(&list->lock, flags);
2047 EXPORT_SYMBOL(skb_queue_head);
2050 * skb_queue_tail - queue a buffer at the list tail
2051 * @list: list to use
2052 * @newsk: buffer to queue
2054 * Queue a buffer at the tail of the list. This function takes the
2055 * list lock and can be used safely with other locking &sk_buff functions
2058 * A buffer cannot be placed on two lists at the same time.
2060 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2062 unsigned long flags;
2064 spin_lock_irqsave(&list->lock, flags);
2065 __skb_queue_tail(list, newsk);
2066 spin_unlock_irqrestore(&list->lock, flags);
2068 EXPORT_SYMBOL(skb_queue_tail);
2071 * skb_unlink - remove a buffer from a list
2072 * @skb: buffer to remove
2073 * @list: list to use
2075 * Remove a packet from a list. The list locks are taken and this
2076 * function is atomic with respect to other list locked calls
2078 * You must know what list the SKB is on.
2080 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2082 unsigned long flags;
2084 spin_lock_irqsave(&list->lock, flags);
2085 __skb_unlink(skb, list);
2086 spin_unlock_irqrestore(&list->lock, flags);
2088 EXPORT_SYMBOL(skb_unlink);
2091 * skb_append - append a buffer
2092 * @old: buffer to insert after
2093 * @newsk: buffer to insert
2094 * @list: list to use
2096 * Place a packet after a given packet in a list. The list locks are taken
2097 * and this function is atomic with respect to other list locked calls.
2098 * A buffer cannot be placed on two lists at the same time.
2100 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2102 unsigned long flags;
2104 spin_lock_irqsave(&list->lock, flags);
2105 __skb_queue_after(list, old, newsk);
2106 spin_unlock_irqrestore(&list->lock, flags);
2108 EXPORT_SYMBOL(skb_append);
2111 * skb_insert - insert a buffer
2112 * @old: buffer to insert before
2113 * @newsk: buffer to insert
2114 * @list: list to use
2116 * Place a packet before a given packet in a list. The list locks are
2117 * taken and this function is atomic with respect to other list locked
2120 * A buffer cannot be placed on two lists at the same time.
2122 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2124 unsigned long flags;
2126 spin_lock_irqsave(&list->lock, flags);
2127 __skb_insert(newsk, old->prev, old, list);
2128 spin_unlock_irqrestore(&list->lock, flags);
2130 EXPORT_SYMBOL(skb_insert);
2132 static inline void skb_split_inside_header(struct sk_buff *skb,
2133 struct sk_buff* skb1,
2134 const u32 len, const int pos)
2138 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2140 /* And move data appendix as is. */
2141 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2142 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2144 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2145 skb_shinfo(skb)->nr_frags = 0;
2146 skb1->data_len = skb->data_len;
2147 skb1->len += skb1->data_len;
2150 skb_set_tail_pointer(skb, len);
2153 static inline void skb_split_no_header(struct sk_buff *skb,
2154 struct sk_buff* skb1,
2155 const u32 len, int pos)
2158 const int nfrags = skb_shinfo(skb)->nr_frags;
2160 skb_shinfo(skb)->nr_frags = 0;
2161 skb1->len = skb1->data_len = skb->len - len;
2163 skb->data_len = len - pos;
2165 for (i = 0; i < nfrags; i++) {
2166 int size = skb_shinfo(skb)->frags[i].size;
2168 if (pos + size > len) {
2169 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2173 * We have two variants in this case:
2174 * 1. Move all the frag to the second
2175 * part, if it is possible. F.e.
2176 * this approach is mandatory for TUX,
2177 * where splitting is expensive.
2178 * 2. Split is accurately. We make this.
2180 skb_frag_ref(skb, i);
2181 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2182 skb_shinfo(skb1)->frags[0].size -= len - pos;
2183 skb_shinfo(skb)->frags[i].size = len - pos;
2184 skb_shinfo(skb)->nr_frags++;
2188 skb_shinfo(skb)->nr_frags++;
2191 skb_shinfo(skb1)->nr_frags = k;
2195 * skb_split - Split fragmented skb to two parts at length len.
