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 <trace/events/skb.h>
70 #include <linux/highmem.h>
72 static struct kmem_cache *skbuff_head_cache __read_mostly;
73 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
75 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
76 struct pipe_buffer *buf)
81 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
82 struct pipe_buffer *buf)
87 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
88 struct pipe_buffer *buf)
94 /* Pipe buffer operations for a socket. */
95 static const struct pipe_buf_operations sock_pipe_buf_ops = {
97 .map = generic_pipe_buf_map,
98 .unmap = generic_pipe_buf_unmap,
99 .confirm = generic_pipe_buf_confirm,
100 .release = sock_pipe_buf_release,
101 .steal = sock_pipe_buf_steal,
102 .get = sock_pipe_buf_get,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
121 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here, skb->len, sz, skb->head, skb->data,
124 (unsigned long)skb->tail, (unsigned long)skb->end,
125 skb->dev ? skb->dev->name : "<NULL>");
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
140 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
148 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
170 int fclone, int node)
172 struct kmem_cache *cache;
173 struct skb_shared_info *shinfo;
177 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
180 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
190 size = SKB_DATA_ALIGN(size);
191 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
192 data = kmalloc_node_track_caller(size, gfp_mask, node);
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
199 size = SKB_WITH_OVERHEAD(ksize(data));
200 prefetchw(data + size);
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
207 memset(skb, 0, offsetof(struct sk_buff, tail));
208 /* Account for allocated memory : skb + skb->head */
209 skb->truesize = SKB_TRUESIZE(size);
210 atomic_set(&skb->users, 1);
213 skb_reset_tail_pointer(skb);
214 skb->end = skb->tail + size;
215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb->mac_header = ~0U;
219 /* make sure we initialize shinfo sequentially */
220 shinfo = skb_shinfo(skb);
221 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
222 atomic_set(&shinfo->dataref, 1);
223 kmemcheck_annotate_variable(shinfo->destructor_arg);
226 struct sk_buff *child = skb + 1;
227 atomic_t *fclone_ref = (atomic_t *) (child + 1);
229 kmemcheck_annotate_bitfield(child, flags1);
230 kmemcheck_annotate_bitfield(child, flags2);
231 skb->fclone = SKB_FCLONE_ORIG;
232 atomic_set(fclone_ref, 1);
234 child->fclone = SKB_FCLONE_UNAVAILABLE;
239 kmem_cache_free(cache, skb);
243 EXPORT_SYMBOL(__alloc_skb);
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
249 * Allocate a new &sk_buff. Caller provides space holding head and
250 * skb_shared_info. @data must have been allocated by kmalloc()
251 * The return is the new skb buffer.
252 * On a failure the return is %NULL, and @data is not freed.
254 * Before IO, driver allocates only data buffer where NIC put incoming frame
255 * Driver should add room at head (NET_SKB_PAD) and
256 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
257 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
258 * before giving packet to stack.
259 * RX rings only contains data buffers, not full skbs.
261 struct sk_buff *build_skb(void *data)
263 struct skb_shared_info *shinfo;
267 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
271 size = ksize(data) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
273 memset(skb, 0, offsetof(struct sk_buff, tail));
274 skb->truesize = SKB_TRUESIZE(size);
275 atomic_set(&skb->users, 1);
278 skb_reset_tail_pointer(skb);
279 skb->end = skb->tail + size;
280 #ifdef NET_SKBUFF_DATA_USES_OFFSET
281 skb->mac_header = ~0U;
284 /* make sure we initialize shinfo sequentially */
285 shinfo = skb_shinfo(skb);
286 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
287 atomic_set(&shinfo->dataref, 1);
288 kmemcheck_annotate_variable(shinfo->destructor_arg);
292 EXPORT_SYMBOL(build_skb);
295 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
296 * @dev: network device to receive on
297 * @length: length to allocate
298 * @gfp_mask: get_free_pages mask, passed to alloc_skb
300 * Allocate a new &sk_buff and assign it a usage count of one. The
301 * buffer has unspecified headroom built in. Users should allocate
302 * the headroom they think they need without accounting for the
303 * built in space. The built in space is used for optimisations.
305 * %NULL is returned if there is no free memory.
307 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
308 unsigned int length, gfp_t gfp_mask)
312 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
314 skb_reserve(skb, NET_SKB_PAD);
319 EXPORT_SYMBOL(__netdev_alloc_skb);
321 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
322 int size, unsigned int truesize)
324 skb_fill_page_desc(skb, i, page, off, size);
326 skb->data_len += size;
327 skb->truesize += truesize;
329 EXPORT_SYMBOL(skb_add_rx_frag);
332 * dev_alloc_skb - allocate an skbuff for receiving
333 * @length: length to allocate
335 * Allocate a new &sk_buff and assign it a usage count of one. The
336 * buffer has unspecified headroom built in. Users should allocate
337 * the headroom they think they need without accounting for the
338 * built in space. The built in space is used for optimisations.
340 * %NULL is returned if there is no free memory. Although this function
341 * allocates memory it can be called from an interrupt.
343 struct sk_buff *dev_alloc_skb(unsigned int length)
346 * There is more code here than it seems:
347 * __dev_alloc_skb is an inline
349 return __dev_alloc_skb(length, GFP_ATOMIC);
351 EXPORT_SYMBOL(dev_alloc_skb);
353 static void skb_drop_list(struct sk_buff **listp)
355 struct sk_buff *list = *listp;
360 struct sk_buff *this = list;
366 static inline void skb_drop_fraglist(struct sk_buff *skb)
368 skb_drop_list(&skb_shinfo(skb)->frag_list);
371 static void skb_clone_fraglist(struct sk_buff *skb)
373 struct sk_buff *list;
375 skb_walk_frags(skb, list)
379 static void skb_release_data(struct sk_buff *skb)
382 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
383 &skb_shinfo(skb)->dataref)) {
384 if (skb_shinfo(skb)->nr_frags) {
386 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
387 skb_frag_unref(skb, i);
391 * If skb buf is from userspace, we need to notify the caller
392 * the lower device DMA has done;
394 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
395 struct ubuf_info *uarg;
397 uarg = skb_shinfo(skb)->destructor_arg;
399 uarg->callback(uarg);
402 if (skb_has_frag_list(skb))
403 skb_drop_fraglist(skb);
410 * Free an skbuff by memory without cleaning the state.
412 static void kfree_skbmem(struct sk_buff *skb)
414 struct sk_buff *other;
415 atomic_t *fclone_ref;
417 switch (skb->fclone) {
418 case SKB_FCLONE_UNAVAILABLE:
419 kmem_cache_free(skbuff_head_cache, skb);
422 case SKB_FCLONE_ORIG:
423 fclone_ref = (atomic_t *) (skb + 2);
424 if (atomic_dec_and_test(fclone_ref))
425 kmem_cache_free(skbuff_fclone_cache, skb);
428 case SKB_FCLONE_CLONE:
429 fclone_ref = (atomic_t *) (skb + 1);
432 /* The clone portion is available for
433 * fast-cloning again.
435 skb->fclone = SKB_FCLONE_UNAVAILABLE;
437 if (atomic_dec_and_test(fclone_ref))
438 kmem_cache_free(skbuff_fclone_cache, other);
443 static void skb_release_head_state(struct sk_buff *skb)
447 secpath_put(skb->sp);
449 if (skb->destructor) {
451 skb->destructor(skb);
453 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
454 nf_conntrack_put(skb->nfct);
456 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
457 nf_conntrack_put_reasm(skb->nfct_reasm);
459 #ifdef CONFIG_BRIDGE_NETFILTER
460 nf_bridge_put(skb->nf_bridge);
462 /* XXX: IS this still necessary? - JHS */
463 #ifdef CONFIG_NET_SCHED
465 #ifdef CONFIG_NET_CLS_ACT
471 /* Free everything but the sk_buff shell. */
472 static void skb_release_all(struct sk_buff *skb)
474 skb_release_head_state(skb);
475 skb_release_data(skb);
479 * __kfree_skb - private function
482 * Free an sk_buff. Release anything attached to the buffer.
