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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 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 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__, here, skb->len, sz, skb->head, skb->data,
125 (unsigned long)skb->tail, (unsigned long)skb->end,
126 skb->dev ? skb->dev->name : "<NULL>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__, 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
248 * @frag_size: size of fragment, or 0 if head was kmalloced
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
262 struct sk_buff *build_skb(void *data, unsigned int frag_size)
264 struct skb_shared_info *shinfo;
266 unsigned int size = frag_size ? : ksize(data);
268 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
272 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
274 memset(skb, 0, offsetof(struct sk_buff, tail));
275 skb->truesize = SKB_TRUESIZE(size);
276 skb->head_frag = frag_size != 0;
277 atomic_set(&skb->users, 1);
280 skb_reset_tail_pointer(skb);
281 skb->end = skb->tail + size;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb->mac_header = ~0U;
286 /* make sure we initialize shinfo sequentially */
287 shinfo = skb_shinfo(skb);
288 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
289 atomic_set(&shinfo->dataref, 1);
290 kmemcheck_annotate_variable(shinfo->destructor_arg);
294 EXPORT_SYMBOL(build_skb);
296 struct netdev_alloc_cache {
300 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
303 * netdev_alloc_frag - allocate a page fragment
304 * @fragsz: fragment size
306 * Allocates a frag from a page for receive buffer.
307 * Uses GFP_ATOMIC allocations.
309 void *netdev_alloc_frag(unsigned int fragsz)
311 struct netdev_alloc_cache *nc;
315 local_irq_save(flags);
316 nc = &__get_cpu_var(netdev_alloc_cache);
317 if (unlikely(!nc->page)) {
319 nc->page = alloc_page(GFP_ATOMIC | __GFP_COLD);
322 if (likely(nc->page)) {
323 if (nc->offset + fragsz > PAGE_SIZE) {
327 data = page_address(nc->page) + nc->offset;
328 nc->offset += fragsz;
331 local_irq_restore(flags);
334 EXPORT_SYMBOL(netdev_alloc_frag);
337 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
338 * @dev: network device to receive on
339 * @length: length to allocate
340 * @gfp_mask: get_free_pages mask, passed to alloc_skb
342 * Allocate a new &sk_buff and assign it a usage count of one. The
343 * buffer has unspecified headroom built in. Users should allocate
344 * the headroom they think they need without accounting for the
345 * built in space. The built in space is used for optimisations.
347 * %NULL is returned if there is no free memory.
349 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
350 unsigned int length, gfp_t gfp_mask)
352 struct sk_buff *skb = NULL;
353 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
354 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
356 if (fragsz <= PAGE_SIZE && !(gfp_mask & __GFP_WAIT)) {
357 void *data = netdev_alloc_frag(fragsz);
360 skb = build_skb(data, fragsz);
362 put_page(virt_to_head_page(data));
365 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
368 skb_reserve(skb, NET_SKB_PAD);
373 EXPORT_SYMBOL(__netdev_alloc_skb);
375 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
376 int size, unsigned int truesize)
378 skb_fill_page_desc(skb, i, page, off, size);
380 skb->data_len += size;
381 skb->truesize += truesize;
383 EXPORT_SYMBOL(skb_add_rx_frag);
385 static void skb_drop_list(struct sk_buff **listp)
387 struct sk_buff *list = *listp;
392 struct sk_buff *this = list;
398 static inline void skb_drop_fraglist(struct sk_buff *skb)
400 skb_drop_list(&skb_shinfo(skb)->frag_list);
403 static void skb_clone_fraglist(struct sk_buff *skb)
405 struct sk_buff *list;
407 skb_walk_frags(skb, list)
411 static void skb_free_head(struct sk_buff *skb)
414 put_page(virt_to_head_page(skb->head));
419 static void skb_release_data(struct sk_buff *skb)
422 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
423 &skb_shinfo(skb)->dataref)) {
424 if (skb_shinfo(skb)->nr_frags) {
426 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
427 skb_frag_unref(skb, i);
431 * If skb buf is from userspace, we need to notify the caller
432 * the lower device DMA has done;
434 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
435 struct ubuf_info *uarg;
437 uarg = skb_shinfo(skb)->destructor_arg;
439 uarg->callback(uarg);
442 if (skb_has_frag_list(skb))
443 skb_drop_fraglist(skb);
450 * Free an skbuff by memory without cleaning the state.
452 static void kfree_skbmem(struct sk_buff *skb)
454 struct sk_buff *other;
455 atomic_t *fclone_ref;
457 switch (skb->fclone) {
458 case SKB_FCLONE_UNAVAILABLE:
459 kmem_cache_free(skbuff_head_cache, skb);
462 case SKB_FCLONE_ORIG:
463 fclone_ref = (atomic_t *) (skb + 2);
464 if (atomic_dec_and_test(fclone_ref))
465 kmem_cache_free(skbuff_fclone_cache, skb);
468 case SKB_FCLONE_CLONE:
469 fclone_ref = (atomic_t *) (skb + 1);
472 /* The clone portion is available for
473 * fast-cloning again.
475 skb->fclone = SKB_FCLONE_UNAVAILABLE;
477 if (atomic_dec_and_test(fclone_ref))
478 kmem_cache_free(skbuff_fclone_cache, other);
483 static void skb_release_head_state(struct sk_buff *skb)
487 secpath_put(skb->sp);
489 if (skb->destructor) {
491 skb->destructor(skb);
493 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
494 nf_conntrack_put(skb->nfct);
496 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
497 nf_conntrack_put_reasm(skb->nfct_reasm);
499 #ifdef CONFIG_BRIDGE_NETFILTER
500 nf_bridge_put(skb->nf_bridge);
502 /* XXX: IS this still necessary? - JHS */
503 #ifdef CONFIG_NET_SCHED
505 #ifdef CONFIG_NET_CLS_ACT
511 /* Free everything but the sk_buff shell. */
512 static void skb_release_all(struct sk_buff *skb)
514 skb_release_head_state(skb);
515 skb_release_data(skb);
519 * __kfree_skb - private function
522 * Free an sk_buff. Release anything attached to the buffer.
523 * Clean the state. This is an internal helper function. Users should
524 * always call kfree_skb
527 void __kfree_skb(struct sk_buff *skb)
529 skb_release_all(skb);
532 EXPORT_SYMBOL(__kfree_skb);
535 * kfree_skb - free an sk_buff
536 * @skb: buffer to free
538 * Drop a reference to the buffer and free it if the usage count has
541 void kfree_skb(struct sk_buff *skb)
545 if (likely(atomic_read(&skb->users) == 1))
547 else if (likely(!atomic_dec_and_test(&skb->users)))
549 trace_kfree_skb(skb, __builtin_return_address(0));
552 EXPORT_SYMBOL(kfree_skb);
555 * consume_skb - free an skbuff
556 * @skb: buffer to free
558 * Drop a ref to the buffer and free it if the usage count has hit zero
559 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
560 * is being dropped after a failure and notes that
562 void consume_skb(struct sk_buff *skb)
566 if (likely(atomic_read(&skb->users) == 1))
568 else if (likely(!atomic_dec_and_test(&skb->users)))
570 trace_consume_skb(skb);
573 EXPORT_SYMBOL(consume_skb);
576 * skb_recycle - clean up an skb for reuse
579 * Recycles the skb to be reused as a receive buffer. This
580 * function does any necessary reference count dropping, and
581 * cleans up the skbuff as if it just came from __alloc_skb().
583 void skb_recycle(struct sk_buff *skb)
585 struct skb_shared_info *shinfo;
587 skb_release_head_state(skb);
589 shinfo = skb_shinfo(skb);
590 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
591 atomic_set(&shinfo->dataref, 1);
593 memset(skb, 0, offsetof(struct sk_buff, tail));
594 skb->data = skb->head + NET_SKB_PAD;
595 skb_reset_tail_pointer(skb);
597 EXPORT_SYMBOL(skb_recycle);
600 * skb_recycle_check - check if skb can be reused for receive
602 * @skb_size: minimum receive buffer size
604 * Checks that the skb passed in is not shared or cloned, and
605 * that it is linear and its head portion at least as large as
606 * skb_size so that it can be recycled as a receive buffer.
607 * If these conditions are met, this function does any necessary
608 * reference count dropping and cleans up the skbuff as if it
609 * just came from __alloc_skb().