2196 * @skb: the buffer to split
2197 * @skb1: the buffer to receive the second part
2198 * @len: new length for skb
2200 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2202 int pos = skb_headlen(skb);
2204 if (len < pos) /* Split line is inside header. */
2205 skb_split_inside_header(skb, skb1, len, pos);
2206 else /* Second chunk has no header, nothing to copy. */
2207 skb_split_no_header(skb, skb1, len, pos);
2209 EXPORT_SYMBOL(skb_split);
2211 /* Shifting from/to a cloned skb is a no-go.
2213 * Caller cannot keep skb_shinfo related pointers past calling here!
2215 static int skb_prepare_for_shift(struct sk_buff *skb)
2217 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2221 * skb_shift - Shifts paged data partially from skb to another
2222 * @tgt: buffer into which tail data gets added
2223 * @skb: buffer from which the paged data comes from
2224 * @shiftlen: shift up to this many bytes
2226 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2227 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2228 * It's up to caller to free skb if everything was shifted.
2230 * If @tgt runs out of frags, the whole operation is aborted.
2232 * Skb cannot include anything else but paged data while tgt is allowed
2233 * to have non-paged data as well.
2235 * TODO: full sized shift could be optimized but that would need
2236 * specialized skb free'er to handle frags without up-to-date nr_frags.
2238 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2240 int from, to, merge, todo;
2241 struct skb_frag_struct *fragfrom, *fragto;
2243 BUG_ON(shiftlen > skb->len);
2244 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2248 to = skb_shinfo(tgt)->nr_frags;
2249 fragfrom = &skb_shinfo(skb)->frags[from];
2251 /* Actual merge is delayed until the point when we know we can
2252 * commit all, so that we don't have to undo partial changes
2255 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2256 fragfrom->page_offset)) {
2261 todo -= fragfrom->size;
2263 if (skb_prepare_for_shift(skb) ||
2264 skb_prepare_for_shift(tgt))
2267 /* All previous frag pointers might be stale! */
2268 fragfrom = &skb_shinfo(skb)->frags[from];
2269 fragto = &skb_shinfo(tgt)->frags[merge];
2271 fragto->size += shiftlen;
2272 fragfrom->size -= shiftlen;
2273 fragfrom->page_offset += shiftlen;
2281 /* Skip full, not-fitting skb to avoid expensive operations */
2282 if ((shiftlen == skb->len) &&
2283 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2286 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2289 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2290 if (to == MAX_SKB_FRAGS)
2293 fragfrom = &skb_shinfo(skb)->frags[from];
2294 fragto = &skb_shinfo(tgt)->frags[to];
2296 if (todo >= fragfrom->size) {
2297 *fragto = *fragfrom;
2298 todo -= fragfrom->size;
2303 __skb_frag_ref(fragfrom);
2304 fragto->page = fragfrom->page;
2305 fragto->page_offset = fragfrom->page_offset;
2306 fragto->size = todo;
2308 fragfrom->page_offset += todo;
2309 fragfrom->size -= todo;
2317 /* Ready to "commit" this state change to tgt */
2318 skb_shinfo(tgt)->nr_frags = to;
2321 fragfrom = &skb_shinfo(skb)->frags[0];
2322 fragto = &skb_shinfo(tgt)->frags[merge];
2324 fragto->size += fragfrom->size;
2325 __skb_frag_unref(fragfrom);
2328 /* Reposition in the original skb */
2330 while (from < skb_shinfo(skb)->nr_frags)
2331 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2332 skb_shinfo(skb)->nr_frags = to;
2334 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2337 /* Most likely the tgt won't ever need its checksum anymore, skb on
2338 * the other hand might need it if it needs to be resent
2340 tgt->ip_summed = CHECKSUM_PARTIAL;
2341 skb->ip_summed = CHECKSUM_PARTIAL;
2343 /* Yak, is it really working this way? Some helper please? */
2344 skb->len -= shiftlen;
2345 skb->data_len -= shiftlen;
2346 skb->truesize -= shiftlen;
2347 tgt->len += shiftlen;
2348 tgt->data_len += shiftlen;
2349 tgt->truesize += shiftlen;
2355 * skb_prepare_seq_read - Prepare a sequential read of skb data
2356 * @skb: the buffer to read
2357 * @from: lower offset of data to be read
2358 * @to: upper offset of data to be read
2359 * @st: state variable
2361 * Initializes the specified state variable. Must be called before
2362 * invoking skb_seq_read() for the first time.