483 * Clean the state. This is an internal helper function. Users should
484 * always call kfree_skb
487 void __kfree_skb(struct sk_buff *skb)
489 skb_release_all(skb);
492 EXPORT_SYMBOL(__kfree_skb);
495 * kfree_skb - free an sk_buff
496 * @skb: buffer to free
498 * Drop a reference to the buffer and free it if the usage count has
501 void kfree_skb(struct sk_buff *skb)
505 if (likely(atomic_read(&skb->users) == 1))
507 else if (likely(!atomic_dec_and_test(&skb->users)))
509 trace_kfree_skb(skb, __builtin_return_address(0));
512 EXPORT_SYMBOL(kfree_skb);
515 * consume_skb - free an skbuff
516 * @skb: buffer to free
518 * Drop a ref to the buffer and free it if the usage count has hit zero
519 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
520 * is being dropped after a failure and notes that
522 void consume_skb(struct sk_buff *skb)
526 if (likely(atomic_read(&skb->users) == 1))
528 else if (likely(!atomic_dec_and_test(&skb->users)))
530 trace_consume_skb(skb);
533 EXPORT_SYMBOL(consume_skb);
536 * skb_recycle - clean up an skb for reuse
539 * Recycles the skb to be reused as a receive buffer. This
540 * function does any necessary reference count dropping, and
541 * cleans up the skbuff as if it just came from __alloc_skb().
543 void skb_recycle(struct sk_buff *skb)
545 struct skb_shared_info *shinfo;
547 skb_release_head_state(skb);
549 shinfo = skb_shinfo(skb);
550 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
551 atomic_set(&shinfo->dataref, 1);
553 memset(skb, 0, offsetof(struct sk_buff, tail));
554 skb->data = skb->head + NET_SKB_PAD;
555 skb_reset_tail_pointer(skb);
557 EXPORT_SYMBOL(skb_recycle);
560 * skb_recycle_check - check if skb can be reused for receive
562 * @skb_size: minimum receive buffer size
564 * Checks that the skb passed in is not shared or cloned, and
565 * that it is linear and its head portion at least as large as
566 * skb_size so that it can be recycled as a receive buffer.
567 * If these conditions are met, this function does any necessary
568 * reference count dropping and cleans up the skbuff as if it
569 * just came from __alloc_skb().
571 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
573 if (!skb_is_recycleable(skb, skb_size))
580 EXPORT_SYMBOL(skb_recycle_check);
582 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
584 new->tstamp = old->tstamp;
586 new->transport_header = old->transport_header;
587 new->network_header = old->network_header;
588 new->mac_header = old->mac_header;
589 skb_dst_copy(new, old);
590 new->rxhash = old->rxhash;
591 new->ooo_okay = old->ooo_okay;
592 new->l4_rxhash = old->l4_rxhash;
593 new->no_fcs = old->no_fcs;
595 new->sp = secpath_get(old->sp);
597 memcpy(new->cb, old->cb, sizeof(old->cb));
598 new->csum = old->csum;
599 new->local_df = old->local_df;
600 new->pkt_type = old->pkt_type;
601 new->ip_summed = old->ip_summed;
602 skb_copy_queue_mapping(new, old);
603 new->priority = old->priority;
604 #if IS_ENABLED(CONFIG_IP_VS)
605 new->ipvs_property = old->ipvs_property;
607 new->protocol = old->protocol;
608 new->mark = old->mark;
609 new->skb_iif = old->skb_iif;
611 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
612 new->nf_trace = old->nf_trace;
614 #ifdef CONFIG_NET_SCHED
615 new->tc_index = old->tc_index;
616 #ifdef CONFIG_NET_CLS_ACT
617 new->tc_verd = old->tc_verd;
620 new->vlan_tci = old->vlan_tci;
622 skb_copy_secmark(new, old);
626 * You should not add any new code to this function. Add it to
627 * __copy_skb_header above instead.
629 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
631 #define C(x) n->x = skb->x
633 n->next = n->prev = NULL;
635 __copy_skb_header(n, skb);
640 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
643 n->destructor = NULL;
649 atomic_set(&n->users, 1);
651 atomic_inc(&(skb_shinfo(skb)->dataref));
659 * skb_morph - morph one skb into another
660 * @dst: the skb to receive the contents
661 * @src: the skb to supply the contents
663 * This is identical to skb_clone except that the target skb is
664 * supplied by the user.
666 * The target skb is returned upon exit.
668 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
670 skb_release_all(dst);
671 return __skb_clone(dst, src);
673 EXPORT_SYMBOL_GPL(skb_morph);
675 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
676 * @skb: the skb to modify
677 * @gfp_mask: allocation priority
679 * This must be called on SKBTX_DEV_ZEROCOPY skb.
680 * It will copy all frags into kernel and drop the reference
681 * to userspace pages.
683 * If this function is called from an interrupt gfp_mask() must be
686 * Returns 0 on success or a negative error code on failure
687 * to allocate kernel memory to copy to.
689 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
692 int num_frags = skb_shinfo(skb)->nr_frags;
693 struct page *page, *head = NULL;
694 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
696 for (i = 0; i < num_frags; i++) {
698 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
700 page = alloc_page(GFP_ATOMIC);
703 struct page *next = (struct page *)head->private;
709 vaddr = kmap_atomic(skb_frag_page(f));
710 memcpy(page_address(page),
711 vaddr + f->page_offset, skb_frag_size(f));
712 kunmap_atomic(vaddr);
713 page->private = (unsigned long)head;
717 /* skb frags release userspace buffers */
718 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
719 skb_frag_unref(skb, i);
721 uarg->callback(uarg);
723 /* skb frags point to kernel buffers */
724 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
725 __skb_fill_page_desc(skb, i-1, head, 0,
726 skb_shinfo(skb)->frags[i - 1].size);
727 head = (struct page *)head->private;
730 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
736 * skb_clone - duplicate an sk_buff
737 * @skb: buffer to clone
738 * @gfp_mask: allocation priority
740 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
741 * copies share the same packet data but not structure. The new
742 * buffer has a reference count of 1. If the allocation fails the
743 * function returns %NULL otherwise the new buffer is returned.
745 * If this function is called from an interrupt gfp_mask() must be
749 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
753 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
754 if (skb_copy_ubufs(skb, gfp_mask))
759 if (skb->fclone == SKB_FCLONE_ORIG &&
760 n->fclone == SKB_FCLONE_UNAVAILABLE) {
761 atomic_t *fclone_ref = (atomic_t *) (n + 1);
762 n->fclone = SKB_FCLONE_CLONE;
763 atomic_inc(fclone_ref);
765 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
769 kmemcheck_annotate_bitfield(n, flags1);
770 kmemcheck_annotate_bitfield(n, flags2);
771 n->fclone = SKB_FCLONE_UNAVAILABLE;
774 return __skb_clone(n, skb);
776 EXPORT_SYMBOL(skb_clone);
778 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
780 #ifndef NET_SKBUFF_DATA_USES_OFFSET
782 * Shift between the two data areas in bytes
784 unsigned long offset = new->data - old->data;
787 __copy_skb_header(new, old);
789 #ifndef NET_SKBUFF_DATA_USES_OFFSET
790 /* {transport,network,mac}_header are relative to skb->head */
791 new->transport_header += offset;
792 new->network_header += offset;
793 if (skb_mac_header_was_set(new))
794 new->mac_header += offset;
796 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
797 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
798 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
802 * skb_copy - create private copy of an sk_buff
803 * @skb: buffer to copy
804 * @gfp_mask: allocation priority
806 * Make a copy of both an &sk_buff and its data. This is used when the
807 * caller wishes to modify the data and needs a private copy of the
808 * data to alter. Returns %NULL on failure or the pointer to the buffer
809 * on success. The returned buffer has a reference count of 1.
811 * As by-product this function converts non-linear &sk_buff to linear
812 * one, so that &sk_buff becomes completely private and caller is allowed
813 * to modify all the data of returned buffer. This means that this
814 * function is not recommended for use in circumstances when only
815 * header is going to be modified. Use pskb_copy() instead.
818 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
820 int headerlen = skb_headroom(skb);
821 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
822 struct sk_buff *n = alloc_skb(size, gfp_mask);
827 /* Set the data pointer */
828 skb_reserve(n, headerlen);
829 /* Set the tail pointer and length */
830 skb_put(n, skb->len);
832 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
835 copy_skb_header(n, skb);
838 EXPORT_SYMBOL(skb_copy);
841 * __pskb_copy - create copy of an sk_buff with private head.
842 * @skb: buffer to copy
843 * @headroom: headroom of new skb
844 * @gfp_mask: allocation priority
846 * Make a copy of both an &sk_buff and part of its data, located
847 * in header. Fragmented data remain shared. This is used when
848 * the caller wishes to modify only header of &sk_buff and needs
849 * private copy of the header to alter. Returns %NULL on failure
850 * or the pointer to the buffer on success.
851 * The returned buffer has a reference count of 1.