611 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
613 if (!skb_is_recycleable(skb, skb_size))
620 EXPORT_SYMBOL(skb_recycle_check);
622 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
624 new->tstamp = old->tstamp;
626 new->transport_header = old->transport_header;
627 new->network_header = old->network_header;
628 new->mac_header = old->mac_header;
629 skb_dst_copy(new, old);
630 new->rxhash = old->rxhash;
631 new->ooo_okay = old->ooo_okay;
632 new->l4_rxhash = old->l4_rxhash;
633 new->no_fcs = old->no_fcs;
635 new->sp = secpath_get(old->sp);
637 memcpy(new->cb, old->cb, sizeof(old->cb));
638 new->csum = old->csum;
639 new->local_df = old->local_df;
640 new->pkt_type = old->pkt_type;
641 new->ip_summed = old->ip_summed;
642 skb_copy_queue_mapping(new, old);
643 new->priority = old->priority;
644 #if IS_ENABLED(CONFIG_IP_VS)
645 new->ipvs_property = old->ipvs_property;
647 new->protocol = old->protocol;
648 new->mark = old->mark;
649 new->skb_iif = old->skb_iif;
651 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
652 new->nf_trace = old->nf_trace;
654 #ifdef CONFIG_NET_SCHED
655 new->tc_index = old->tc_index;
656 #ifdef CONFIG_NET_CLS_ACT
657 new->tc_verd = old->tc_verd;
660 new->vlan_tci = old->vlan_tci;
662 skb_copy_secmark(new, old);
666 * You should not add any new code to this function. Add it to
667 * __copy_skb_header above instead.
669 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
671 #define C(x) n->x = skb->x
673 n->next = n->prev = NULL;
675 __copy_skb_header(n, skb);
680 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
683 n->destructor = NULL;
690 atomic_set(&n->users, 1);
692 atomic_inc(&(skb_shinfo(skb)->dataref));
700 * skb_morph - morph one skb into another
701 * @dst: the skb to receive the contents
702 * @src: the skb to supply the contents
704 * This is identical to skb_clone except that the target skb is
705 * supplied by the user.
707 * The target skb is returned upon exit.
709 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
711 skb_release_all(dst);
712 return __skb_clone(dst, src);
714 EXPORT_SYMBOL_GPL(skb_morph);
716 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
717 * @skb: the skb to modify
718 * @gfp_mask: allocation priority
720 * This must be called on SKBTX_DEV_ZEROCOPY skb.
721 * It will copy all frags into kernel and drop the reference
722 * to userspace pages.
724 * If this function is called from an interrupt gfp_mask() must be
727 * Returns 0 on success or a negative error code on failure
728 * to allocate kernel memory to copy to.
730 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
733 int num_frags = skb_shinfo(skb)->nr_frags;
734 struct page *page, *head = NULL;
735 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
737 for (i = 0; i < num_frags; i++) {
739 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
741 page = alloc_page(GFP_ATOMIC);
744 struct page *next = (struct page *)head->private;
750 vaddr = kmap_atomic(skb_frag_page(f));
751 memcpy(page_address(page),
752 vaddr + f->page_offset, skb_frag_size(f));
753 kunmap_atomic(vaddr);
754 page->private = (unsigned long)head;
758 /* skb frags release userspace buffers */
759 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
760 skb_frag_unref(skb, i);
762 uarg->callback(uarg);
764 /* skb frags point to kernel buffers */
765 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
766 __skb_fill_page_desc(skb, i-1, head, 0,
767 skb_shinfo(skb)->frags[i - 1].size);
768 head = (struct page *)head->private;
771 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
777 * skb_clone - duplicate an sk_buff
778 * @skb: buffer to clone
779 * @gfp_mask: allocation priority
781 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
782 * copies share the same packet data but not structure. The new
783 * buffer has a reference count of 1. If the allocation fails the
784 * function returns %NULL otherwise the new buffer is returned.
786 * If this function is called from an interrupt gfp_mask() must be
790 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
794 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
795 if (skb_copy_ubufs(skb, gfp_mask))
800 if (skb->fclone == SKB_FCLONE_ORIG &&
801 n->fclone == SKB_FCLONE_UNAVAILABLE) {
802 atomic_t *fclone_ref = (atomic_t *) (n + 1);
803 n->fclone = SKB_FCLONE_CLONE;
804 atomic_inc(fclone_ref);
806 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
810 kmemcheck_annotate_bitfield(n, flags1);
811 kmemcheck_annotate_bitfield(n, flags2);
812 n->fclone = SKB_FCLONE_UNAVAILABLE;
815 return __skb_clone(n, skb);
817 EXPORT_SYMBOL(skb_clone);
819 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
821 #ifndef NET_SKBUFF_DATA_USES_OFFSET
823 * Shift between the two data areas in bytes
825 unsigned long offset = new->data - old->data;
828 __copy_skb_header(new, old);
830 #ifndef NET_SKBUFF_DATA_USES_OFFSET
831 /* {transport,network,mac}_header are relative to skb->head */
832 new->transport_header += offset;
833 new->network_header += offset;
834 if (skb_mac_header_was_set(new))
835 new->mac_header += offset;
837 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
838 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
839 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
843 * skb_copy - create private copy of an sk_buff
844 * @skb: buffer to copy
845 * @gfp_mask: allocation priority
847 * Make a copy of both an &sk_buff and its data. This is used when the
848 * caller wishes to modify the data and needs a private copy of the
849 * data to alter. Returns %NULL on failure or the pointer to the buffer
850 * on success. The returned buffer has a reference count of 1.
852 * As by-product this function converts non-linear &sk_buff to linear
853 * one, so that &sk_buff becomes completely private and caller is allowed
854 * to modify all the data of returned buffer. This means that this
855 * function is not recommended for use in circumstances when only
856 * header is going to be modified. Use pskb_copy() instead.
859 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
861 int headerlen = skb_headroom(skb);
862 unsigned int size = skb_end_offset(skb) + skb->data_len;
863 struct sk_buff *n = alloc_skb(size, gfp_mask);
868 /* Set the data pointer */
869 skb_reserve(n, headerlen);
870 /* Set the tail pointer and length */
871 skb_put(n, skb->len);
873 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
876 copy_skb_header(n, skb);
879 EXPORT_SYMBOL(skb_copy);
882 * __pskb_copy - create copy of an sk_buff with private head.
883 * @skb: buffer to copy
884 * @headroom: headroom of new skb
885 * @gfp_mask: allocation priority
887 * Make a copy of both an &sk_buff and part of its data, located
888 * in header. Fragmented data remain shared. This is used when
889 * the caller wishes to modify only header of &sk_buff and needs
890 * private copy of the header to alter. Returns %NULL on failure
891 * or the pointer to the buffer on success.
892 * The returned buffer has a reference count of 1.
895 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
897 unsigned int size = skb_headlen(skb) + headroom;
898 struct sk_buff *n = alloc_skb(size, gfp_mask);
903 /* Set the data pointer */
904 skb_reserve(n, headroom);
905 /* Set the tail pointer and length */
906 skb_put(n, skb_headlen(skb));
908 skb_copy_from_linear_data(skb, n->data, n->len);
910 n->truesize += skb->data_len;
911 n->data_len = skb->data_len;
914 if (skb_shinfo(skb)->nr_frags) {
917 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
918 if (skb_copy_ubufs(skb, gfp_mask)) {
924 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
925 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
926 skb_frag_ref(skb, i);
928 skb_shinfo(n)->nr_frags = i;
931 if (skb_has_frag_list(skb)) {
932 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
933 skb_clone_fraglist(n);
936 copy_skb_header(n, skb);
940 EXPORT_SYMBOL(__pskb_copy);
943 * pskb_expand_head - reallocate header of &sk_buff
944 * @skb: buffer to reallocate
945 * @nhead: room to add at head
946 * @ntail: room to add at tail
947 * @gfp_mask: allocation priority
949 * Expands (or creates identical copy, if &nhead and &ntail are zero)
950 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
951 * reference count of 1. Returns zero in the case of success or error,
952 * if expansion failed. In the last case, &sk_buff is not changed.
954 * All the pointers pointing into skb header may change and must be
955 * reloaded after call to this function.
958 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
963 int size = nhead + skb_end_offset(skb) + ntail;
971 size = SKB_DATA_ALIGN(size);
973 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
977 size = SKB_WITH_OVERHEAD(ksize(data));
979 /* Copy only real data... and, alas, header. This should be
980 * optimized for the cases when header is void.