2364 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2365 unsigned int to, struct skb_seq_state *st)
2367 st->lower_offset = from;
2368 st->upper_offset = to;
2369 st->root_skb = st->cur_skb = skb;
2370 st->frag_idx = st->stepped_offset = 0;
2371 st->frag_data = NULL;
2373 EXPORT_SYMBOL(skb_prepare_seq_read);
2376 * skb_seq_read - Sequentially read skb data
2377 * @consumed: number of bytes consumed by the caller so far
2378 * @data: destination pointer for data to be returned
2379 * @st: state variable
2381 * Reads a block of skb data at &consumed relative to the
2382 * lower offset specified to skb_prepare_seq_read(). Assigns
2383 * the head of the data block to &data and returns the length
2384 * of the block or 0 if the end of the skb data or the upper
2385 * offset has been reached.
2387 * The caller is not required to consume all of the data
2388 * returned, i.e. &consumed is typically set to the number
2389 * of bytes already consumed and the next call to
2390 * skb_seq_read() will return the remaining part of the block.
2392 * Note 1: The size of each block of data returned can be arbitrary,
2393 * this limitation is the cost for zerocopy seqeuental
2394 * reads of potentially non linear data.
2396 * Note 2: Fragment lists within fragments are not implemented
2397 * at the moment, state->root_skb could be replaced with
2398 * a stack for this purpose.
2400 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2401 struct skb_seq_state *st)
2403 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2406 if (unlikely(abs_offset >= st->upper_offset))
2410 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2412 if (abs_offset < block_limit && !st->frag_data) {
2413 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2414 return block_limit - abs_offset;
2417 if (st->frag_idx == 0 && !st->frag_data)
2418 st->stepped_offset += skb_headlen(st->cur_skb);
2420 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2421 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2422 block_limit = frag->size + st->stepped_offset;
2424 if (abs_offset < block_limit) {
2426 st->frag_data = kmap_skb_frag(frag);
2428 *data = (u8 *) st->frag_data + frag->page_offset +
2429 (abs_offset - st->stepped_offset);
2431 return block_limit - abs_offset;
2434 if (st->frag_data) {
2435 kunmap_skb_frag(st->frag_data);
2436 st->frag_data = NULL;
2440 st->stepped_offset += frag->size;
2443 if (st->frag_data) {
2444 kunmap_skb_frag(st->frag_data);
2445 st->frag_data = NULL;
2448 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2449 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2452 } else if (st->cur_skb->next) {
2453 st->cur_skb = st->cur_skb->next;
2460 EXPORT_SYMBOL(skb_seq_read);
2463 * skb_abort_seq_read - Abort a sequential read of skb data
2464 * @st: state variable
2466 * Must be called if skb_seq_read() was not called until it
2469 void skb_abort_seq_read(struct skb_seq_state *st)
2472 kunmap_skb_frag(st->frag_data);
2474 EXPORT_SYMBOL(skb_abort_seq_read);
2476 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2478 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2479 struct ts_config *conf,
2480 struct ts_state *state)
2482 return skb_seq_read(offset, text, TS_SKB_CB(state));
2485 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2487 skb_abort_seq_read(TS_SKB_CB(state));
2491 * skb_find_text - Find a text pattern in skb data
2492 * @skb: the buffer to look in
2493 * @from: search offset
2495 * @config: textsearch configuration
2496 * @state: uninitialized textsearch state variable
2498 * Finds a pattern in the skb data according to the specified
2499 * textsearch configuration. Use textsearch_next() to retrieve
2500 * subsequent occurrences of the pattern. Returns the offset
2501 * to the first occurrence or UINT_MAX if no match was found.