854 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
856 unsigned int size = skb_headlen(skb) + headroom;
857 struct sk_buff *n = alloc_skb(size, gfp_mask);
862 /* Set the data pointer */
863 skb_reserve(n, headroom);
864 /* Set the tail pointer and length */
865 skb_put(n, skb_headlen(skb));
867 skb_copy_from_linear_data(skb, n->data, n->len);
869 n->truesize += skb->data_len;
870 n->data_len = skb->data_len;
873 if (skb_shinfo(skb)->nr_frags) {
876 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
877 if (skb_copy_ubufs(skb, gfp_mask)) {
883 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
884 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
885 skb_frag_ref(skb, i);
887 skb_shinfo(n)->nr_frags = i;
890 if (skb_has_frag_list(skb)) {
891 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
892 skb_clone_fraglist(n);
895 copy_skb_header(n, skb);
899 EXPORT_SYMBOL(__pskb_copy);
902 * pskb_expand_head - reallocate header of &sk_buff
903 * @skb: buffer to reallocate
904 * @nhead: room to add at head
905 * @ntail: room to add at tail
906 * @gfp_mask: allocation priority
908 * Expands (or creates identical copy, if &nhead and &ntail are zero)
909 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
910 * reference count of 1. Returns zero in the case of success or error,
911 * if expansion failed. In the last case, &sk_buff is not changed.
913 * All the pointers pointing into skb header may change and must be
914 * reloaded after call to this function.
917 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
922 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
931 size = SKB_DATA_ALIGN(size);
933 /* Check if we can avoid taking references on fragments if we own
934 * the last reference on skb->head. (see skb_release_data())
939 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
940 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
944 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
945 memmove(skb->head + size, skb_shinfo(skb),
946 offsetof(struct skb_shared_info,
947 frags[skb_shinfo(skb)->nr_frags]));
948 memmove(skb->head + nhead, skb->head,
949 skb_tail_pointer(skb) - skb->head);
954 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
958 /* Copy only real data... and, alas, header. This should be
959 * optimized for the cases when header is void.
961 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
963 memcpy((struct skb_shared_info *)(data + size),
965 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
970 /* copy this zero copy skb frags */
971 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
972 if (skb_copy_ubufs(skb, gfp_mask))
975 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
976 skb_frag_ref(skb, i);
978 if (skb_has_frag_list(skb))
979 skb_clone_fraglist(skb);
981 skb_release_data(skb);
983 off = (data + nhead) - skb->head;
988 #ifdef NET_SKBUFF_DATA_USES_OFFSET
992 skb->end = skb->head + size;
994 /* {transport,network,mac}_header and tail are relative to skb->head */
996 skb->transport_header += off;
997 skb->network_header += off;
998 if (skb_mac_header_was_set(skb))
999 skb->mac_header += off;
1000 /* Only adjust this if it actually is csum_start rather than csum */
1001 if (skb->ip_summed == CHECKSUM_PARTIAL)
1002 skb->csum_start += nhead;
1006 atomic_set(&skb_shinfo(skb)->dataref, 1);
1014 EXPORT_SYMBOL(pskb_expand_head);
1016 /* Make private copy of skb with writable head and some headroom */
1018 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1020 struct sk_buff *skb2;
1021 int delta = headroom - skb_headroom(skb);
1024 skb2 = pskb_copy(skb, GFP_ATOMIC);
1026 skb2 = skb_clone(skb, GFP_ATOMIC);
1027 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1035 EXPORT_SYMBOL(skb_realloc_headroom);
1038 * skb_copy_expand - copy and expand sk_buff
1039 * @skb: buffer to copy
1040 * @newheadroom: new free bytes at head
1041 * @newtailroom: new free bytes at tail
1042 * @gfp_mask: allocation priority
1044 * Make a copy of both an &sk_buff and its data and while doing so
1045 * allocate additional space.
1047 * This is used when the caller wishes to modify the data and needs a
1048 * private copy of the data to alter as well as more space for new fields.
1049 * Returns %NULL on failure or the pointer to the buffer
1050 * on success. The returned buffer has a reference count of 1.
1052 * You must pass %GFP_ATOMIC as the allocation priority if this function
1053 * is called from an interrupt.
1055 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1056 int newheadroom, int newtailroom,
1060 * Allocate the copy buffer
1062 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1064 int oldheadroom = skb_headroom(skb);
1065 int head_copy_len, head_copy_off;
1071 skb_reserve(n, newheadroom);
1073 /* Set the tail pointer and length */
1074 skb_put(n, skb->len);
1076 head_copy_len = oldheadroom;
1078 if (newheadroom <= head_copy_len)
1079 head_copy_len = newheadroom;
1081 head_copy_off = newheadroom - head_copy_len;
1083 /* Copy the linear header and data. */
1084 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1085 skb->len + head_copy_len))
1088 copy_skb_header(n, skb);
1090 off = newheadroom - oldheadroom;
1091 if (n->ip_summed == CHECKSUM_PARTIAL)
1092 n->csum_start += off;
1093 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1094 n->transport_header += off;
1095 n->network_header += off;
1096 if (skb_mac_header_was_set(skb))
1097 n->mac_header += off;
1102 EXPORT_SYMBOL(skb_copy_expand);
1105 * skb_pad - zero pad the tail of an skb
1106 * @skb: buffer to pad
1107 * @pad: space to pad
1109 * Ensure that a buffer is followed by a padding area that is zero
1110 * filled. Used by network drivers which may DMA or transfer data
1111 * beyond the buffer end onto the wire.
1113 * May return error in out of memory cases. The skb is freed on error.
1116 int skb_pad(struct sk_buff *skb, int pad)
1121 /* If the skbuff is non linear tailroom is always zero.. */
1122 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1123 memset(skb->data+skb->len, 0, pad);
1127 ntail = skb->data_len + pad - (skb->end - skb->tail);
1128 if (likely(skb_cloned(skb) || ntail > 0)) {
1129 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1134 /* FIXME: The use of this function with non-linear skb's really needs
1137 err = skb_linearize(skb);
1141 memset(skb->data + skb->len, 0, pad);
1148 EXPORT_SYMBOL(skb_pad);
1151 * skb_put - add data to a buffer
1152 * @skb: buffer to use
1153 * @len: amount of data to add
1155 * This function extends the used data area of the buffer. If this would
1156 * exceed the total buffer size the kernel will panic. A pointer to the
1157 * first byte of the extra data is returned.
1159 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1161 unsigned char *tmp = skb_tail_pointer(skb);
1162 SKB_LINEAR_ASSERT(skb);
1165 if (unlikely(skb->tail > skb->end))
1166 skb_over_panic(skb, len, __builtin_return_address(0));
1169 EXPORT_SYMBOL(skb_put);
1172 * skb_push - add data to the start of a buffer
1173 * @skb: buffer to use
1174 * @len: amount of data to add
1176 * This function extends the used data area of the buffer at the buffer
1177 * start. If this would exceed the total buffer headroom the kernel will
1178 * panic. A pointer to the first byte of the extra data is returned.
1180 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1184 if (unlikely(skb->data<skb->head))
1185 skb_under_panic(skb, len, __builtin_return_address(0));
1188 EXPORT_SYMBOL(skb_push);
1191 * skb_pull - remove data from the start of a buffer
1192 * @skb: buffer to use
1193 * @len: amount of data to remove
1195 * This function removes data from the start of a buffer, returning
1196 * the memory to the headroom. A pointer to the next data in the buffer
1197 * is returned. Once the data has been pulled future pushes will overwrite
1200 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1202 return skb_pull_inline(skb, len);
1204 EXPORT_SYMBOL(skb_pull);
1207 * skb_trim - remove end from a buffer
1208 * @skb: buffer to alter
1211 * Cut the length of a buffer down by removing data from the tail. If
1212 * the buffer is already under the length specified it is not modified.
1213 * The skb must be linear.
1215 void skb_trim(struct sk_buff *skb, unsigned int len)
1218 __skb_trim(skb, len);
1220 EXPORT_SYMBOL(skb_trim);
1222 /* Trims skb to length len. It can change skb pointers.
