2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
74 static struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here, skb->len, sz, skb->head, skb->data,
126 (unsigned long)skb->tail, (unsigned long)skb->end,
127 skb->dev ? skb->dev->name : "<NULL>");
132 * skb_under_panic - private function
137 * Out of line support code for skb_push(). Not user callable.
140 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here, skb->len, sz, skb->head, skb->data,
145 (unsigned long)skb->tail, (unsigned long)skb->end,
146 skb->dev ? skb->dev->name : "<NULL>");
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
187 /* We do our best to align skb_shared_info on a separate cache
188 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
189 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
190 * Both skb->head and skb_shared_info are cache line aligned.
192 size = SKB_DATA_ALIGN(size);
193 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
194 data = kmalloc_node_track_caller(size, gfp_mask, node);
197 /* kmalloc(size) might give us more room than requested.
198 * Put skb_shared_info exactly at the end of allocated zone,
199 * to allow max possible filling before reallocation.
201 size = SKB_WITH_OVERHEAD(ksize(data));
202 prefetchw(data + size);
205 * Only clear those fields we need to clear, not those that we will
206 * actually initialise below. Hence, don't put any more fields after
207 * the tail pointer in struct sk_buff!
209 memset(skb, 0, offsetof(struct sk_buff, tail));
210 /* Account for allocated memory : skb + skb->head */
211 skb->truesize = SKB_TRUESIZE(size);
212 atomic_set(&skb->users, 1);
215 skb_reset_tail_pointer(skb);
216 skb->end = skb->tail + size;
217 #ifdef NET_SKBUFF_DATA_USES_OFFSET
218 skb->mac_header = ~0U;
221 /* make sure we initialize shinfo sequentially */
222 shinfo = skb_shinfo(skb);
223 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
224 atomic_set(&shinfo->dataref, 1);
225 kmemcheck_annotate_variable(shinfo->destructor_arg);
228 struct sk_buff *child = skb + 1;
229 atomic_t *fclone_ref = (atomic_t *) (child + 1);
231 kmemcheck_annotate_bitfield(child, flags1);
232 kmemcheck_annotate_bitfield(child, flags2);
233 skb->fclone = SKB_FCLONE_ORIG;
234 atomic_set(fclone_ref, 1);
236 child->fclone = SKB_FCLONE_UNAVAILABLE;
241 kmem_cache_free(cache, skb);
245 EXPORT_SYMBOL(__alloc_skb);
248 * build_skb - build a network buffer
249 * @data: data buffer provided by caller
251 * Allocate a new &sk_buff. Caller provides space holding head and
252 * skb_shared_info. @data must have been allocated by kmalloc()
253 * The return is the new skb buffer.
254 * On a failure the return is %NULL, and @data is not freed.
256 * Before IO, driver allocates only data buffer where NIC put incoming frame
257 * Driver should add room at head (NET_SKB_PAD) and
258 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
259 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
260 * before giving packet to stack.
261 * RX rings only contains data buffers, not full skbs.
263 struct sk_buff *build_skb(void *data)
265 struct skb_shared_info *shinfo;
269 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
273 size = ksize(data) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
275 memset(skb, 0, offsetof(struct sk_buff, tail));
276 skb->truesize = SKB_TRUESIZE(size);
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);
297 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
298 * @dev: network device to receive on
299 * @length: length to allocate
300 * @gfp_mask: get_free_pages mask, passed to alloc_skb
302 * Allocate a new &sk_buff and assign it a usage count of one. The
303 * buffer has unspecified headroom built in. Users should allocate
304 * the headroom they think they need without accounting for the
305 * built in space. The built in space is used for optimisations.
307 * %NULL is returned if there is no free memory.
309 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
310 unsigned int length, gfp_t gfp_mask)
314 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
316 skb_reserve(skb, NET_SKB_PAD);
321 EXPORT_SYMBOL(__netdev_alloc_skb);
323 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
326 skb_fill_page_desc(skb, i, page, off, size);
328 skb->data_len += size;
329 skb->truesize += size;
331 EXPORT_SYMBOL(skb_add_rx_frag);
334 * dev_alloc_skb - allocate an skbuff for receiving
335 * @length: length to allocate
337 * Allocate a new &sk_buff and assign it a usage count of one. The
338 * buffer has unspecified headroom built in. Users should allocate
339 * the headroom they think they need without accounting for the
340 * built in space. The built in space is used for optimisations.