2503 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2504 unsigned int to, struct ts_config *config,
2505 struct ts_state *state)
2509 config->get_next_block = skb_ts_get_next_block;
2510 config->finish = skb_ts_finish;
2512 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2514 ret = textsearch_find(config, state);
2515 return (ret <= to - from ? ret : UINT_MAX);
2517 EXPORT_SYMBOL(skb_find_text);
2520 * skb_append_datato_frags: - append the user data to a skb
2521 * @sk: sock structure
2522 * @skb: skb structure to be appened with user data.
2523 * @getfrag: call back function to be used for getting the user data
2524 * @from: pointer to user message iov
2525 * @length: length of the iov message
2527 * Description: This procedure append the user data in the fragment part
2528 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2530 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2531 int (*getfrag)(void *from, char *to, int offset,
2532 int len, int odd, struct sk_buff *skb),
2533 void *from, int length)
2536 skb_frag_t *frag = NULL;
2537 struct page *page = NULL;
2543 /* Return error if we don't have space for new frag */
2544 frg_cnt = skb_shinfo(skb)->nr_frags;
2545 if (frg_cnt >= MAX_SKB_FRAGS)
2548 /* allocate a new page for next frag */
2549 page = alloc_pages(sk->sk_allocation, 0);
2551 /* If alloc_page fails just return failure and caller will
2552 * free previous allocated pages by doing kfree_skb()
2557 /* initialize the next frag */
2558 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2559 skb->truesize += PAGE_SIZE;
2560 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2562 /* get the new initialized frag */
2563 frg_cnt = skb_shinfo(skb)->nr_frags;
2564 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2566 /* copy the user data to page */
2567 left = PAGE_SIZE - frag->page_offset;
2568 copy = (length > left)? left : length;
2570 ret = getfrag(from, skb_frag_address(frag) + frag->size,
2571 offset, copy, 0, skb);
2575 /* copy was successful so update the size parameters */
2578 skb->data_len += copy;
2582 } while (length > 0);
2586 EXPORT_SYMBOL(skb_append_datato_frags);
2589 * skb_pull_rcsum - pull skb and update receive checksum
2590 * @skb: buffer to update
2591 * @len: length of data pulled
2593 * This function performs an skb_pull on the packet and updates
2594 * the CHECKSUM_COMPLETE checksum. It should be used on
2595 * receive path processing instead of skb_pull unless you know
2596 * that the checksum difference is zero (e.g., a valid IP header)
2597 * or you are setting ip_summed to CHECKSUM_NONE.
2599 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2601 BUG_ON(len > skb->len);
2603 BUG_ON(skb->len < skb->data_len);
2604 skb_postpull_rcsum(skb, skb->data, len);
2605 return skb->data += len;
2607 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2610 * skb_segment - Perform protocol segmentation on skb.
2611 * @skb: buffer to segment
2612 * @features: features for the output path (see dev->features)
2614 * This function performs segmentation on the given skb. It returns
2615 * a pointer to the first in a list of new skbs for the segments.
2616 * In case of error it returns ERR_PTR(err).