982 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
984 memcpy((struct skb_shared_info *)(data + size),
986 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
989 * if shinfo is shared we must drop the old head gracefully, but if it
990 * is not we can just drop the old head and let the existing refcount
991 * be since all we did is relocate the values
993 if (skb_cloned(skb)) {
994 /* copy this zero copy skb frags */
995 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
996 if (skb_copy_ubufs(skb, gfp_mask))
999 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1000 skb_frag_ref(skb, i);
1002 if (skb_has_frag_list(skb))
1003 skb_clone_fraglist(skb);
1005 skb_release_data(skb);
1009 off = (data + nhead) - skb->head;
1014 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1018 skb->end = skb->head + size;
1020 /* {transport,network,mac}_header and tail are relative to skb->head */
1022 skb->transport_header += off;
1023 skb->network_header += off;
1024 if (skb_mac_header_was_set(skb))
1025 skb->mac_header += off;
1026 /* Only adjust this if it actually is csum_start rather than csum */
1027 if (skb->ip_summed == CHECKSUM_PARTIAL)
1028 skb->csum_start += nhead;
1032 atomic_set(&skb_shinfo(skb)->dataref, 1);
1040 EXPORT_SYMBOL(pskb_expand_head);
1042 /* Make private copy of skb with writable head and some headroom */
1044 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1046 struct sk_buff *skb2;
1047 int delta = headroom - skb_headroom(skb);
1050 skb2 = pskb_copy(skb, GFP_ATOMIC);
1052 skb2 = skb_clone(skb, GFP_ATOMIC);
1053 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1061 EXPORT_SYMBOL(skb_realloc_headroom);
1064 * skb_copy_expand - copy and expand sk_buff
1065 * @skb: buffer to copy
1066 * @newheadroom: new free bytes at head
1067 * @newtailroom: new free bytes at tail
1068 * @gfp_mask: allocation priority
1070 * Make a copy of both an &sk_buff and its data and while doing so
1071 * allocate additional space.
1073 * This is used when the caller wishes to modify the data and needs a
1074 * private copy of the data to alter as well as more space for new fields.
1075 * Returns %NULL on failure or the pointer to the buffer
1076 * on success. The returned buffer has a reference count of 1.
1078 * You must pass %GFP_ATOMIC as the allocation priority if this function
1079 * is called from an interrupt.
1081 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1082 int newheadroom, int newtailroom,
1086 * Allocate the copy buffer
1088 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1090 int oldheadroom = skb_headroom(skb);
1091 int head_copy_len, head_copy_off;
1097 skb_reserve(n, newheadroom);
1099 /* Set the tail pointer and length */
1100 skb_put(n, skb->len);
1102 head_copy_len = oldheadroom;
1104 if (newheadroom <= head_copy_len)
1105 head_copy_len = newheadroom;
1107 head_copy_off = newheadroom - head_copy_len;
1109 /* Copy the linear header and data. */
1110 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1111 skb->len + head_copy_len))
1114 copy_skb_header(n, skb);
1116 off = newheadroom - oldheadroom;
1117 if (n->ip_summed == CHECKSUM_PARTIAL)
1118 n->csum_start += off;
1119 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1120 n->transport_header += off;
1121 n->network_header += off;
1122 if (skb_mac_header_was_set(skb))
1123 n->mac_header += off;
1128 EXPORT_SYMBOL(skb_copy_expand);
1131 * skb_pad - zero pad the tail of an skb
1132 * @skb: buffer to pad
1133 * @pad: space to pad
1135 * Ensure that a buffer is followed by a padding area that is zero
1136 * filled. Used by network drivers which may DMA or transfer data
1137 * beyond the buffer end onto the wire.
1139 * May return error in out of memory cases. The skb is freed on error.
1142 int skb_pad(struct sk_buff *skb, int pad)
1147 /* If the skbuff is non linear tailroom is always zero.. */
1148 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1149 memset(skb->data+skb->len, 0, pad);
1153 ntail = skb->data_len + pad - (skb->end - skb->tail);
1154 if (likely(skb_cloned(skb) || ntail > 0)) {
1155 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1160 /* FIXME: The use of this function with non-linear skb's really needs
1163 err = skb_linearize(skb);
1167 memset(skb->data + skb->len, 0, pad);
1174 EXPORT_SYMBOL(skb_pad);
1177 * skb_put - add data to a buffer
1178 * @skb: buffer to use
1179 * @len: amount of data to add
1181 * This function extends the used data area of the buffer. If this would
1182 * exceed the total buffer size the kernel will panic. A pointer to the
1183 * first byte of the extra data is returned.
1185 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1187 unsigned char *tmp = skb_tail_pointer(skb);
1188 SKB_LINEAR_ASSERT(skb);
1191 if (unlikely(skb->tail > skb->end))
1192 skb_over_panic(skb, len, __builtin_return_address(0));
1195 EXPORT_SYMBOL(skb_put);
1198 * skb_push - add data to the start of a buffer
1199 * @skb: buffer to use
1200 * @len: amount of data to add
1202 * This function extends the used data area of the buffer at the buffer
1203 * start. If this would exceed the total buffer headroom the kernel will
1204 * panic. A pointer to the first byte of the extra data is returned.
1206 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1210 if (unlikely(skb->data<skb->head))
1211 skb_under_panic(skb, len, __builtin_return_address(0));
1214 EXPORT_SYMBOL(skb_push);
1217 * skb_pull - remove data from the start of a buffer
1218 * @skb: buffer to use
1219 * @len: amount of data to remove
1221 * This function removes data from the start of a buffer, returning
1222 * the memory to the headroom. A pointer to the next data in the buffer
1223 * is returned. Once the data has been pulled future pushes will overwrite
1226 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1228 return skb_pull_inline(skb, len);
1230 EXPORT_SYMBOL(skb_pull);
1233 * skb_trim - remove end from a buffer
1234 * @skb: buffer to alter
1237 * Cut the length of a buffer down by removing data from the tail. If
1238 * the buffer is already under the length specified it is not modified.
1239 * The skb must be linear.
1241 void skb_trim(struct sk_buff *skb, unsigned int len)
1244 __skb_trim(skb, len);
1246 EXPORT_SYMBOL(skb_trim);
1248 /* Trims skb to length len. It can change skb pointers.
1251 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1253 struct sk_buff **fragp;
1254 struct sk_buff *frag;
1255 int offset = skb_headlen(skb);
1256 int nfrags = skb_shinfo(skb)->nr_frags;
1260 if (skb_cloned(skb) &&
1261 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1268 for (; i < nfrags; i++) {
1269 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1276 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1279 skb_shinfo(skb)->nr_frags = i;
1281 for (; i < nfrags; i++)
1282 skb_frag_unref(skb, i);
1284 if (skb_has_frag_list(skb))
1285 skb_drop_fraglist(skb);
1289 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1290 fragp = &frag->next) {
1291 int end = offset + frag->len;
1293 if (skb_shared(frag)) {
1294 struct sk_buff *nfrag;
1296 nfrag = skb_clone(frag, GFP_ATOMIC);
1297 if (unlikely(!nfrag))
1300 nfrag->next = frag->next;
1312 unlikely((err = pskb_trim(frag, len - offset))))
1316 skb_drop_list(&frag->next);
1321 if (len > skb_headlen(skb)) {
1322 skb->data_len -= skb->len - len;
1327 skb_set_tail_pointer(skb, len);
1332 EXPORT_SYMBOL(___pskb_trim);
1335 * __pskb_pull_tail - advance tail of skb header
1336 * @skb: buffer to reallocate
1337 * @delta: number of bytes to advance tail
1339 * The function makes a sense only on a fragmented &sk_buff,
1340 * it expands header moving its tail forward and copying necessary
1341 * data from fragmented part.
1343 * &sk_buff MUST have reference count of 1.
1345 * Returns %NULL (and &sk_buff does not change) if pull failed
1346 * or value of new tail of skb in the case of success.
1348 * All the pointers pointing into skb header may change and must be
1349 * reloaded after call to this function.
1352 /* Moves tail of skb head forward, copying data from fragmented part,
1353 * when it is necessary.
1354 * 1. It may fail due to malloc failure.
1355 * 2. It may change skb pointers.
1357 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1359 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1361 /* If skb has not enough free space at tail, get new one
1362 * plus 128 bytes for future expansions. If we have enough
1363 * room at tail, reallocate without expansion only if skb is cloned.
1365 int i, k, eat = (skb->tail + delta) - skb->end;
1367 if (eat > 0 || skb_cloned(skb)) {
1368 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1373 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1376 /* Optimization: no fragments, no reasons to preestimate
1377 * size of pulled pages. Superb.
1379 if (!skb_has_frag_list(skb))
1382 /* Estimate size of pulled pages. */
1384 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1385 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1392 /* If we need update frag list, we are in troubles.
1393 * Certainly, it possible to add an offset to skb data,
1394 * but taking into account that pulling is expected to
1395 * be very rare operation, it is worth to fight against
1396 * further bloating skb head and crucify ourselves here instead.