1225 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1227 struct sk_buff **fragp;
1228 struct sk_buff *frag;
1229 int offset = skb_headlen(skb);
1230 int nfrags = skb_shinfo(skb)->nr_frags;
1234 if (skb_cloned(skb) &&
1235 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1242 for (; i < nfrags; i++) {
1243 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1250 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1253 skb_shinfo(skb)->nr_frags = i;
1255 for (; i < nfrags; i++)
1256 skb_frag_unref(skb, i);
1258 if (skb_has_frag_list(skb))
1259 skb_drop_fraglist(skb);
1263 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1264 fragp = &frag->next) {
1265 int end = offset + frag->len;
1267 if (skb_shared(frag)) {
1268 struct sk_buff *nfrag;
1270 nfrag = skb_clone(frag, GFP_ATOMIC);
1271 if (unlikely(!nfrag))
1274 nfrag->next = frag->next;
1286 unlikely((err = pskb_trim(frag, len - offset))))
1290 skb_drop_list(&frag->next);
1295 if (len > skb_headlen(skb)) {
1296 skb->data_len -= skb->len - len;
1301 skb_set_tail_pointer(skb, len);
1306 EXPORT_SYMBOL(___pskb_trim);
1309 * __pskb_pull_tail - advance tail of skb header
1310 * @skb: buffer to reallocate
1311 * @delta: number of bytes to advance tail
1313 * The function makes a sense only on a fragmented &sk_buff,
1314 * it expands header moving its tail forward and copying necessary
1315 * data from fragmented part.
1317 * &sk_buff MUST have reference count of 1.
1319 * Returns %NULL (and &sk_buff does not change) if pull failed
1320 * or value of new tail of skb in the case of success.
1322 * All the pointers pointing into skb header may change and must be
1323 * reloaded after call to this function.
1326 /* Moves tail of skb head forward, copying data from fragmented part,
1327 * when it is necessary.
1328 * 1. It may fail due to malloc failure.
1329 * 2. It may change skb pointers.
1331 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1333 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1335 /* If skb has not enough free space at tail, get new one
1336 * plus 128 bytes for future expansions. If we have enough
1337 * room at tail, reallocate without expansion only if skb is cloned.
1339 int i, k, eat = (skb->tail + delta) - skb->end;
1341 if (eat > 0 || skb_cloned(skb)) {
1342 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1347 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1350 /* Optimization: no fragments, no reasons to preestimate
1351 * size of pulled pages. Superb.
1353 if (!skb_has_frag_list(skb))
1356 /* Estimate size of pulled pages. */
1358 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1359 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1366 /* If we need update frag list, we are in troubles.
1367 * Certainly, it possible to add an offset to skb data,
1368 * but taking into account that pulling is expected to
1369 * be very rare operation, it is worth to fight against
1370 * further bloating skb head and crucify ourselves here instead.
1371 * Pure masohism, indeed. 8)8)
1374 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1375 struct sk_buff *clone = NULL;
1376 struct sk_buff *insp = NULL;
1381 if (list->len <= eat) {
1382 /* Eaten as whole. */
1387 /* Eaten partially. */
1389 if (skb_shared(list)) {
1390 /* Sucks! We need to fork list. :-( */
1391 clone = skb_clone(list, GFP_ATOMIC);
1397 /* This may be pulled without
1401 if (!pskb_pull(list, eat)) {
1409 /* Free pulled out fragments. */
1410 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1411 skb_shinfo(skb)->frag_list = list->next;
1414 /* And insert new clone at head. */
1417 skb_shinfo(skb)->frag_list = clone;
1420 /* Success! Now we may commit changes to skb data. */
1425 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1426 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1429 skb_frag_unref(skb, i);
1432 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1434 skb_shinfo(skb)->frags[k].page_offset += eat;
1435 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1441 skb_shinfo(skb)->nr_frags = k;
1444 skb->data_len -= delta;
1446 return skb_tail_pointer(skb);
1448 EXPORT_SYMBOL(__pskb_pull_tail);
1451 * skb_copy_bits - copy bits from skb to kernel buffer
1453 * @offset: offset in source
1454 * @to: destination buffer
1455 * @len: number of bytes to copy
1457 * Copy the specified number of bytes from the source skb to the
1458 * destination buffer.
1461 * If its prototype is ever changed,
1462 * check arch/{*}/net/{*}.S files,
1463 * since it is called from BPF assembly code.
1465 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1467 int start = skb_headlen(skb);
1468 struct sk_buff *frag_iter;
1471 if (offset > (int)skb->len - len)
1475 if ((copy = start - offset) > 0) {
1478 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1479 if ((len -= copy) == 0)
1485 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1487 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1489 WARN_ON(start > offset + len);
1491 end = start + skb_frag_size(f);
1492 if ((copy = end - offset) > 0) {
1498 vaddr = kmap_atomic(skb_frag_page(f));
1500 vaddr + f->page_offset + offset - start,
1502 kunmap_atomic(vaddr);
1504 if ((len -= copy) == 0)
1512 skb_walk_frags(skb, frag_iter) {
1515 WARN_ON(start > offset + len);
1517 end = start + frag_iter->len;
1518 if ((copy = end - offset) > 0) {
1521 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1523 if ((len -= copy) == 0)
1537 EXPORT_SYMBOL(skb_copy_bits);
1540 * Callback from splice_to_pipe(), if we need to release some pages
1541 * at the end of the spd in case we error'ed out in filling the pipe.
1543 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1545 put_page(spd->pages[i]);
1548 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1549 unsigned int *offset,
1550 struct sk_buff *skb, struct sock *sk)
1552 struct page *p = sk->sk_sndmsg_page;
1557 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1561 off = sk->sk_sndmsg_off = 0;
1562 /* hold one ref to this page until it's full */
1566 off = sk->sk_sndmsg_off;
1567 mlen = PAGE_SIZE - off;
1568 if (mlen < 64 && mlen < *len) {
1573 *len = min_t(unsigned int, *len, mlen);
1576 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1577 sk->sk_sndmsg_off += *len;
1585 * Fill page/offset/length into spd, if it can hold more pages.
1587 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1588 struct pipe_inode_info *pipe, struct page *page,
1589 unsigned int *len, unsigned int offset,
1590 struct sk_buff *skb, int linear,
1593 if (unlikely(spd->nr_pages == pipe->buffers))
1597 page = linear_to_page(page, len, &offset, skb, sk);
1603 spd->pages[spd->nr_pages] = page;
1604 spd->partial[spd->nr_pages].len = *len;
1605 spd->partial[spd->nr_pages].offset = offset;
1611 static inline void __segment_seek(struct page **page, unsigned int *poff,
1612 unsigned int *plen, unsigned int off)
1617 n = *poff / PAGE_SIZE;
1619 *page = nth_page(*page, n);
1621 *poff = *poff % PAGE_SIZE;
1625 static inline int __splice_segment(struct page *page, unsigned int poff,
1626 unsigned int plen, unsigned int *off,
1627 unsigned int *len, struct sk_buff *skb,
1628 struct splice_pipe_desc *spd, int linear,
1630 struct pipe_inode_info *pipe)
1635 /* skip this segment if already processed */
1641 /* ignore any bits we already processed */
1643 __segment_seek(&page, &poff, &plen, *off);
1648 unsigned int flen = min(*len, plen);
1650 /* the linear region may spread across several pages */
1651 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1653 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1656 __segment_seek(&page, &poff, &plen, flen);
1659 } while (*len && plen);
1665 * Map linear and fragment data from the skb to spd. It reports failure if the
1666 * pipe is full or if we already spliced the requested length.
1668 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1669 unsigned int *offset, unsigned int *len,
1670 struct splice_pipe_desc *spd, struct sock *sk)
1675 * map the linear part
1677 if (__splice_segment(virt_to_page(skb->data),
1678 (unsigned long) skb->data & (PAGE_SIZE - 1),
1680 offset, len, skb, spd, 1, sk, pipe))
1684 * then map the fragments
1686 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1687 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1689 if (__splice_segment(skb_frag_page(f),
1690 f->page_offset, skb_frag_size(f),
1691 offset, len, skb, spd, 0, sk, pipe))
1699 * Map data from the skb to a pipe. Should handle both the linear part,
1700 * the fragments, and the frag list. It does NOT handle frag lists within
1701 * the frag list, if such a thing exists. We'd probably need to recurse to
1702 * handle that cleanly.
1704 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1705 struct pipe_inode_info *pipe, unsigned int tlen,
1708 struct partial_page partial[PIPE_DEF_BUFFERS];
1709 struct page *pages[PIPE_DEF_BUFFERS];
1710 struct splice_pipe_desc spd = {
1714 .ops = &sock_pipe_buf_ops,
1715 .spd_release = sock_spd_release,
1717 struct sk_buff *frag_iter;
1718 struct sock *sk = skb->sk;
1721 if (splice_grow_spd(pipe, &spd))
1725 * __skb_splice_bits() only fails if the output has no room left,
1726 * so no point in going over the frag_list for the error case.