342 * %NULL is returned if there is no free memory. Although this function
343 * allocates memory it can be called from an interrupt.
345 struct sk_buff *dev_alloc_skb(unsigned int length)
348 * There is more code here than it seems:
349 * __dev_alloc_skb is an inline
351 return __dev_alloc_skb(length, GFP_ATOMIC);
353 EXPORT_SYMBOL(dev_alloc_skb);
355 static void skb_drop_list(struct sk_buff **listp)
357 struct sk_buff *list = *listp;
362 struct sk_buff *this = list;
368 static inline void skb_drop_fraglist(struct sk_buff *skb)
370 skb_drop_list(&skb_shinfo(skb)->frag_list);
373 static void skb_clone_fraglist(struct sk_buff *skb)
375 struct sk_buff *list;
377 skb_walk_frags(skb, list)
381 static void skb_release_data(struct sk_buff *skb)
384 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
385 &skb_shinfo(skb)->dataref)) {
386 if (skb_shinfo(skb)->nr_frags) {
388 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
389 skb_frag_unref(skb, i);
393 * If skb buf is from userspace, we need to notify the caller
394 * the lower device DMA has done;
396 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
397 struct ubuf_info *uarg;
399 uarg = skb_shinfo(skb)->destructor_arg;
401 uarg->callback(uarg);
404 if (skb_has_frag_list(skb))
405 skb_drop_fraglist(skb);
412 * Free an skbuff by memory without cleaning the state.
414 static void kfree_skbmem(struct sk_buff *skb)
416 struct sk_buff *other;
417 atomic_t *fclone_ref;
419 switch (skb->fclone) {
420 case SKB_FCLONE_UNAVAILABLE:
421 kmem_cache_free(skbuff_head_cache, skb);
424 case SKB_FCLONE_ORIG:
425 fclone_ref = (atomic_t *) (skb + 2);
426 if (atomic_dec_and_test(fclone_ref))
427 kmem_cache_free(skbuff_fclone_cache, skb);
430 case SKB_FCLONE_CLONE:
431 fclone_ref = (atomic_t *) (skb + 1);
434 /* The clone portion is available for
435 * fast-cloning again.
437 skb->fclone = SKB_FCLONE_UNAVAILABLE;
439 if (atomic_dec_and_test(fclone_ref))
440 kmem_cache_free(skbuff_fclone_cache, other);
445 static void skb_release_head_state(struct sk_buff *skb)
449 secpath_put(skb->sp);
451 if (skb->destructor) {
453 skb->destructor(skb);
455 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
456 nf_conntrack_put(skb->nfct);
458 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
459 nf_conntrack_put_reasm(skb->nfct_reasm);
461 #ifdef CONFIG_BRIDGE_NETFILTER
462 nf_bridge_put(skb->nf_bridge);
464 /* XXX: IS this still necessary? - JHS */
465 #ifdef CONFIG_NET_SCHED
467 #ifdef CONFIG_NET_CLS_ACT
473 /* Free everything but the sk_buff shell. */
474 static void skb_release_all(struct sk_buff *skb)
476 skb_release_head_state(skb);
477 skb_release_data(skb);
481 * __kfree_skb - private function
484 * Free an sk_buff. Release anything attached to the buffer.
485 * Clean the state. This is an internal helper function. Users should
486 * always call kfree_skb
489 void __kfree_skb(struct sk_buff *skb)
491 skb_release_all(skb);
494 EXPORT_SYMBOL(__kfree_skb);
497 * kfree_skb - free an sk_buff
498 * @skb: buffer to free
500 * Drop a reference to the buffer and free it if the usage count has
503 void kfree_skb(struct sk_buff *skb)
507 if (likely(atomic_read(&skb->users) == 1))
509 else if (likely(!atomic_dec_and_test(&skb->users)))
511 trace_kfree_skb(skb, __builtin_return_address(0));
514 EXPORT_SYMBOL(kfree_skb);
517 * consume_skb - free an skbuff
518 * @skb: buffer to free
520 * Drop a ref to the buffer and free it if the usage count has hit zero
521 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
522 * is being dropped after a failure and notes that
524 void consume_skb(struct sk_buff *skb)
528 if (likely(atomic_read(&skb->users) == 1))
530 else if (likely(!atomic_dec_and_test(&skb->users)))
532 trace_consume_skb(skb);
535 EXPORT_SYMBOL(consume_skb);
538 * skb_recycle - clean up an skb for reuse
541 * Recycles the skb to be reused as a receive buffer. This
542 * function does any necessary reference count dropping, and
543 * cleans up the skbuff as if it just came from __alloc_skb().