1397 * Pure masohism, indeed. 8)8)
1400 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1401 struct sk_buff *clone = NULL;
1402 struct sk_buff *insp = NULL;
1407 if (list->len <= eat) {
1408 /* Eaten as whole. */
1413 /* Eaten partially. */
1415 if (skb_shared(list)) {
1416 /* Sucks! We need to fork list. :-( */
1417 clone = skb_clone(list, GFP_ATOMIC);
1423 /* This may be pulled without
1427 if (!pskb_pull(list, eat)) {
1435 /* Free pulled out fragments. */
1436 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1437 skb_shinfo(skb)->frag_list = list->next;
1440 /* And insert new clone at head. */
1443 skb_shinfo(skb)->frag_list = clone;
1446 /* Success! Now we may commit changes to skb data. */
1451 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1452 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1455 skb_frag_unref(skb, i);
1458 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1460 skb_shinfo(skb)->frags[k].page_offset += eat;
1461 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1467 skb_shinfo(skb)->nr_frags = k;
1470 skb->data_len -= delta;
1472 return skb_tail_pointer(skb);
1474 EXPORT_SYMBOL(__pskb_pull_tail);
1477 * skb_copy_bits - copy bits from skb to kernel buffer
1479 * @offset: offset in source
1480 * @to: destination buffer
1481 * @len: number of bytes to copy
1483 * Copy the specified number of bytes from the source skb to the
1484 * destination buffer.
1487 * If its prototype is ever changed,
1488 * check arch/{*}/net/{*}.S files,
1489 * since it is called from BPF assembly code.
1491 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1493 int start = skb_headlen(skb);
1494 struct sk_buff *frag_iter;
1497 if (offset > (int)skb->len - len)
1501 if ((copy = start - offset) > 0) {
1504 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1505 if ((len -= copy) == 0)
1511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1513 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1515 WARN_ON(start > offset + len);
1517 end = start + skb_frag_size(f);
1518 if ((copy = end - offset) > 0) {
1524 vaddr = kmap_atomic(skb_frag_page(f));
1526 vaddr + f->page_offset + offset - start,
1528 kunmap_atomic(vaddr);
1530 if ((len -= copy) == 0)
1538 skb_walk_frags(skb, frag_iter) {
1541 WARN_ON(start > offset + len);
1543 end = start + frag_iter->len;
1544 if ((copy = end - offset) > 0) {
1547 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1549 if ((len -= copy) == 0)
1563 EXPORT_SYMBOL(skb_copy_bits);
1566 * Callback from splice_to_pipe(), if we need to release some pages
1567 * at the end of the spd in case we error'ed out in filling the pipe.
1569 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1571 put_page(spd->pages[i]);
1574 static struct page *linear_to_page(struct page *page, unsigned int *len,
1575 unsigned int *offset,
1576 struct sk_buff *skb, struct sock *sk)
1578 struct page *p = sk->sk_sndmsg_page;
1583 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1587 off = sk->sk_sndmsg_off = 0;
1588 /* hold one ref to this page until it's full */
1592 /* If we are the only user of the page, we can reset offset */
1593 if (page_count(p) == 1)
1594 sk->sk_sndmsg_off = 0;
1595 off = sk->sk_sndmsg_off;
1596 mlen = PAGE_SIZE - off;
1597 if (mlen < 64 && mlen < *len) {
1602 *len = min_t(unsigned int, *len, mlen);
1605 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1606 sk->sk_sndmsg_off += *len;
1612 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1614 unsigned int offset)
1616 return spd->nr_pages &&
1617 spd->pages[spd->nr_pages - 1] == page &&
1618 (spd->partial[spd->nr_pages - 1].offset +
1619 spd->partial[spd->nr_pages - 1].len == offset);
1623 * Fill page/offset/length into spd, if it can hold more pages.
1625 static bool spd_fill_page(struct splice_pipe_desc *spd,
1626 struct pipe_inode_info *pipe, struct page *page,
1627 unsigned int *len, unsigned int offset,
1628 struct sk_buff *skb, bool linear,
1631 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1635 page = linear_to_page(page, len, &offset, skb, sk);
1639 if (spd_can_coalesce(spd, page, offset)) {
1640 spd->partial[spd->nr_pages - 1].len += *len;
1644 spd->pages[spd->nr_pages] = page;
1645 spd->partial[spd->nr_pages].len = *len;
1646 spd->partial[spd->nr_pages].offset = offset;
1652 static inline void __segment_seek(struct page **page, unsigned int *poff,
1653 unsigned int *plen, unsigned int off)
1658 n = *poff / PAGE_SIZE;
1660 *page = nth_page(*page, n);
1662 *poff = *poff % PAGE_SIZE;
1666 static bool __splice_segment(struct page *page, unsigned int poff,
1667 unsigned int plen, unsigned int *off,
1668 unsigned int *len, struct sk_buff *skb,
1669 struct splice_pipe_desc *spd, bool linear,
1671 struct pipe_inode_info *pipe)
1676 /* skip this segment if already processed */
1682 /* ignore any bits we already processed */
1684 __segment_seek(&page, &poff, &plen, *off);
1689 unsigned int flen = min(*len, plen);
1691 /* the linear region may spread across several pages */
1692 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1694 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1697 __segment_seek(&page, &poff, &plen, flen);
1700 } while (*len && plen);
1706 * Map linear and fragment data from the skb to spd. It reports true if the
1707 * pipe is full or if we already spliced the requested length.
1709 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1710 unsigned int *offset, unsigned int *len,
1711 struct splice_pipe_desc *spd, struct sock *sk)
1715 /* map the linear part :
1716 * If skb->head_frag is set, this 'linear' part is backed by a
1717 * fragment, and if the head is not shared with any clones then
1718 * we can avoid a copy since we own the head portion of this page.
1720 if (__splice_segment(virt_to_page(skb->data),
1721 (unsigned long) skb->data & (PAGE_SIZE - 1),
1723 offset, len, skb, spd,
1724 skb_head_is_locked(skb),
1729 * then map the fragments
1731 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1732 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1734 if (__splice_segment(skb_frag_page(f),
1735 f->page_offset, skb_frag_size(f),
1736 offset, len, skb, spd, false, sk, pipe))
1744 * Map data from the skb to a pipe. Should handle both the linear part,
1745 * the fragments, and the frag list. It does NOT handle frag lists within
1746 * the frag list, if such a thing exists. We'd probably need to recurse to
1747 * handle that cleanly.
1749 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1750 struct pipe_inode_info *pipe, unsigned int tlen,
1753 struct partial_page partial[MAX_SKB_FRAGS];
1754 struct page *pages[MAX_SKB_FRAGS];
1755 struct splice_pipe_desc spd = {
1758 .nr_pages_max = MAX_SKB_FRAGS,
1760 .ops = &sock_pipe_buf_ops,
1761 .spd_release = sock_spd_release,
1763 struct sk_buff *frag_iter;
1764 struct sock *sk = skb->sk;
1768 * __skb_splice_bits() only fails if the output has no room left,
1769 * so no point in going over the frag_list for the error case.
1771 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1777 * now see if we have a frag_list to map
1779 skb_walk_frags(skb, frag_iter) {
1782 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1789 * Drop the socket lock, otherwise we have reverse
1790 * locking dependencies between sk_lock and i_mutex
1791 * here as compared to sendfile(). We enter here
1792 * with the socket lock held, and splice_to_pipe() will
1793 * grab the pipe inode lock. For sendfile() emulation,
1794 * we call into ->sendpage() with the i_mutex lock held
1795 * and networking will grab the socket lock.
1798 ret = splice_to_pipe(pipe, &spd);
1806 * skb_store_bits - store bits from kernel buffer to skb
1807 * @skb: destination buffer
1808 * @offset: offset in destination
1809 * @from: source buffer
1810 * @len: number of bytes to copy
1812 * Copy the specified number of bytes from the source buffer to the
1813 * destination skb. This function handles all the messy bits of
1814 * traversing fragment lists and such.