1728 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1734 * now see if we have a frag_list to map
1736 skb_walk_frags(skb, frag_iter) {
1739 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1746 * Drop the socket lock, otherwise we have reverse
1747 * locking dependencies between sk_lock and i_mutex
1748 * here as compared to sendfile(). We enter here
1749 * with the socket lock held, and splice_to_pipe() will
1750 * grab the pipe inode lock. For sendfile() emulation,
1751 * we call into ->sendpage() with the i_mutex lock held
1752 * and networking will grab the socket lock.
1755 ret = splice_to_pipe(pipe, &spd);
1759 splice_shrink_spd(pipe, &spd);
1764 * skb_store_bits - store bits from kernel buffer to skb
1765 * @skb: destination buffer
1766 * @offset: offset in destination
1767 * @from: source buffer
1768 * @len: number of bytes to copy
1770 * Copy the specified number of bytes from the source buffer to the
1771 * destination skb. This function handles all the messy bits of
1772 * traversing fragment lists and such.
1775 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1777 int start = skb_headlen(skb);
1778 struct sk_buff *frag_iter;
1781 if (offset > (int)skb->len - len)
1784 if ((copy = start - offset) > 0) {
1787 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1788 if ((len -= copy) == 0)
1794 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1795 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1798 WARN_ON(start > offset + len);
1800 end = start + skb_frag_size(frag);
1801 if ((copy = end - offset) > 0) {
1807 vaddr = kmap_atomic(skb_frag_page(frag));
1808 memcpy(vaddr + frag->page_offset + offset - start,
1810 kunmap_atomic(vaddr);
1812 if ((len -= copy) == 0)
1820 skb_walk_frags(skb, frag_iter) {
1823 WARN_ON(start > offset + len);
1825 end = start + frag_iter->len;
1826 if ((copy = end - offset) > 0) {
1829 if (skb_store_bits(frag_iter, offset - start,
1832 if ((len -= copy) == 0)
1845 EXPORT_SYMBOL(skb_store_bits);
1847 /* Checksum skb data. */
1849 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1850 int len, __wsum csum)
1852 int start = skb_headlen(skb);
1853 int i, copy = start - offset;
1854 struct sk_buff *frag_iter;
1857 /* Checksum header. */
1861 csum = csum_partial(skb->data + offset, copy, csum);
1862 if ((len -= copy) == 0)
1868 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1870 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1872 WARN_ON(start > offset + len);
1874 end = start + skb_frag_size(frag);
1875 if ((copy = end - offset) > 0) {
1881 vaddr = kmap_atomic(skb_frag_page(frag));
1882 csum2 = csum_partial(vaddr + frag->page_offset +
1883 offset - start, copy, 0);
1884 kunmap_atomic(vaddr);
1885 csum = csum_block_add(csum, csum2, pos);
1894 skb_walk_frags(skb, frag_iter) {
1897 WARN_ON(start > offset + len);
1899 end = start + frag_iter->len;
1900 if ((copy = end - offset) > 0) {
1904 csum2 = skb_checksum(frag_iter, offset - start,
1906 csum = csum_block_add(csum, csum2, pos);
1907 if ((len -= copy) == 0)
1918 EXPORT_SYMBOL(skb_checksum);
1920 /* Both of above in one bottle. */
1922 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1923 u8 *to, int len, __wsum csum)
1925 int start = skb_headlen(skb);
1926 int i, copy = start - offset;
1927 struct sk_buff *frag_iter;
1934 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1936 if ((len -= copy) == 0)
1943 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1946 WARN_ON(start > offset + len);
1948 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1949 if ((copy = end - offset) > 0) {
1952 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1956 vaddr = kmap_atomic(skb_frag_page(frag));
1957 csum2 = csum_partial_copy_nocheck(vaddr +
1961 kunmap_atomic(vaddr);
1962 csum = csum_block_add(csum, csum2, pos);
1972 skb_walk_frags(skb, frag_iter) {
1976 WARN_ON(start > offset + len);
1978 end = start + frag_iter->len;
1979 if ((copy = end - offset) > 0) {
1982 csum2 = skb_copy_and_csum_bits(frag_iter,
1985 csum = csum_block_add(csum, csum2, pos);
1986 if ((len -= copy) == 0)
1997 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1999 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2004 if (skb->ip_summed == CHECKSUM_PARTIAL)
2005 csstart = skb_checksum_start_offset(skb);
2007 csstart = skb_headlen(skb);
2009 BUG_ON(csstart > skb_headlen(skb));
2011 skb_copy_from_linear_data(skb, to, csstart);
2014 if (csstart != skb->len)
2015 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2016 skb->len - csstart, 0);
2018 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2019 long csstuff = csstart + skb->csum_offset;
2021 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2024 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2027 * skb_dequeue - remove from the head of the queue
2028 * @list: list to dequeue from
2030 * Remove the head of the list. The list lock is taken so the function
2031 * may be used safely with other locking list functions. The head item is
2032 * returned or %NULL if the list is empty.
2035 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2037 unsigned long flags;
2038 struct sk_buff *result;
2040 spin_lock_irqsave(&list->lock, flags);
2041 result = __skb_dequeue(list);
2042 spin_unlock_irqrestore(&list->lock, flags);
2045 EXPORT_SYMBOL(skb_dequeue);
2048 * skb_dequeue_tail - remove from the tail of the queue
2049 * @list: list to dequeue from
2051 * Remove the tail of the list. The list lock is taken so the function
2052 * may be used safely with other locking list functions. The tail item is
2053 * returned or %NULL if the list is empty.
2055 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2057 unsigned long flags;
2058 struct sk_buff *result;
2060 spin_lock_irqsave(&list->lock, flags);
2061 result = __skb_dequeue_tail(list);
2062 spin_unlock_irqrestore(&list->lock, flags);
2065 EXPORT_SYMBOL(skb_dequeue_tail);
2068 * skb_queue_purge - empty a list
2069 * @list: list to empty
2071 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2072 * the list and one reference dropped. This function takes the list
2073 * lock and is atomic with respect to other list locking functions.
2075 void skb_queue_purge(struct sk_buff_head *list)
2077 struct sk_buff *skb;
2078 while ((skb = skb_dequeue(list)) != NULL)
2081 EXPORT_SYMBOL(skb_queue_purge);
2084 * skb_queue_head - queue a buffer at the list head
2085 * @list: list to use
2086 * @newsk: buffer to queue
2088 * Queue a buffer at the start of the list. This function takes the
2089 * list lock and can be used safely with other locking &sk_buff functions
2092 * A buffer cannot be placed on two lists at the same time.
2094 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2096 unsigned long flags;
2098 spin_lock_irqsave(&list->lock, flags);
2099 __skb_queue_head(list, newsk);
2100 spin_unlock_irqrestore(&list->lock, flags);
2102 EXPORT_SYMBOL(skb_queue_head);
2105 * skb_queue_tail - queue a buffer at the list tail
2106 * @list: list to use
2107 * @newsk: buffer to queue
2109 * Queue a buffer at the tail of the list. This function takes the
2110 * list lock and can be used safely with other locking &sk_buff functions
2113 * A buffer cannot be placed on two lists at the same time.
2115 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2117 unsigned long flags;
2119 spin_lock_irqsave(&list->lock, flags);
2120 __skb_queue_tail(list, newsk);
2121 spin_unlock_irqrestore(&list->lock, flags);
2123 EXPORT_SYMBOL(skb_queue_tail);
2126 * skb_unlink - remove a buffer from a list
2127 * @skb: buffer to remove
2128 * @list: list to use
2130 * Remove a packet from a list. The list locks are taken and this
2131 * function is atomic with respect to other list locked calls
2133 * You must know what list the SKB is on.
2135 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2137 unsigned long flags;
2139 spin_lock_irqsave(&list->lock, flags);
2140 __skb_unlink(skb, list);
2141 spin_unlock_irqrestore(&list->lock, flags);
2143 EXPORT_SYMBOL(skb_unlink);
2146 * skb_append - append a buffer
2147 * @old: buffer to insert after
2148 * @newsk: buffer to insert
2149 * @list: list to use
2151 * Place a packet after a given packet in a list. The list locks are taken
2152 * and this function is atomic with respect to other list locked calls.
2153 * A buffer cannot be placed on two lists at the same time.
2155 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2157 unsigned long flags;
2159 spin_lock_irqsave(&list->lock, flags);
2160 __skb_queue_after(list, old, newsk);
2161 spin_unlock_irqrestore(&list->lock, flags);
2163 EXPORT_SYMBOL(skb_append);
2166 * skb_insert - insert a buffer
2167 * @old: buffer to insert before
2168 * @newsk: buffer to insert
2169 * @list: list to use
2171 * Place a packet before a given packet in a list. The list locks are
2172 * taken and this function is atomic with respect to other list locked
2175 * A buffer cannot be placed on two lists at the same time.