545 void skb_recycle(struct sk_buff *skb)
547 struct skb_shared_info *shinfo;
549 skb_release_head_state(skb);
551 shinfo = skb_shinfo(skb);
552 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
553 atomic_set(&shinfo->dataref, 1);
555 memset(skb, 0, offsetof(struct sk_buff, tail));
556 skb->data = skb->head + NET_SKB_PAD;
557 skb_reset_tail_pointer(skb);
559 EXPORT_SYMBOL(skb_recycle);
562 * skb_recycle_check - check if skb can be reused for receive
564 * @skb_size: minimum receive buffer size
566 * Checks that the skb passed in is not shared or cloned, and
567 * that it is linear and its head portion at least as large as
568 * skb_size so that it can be recycled as a receive buffer.
569 * If these conditions are met, this function does any necessary
570 * reference count dropping and cleans up the skbuff as if it
571 * just came from __alloc_skb().
573 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
575 if (!skb_is_recycleable(skb, skb_size))
582 EXPORT_SYMBOL(skb_recycle_check);
584 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
586 new->tstamp = old->tstamp;
588 new->transport_header = old->transport_header;
589 new->network_header = old->network_header;
590 new->mac_header = old->mac_header;
591 skb_dst_copy(new, old);
592 new->rxhash = old->rxhash;
593 new->ooo_okay = old->ooo_okay;
594 new->l4_rxhash = old->l4_rxhash;
595 new->no_fcs = old->no_fcs;
597 new->sp = secpath_get(old->sp);
599 memcpy(new->cb, old->cb, sizeof(old->cb));
600 new->csum = old->csum;
601 new->local_df = old->local_df;
602 new->pkt_type = old->pkt_type;
603 new->ip_summed = old->ip_summed;
604 skb_copy_queue_mapping(new, old);
605 new->priority = old->priority;
606 #if IS_ENABLED(CONFIG_IP_VS)
607 new->ipvs_property = old->ipvs_property;
609 new->protocol = old->protocol;
610 new->mark = old->mark;
611 new->skb_iif = old->skb_iif;
613 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
614 new->nf_trace = old->nf_trace;
616 #ifdef CONFIG_NET_SCHED
617 new->tc_index = old->tc_index;
618 #ifdef CONFIG_NET_CLS_ACT
619 new->tc_verd = old->tc_verd;
622 new->vlan_tci = old->vlan_tci;
624 skb_copy_secmark(new, old);
628 * You should not add any new code to this function. Add it to
629 * __copy_skb_header above instead.
631 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
633 #define C(x) n->x = skb->x
635 n->next = n->prev = NULL;
637 __copy_skb_header(n, skb);
642 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
645 n->destructor = NULL;
651 atomic_set(&n->users, 1);
653 atomic_inc(&(skb_shinfo(skb)->dataref));
661 * skb_morph - morph one skb into another
662 * @dst: the skb to receive the contents
663 * @src: the skb to supply the contents
665 * This is identical to skb_clone except that the target skb is
666 * supplied by the user.
668 * The target skb is returned upon exit.
670 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
672 skb_release_all(dst);
673 return __skb_clone(dst, src);
675 EXPORT_SYMBOL_GPL(skb_morph);
677 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
678 * @skb: the skb to modify
679 * @gfp_mask: allocation priority
681 * This must be called on SKBTX_DEV_ZEROCOPY skb.
682 * It will copy all frags into kernel and drop the reference
683 * to userspace pages.
685 * If this function is called from an interrupt gfp_mask() must be
688 * Returns 0 on success or a negative error code on failure
689 * to allocate kernel memory to copy to.
691 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
694 int num_frags = skb_shinfo(skb)->nr_frags;
695 struct page *page, *head = NULL;
696 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
698 for (i = 0; i < num_frags; i++) {
700 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
702 page = alloc_page(GFP_ATOMIC);
705 struct page *next = (struct page *)head->private;
711 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
712 memcpy(page_address(page),
713 vaddr + f->page_offset, skb_frag_size(f));
714 kunmap_skb_frag(vaddr);
715 page->private = (unsigned long)head;
719 /* skb frags release userspace buffers */
720 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
721 skb_frag_unref(skb, i);
723 uarg->callback(uarg);
725 /* skb frags point to kernel buffers */
726 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
727 __skb_fill_page_desc(skb, i-1, head, 0,
728 skb_shinfo(skb)->frags[i - 1].size);
729 head = (struct page *)head->private;
732 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
738 * skb_clone - duplicate an sk_buff
739 * @skb: buffer to clone
740 * @gfp_mask: allocation priority
742 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
743 * copies share the same packet data but not structure. The new
744 * buffer has a reference count of 1. If the allocation fails the
745 * function returns %NULL otherwise the new buffer is returned.