1817 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1819 int start = skb_headlen(skb);
1820 struct sk_buff *frag_iter;
1823 if (offset > (int)skb->len - len)
1826 if ((copy = start - offset) > 0) {
1829 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1830 if ((len -= copy) == 0)
1836 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1837 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1840 WARN_ON(start > offset + len);
1842 end = start + skb_frag_size(frag);
1843 if ((copy = end - offset) > 0) {
1849 vaddr = kmap_atomic(skb_frag_page(frag));
1850 memcpy(vaddr + frag->page_offset + offset - start,
1852 kunmap_atomic(vaddr);
1854 if ((len -= copy) == 0)
1862 skb_walk_frags(skb, frag_iter) {
1865 WARN_ON(start > offset + len);
1867 end = start + frag_iter->len;
1868 if ((copy = end - offset) > 0) {
1871 if (skb_store_bits(frag_iter, offset - start,
1874 if ((len -= copy) == 0)
1887 EXPORT_SYMBOL(skb_store_bits);
1889 /* Checksum skb data. */
1891 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1892 int len, __wsum csum)
1894 int start = skb_headlen(skb);
1895 int i, copy = start - offset;
1896 struct sk_buff *frag_iter;
1899 /* Checksum header. */
1903 csum = csum_partial(skb->data + offset, copy, csum);
1904 if ((len -= copy) == 0)
1910 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1912 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1914 WARN_ON(start > offset + len);
1916 end = start + skb_frag_size(frag);
1917 if ((copy = end - offset) > 0) {
1923 vaddr = kmap_atomic(skb_frag_page(frag));
1924 csum2 = csum_partial(vaddr + frag->page_offset +
1925 offset - start, copy, 0);
1926 kunmap_atomic(vaddr);
1927 csum = csum_block_add(csum, csum2, pos);
1936 skb_walk_frags(skb, frag_iter) {
1939 WARN_ON(start > offset + len);
1941 end = start + frag_iter->len;
1942 if ((copy = end - offset) > 0) {
1946 csum2 = skb_checksum(frag_iter, offset - start,
1948 csum = csum_block_add(csum, csum2, pos);
1949 if ((len -= copy) == 0)
1960 EXPORT_SYMBOL(skb_checksum);
1962 /* Both of above in one bottle. */
1964 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1965 u8 *to, int len, __wsum csum)
1967 int start = skb_headlen(skb);
1968 int i, copy = start - offset;
1969 struct sk_buff *frag_iter;
1976 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1978 if ((len -= copy) == 0)
1985 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1988 WARN_ON(start > offset + len);
1990 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1991 if ((copy = end - offset) > 0) {
1994 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1998 vaddr = kmap_atomic(skb_frag_page(frag));
1999 csum2 = csum_partial_copy_nocheck(vaddr +
2003 kunmap_atomic(vaddr);
2004 csum = csum_block_add(csum, csum2, pos);
2014 skb_walk_frags(skb, frag_iter) {
2018 WARN_ON(start > offset + len);
2020 end = start + frag_iter->len;
2021 if ((copy = end - offset) > 0) {
2024 csum2 = skb_copy_and_csum_bits(frag_iter,
2027 csum = csum_block_add(csum, csum2, pos);
2028 if ((len -= copy) == 0)
2039 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2041 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2046 if (skb->ip_summed == CHECKSUM_PARTIAL)
2047 csstart = skb_checksum_start_offset(skb);
2049 csstart = skb_headlen(skb);
2051 BUG_ON(csstart > skb_headlen(skb));
2053 skb_copy_from_linear_data(skb, to, csstart);
2056 if (csstart != skb->len)
2057 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2058 skb->len - csstart, 0);
2060 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2061 long csstuff = csstart + skb->csum_offset;
2063 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2066 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2069 * skb_dequeue - remove from the head of the queue
2070 * @list: list to dequeue from
2072 * Remove the head of the list. The list lock is taken so the function
2073 * may be used safely with other locking list functions. The head item is
2074 * returned or %NULL if the list is empty.
2077 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2079 unsigned long flags;
2080 struct sk_buff *result;
2082 spin_lock_irqsave(&list->lock, flags);
2083 result = __skb_dequeue(list);
2084 spin_unlock_irqrestore(&list->lock, flags);
2087 EXPORT_SYMBOL(skb_dequeue);
2090 * skb_dequeue_tail - remove from the tail of the queue
2091 * @list: list to dequeue from
2093 * Remove the tail of the list. The list lock is taken so the function
2094 * may be used safely with other locking list functions. The tail item is
2095 * returned or %NULL if the list is empty.
2097 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2099 unsigned long flags;
2100 struct sk_buff *result;
2102 spin_lock_irqsave(&list->lock, flags);
2103 result = __skb_dequeue_tail(list);
2104 spin_unlock_irqrestore(&list->lock, flags);
2107 EXPORT_SYMBOL(skb_dequeue_tail);
2110 * skb_queue_purge - empty a list
2111 * @list: list to empty
2113 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2114 * the list and one reference dropped. This function takes the list
2115 * lock and is atomic with respect to other list locking functions.
2117 void skb_queue_purge(struct sk_buff_head *list)
2119 struct sk_buff *skb;
2120 while ((skb = skb_dequeue(list)) != NULL)
2123 EXPORT_SYMBOL(skb_queue_purge);
2126 * skb_queue_head - queue a buffer at the list head
2127 * @list: list to use
2128 * @newsk: buffer to queue
2130 * Queue a buffer at the start of the list. This function takes the
2131 * list lock and can be used safely with other locking &sk_buff functions
2134 * A buffer cannot be placed on two lists at the same time.
2136 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2138 unsigned long flags;
2140 spin_lock_irqsave(&list->lock, flags);
2141 __skb_queue_head(list, newsk);
2142 spin_unlock_irqrestore(&list->lock, flags);
2144 EXPORT_SYMBOL(skb_queue_head);
2147 * skb_queue_tail - queue a buffer at the list tail
2148 * @list: list to use
2149 * @newsk: buffer to queue
2151 * Queue a buffer at the tail of the list. This function takes the
2152 * list lock and can be used safely with other locking &sk_buff functions
2155 * A buffer cannot be placed on two lists at the same time.
2157 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2159 unsigned long flags;
2161 spin_lock_irqsave(&list->lock, flags);
2162 __skb_queue_tail(list, newsk);
2163 spin_unlock_irqrestore(&list->lock, flags);
2165 EXPORT_SYMBOL(skb_queue_tail);
2168 * skb_unlink - remove a buffer from a list
2169 * @skb: buffer to remove
2170 * @list: list to use
2172 * Remove a packet from a list. The list locks are taken and this
2173 * function is atomic with respect to other list locked calls
2175 * You must know what list the SKB is on.
2177 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2179 unsigned long flags;
2181 spin_lock_irqsave(&list->lock, flags);
2182 __skb_unlink(skb, list);
2183 spin_unlock_irqrestore(&list->lock, flags);
2185 EXPORT_SYMBOL(skb_unlink);
2188 * skb_append - append a buffer
2189 * @old: buffer to insert after
2190 * @newsk: buffer to insert
2191 * @list: list to use
2193 * Place a packet after a given packet in a list. The list locks are taken
2194 * and this function is atomic with respect to other list locked calls.
2195 * A buffer cannot be placed on two lists at the same time.
2197 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2199 unsigned long flags;
2201 spin_lock_irqsave(&list->lock, flags);
2202 __skb_queue_after(list, old, newsk);
2203 spin_unlock_irqrestore(&list->lock, flags);
2205 EXPORT_SYMBOL(skb_append);
2208 * skb_insert - insert a buffer
2209 * @old: buffer to insert before
2210 * @newsk: buffer to insert
2211 * @list: list to use
2213 * Place a packet before a given packet in a list. The list locks are
2214 * taken and this function is atomic with respect to other list locked
2217 * A buffer cannot be placed on two lists at the same time.
2219 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2221 unsigned long flags;
2223 spin_lock_irqsave(&list->lock, flags);
2224 __skb_insert(newsk, old->prev, old, list);
2225 spin_unlock_irqrestore(&list->lock, flags);
2227 EXPORT_SYMBOL(skb_insert);
2229 static inline void skb_split_inside_header(struct sk_buff *skb,
2230 struct sk_buff* skb1,
2231 const u32 len, const int pos)
2235 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2237 /* And move data appendix as is. */
2238 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2239 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2241 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2242 skb_shinfo(skb)->nr_frags = 0;
2243 skb1->data_len = skb->data_len;
2244 skb1->len += skb1->data_len;
2247 skb_set_tail_pointer(skb, len);
2250 static inline void skb_split_no_header(struct sk_buff *skb,
2251 struct sk_buff* skb1,
2252 const u32 len, int pos)
2255 const int nfrags = skb_shinfo(skb)->nr_frags;
2257 skb_shinfo(skb)->nr_frags = 0;
2258 skb1->len = skb1->data_len = skb->len - len;
2260 skb->data_len = len - pos;
2262 for (i = 0; i < nfrags; i++) {
2263 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2265 if (pos + size > len) {
2266 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2270 * We have two variants in this case:
2271 * 1. Move all the frag to the second
2272 * part, if it is possible. F.e.
2273 * this approach is mandatory for TUX,
2274 * where splitting is expensive.
2275 * 2. Split is accurately. We make this.
2277 skb_frag_ref(skb, i);
2278 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2279 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2280 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2281 skb_shinfo(skb)->nr_frags++;
2285 skb_shinfo(skb)->nr_frags++;
2288 skb_shinfo(skb1)->nr_frags = k;
2292 * skb_split - Split fragmented skb to two parts at length len.