2177 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2179 unsigned long flags;
2181 spin_lock_irqsave(&list->lock, flags);
2182 __skb_insert(newsk, old->prev, old, list);
2183 spin_unlock_irqrestore(&list->lock, flags);
2185 EXPORT_SYMBOL(skb_insert);
2187 static inline void skb_split_inside_header(struct sk_buff *skb,
2188 struct sk_buff* skb1,
2189 const u32 len, const int pos)
2193 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2195 /* And move data appendix as is. */
2196 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2197 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2199 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2200 skb_shinfo(skb)->nr_frags = 0;
2201 skb1->data_len = skb->data_len;
2202 skb1->len += skb1->data_len;
2205 skb_set_tail_pointer(skb, len);
2208 static inline void skb_split_no_header(struct sk_buff *skb,
2209 struct sk_buff* skb1,
2210 const u32 len, int pos)
2213 const int nfrags = skb_shinfo(skb)->nr_frags;
2215 skb_shinfo(skb)->nr_frags = 0;
2216 skb1->len = skb1->data_len = skb->len - len;
2218 skb->data_len = len - pos;
2220 for (i = 0; i < nfrags; i++) {
2221 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2223 if (pos + size > len) {
2224 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2228 * We have two variants in this case:
2229 * 1. Move all the frag to the second
2230 * part, if it is possible. F.e.
2231 * this approach is mandatory for TUX,
2232 * where splitting is expensive.
2233 * 2. Split is accurately. We make this.
2235 skb_frag_ref(skb, i);
2236 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2237 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2238 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2239 skb_shinfo(skb)->nr_frags++;
2243 skb_shinfo(skb)->nr_frags++;
2246 skb_shinfo(skb1)->nr_frags = k;
2250 * skb_split - Split fragmented skb to two parts at length len.
2251 * @skb: the buffer to split
2252 * @skb1: the buffer to receive the second part
2253 * @len: new length for skb
2255 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2257 int pos = skb_headlen(skb);
2259 if (len < pos) /* Split line is inside header. */
2260 skb_split_inside_header(skb, skb1, len, pos);
2261 else /* Second chunk has no header, nothing to copy. */
2262 skb_split_no_header(skb, skb1, len, pos);
2264 EXPORT_SYMBOL(skb_split);
2266 /* Shifting from/to a cloned skb is a no-go.
2268 * Caller cannot keep skb_shinfo related pointers past calling here!
2270 static int skb_prepare_for_shift(struct sk_buff *skb)
2272 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2276 * skb_shift - Shifts paged data partially from skb to another
2277 * @tgt: buffer into which tail data gets added
2278 * @skb: buffer from which the paged data comes from
2279 * @shiftlen: shift up to this many bytes
2281 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2282 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2283 * It's up to caller to free skb if everything was shifted.
2285 * If @tgt runs out of frags, the whole operation is aborted.
2287 * Skb cannot include anything else but paged data while tgt is allowed
2288 * to have non-paged data as well.
2290 * TODO: full sized shift could be optimized but that would need
2291 * specialized skb free'er to handle frags without up-to-date nr_frags.
2293 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2295 int from, to, merge, todo;
2296 struct skb_frag_struct *fragfrom, *fragto;
2298 BUG_ON(shiftlen > skb->len);
2299 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2303 to = skb_shinfo(tgt)->nr_frags;
2304 fragfrom = &skb_shinfo(skb)->frags[from];
2306 /* Actual merge is delayed until the point when we know we can
2307 * commit all, so that we don't have to undo partial changes
2310 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2311 fragfrom->page_offset)) {
2316 todo -= skb_frag_size(fragfrom);
2318 if (skb_prepare_for_shift(skb) ||
2319 skb_prepare_for_shift(tgt))
2322 /* All previous frag pointers might be stale! */
2323 fragfrom = &skb_shinfo(skb)->frags[from];
2324 fragto = &skb_shinfo(tgt)->frags[merge];
2326 skb_frag_size_add(fragto, shiftlen);
2327 skb_frag_size_sub(fragfrom, shiftlen);
2328 fragfrom->page_offset += shiftlen;
2336 /* Skip full, not-fitting skb to avoid expensive operations */
2337 if ((shiftlen == skb->len) &&
2338 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2341 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2344 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2345 if (to == MAX_SKB_FRAGS)
2348 fragfrom = &skb_shinfo(skb)->frags[from];
2349 fragto = &skb_shinfo(tgt)->frags[to];
2351 if (todo >= skb_frag_size(fragfrom)) {
2352 *fragto = *fragfrom;
2353 todo -= skb_frag_size(fragfrom);
2358 __skb_frag_ref(fragfrom);
2359 fragto->page = fragfrom->page;
2360 fragto->page_offset = fragfrom->page_offset;
2361 skb_frag_size_set(fragto, todo);
2363 fragfrom->page_offset += todo;
2364 skb_frag_size_sub(fragfrom, todo);
2372 /* Ready to "commit" this state change to tgt */
2373 skb_shinfo(tgt)->nr_frags = to;
2376 fragfrom = &skb_shinfo(skb)->frags[0];
2377 fragto = &skb_shinfo(tgt)->frags[merge];
2379 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2380 __skb_frag_unref(fragfrom);
2383 /* Reposition in the original skb */
2385 while (from < skb_shinfo(skb)->nr_frags)
2386 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2387 skb_shinfo(skb)->nr_frags = to;
2389 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2392 /* Most likely the tgt won't ever need its checksum anymore, skb on
2393 * the other hand might need it if it needs to be resent
2395 tgt->ip_summed = CHECKSUM_PARTIAL;
2396 skb->ip_summed = CHECKSUM_PARTIAL;
2398 /* Yak, is it really working this way? Some helper please? */
2399 skb->len -= shiftlen;
2400 skb->data_len -= shiftlen;
2401 skb->truesize -= shiftlen;
2402 tgt->len += shiftlen;
2403 tgt->data_len += shiftlen;
2404 tgt->truesize += shiftlen;
2410 * skb_prepare_seq_read - Prepare a sequential read of skb data
2411 * @skb: the buffer to read
2412 * @from: lower offset of data to be read
2413 * @to: upper offset of data to be read
2414 * @st: state variable
2416 * Initializes the specified state variable. Must be called before
2417 * invoking skb_seq_read() for the first time.
2419 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2420 unsigned int to, struct skb_seq_state *st)
2422 st->lower_offset = from;
2423 st->upper_offset = to;
2424 st->root_skb = st->cur_skb = skb;
2425 st->frag_idx = st->stepped_offset = 0;
2426 st->frag_data = NULL;
2428 EXPORT_SYMBOL(skb_prepare_seq_read);
2431 * skb_seq_read - Sequentially read skb data
2432 * @consumed: number of bytes consumed by the caller so far
2433 * @data: destination pointer for data to be returned
2434 * @st: state variable
2436 * Reads a block of skb data at &consumed relative to the
2437 * lower offset specified to skb_prepare_seq_read(). Assigns
2438 * the head of the data block to &data and returns the length
2439 * of the block or 0 if the end of the skb data or the upper
2440 * offset has been reached.
2442 * The caller is not required to consume all of the data
2443 * returned, i.e. &consumed is typically set to the number
2444 * of bytes already consumed and the next call to
2445 * skb_seq_read() will return the remaining part of the block.
2447 * Note 1: The size of each block of data returned can be arbitrary,
2448 * this limitation is the cost for zerocopy seqeuental
2449 * reads of potentially non linear data.
2451 * Note 2: Fragment lists within fragments are not implemented
2452 * at the moment, state->root_skb could be replaced with
2453 * a stack for this purpose.