747 * If this function is called from an interrupt gfp_mask() must be
751 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
755 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
756 if (skb_copy_ubufs(skb, gfp_mask))
761 if (skb->fclone == SKB_FCLONE_ORIG &&
762 n->fclone == SKB_FCLONE_UNAVAILABLE) {
763 atomic_t *fclone_ref = (atomic_t *) (n + 1);
764 n->fclone = SKB_FCLONE_CLONE;
765 atomic_inc(fclone_ref);
767 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
771 kmemcheck_annotate_bitfield(n, flags1);
772 kmemcheck_annotate_bitfield(n, flags2);
773 n->fclone = SKB_FCLONE_UNAVAILABLE;
776 return __skb_clone(n, skb);
778 EXPORT_SYMBOL(skb_clone);
780 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
782 #ifndef NET_SKBUFF_DATA_USES_OFFSET
784 * Shift between the two data areas in bytes
786 unsigned long offset = new->data - old->data;
789 __copy_skb_header(new, old);
791 #ifndef NET_SKBUFF_DATA_USES_OFFSET
792 /* {transport,network,mac}_header are relative to skb->head */
793 new->transport_header += offset;
794 new->network_header += offset;
795 if (skb_mac_header_was_set(new))
796 new->mac_header += offset;
798 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
799 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
800 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
804 * skb_copy - create private copy of an sk_buff
805 * @skb: buffer to copy
806 * @gfp_mask: allocation priority
808 * Make a copy of both an &sk_buff and its data. This is used when the
809 * caller wishes to modify the data and needs a private copy of the
810 * data to alter. Returns %NULL on failure or the pointer to the buffer
811 * on success. The returned buffer has a reference count of 1.
813 * As by-product this function converts non-linear &sk_buff to linear
814 * one, so that &sk_buff becomes completely private and caller is allowed
815 * to modify all the data of returned buffer. This means that this
816 * function is not recommended for use in circumstances when only
817 * header is going to be modified. Use pskb_copy() instead.
820 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
822 int headerlen = skb_headroom(skb);
823 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
824 struct sk_buff *n = alloc_skb(size, gfp_mask);
829 /* Set the data pointer */
830 skb_reserve(n, headerlen);
831 /* Set the tail pointer and length */
832 skb_put(n, skb->len);
834 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
837 copy_skb_header(n, skb);
840 EXPORT_SYMBOL(skb_copy);
843 * __pskb_copy - create copy of an sk_buff with private head.
844 * @skb: buffer to copy
845 * @headroom: headroom of new skb
846 * @gfp_mask: allocation priority
848 * Make a copy of both an &sk_buff and part of its data, located
849 * in header. Fragmented data remain shared. This is used when
850 * the caller wishes to modify only header of &sk_buff and needs
851 * private copy of the header to alter. Returns %NULL on failure
852 * or the pointer to the buffer on success.
853 * The returned buffer has a reference count of 1.
856 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
858 unsigned int size = skb_headlen(skb) + headroom;
859 struct sk_buff *n = alloc_skb(size, gfp_mask);
864 /* Set the data pointer */
865 skb_reserve(n, headroom);
866 /* Set the tail pointer and length */
867 skb_put(n, skb_headlen(skb));
869 skb_copy_from_linear_data(skb, n->data, n->len);
871 n->truesize += skb->data_len;
872 n->data_len = skb->data_len;
875 if (skb_shinfo(skb)->nr_frags) {
878 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
879 if (skb_copy_ubufs(skb, gfp_mask)) {
885 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
886 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
887 skb_frag_ref(skb, i);
889 skb_shinfo(n)->nr_frags = i;
892 if (skb_has_frag_list(skb)) {
893 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
894 skb_clone_fraglist(n);
897 copy_skb_header(n, skb);
901 EXPORT_SYMBOL(__pskb_copy);
904 * pskb_expand_head - reallocate header of &sk_buff
905 * @skb: buffer to reallocate
906 * @nhead: room to add at head
907 * @ntail: room to add at tail
908 * @gfp_mask: allocation priority
910 * Expands (or creates identical copy, if &nhead and &ntail are zero)
911 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
912 * reference count of 1. Returns zero in the case of success or error,
913 * if expansion failed. In the last case, &sk_buff is not changed.