2618 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2620 struct sk_buff *segs = NULL;
2621 struct sk_buff *tail = NULL;
2622 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2623 unsigned int mss = skb_shinfo(skb)->gso_size;
2624 unsigned int doffset = skb->data - skb_mac_header(skb);
2625 unsigned int offset = doffset;
2626 unsigned int headroom;
2628 int sg = !!(features & NETIF_F_SG);
2629 int nfrags = skb_shinfo(skb)->nr_frags;
2634 __skb_push(skb, doffset);
2635 headroom = skb_headroom(skb);
2636 pos = skb_headlen(skb);
2639 struct sk_buff *nskb;
2644 len = skb->len - offset;
2648 hsize = skb_headlen(skb) - offset;
2651 if (hsize > len || !sg)
2654 if (!hsize && i >= nfrags) {
2655 BUG_ON(fskb->len != len);
2658 nskb = skb_clone(fskb, GFP_ATOMIC);
2661 if (unlikely(!nskb))
2664 hsize = skb_end_pointer(nskb) - nskb->head;
2665 if (skb_cow_head(nskb, doffset + headroom)) {
2670 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2672 skb_release_head_state(nskb);
2673 __skb_push(nskb, doffset);
2675 nskb = alloc_skb(hsize + doffset + headroom,
2678 if (unlikely(!nskb))
2681 skb_reserve(nskb, headroom);
2682 __skb_put(nskb, doffset);
2691 __copy_skb_header(nskb, skb);
2692 nskb->mac_len = skb->mac_len;
2694 /* nskb and skb might have different headroom */
2695 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2696 nskb->csum_start += skb_headroom(nskb) - headroom;
2698 skb_reset_mac_header(nskb);
2699 skb_set_network_header(nskb, skb->mac_len);
2700 nskb->transport_header = (nskb->network_header +
2701 skb_network_header_len(skb));
2702 skb_copy_from_linear_data(skb, nskb->data, doffset);
2704 if (fskb != skb_shinfo(skb)->frag_list)
2708 nskb->ip_summed = CHECKSUM_NONE;
2709 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2715 frag = skb_shinfo(nskb)->frags;
2717 skb_copy_from_linear_data_offset(skb, offset,
2718 skb_put(nskb, hsize), hsize);
2720 while (pos < offset + len && i < nfrags) {
2721 *frag = skb_shinfo(skb)->frags[i];
2722 __skb_frag_ref(frag);
2726 frag->page_offset += offset - pos;
2727 frag->size -= offset - pos;
2730 skb_shinfo(nskb)->nr_frags++;
2732 if (pos + size <= offset + len) {
2736 frag->size -= pos + size - (offset + len);
2743 if (pos < offset + len) {
2744 struct sk_buff *fskb2 = fskb;
2746 BUG_ON(pos + fskb->len != offset + len);
2752 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2758 SKB_FRAG_ASSERT(nskb);
2759 skb_shinfo(nskb)->frag_list = fskb2;
2763 nskb->data_len = len - hsize;
2764 nskb->len += nskb->data_len;
2765 nskb->truesize += nskb->data_len;
2766 } while ((offset += len) < skb->len);
2771 while ((skb = segs)) {
2775 return ERR_PTR(err);
2777 EXPORT_SYMBOL_GPL(skb_segment);
2779 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2781 struct sk_buff *p = *head;
2782 struct sk_buff *nskb;
2783 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2784 struct skb_shared_info *pinfo = skb_shinfo(p);
2785 unsigned int headroom;
2786 unsigned int len = skb_gro_len(skb);
2787 unsigned int offset = skb_gro_offset(skb);
2788 unsigned int headlen = skb_headlen(skb);
2790 if (p->len + len >= 65536)
2793 if (pinfo->frag_list)
2795 else if (headlen <= offset) {
2798 int i = skbinfo->nr_frags;
2799 int nr_frags = pinfo->nr_frags + i;
2803 if (nr_frags > MAX_SKB_FRAGS)
2806 pinfo->nr_frags = nr_frags;
2807 skbinfo->nr_frags = 0;
2809 frag = pinfo->frags + nr_frags;
2810 frag2 = skbinfo->frags + i;
2815 frag->page_offset += offset;
2816 frag->size -= offset;
2818 skb->truesize -= skb->data_len;
2819 skb->len -= skb->data_len;
2822 NAPI_GRO_CB(skb)->free = 1;
2824 } else if (skb_gro_len(p) != pinfo->gso_size)
2827 headroom = skb_headroom(p);
2828 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2829 if (unlikely(!nskb))
2832 __copy_skb_header(nskb, p);
2833 nskb->mac_len = p->mac_len;
2835 skb_reserve(nskb, headroom);
2836 __skb_put(nskb, skb_gro_offset(p));
2838 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2839 skb_set_network_header(nskb, skb_network_offset(p));