2293 * @skb: the buffer to split
2294 * @skb1: the buffer to receive the second part
2295 * @len: new length for skb
2297 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2299 int pos = skb_headlen(skb);
2301 if (len < pos) /* Split line is inside header. */
2302 skb_split_inside_header(skb, skb1, len, pos);
2303 else /* Second chunk has no header, nothing to copy. */
2304 skb_split_no_header(skb, skb1, len, pos);
2306 EXPORT_SYMBOL(skb_split);
2308 /* Shifting from/to a cloned skb is a no-go.
2310 * Caller cannot keep skb_shinfo related pointers past calling here!
2312 static int skb_prepare_for_shift(struct sk_buff *skb)
2314 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2318 * skb_shift - Shifts paged data partially from skb to another
2319 * @tgt: buffer into which tail data gets added
2320 * @skb: buffer from which the paged data comes from
2321 * @shiftlen: shift up to this many bytes
2323 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2324 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2325 * It's up to caller to free skb if everything was shifted.
2327 * If @tgt runs out of frags, the whole operation is aborted.
2329 * Skb cannot include anything else but paged data while tgt is allowed
2330 * to have non-paged data as well.
2332 * TODO: full sized shift could be optimized but that would need
2333 * specialized skb free'er to handle frags without up-to-date nr_frags.
2335 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2337 int from, to, merge, todo;
2338 struct skb_frag_struct *fragfrom, *fragto;
2340 BUG_ON(shiftlen > skb->len);
2341 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2345 to = skb_shinfo(tgt)->nr_frags;
2346 fragfrom = &skb_shinfo(skb)->frags[from];
2348 /* Actual merge is delayed until the point when we know we can
2349 * commit all, so that we don't have to undo partial changes
2352 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2353 fragfrom->page_offset)) {
2358 todo -= skb_frag_size(fragfrom);
2360 if (skb_prepare_for_shift(skb) ||
2361 skb_prepare_for_shift(tgt))
2364 /* All previous frag pointers might be stale! */
2365 fragfrom = &skb_shinfo(skb)->frags[from];
2366 fragto = &skb_shinfo(tgt)->frags[merge];
2368 skb_frag_size_add(fragto, shiftlen);
2369 skb_frag_size_sub(fragfrom, shiftlen);
2370 fragfrom->page_offset += shiftlen;
2378 /* Skip full, not-fitting skb to avoid expensive operations */
2379 if ((shiftlen == skb->len) &&
2380 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2383 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2386 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2387 if (to == MAX_SKB_FRAGS)
2390 fragfrom = &skb_shinfo(skb)->frags[from];
2391 fragto = &skb_shinfo(tgt)->frags[to];
2393 if (todo >= skb_frag_size(fragfrom)) {
2394 *fragto = *fragfrom;
2395 todo -= skb_frag_size(fragfrom);
2400 __skb_frag_ref(fragfrom);
2401 fragto->page = fragfrom->page;
2402 fragto->page_offset = fragfrom->page_offset;
2403 skb_frag_size_set(fragto, todo);
2405 fragfrom->page_offset += todo;
2406 skb_frag_size_sub(fragfrom, todo);
2414 /* Ready to "commit" this state change to tgt */
2415 skb_shinfo(tgt)->nr_frags = to;
2418 fragfrom = &skb_shinfo(skb)->frags[0];
2419 fragto = &skb_shinfo(tgt)->frags[merge];
2421 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2422 __skb_frag_unref(fragfrom);
2425 /* Reposition in the original skb */
2427 while (from < skb_shinfo(skb)->nr_frags)
2428 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2429 skb_shinfo(skb)->nr_frags = to;
2431 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2434 /* Most likely the tgt won't ever need its checksum anymore, skb on
2435 * the other hand might need it if it needs to be resent
2437 tgt->ip_summed = CHECKSUM_PARTIAL;
2438 skb->ip_summed = CHECKSUM_PARTIAL;
2440 /* Yak, is it really working this way? Some helper please? */
2441 skb->len -= shiftlen;
2442 skb->data_len -= shiftlen;
2443 skb->truesize -= shiftlen;
2444 tgt->len += shiftlen;
2445 tgt->data_len += shiftlen;
2446 tgt->truesize += shiftlen;
2452 * skb_prepare_seq_read - Prepare a sequential read of skb data
2453 * @skb: the buffer to read
2454 * @from: lower offset of data to be read
2455 * @to: upper offset of data to be read
2456 * @st: state variable
2458 * Initializes the specified state variable. Must be called before
2459 * invoking skb_seq_read() for the first time.
2461 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2462 unsigned int to, struct skb_seq_state *st)
2464 st->lower_offset = from;
2465 st->upper_offset = to;
2466 st->root_skb = st->cur_skb = skb;
2467 st->frag_idx = st->stepped_offset = 0;
2468 st->frag_data = NULL;
2470 EXPORT_SYMBOL(skb_prepare_seq_read);
2473 * skb_seq_read - Sequentially read skb data
2474 * @consumed: number of bytes consumed by the caller so far
2475 * @data: destination pointer for data to be returned
2476 * @st: state variable
2478 * Reads a block of skb data at &consumed relative to the
2479 * lower offset specified to skb_prepare_seq_read(). Assigns
2480 * the head of the data block to &data and returns the length
2481 * of the block or 0 if the end of the skb data or the upper
2482 * offset has been reached.
2484 * The caller is not required to consume all of the data
2485 * returned, i.e. &consumed is typically set to the number
2486 * of bytes already consumed and the next call to
2487 * skb_seq_read() will return the remaining part of the block.
2489 * Note 1: The size of each block of data returned can be arbitrary,
2490 * this limitation is the cost for zerocopy seqeuental
2491 * reads of potentially non linear data.
2493 * Note 2: Fragment lists within fragments are not implemented
2494 * at the moment, state->root_skb could be replaced with
2495 * a stack for this purpose.
2497 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2498 struct skb_seq_state *st)
2500 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2503 if (unlikely(abs_offset >= st->upper_offset))
2507 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2509 if (abs_offset < block_limit && !st->frag_data) {
2510 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2511 return block_limit - abs_offset;
2514 if (st->frag_idx == 0 && !st->frag_data)
2515 st->stepped_offset += skb_headlen(st->cur_skb);
2517 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2518 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2519 block_limit = skb_frag_size(frag) + st->stepped_offset;
2521 if (abs_offset < block_limit) {
2523 st->frag_data = kmap_atomic(skb_frag_page(frag));
2525 *data = (u8 *) st->frag_data + frag->page_offset +
2526 (abs_offset - st->stepped_offset);
2528 return block_limit - abs_offset;
2531 if (st->frag_data) {
2532 kunmap_atomic(st->frag_data);
2533 st->frag_data = NULL;
2537 st->stepped_offset += skb_frag_size(frag);
2540 if (st->frag_data) {
2541 kunmap_atomic(st->frag_data);
2542 st->frag_data = NULL;
2545 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2546 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2549 } else if (st->cur_skb->next) {
2550 st->cur_skb = st->cur_skb->next;
2557 EXPORT_SYMBOL(skb_seq_read);
2560 * skb_abort_seq_read - Abort a sequential read of skb data
2561 * @st: state variable
2563 * Must be called if skb_seq_read() was not called until it
2566 void skb_abort_seq_read(struct skb_seq_state *st)
2569 kunmap_atomic(st->frag_data);
2571 EXPORT_SYMBOL(skb_abort_seq_read);
2573 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2575 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2576 struct ts_config *conf,
2577 struct ts_state *state)
2579 return skb_seq_read(offset, text, TS_SKB_CB(state));
2582 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2584 skb_abort_seq_read(TS_SKB_CB(state));
2588 * skb_find_text - Find a text pattern in skb data
2589 * @skb: the buffer to look in
2590 * @from: search offset
2592 * @config: textsearch configuration
2593 * @state: uninitialized textsearch state variable
2595 * Finds a pattern in the skb data according to the specified
2596 * textsearch configuration. Use textsearch_next() to retrieve
2597 * subsequent occurrences of the pattern. Returns the offset
2598 * to the first occurrence or UINT_MAX if no match was found.
2600 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2601 unsigned int to, struct ts_config *config,
2602 struct ts_state *state)
2606 config->get_next_block = skb_ts_get_next_block;
2607 config->finish = skb_ts_finish;
2609 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2611 ret = textsearch_find(config, state);
2612 return (ret <= to - from ? ret : UINT_MAX);
2614 EXPORT_SYMBOL(skb_find_text);
2617 * skb_append_datato_frags: - append the user data to a skb
2618 * @sk: sock structure
2619 * @skb: skb structure to be appened with user data.