915 * All the pointers pointing into skb header may change and must be
916 * reloaded after call to this function.
919 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
924 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
933 size = SKB_DATA_ALIGN(size);
935 /* Check if we can avoid taking references on fragments if we own
936 * the last reference on skb->head. (see skb_release_data())
941 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
942 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
946 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
947 memmove(skb->head + size, skb_shinfo(skb),
948 offsetof(struct skb_shared_info,
949 frags[skb_shinfo(skb)->nr_frags]));
950 memmove(skb->head + nhead, skb->head,
951 skb_tail_pointer(skb) - skb->head);
956 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
960 /* Copy only real data... and, alas, header. This should be
961 * optimized for the cases when header is void.
963 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
965 memcpy((struct skb_shared_info *)(data + size),
967 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
972 /* copy this zero copy skb frags */
973 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
974 if (skb_copy_ubufs(skb, gfp_mask))
977 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
978 skb_frag_ref(skb, i);
980 if (skb_has_frag_list(skb))
981 skb_clone_fraglist(skb);
983 skb_release_data(skb);
985 off = (data + nhead) - skb->head;
990 #ifdef NET_SKBUFF_DATA_USES_OFFSET
994 skb->end = skb->head + size;
996 /* {transport,network,mac}_header and tail are relative to skb->head */
998 skb->transport_header += off;
999 skb->network_header += off;
1000 if (skb_mac_header_was_set(skb))
1001 skb->mac_header += off;
1002 /* Only adjust this if it actually is csum_start rather than csum */
1003 if (skb->ip_summed == CHECKSUM_PARTIAL)
1004 skb->csum_start += nhead;
1008 atomic_set(&skb_shinfo(skb)->dataref, 1);
1016 EXPORT_SYMBOL(pskb_expand_head);
1018 /* Make private copy of skb with writable head and some headroom */
1020 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1022 struct sk_buff *skb2;
1023 int delta = headroom - skb_headroom(skb);
1026 skb2 = pskb_copy(skb, GFP_ATOMIC);
1028 skb2 = skb_clone(skb, GFP_ATOMIC);
1029 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1037 EXPORT_SYMBOL(skb_realloc_headroom);
1040 * skb_copy_expand - copy and expand sk_buff
1041 * @skb: buffer to copy
1042 * @newheadroom: new free bytes at head
1043 * @newtailroom: new free bytes at tail
1044 * @gfp_mask: allocation priority
1046 * Make a copy of both an &sk_buff and its data and while doing so
1047 * allocate additional space.
1049 * This is used when the caller wishes to modify the data and needs a
1050 * private copy of the data to alter as well as more space for new fields.
1051 * Returns %NULL on failure or the pointer to the buffer
1052 * on success. The returned buffer has a reference count of 1.
1054 * You must pass %GFP_ATOMIC as the allocation priority if this function
1055 * is called from an interrupt.
1057 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1058 int newheadroom, int newtailroom,
1062 * Allocate the copy buffer
1064 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1066 int oldheadroom = skb_headroom(skb);
1067 int head_copy_len, head_copy_off;
1073 skb_reserve(n, newheadroom);
1075 /* Set the tail pointer and length */
1076 skb_put(n, skb->len);
1078 head_copy_len = oldheadroom;
1080 if (newheadroom <= head_copy_len)
1081 head_copy_len = newheadroom;
1083 head_copy_off = newheadroom - head_copy_len;
1085 /* Copy the linear header and data. */
1086 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1087 skb->len + head_copy_len))
1090 copy_skb_header(n, skb);
1092 off = newheadroom - oldheadroom;
1093 if (n->ip_summed == CHECKSUM_PARTIAL)
1094 n->csum_start += off;
1095 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1096 n->transport_header += off;
1097 n->network_header += off;
1098 if (skb_mac_header_was_set(skb))
1099 n->mac_header += off;
1104 EXPORT_SYMBOL(skb_copy_expand);
1107 * skb_pad - zero pad the tail of an skb
1108 * @skb: buffer to pad
1109 * @pad: space to pad
1111 * Ensure that a buffer is followed by a padding area that is zero
1112 * filled. Used by network drivers which may DMA or transfer data
1113 * beyond the buffer end onto the wire.