2840 skb_set_transport_header(nskb, skb_transport_offset(p));
2842 __skb_pull(p, skb_gro_offset(p));
2843 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2844 p->data - skb_mac_header(p));
2846 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2847 skb_shinfo(nskb)->frag_list = p;
2848 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2849 pinfo->gso_size = 0;
2850 skb_header_release(p);
2853 nskb->data_len += p->len;
2854 nskb->truesize += p->len;
2855 nskb->len += p->len;
2858 nskb->next = p->next;
2864 if (offset > headlen) {
2865 unsigned int eat = offset - headlen;
2867 skbinfo->frags[0].page_offset += eat;
2868 skbinfo->frags[0].size -= eat;
2869 skb->data_len -= eat;
2874 __skb_pull(skb, offset);
2876 p->prev->next = skb;
2878 skb_header_release(skb);
2881 NAPI_GRO_CB(p)->count++;
2886 NAPI_GRO_CB(skb)->same_flow = 1;
2889 EXPORT_SYMBOL_GPL(skb_gro_receive);
2891 void __init skb_init(void)
2893 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2894 sizeof(struct sk_buff),
2896 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2898 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2899 (2*sizeof(struct sk_buff)) +
2902 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2907 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2908 * @skb: Socket buffer containing the buffers to be mapped
2909 * @sg: The scatter-gather list to map into
2910 * @offset: The offset into the buffer's contents to start mapping
2911 * @len: Length of buffer space to be mapped
2913 * Fill the specified scatter-gather list with mappings/pointers into a
2914 * region of the buffer space attached to a socket buffer.
2917 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2919 int start = skb_headlen(skb);
2920 int i, copy = start - offset;
2921 struct sk_buff *frag_iter;
2927 sg_set_buf(sg, skb->data + offset, copy);
2929 if ((len -= copy) == 0)
2934 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2937 WARN_ON(start > offset + len);
2939 end = start + skb_shinfo(skb)->frags[i].size;
2940 if ((copy = end - offset) > 0) {
2941 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2945 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
2946 frag->page_offset+offset-start);
2955 skb_walk_frags(skb, frag_iter) {
2958 WARN_ON(start > offset + len);
2960 end = start + frag_iter->len;
2961 if ((copy = end - offset) > 0) {
2964 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2966 if ((len -= copy) == 0)
2976 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2978 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2980 sg_mark_end(&sg[nsg - 1]);
2984 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2987 * skb_cow_data - Check that a socket buffer's data buffers are writable
2988 * @skb: The socket buffer to check.
2989 * @tailbits: Amount of trailing space to be added
2990 * @trailer: Returned pointer to the skb where the @tailbits space begins
2992 * Make sure that the data buffers attached to a socket buffer are
2993 * writable. If they are not, private copies are made of the data buffers
2994 * and the socket buffer is set to use these instead.
2996 * If @tailbits is given, make sure that there is space to write @tailbits
2997 * bytes of data beyond current end of socket buffer. @trailer will be
2998 * set to point to the skb in which this space begins.
3000 * The number of scatterlist elements required to completely map the
3001 * COW'd and extended socket buffer will be returned.
3003 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3007 struct sk_buff *skb1, **skb_p;
3009 /* If skb is cloned or its head is paged, reallocate
3010 * head pulling out all the pages (pages are considered not writable
3011 * at the moment even if they are anonymous).
3013 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3014 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3017 /* Easy case. Most of packets will go this way. */
3018 if (!skb_has_frag_list(skb)) {
3019 /* A little of trouble, not enough of space for trailer.