2620 * @getfrag: call back function to be used for getting the user data
2621 * @from: pointer to user message iov
2622 * @length: length of the iov message
2624 * Description: This procedure append the user data in the fragment part
2625 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2627 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2628 int (*getfrag)(void *from, char *to, int offset,
2629 int len, int odd, struct sk_buff *skb),
2630 void *from, int length)
2633 skb_frag_t *frag = NULL;
2634 struct page *page = NULL;
2640 /* Return error if we don't have space for new frag */
2641 frg_cnt = skb_shinfo(skb)->nr_frags;
2642 if (frg_cnt >= MAX_SKB_FRAGS)
2645 /* allocate a new page for next frag */
2646 page = alloc_pages(sk->sk_allocation, 0);
2648 /* If alloc_page fails just return failure and caller will
2649 * free previous allocated pages by doing kfree_skb()
2654 /* initialize the next frag */
2655 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2656 skb->truesize += PAGE_SIZE;
2657 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2659 /* get the new initialized frag */
2660 frg_cnt = skb_shinfo(skb)->nr_frags;
2661 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2663 /* copy the user data to page */
2664 left = PAGE_SIZE - frag->page_offset;
2665 copy = (length > left)? left : length;
2667 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2668 offset, copy, 0, skb);
2672 /* copy was successful so update the size parameters */
2673 skb_frag_size_add(frag, copy);
2675 skb->data_len += copy;
2679 } while (length > 0);
2683 EXPORT_SYMBOL(skb_append_datato_frags);
2686 * skb_pull_rcsum - pull skb and update receive checksum
2687 * @skb: buffer to update
2688 * @len: length of data pulled
2690 * This function performs an skb_pull on the packet and updates
2691 * the CHECKSUM_COMPLETE checksum. It should be used on
2692 * receive path processing instead of skb_pull unless you know
2693 * that the checksum difference is zero (e.g., a valid IP header)
2694 * or you are setting ip_summed to CHECKSUM_NONE.
2696 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2698 BUG_ON(len > skb->len);
2700 BUG_ON(skb->len < skb->data_len);
2701 skb_postpull_rcsum(skb, skb->data, len);
2702 return skb->data += len;
2704 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2707 * skb_segment - Perform protocol segmentation on skb.
2708 * @skb: buffer to segment
2709 * @features: features for the output path (see dev->features)
2711 * This function performs segmentation on the given skb. It returns
2712 * a pointer to the first in a list of new skbs for the segments.
2713 * In case of error it returns ERR_PTR(err).
2715 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2717 struct sk_buff *segs = NULL;
2718 struct sk_buff *tail = NULL;
2719 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2720 unsigned int mss = skb_shinfo(skb)->gso_size;
2721 unsigned int doffset = skb->data - skb_mac_header(skb);
2722 unsigned int offset = doffset;
2723 unsigned int headroom;
2725 int sg = !!(features & NETIF_F_SG);
2726 int nfrags = skb_shinfo(skb)->nr_frags;
2731 __skb_push(skb, doffset);
2732 headroom = skb_headroom(skb);
2733 pos = skb_headlen(skb);
2736 struct sk_buff *nskb;
2741 len = skb->len - offset;
2745 hsize = skb_headlen(skb) - offset;
2748 if (hsize > len || !sg)
2751 if (!hsize && i >= nfrags) {
2752 BUG_ON(fskb->len != len);
2755 nskb = skb_clone(fskb, GFP_ATOMIC);
2758 if (unlikely(!nskb))
2761 hsize = skb_end_offset(nskb);
2762 if (skb_cow_head(nskb, doffset + headroom)) {
2767 nskb->truesize += skb_end_offset(nskb) - hsize;
2768 skb_release_head_state(nskb);
2769 __skb_push(nskb, doffset);
2771 nskb = alloc_skb(hsize + doffset + headroom,
2774 if (unlikely(!nskb))
2777 skb_reserve(nskb, headroom);
2778 __skb_put(nskb, doffset);
2787 __copy_skb_header(nskb, skb);
2788 nskb->mac_len = skb->mac_len;
2790 /* nskb and skb might have different headroom */
2791 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2792 nskb->csum_start += skb_headroom(nskb) - headroom;
2794 skb_reset_mac_header(nskb);
2795 skb_set_network_header(nskb, skb->mac_len);
2796 nskb->transport_header = (nskb->network_header +
2797 skb_network_header_len(skb));
2798 skb_copy_from_linear_data(skb, nskb->data, doffset);
2800 if (fskb != skb_shinfo(skb)->frag_list)
2804 nskb->ip_summed = CHECKSUM_NONE;
2805 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2811 frag = skb_shinfo(nskb)->frags;
2813 skb_copy_from_linear_data_offset(skb, offset,
2814 skb_put(nskb, hsize), hsize);
2816 while (pos < offset + len && i < nfrags) {
2817 *frag = skb_shinfo(skb)->frags[i];
2818 __skb_frag_ref(frag);
2819 size = skb_frag_size(frag);
2822 frag->page_offset += offset - pos;
2823 skb_frag_size_sub(frag, offset - pos);
2826 skb_shinfo(nskb)->nr_frags++;
2828 if (pos + size <= offset + len) {
2832 skb_frag_size_sub(frag, pos + size - (offset + len));
2839 if (pos < offset + len) {
2840 struct sk_buff *fskb2 = fskb;
2842 BUG_ON(pos + fskb->len != offset + len);
2848 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2854 SKB_FRAG_ASSERT(nskb);
2855 skb_shinfo(nskb)->frag_list = fskb2;
2859 nskb->data_len = len - hsize;
2860 nskb->len += nskb->data_len;
2861 nskb->truesize += nskb->data_len;
2862 } while ((offset += len) < skb->len);
2867 while ((skb = segs)) {
2871 return ERR_PTR(err);
2873 EXPORT_SYMBOL_GPL(skb_segment);
2875 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2877 struct sk_buff *p = *head;
2878 struct sk_buff *nskb;
2879 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2880 struct skb_shared_info *pinfo = skb_shinfo(p);
2881 unsigned int headroom;
2882 unsigned int len = skb_gro_len(skb);
2883 unsigned int offset = skb_gro_offset(skb);
2884 unsigned int headlen = skb_headlen(skb);
2885 unsigned int delta_truesize;
2887 if (p->len + len >= 65536)
2890 if (pinfo->frag_list)
2892 else if (headlen <= offset) {
2895 int i = skbinfo->nr_frags;
2896 int nr_frags = pinfo->nr_frags + i;
2900 if (nr_frags > MAX_SKB_FRAGS)
2903 pinfo->nr_frags = nr_frags;
2904 skbinfo->nr_frags = 0;
2906 frag = pinfo->frags + nr_frags;
2907 frag2 = skbinfo->frags + i;
2912 frag->page_offset += offset;
2913 skb_frag_size_sub(frag, offset);
2915 /* all fragments truesize : remove (head size + sk_buff) */
2916 delta_truesize = skb->truesize -
2917 SKB_TRUESIZE(skb_end_offset(skb));
2919 skb->truesize -= skb->data_len;
2920 skb->len -= skb->data_len;
2923 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2925 } else if (skb->head_frag) {
2926 int nr_frags = pinfo->nr_frags;
2927 skb_frag_t *frag = pinfo->frags + nr_frags;
2928 struct page *page = virt_to_head_page(skb->head);
2929 unsigned int first_size = headlen - offset;
2930 unsigned int first_offset;
2932 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2935 first_offset = skb->data -
2936 (unsigned char *)page_address(page) +
2939 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2941 frag->page.p = page;
2942 frag->page_offset = first_offset;
2943 skb_frag_size_set(frag, first_size);
2945 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2946 /* We dont need to clear skbinfo->nr_frags here */
2948 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
2949 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
2951 } else if (skb_gro_len(p) != pinfo->gso_size)
2954 headroom = skb_headroom(p);
2955 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2956 if (unlikely(!nskb))
2959 __copy_skb_header(nskb, p);
2960 nskb->mac_len = p->mac_len;
2962 skb_reserve(nskb, headroom);
2963 __skb_put(nskb, skb_gro_offset(p));
2965 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2966 skb_set_network_header(nskb, skb_network_offset(p));
2967 skb_set_transport_header(nskb, skb_transport_offset(p));
2969 __skb_pull(p, skb_gro_offset(p));
2970 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2971 p->data - skb_mac_header(p));
2973 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2974 skb_shinfo(nskb)->frag_list = p;
2975 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2976 pinfo->gso_size = 0;
2977 skb_header_release(p);
2980 nskb->data_len += p->len;
2981 nskb->truesize += p->truesize;
2982 nskb->len += p->len;
2985 nskb->next = p->next;
2991 delta_truesize = skb->truesize;
2992 if (offset > headlen) {
2993 unsigned int eat = offset - headlen;
2995 skbinfo->frags[0].page_offset += eat;
2996 skb_frag_size_sub(&skbinfo->frags[0], eat);
2997 skb->data_len -= eat;
3002 __skb_pull(skb, offset);
3004 p->prev->next = skb;
3006 skb_header_release(skb);
3009 NAPI_GRO_CB(p)->count++;
3011 p->truesize += delta_truesize;
3014 NAPI_GRO_CB(skb)->same_flow = 1;
3017 EXPORT_SYMBOL_GPL(skb_gro_receive);
3019 void __init skb_init(void)
3021 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3022 sizeof(struct sk_buff),
3024 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3026 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3027 (2*sizeof(struct sk_buff)) +
3030 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3035 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3036 * @skb: Socket buffer containing the buffers to be mapped
3037 * @sg: The scatter-gather list to map into
3038 * @offset: The offset into the buffer's contents to start mapping
3039 * @len: Length of buffer space to be mapped
3041 * Fill the specified scatter-gather list with mappings/pointers into a
3042 * region of the buffer space attached to a socket buffer.