2455 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2456 struct skb_seq_state *st)
2458 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2461 if (unlikely(abs_offset >= st->upper_offset))
2465 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2467 if (abs_offset < block_limit && !st->frag_data) {
2468 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2469 return block_limit - abs_offset;
2472 if (st->frag_idx == 0 && !st->frag_data)
2473 st->stepped_offset += skb_headlen(st->cur_skb);
2475 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2476 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2477 block_limit = skb_frag_size(frag) + st->stepped_offset;
2479 if (abs_offset < block_limit) {
2481 st->frag_data = kmap_atomic(skb_frag_page(frag));
2483 *data = (u8 *) st->frag_data + frag->page_offset +
2484 (abs_offset - st->stepped_offset);
2486 return block_limit - abs_offset;
2489 if (st->frag_data) {
2490 kunmap_atomic(st->frag_data);
2491 st->frag_data = NULL;
2495 st->stepped_offset += skb_frag_size(frag);
2498 if (st->frag_data) {
2499 kunmap_atomic(st->frag_data);
2500 st->frag_data = NULL;
2503 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2504 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2507 } else if (st->cur_skb->next) {
2508 st->cur_skb = st->cur_skb->next;
2515 EXPORT_SYMBOL(skb_seq_read);
2518 * skb_abort_seq_read - Abort a sequential read of skb data
2519 * @st: state variable
2521 * Must be called if skb_seq_read() was not called until it
2524 void skb_abort_seq_read(struct skb_seq_state *st)
2527 kunmap_atomic(st->frag_data);
2529 EXPORT_SYMBOL(skb_abort_seq_read);
2531 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2533 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2534 struct ts_config *conf,
2535 struct ts_state *state)
2537 return skb_seq_read(offset, text, TS_SKB_CB(state));
2540 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2542 skb_abort_seq_read(TS_SKB_CB(state));
2546 * skb_find_text - Find a text pattern in skb data
2547 * @skb: the buffer to look in
2548 * @from: search offset
2550 * @config: textsearch configuration
2551 * @state: uninitialized textsearch state variable
2553 * Finds a pattern in the skb data according to the specified
2554 * textsearch configuration. Use textsearch_next() to retrieve
2555 * subsequent occurrences of the pattern. Returns the offset
2556 * to the first occurrence or UINT_MAX if no match was found.
2558 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2559 unsigned int to, struct ts_config *config,
2560 struct ts_state *state)
2564 config->get_next_block = skb_ts_get_next_block;
2565 config->finish = skb_ts_finish;
2567 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2569 ret = textsearch_find(config, state);
2570 return (ret <= to - from ? ret : UINT_MAX);
2572 EXPORT_SYMBOL(skb_find_text);
2575 * skb_append_datato_frags: - append the user data to a skb
2576 * @sk: sock structure
2577 * @skb: skb structure to be appened with user data.
2578 * @getfrag: call back function to be used for getting the user data
2579 * @from: pointer to user message iov
2580 * @length: length of the iov message
2582 * Description: This procedure append the user data in the fragment part
2583 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2585 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2586 int (*getfrag)(void *from, char *to, int offset,
2587 int len, int odd, struct sk_buff *skb),
2588 void *from, int length)
2591 skb_frag_t *frag = NULL;
2592 struct page *page = NULL;
2598 /* Return error if we don't have space for new frag */
2599 frg_cnt = skb_shinfo(skb)->nr_frags;
2600 if (frg_cnt >= MAX_SKB_FRAGS)
2603 /* allocate a new page for next frag */
2604 page = alloc_pages(sk->sk_allocation, 0);
2606 /* If alloc_page fails just return failure and caller will
2607 * free previous allocated pages by doing kfree_skb()
2612 /* initialize the next frag */
2613 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2614 skb->truesize += PAGE_SIZE;
2615 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2617 /* get the new initialized frag */
2618 frg_cnt = skb_shinfo(skb)->nr_frags;
2619 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2621 /* copy the user data to page */
2622 left = PAGE_SIZE - frag->page_offset;
2623 copy = (length > left)? left : length;
2625 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2626 offset, copy, 0, skb);
2630 /* copy was successful so update the size parameters */
2631 skb_frag_size_add(frag, copy);
2633 skb->data_len += copy;
2637 } while (length > 0);
2641 EXPORT_SYMBOL(skb_append_datato_frags);
2644 * skb_pull_rcsum - pull skb and update receive checksum
2645 * @skb: buffer to update
2646 * @len: length of data pulled
2648 * This function performs an skb_pull on the packet and updates
2649 * the CHECKSUM_COMPLETE checksum. It should be used on
2650 * receive path processing instead of skb_pull unless you know
2651 * that the checksum difference is zero (e.g., a valid IP header)
2652 * or you are setting ip_summed to CHECKSUM_NONE.
2654 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2656 BUG_ON(len > skb->len);
2658 BUG_ON(skb->len < skb->data_len);
2659 skb_postpull_rcsum(skb, skb->data, len);
2660 return skb->data += len;
2662 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2665 * skb_segment - Perform protocol segmentation on skb.
2666 * @skb: buffer to segment
2667 * @features: features for the output path (see dev->features)
2669 * This function performs segmentation on the given skb. It returns
2670 * a pointer to the first in a list of new skbs for the segments.
2671 * In case of error it returns ERR_PTR(err).
2673 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2675 struct sk_buff *segs = NULL;
2676 struct sk_buff *tail = NULL;
2677 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2678 unsigned int mss = skb_shinfo(skb)->gso_size;
2679 unsigned int doffset = skb->data - skb_mac_header(skb);
2680 unsigned int offset = doffset;
2681 unsigned int headroom;
2683 int sg = !!(features & NETIF_F_SG);
2684 int nfrags = skb_shinfo(skb)->nr_frags;
2689 __skb_push(skb, doffset);
2690 headroom = skb_headroom(skb);
2691 pos = skb_headlen(skb);
2694 struct sk_buff *nskb;
2699 len = skb->len - offset;
2703 hsize = skb_headlen(skb) - offset;
2706 if (hsize > len || !sg)
2709 if (!hsize && i >= nfrags) {
2710 BUG_ON(fskb->len != len);
2713 nskb = skb_clone(fskb, GFP_ATOMIC);
2716 if (unlikely(!nskb))
2719 hsize = skb_end_pointer(nskb) - nskb->head;
2720 if (skb_cow_head(nskb, doffset + headroom)) {
2725 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2727 skb_release_head_state(nskb);
2728 __skb_push(nskb, doffset);
2730 nskb = alloc_skb(hsize + doffset + headroom,
2733 if (unlikely(!nskb))
2736 skb_reserve(nskb, headroom);
2737 __skb_put(nskb, doffset);
2746 __copy_skb_header(nskb, skb);
2747 nskb->mac_len = skb->mac_len;
2749 /* nskb and skb might have different headroom */
2750 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2751 nskb->csum_start += skb_headroom(nskb) - headroom;
2753 skb_reset_mac_header(nskb);
2754 skb_set_network_header(nskb, skb->mac_len);
2755 nskb->transport_header = (nskb->network_header +
2756 skb_network_header_len(skb));
2757 skb_copy_from_linear_data(skb, nskb->data, doffset);
2759 if (fskb != skb_shinfo(skb)->frag_list)
2763 nskb->ip_summed = CHECKSUM_NONE;
2764 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2770 frag = skb_shinfo(nskb)->frags;
2772 skb_copy_from_linear_data_offset(skb, offset,
2773 skb_put(nskb, hsize), hsize);
2775 while (pos < offset + len && i < nfrags) {
2776 *frag = skb_shinfo(skb)->frags[i];
2777 __skb_frag_ref(frag);
2778 size = skb_frag_size(frag);
2781 frag->page_offset += offset - pos;
2782 skb_frag_size_sub(frag, offset - pos);
2785 skb_shinfo(nskb)->nr_frags++;
2787 if (pos + size <= offset + len) {
2791 skb_frag_size_sub(frag, pos + size - (offset + len));
2798 if (pos < offset + len) {
2799 struct sk_buff *fskb2 = fskb;
2801 BUG_ON(pos + fskb->len != offset + len);
2807 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2813 SKB_FRAG_ASSERT(nskb);
2814 skb_shinfo(nskb)->frag_list = fskb2;
2818 nskb->data_len = len - hsize;
2819 nskb->len += nskb->data_len;
2820 nskb->truesize += nskb->data_len;
2821 } while ((offset += len) < skb->len);
2826 while ((skb = segs)) {
2830 return ERR_PTR(err);
2832 EXPORT_SYMBOL_GPL(skb_segment);
2834 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2836 struct sk_buff *p = *head;
2837 struct sk_buff *nskb;
2838 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2839 struct skb_shared_info *pinfo = skb_shinfo(p);
2840 unsigned int headroom;
2841 unsigned int len = skb_gro_len(skb);
2842 unsigned int offset = skb_gro_offset(skb);
2843 unsigned int headlen = skb_headlen(skb);
2845 if (p->len + len >= 65536)
2848 if (pinfo->frag_list)
2850 else if (headlen <= offset) {
2853 int i = skbinfo->nr_frags;
2854 int nr_frags = pinfo->nr_frags + i;
2858 if (nr_frags > MAX_SKB_FRAGS)
2861 pinfo->nr_frags = nr_frags;
2862 skbinfo->nr_frags = 0;
2864 frag = pinfo->frags + nr_frags;
2865 frag2 = skbinfo->frags + i;
2870 frag->page_offset += offset;
2871 skb_frag_size_sub(frag, offset);
2873 skb->truesize -= skb->data_len;
2874 skb->len -= skb->data_len;
2877 NAPI_GRO_CB(skb)->free = 1;
2879 } else if (skb_gro_len(p) != pinfo->gso_size)
2882 headroom = skb_headroom(p);
2883 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2884 if (unlikely(!nskb))
2887 __copy_skb_header(nskb, p);
2888 nskb->mac_len = p->mac_len;
2890 skb_reserve(nskb, headroom);
2891 __skb_put(nskb, skb_gro_offset(p));
2893 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2894 skb_set_network_header(nskb, skb_network_offset(p));
2895 skb_set_transport_header(nskb, skb_transport_offset(p));
2897 __skb_pull(p, skb_gro_offset(p));
2898 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2899 p->data - skb_mac_header(p));
2901 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2902 skb_shinfo(nskb)->frag_list = p;
2903 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2904 pinfo->gso_size = 0;
2905 skb_header_release(p);
2908 nskb->data_len += p->len;
2909 nskb->truesize += p->truesize;
2910 nskb->len += p->len;
2913 nskb->next = p->next;
2919 p->truesize += skb->truesize - len;
2920 if (offset > headlen) {
2921 unsigned int eat = offset - headlen;
2923 skbinfo->frags[0].page_offset += eat;
2924 skb_frag_size_sub(&skbinfo->frags[0], eat);
2925 skb->data_len -= eat;
2930 __skb_pull(skb, offset);
2932 p->prev->next = skb;
2934 skb_header_release(skb);
2937 NAPI_GRO_CB(p)->count++;
2942 NAPI_GRO_CB(skb)->same_flow = 1;
2945 EXPORT_SYMBOL_GPL(skb_gro_receive);
2947 void __init skb_init(void)
2949 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2950 sizeof(struct sk_buff),
2952 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2954 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2955 (2*sizeof(struct sk_buff)) +
2958 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2963 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2964 * @skb: Socket buffer containing the buffers to be mapped
2965 * @sg: The scatter-gather list to map into
2966 * @offset: The offset into the buffer's contents to start mapping
2967 * @len: Length of buffer space to be mapped
2969 * Fill the specified scatter-gather list with mappings/pointers into a
2970 * region of the buffer space attached to a socket buffer.