3020 * This should not happen, when stack is tuned to generate
3021 * good frames. OK, on miss we reallocate and reserve even more
3022 * space, 128 bytes is fair. */
3024 if (skb_tailroom(skb) < tailbits &&
3025 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3033 /* Misery. We are in troubles, going to mincer fragments... */
3036 skb_p = &skb_shinfo(skb)->frag_list;
3039 while ((skb1 = *skb_p) != NULL) {
3042 /* The fragment is partially pulled by someone,
3043 * this can happen on input. Copy it and everything
3046 if (skb_shared(skb1))
3049 /* If the skb is the last, worry about trailer. */
3051 if (skb1->next == NULL && tailbits) {
3052 if (skb_shinfo(skb1)->nr_frags ||
3053 skb_has_frag_list(skb1) ||
3054 skb_tailroom(skb1) < tailbits)
3055 ntail = tailbits + 128;
3061 skb_shinfo(skb1)->nr_frags ||
3062 skb_has_frag_list(skb1)) {
3063 struct sk_buff *skb2;
3065 /* Fuck, we are miserable poor guys... */
3067 skb2 = skb_copy(skb1, GFP_ATOMIC);
3069 skb2 = skb_copy_expand(skb1,
3073 if (unlikely(skb2 == NULL))
3077 skb_set_owner_w(skb2, skb1->sk);
3079 /* Looking around. Are we still alive?
3080 * OK, link new skb, drop old one */
3082 skb2->next = skb1->next;
3089 skb_p = &skb1->next;
3094 EXPORT_SYMBOL_GPL(skb_cow_data);
3096 static void sock_rmem_free(struct sk_buff *skb)
3098 struct sock *sk = skb->sk;
3100 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3104 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3106 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3108 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3109 (unsigned)sk->sk_rcvbuf)
3114 skb->destructor = sock_rmem_free;
3115 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3117 /* before exiting rcu section, make sure dst is refcounted */
3120 skb_queue_tail(&sk->sk_error_queue, skb);
3121 if (!sock_flag(sk, SOCK_DEAD))
3122 sk->sk_data_ready(sk, skb->len);
3125 EXPORT_SYMBOL(sock_queue_err_skb);
3127 void skb_tstamp_tx(struct sk_buff *orig_skb,
3128 struct skb_shared_hwtstamps *hwtstamps)
3130 struct sock *sk = orig_skb->sk;
3131 struct sock_exterr_skb *serr;
3132 struct sk_buff *skb;
3138 skb = skb_clone(orig_skb, GFP_ATOMIC);
3143 *skb_hwtstamps(skb) =
3147 * no hardware time stamps available,
3148 * so keep the shared tx_flags and only
3149 * store software time stamp
3151 skb->tstamp = ktime_get_real();
3154 serr = SKB_EXT_ERR(skb);
3155 memset(serr, 0, sizeof(*serr));
3156 serr->ee.ee_errno = ENOMSG;
3157 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3159 err = sock_queue_err_skb(sk, skb);
3164 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3168 * skb_partial_csum_set - set up and verify partial csum values for packet
3169 * @skb: the skb to set
3170 * @start: the number of bytes after skb->data to start checksumming.
3171 * @off: the offset from start to place the checksum.
3173 * For untrusted partially-checksummed packets, we need to make sure the values
3174 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3176 * This function checks and sets those values and skb->ip_summed: if this
3177 * returns false you should drop the packet.
3179 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3181 if (unlikely(start > skb_headlen(skb)) ||
3182 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3183 if (net_ratelimit())
3185 "bad partial csum: csum=%u/%u len=%u\n",
3186 start, off, skb_headlen(skb));
3189 skb->ip_summed = CHECKSUM_PARTIAL;
3190 skb->csum_start = skb_headroom(skb) + start;
3191 skb->csum_offset = off;
3194 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3196 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3198 if (net_ratelimit())
3199 pr_warning("%s: received packets cannot be forwarded"
3200 " while LRO is enabled\n", skb->dev->name);
3202 EXPORT_SYMBOL(__skb_warn_lro_forwarding);