3045 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3047 int start = skb_headlen(skb);
3048 int i, copy = start - offset;
3049 struct sk_buff *frag_iter;
3055 sg_set_buf(sg, skb->data + offset, copy);
3057 if ((len -= copy) == 0)
3062 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3065 WARN_ON(start > offset + len);
3067 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3068 if ((copy = end - offset) > 0) {
3069 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3073 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3074 frag->page_offset+offset-start);
3083 skb_walk_frags(skb, frag_iter) {
3086 WARN_ON(start > offset + len);
3088 end = start + frag_iter->len;
3089 if ((copy = end - offset) > 0) {
3092 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3094 if ((len -= copy) == 0)
3104 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3106 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3108 sg_mark_end(&sg[nsg - 1]);
3112 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3115 * skb_cow_data - Check that a socket buffer's data buffers are writable
3116 * @skb: The socket buffer to check.
3117 * @tailbits: Amount of trailing space to be added
3118 * @trailer: Returned pointer to the skb where the @tailbits space begins
3120 * Make sure that the data buffers attached to a socket buffer are
3121 * writable. If they are not, private copies are made of the data buffers
3122 * and the socket buffer is set to use these instead.
3124 * If @tailbits is given, make sure that there is space to write @tailbits
3125 * bytes of data beyond current end of socket buffer. @trailer will be
3126 * set to point to the skb in which this space begins.
3128 * The number of scatterlist elements required to completely map the
3129 * COW'd and extended socket buffer will be returned.
3131 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3135 struct sk_buff *skb1, **skb_p;
3137 /* If skb is cloned or its head is paged, reallocate
3138 * head pulling out all the pages (pages are considered not writable
3139 * at the moment even if they are anonymous).
3141 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3142 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3145 /* Easy case. Most of packets will go this way. */
3146 if (!skb_has_frag_list(skb)) {
3147 /* A little of trouble, not enough of space for trailer.
3148 * This should not happen, when stack is tuned to generate
3149 * good frames. OK, on miss we reallocate and reserve even more
3150 * space, 128 bytes is fair. */
3152 if (skb_tailroom(skb) < tailbits &&
3153 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3161 /* Misery. We are in troubles, going to mincer fragments... */
3164 skb_p = &skb_shinfo(skb)->frag_list;
3167 while ((skb1 = *skb_p) != NULL) {
3170 /* The fragment is partially pulled by someone,
3171 * this can happen on input. Copy it and everything
3174 if (skb_shared(skb1))
3177 /* If the skb is the last, worry about trailer. */
3179 if (skb1->next == NULL && tailbits) {
3180 if (skb_shinfo(skb1)->nr_frags ||
3181 skb_has_frag_list(skb1) ||
3182 skb_tailroom(skb1) < tailbits)
3183 ntail = tailbits + 128;
3189 skb_shinfo(skb1)->nr_frags ||
3190 skb_has_frag_list(skb1)) {
3191 struct sk_buff *skb2;
3193 /* Fuck, we are miserable poor guys... */
3195 skb2 = skb_copy(skb1, GFP_ATOMIC);
3197 skb2 = skb_copy_expand(skb1,
3201 if (unlikely(skb2 == NULL))
3205 skb_set_owner_w(skb2, skb1->sk);
3207 /* Looking around. Are we still alive?
3208 * OK, link new skb, drop old one */
3210 skb2->next = skb1->next;
3217 skb_p = &skb1->next;
3222 EXPORT_SYMBOL_GPL(skb_cow_data);
3224 static void sock_rmem_free(struct sk_buff *skb)
3226 struct sock *sk = skb->sk;
3228 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3232 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3234 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3238 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3239 (unsigned int)sk->sk_rcvbuf)
3244 skb->destructor = sock_rmem_free;
3245 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3247 /* before exiting rcu section, make sure dst is refcounted */
3250 skb_queue_tail(&sk->sk_error_queue, skb);
3251 if (!sock_flag(sk, SOCK_DEAD))
3252 sk->sk_data_ready(sk, len);
3255 EXPORT_SYMBOL(sock_queue_err_skb);
3257 void skb_tstamp_tx(struct sk_buff *orig_skb,
3258 struct skb_shared_hwtstamps *hwtstamps)
3260 struct sock *sk = orig_skb->sk;
3261 struct sock_exterr_skb *serr;
3262 struct sk_buff *skb;
3268 skb = skb_clone(orig_skb, GFP_ATOMIC);
3273 *skb_hwtstamps(skb) =
3277 * no hardware time stamps available,
3278 * so keep the shared tx_flags and only
3279 * store software time stamp
3281 skb->tstamp = ktime_get_real();
3284 serr = SKB_EXT_ERR(skb);
3285 memset(serr, 0, sizeof(*serr));
3286 serr->ee.ee_errno = ENOMSG;
3287 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3289 err = sock_queue_err_skb(sk, skb);
3294 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3296 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3298 struct sock *sk = skb->sk;
3299 struct sock_exterr_skb *serr;
3302 skb->wifi_acked_valid = 1;
3303 skb->wifi_acked = acked;
3305 serr = SKB_EXT_ERR(skb);
3306 memset(serr, 0, sizeof(*serr));
3307 serr->ee.ee_errno = ENOMSG;
3308 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3310 err = sock_queue_err_skb(sk, skb);
3314 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3318 * skb_partial_csum_set - set up and verify partial csum values for packet
3319 * @skb: the skb to set
3320 * @start: the number of bytes after skb->data to start checksumming.
3321 * @off: the offset from start to place the checksum.
3323 * For untrusted partially-checksummed packets, we need to make sure the values
3324 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3326 * This function checks and sets those values and skb->ip_summed: if this
3327 * returns false you should drop the packet.
3329 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3331 if (unlikely(start > skb_headlen(skb)) ||
3332 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3333 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3334 start, off, skb_headlen(skb));
3337 skb->ip_summed = CHECKSUM_PARTIAL;
3338 skb->csum_start = skb_headroom(skb) + start;
3339 skb->csum_offset = off;
3342 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3344 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3346 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3349 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3351 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3354 kmem_cache_free(skbuff_head_cache, skb);
3358 EXPORT_SYMBOL(kfree_skb_partial);
3361 * skb_try_coalesce - try to merge skb to prior one
3363 * @from: buffer to add
3364 * @fragstolen: pointer to boolean
3365 * @delta_truesize: how much more was allocated than was requested
3367 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3368 bool *fragstolen, int *delta_truesize)
3370 int i, delta, len = from->len;
3372 *fragstolen = false;
3377 if (len <= skb_tailroom(to)) {
3378 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3379 *delta_truesize = 0;
3383 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3386 if (skb_headlen(from) != 0) {
3388 unsigned int offset;
3390 if (skb_shinfo(to)->nr_frags +
3391 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3394 if (skb_head_is_locked(from))
3397 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3399 page = virt_to_head_page(from->head);
3400 offset = from->data - (unsigned char *)page_address(page);
3402 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3403 page, offset, skb_headlen(from));
3406 if (skb_shinfo(to)->nr_frags +
3407 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3410 delta = from->truesize -
3411 SKB_TRUESIZE(skb_end_pointer(from) - from->head);
3414 WARN_ON_ONCE(delta < len);
3416 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3417 skb_shinfo(from)->frags,
3418 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3419 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3421 if (!skb_cloned(from))
3422 skb_shinfo(from)->nr_frags = 0;
3424 /* if the skb is cloned this does nothing since we set nr_frags to 0 */
3425 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3426 skb_frag_ref(from, i);
3428 to->truesize += delta;
3430 to->data_len += len;
3432 *delta_truesize = delta;
3435 EXPORT_SYMBOL(skb_try_coalesce);