2973 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2975 int start = skb_headlen(skb);
2976 int i, copy = start - offset;
2977 struct sk_buff *frag_iter;
2983 sg_set_buf(sg, skb->data + offset, copy);
2985 if ((len -= copy) == 0)
2990 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2993 WARN_ON(start > offset + len);
2995 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2996 if ((copy = end - offset) > 0) {
2997 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3001 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3002 frag->page_offset+offset-start);
3011 skb_walk_frags(skb, frag_iter) {
3014 WARN_ON(start > offset + len);
3016 end = start + frag_iter->len;
3017 if ((copy = end - offset) > 0) {
3020 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3022 if ((len -= copy) == 0)
3032 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3034 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3036 sg_mark_end(&sg[nsg - 1]);
3040 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3043 * skb_cow_data - Check that a socket buffer's data buffers are writable
3044 * @skb: The socket buffer to check.
3045 * @tailbits: Amount of trailing space to be added
3046 * @trailer: Returned pointer to the skb where the @tailbits space begins
3048 * Make sure that the data buffers attached to a socket buffer are
3049 * writable. If they are not, private copies are made of the data buffers
3050 * and the socket buffer is set to use these instead.
3052 * If @tailbits is given, make sure that there is space to write @tailbits
3053 * bytes of data beyond current end of socket buffer. @trailer will be
3054 * set to point to the skb in which this space begins.
3056 * The number of scatterlist elements required to completely map the
3057 * COW'd and extended socket buffer will be returned.
3059 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3063 struct sk_buff *skb1, **skb_p;
3065 /* If skb is cloned or its head is paged, reallocate
3066 * head pulling out all the pages (pages are considered not writable
3067 * at the moment even if they are anonymous).
3069 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3070 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3073 /* Easy case. Most of packets will go this way. */
3074 if (!skb_has_frag_list(skb)) {
3075 /* A little of trouble, not enough of space for trailer.
3076 * This should not happen, when stack is tuned to generate
3077 * good frames. OK, on miss we reallocate and reserve even more
3078 * space, 128 bytes is fair. */
3080 if (skb_tailroom(skb) < tailbits &&
3081 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3089 /* Misery. We are in troubles, going to mincer fragments... */
3092 skb_p = &skb_shinfo(skb)->frag_list;
3095 while ((skb1 = *skb_p) != NULL) {
3098 /* The fragment is partially pulled by someone,
3099 * this can happen on input. Copy it and everything
3102 if (skb_shared(skb1))
3105 /* If the skb is the last, worry about trailer. */
3107 if (skb1->next == NULL && tailbits) {
3108 if (skb_shinfo(skb1)->nr_frags ||
3109 skb_has_frag_list(skb1) ||
3110 skb_tailroom(skb1) < tailbits)
3111 ntail = tailbits + 128;
3117 skb_shinfo(skb1)->nr_frags ||
3118 skb_has_frag_list(skb1)) {
3119 struct sk_buff *skb2;
3121 /* Fuck, we are miserable poor guys... */
3123 skb2 = skb_copy(skb1, GFP_ATOMIC);
3125 skb2 = skb_copy_expand(skb1,
3129 if (unlikely(skb2 == NULL))
3133 skb_set_owner_w(skb2, skb1->sk);
3135 /* Looking around. Are we still alive?
3136 * OK, link new skb, drop old one */
3138 skb2->next = skb1->next;
3145 skb_p = &skb1->next;
3150 EXPORT_SYMBOL_GPL(skb_cow_data);
3152 static void sock_rmem_free(struct sk_buff *skb)
3154 struct sock *sk = skb->sk;
3156 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3160 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3162 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3166 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3167 (unsigned)sk->sk_rcvbuf)
3172 skb->destructor = sock_rmem_free;
3173 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3175 /* before exiting rcu section, make sure dst is refcounted */
3178 skb_queue_tail(&sk->sk_error_queue, skb);
3179 if (!sock_flag(sk, SOCK_DEAD))
3180 sk->sk_data_ready(sk, len);
3183 EXPORT_SYMBOL(sock_queue_err_skb);
3185 void skb_tstamp_tx(struct sk_buff *orig_skb,
3186 struct skb_shared_hwtstamps *hwtstamps)
3188 struct sock *sk = orig_skb->sk;
3189 struct sock_exterr_skb *serr;
3190 struct sk_buff *skb;
3196 skb = skb_clone(orig_skb, GFP_ATOMIC);
3201 *skb_hwtstamps(skb) =
3205 * no hardware time stamps available,
3206 * so keep the shared tx_flags and only
3207 * store software time stamp
3209 skb->tstamp = ktime_get_real();
3212 serr = SKB_EXT_ERR(skb);
3213 memset(serr, 0, sizeof(*serr));
3214 serr->ee.ee_errno = ENOMSG;
3215 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3217 err = sock_queue_err_skb(sk, skb);
3222 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3224 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3226 struct sock *sk = skb->sk;
3227 struct sock_exterr_skb *serr;
3230 skb->wifi_acked_valid = 1;
3231 skb->wifi_acked = acked;
3233 serr = SKB_EXT_ERR(skb);
3234 memset(serr, 0, sizeof(*serr));
3235 serr->ee.ee_errno = ENOMSG;
3236 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3238 err = sock_queue_err_skb(sk, skb);
3242 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3246 * skb_partial_csum_set - set up and verify partial csum values for packet
3247 * @skb: the skb to set
3248 * @start: the number of bytes after skb->data to start checksumming.
3249 * @off: the offset from start to place the checksum.
3251 * For untrusted partially-checksummed packets, we need to make sure the values
3252 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3254 * This function checks and sets those values and skb->ip_summed: if this
3255 * returns false you should drop the packet.
3257 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3259 if (unlikely(start > skb_headlen(skb)) ||
3260 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3261 if (net_ratelimit())
3263 "bad partial csum: csum=%u/%u len=%u\n",
3264 start, off, skb_headlen(skb));
3267 skb->ip_summed = CHECKSUM_PARTIAL;
3268 skb->csum_start = skb_headroom(skb) + start;
3269 skb->csum_offset = off;
3272 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3274 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3276 if (net_ratelimit())
3277 pr_warning("%s: received packets cannot be forwarded"
3278 " while LRO is enabled\n", skb->dev->name);
3280 EXPORT_SYMBOL(__skb_warn_lro_forwarding);