1115 * May return error in out of memory cases. The skb is freed on error.
1118 int skb_pad(struct sk_buff *skb, int pad)
1123 /* If the skbuff is non linear tailroom is always zero.. */
1124 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1125 memset(skb->data+skb->len, 0, pad);
1129 ntail = skb->data_len + pad - (skb->end - skb->tail);
1130 if (likely(skb_cloned(skb) || ntail > 0)) {
1131 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1136 /* FIXME: The use of this function with non-linear skb's really needs
1139 err = skb_linearize(skb);
1143 memset(skb->data + skb->len, 0, pad);
1150 EXPORT_SYMBOL(skb_pad);
1153 * skb_put - add data to a buffer
1154 * @skb: buffer to use
1155 * @len: amount of data to add
1157 * This function extends the used data area of the buffer. If this would
1158 * exceed the total buffer size the kernel will panic. A pointer to the
1159 * first byte of the extra data is returned.
1161 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1163 unsigned char *tmp = skb_tail_pointer(skb);
1164 SKB_LINEAR_ASSERT(skb);
1167 if (unlikely(skb->tail > skb->end))
1168 skb_over_panic(skb, len, __builtin_return_address(0));
1171 EXPORT_SYMBOL(skb_put);
1174 * skb_push - add data to the start of a buffer
1175 * @skb: buffer to use
1176 * @len: amount of data to add
1178 * This function extends the used data area of the buffer at the buffer
1179 * start. If this would exceed the total buffer headroom the kernel will
1180 * panic. A pointer to the first byte of the extra data is returned.
1182 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1186 if (unlikely(skb->data<skb->head))
1187 skb_under_panic(skb, len, __builtin_return_address(0));
1190 EXPORT_SYMBOL(skb_push);
1193 * skb_pull - remove data from the start of a buffer
1194 * @skb: buffer to use
1195 * @len: amount of data to remove
1197 * This function removes data from the start of a buffer, returning
1198 * the memory to the headroom. A pointer to the next data in the buffer
1199 * is returned. Once the data has been pulled future pushes will overwrite
1202 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1204 return skb_pull_inline(skb, len);
1206 EXPORT_SYMBOL(skb_pull);
1209 * skb_trim - remove end from a buffer
1210 * @skb: buffer to alter
1213 * Cut the length of a buffer down by removing data from the tail. If
1214 * the buffer is already under the length specified it is not modified.
1215 * The skb must be linear.
1217 void skb_trim(struct sk_buff *skb, unsigned int len)
1220 __skb_trim(skb, len);
1222 EXPORT_SYMBOL(skb_trim);
1224 /* Trims skb to length len. It can change skb pointers.
1227 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1229 struct sk_buff **fragp;
1230 struct sk_buff *frag;
1231 int offset = skb_headlen(skb);
1232 int nfrags = skb_shinfo(skb)->nr_frags;
1236 if (skb_cloned(skb) &&
1237 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1244 for (; i < nfrags; i++) {
1245 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1252 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1255 skb_shinfo(skb)->nr_frags = i;
1257 for (; i < nfrags; i++)
1258 skb_frag_unref(skb, i);
1260 if (skb_has_frag_list(skb))
1261 skb_drop_fraglist(skb);
1265 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1266 fragp = &frag->next) {
1267 int end = offset + frag->len;
1269 if (skb_shared(frag)) {
1270 struct sk_buff *nfrag;
1272 nfrag = skb_clone(frag, GFP_ATOMIC);
1273 if (unlikely(!nfrag))
1276 nfrag->next = frag->next;
1288 unlikely((err = pskb_trim(frag, len - offset))))
1292 skb_drop_list(&frag->next);
1297 if (len > skb_headlen(skb)) {
1298 skb->data_len -= skb->len - len;
1303 skb_set_tail_pointer(skb, len);
1308 EXPORT_SYMBOL(___pskb_trim);
1311 * __pskb_pull_tail - advance tail of skb header
1312 * @skb: buffer to reallocate
1313 * @delta: number of bytes to advance tail
1315 * The function makes a sense only on a fragmented &sk_buff,
1316 * it expands header moving its tail forward and copying necessary
1317 * data from fragmented part.
1319 * &sk_buff MUST have reference count of 1.
1321 * Returns %NULL (and &sk_buff does not change) if pull failed
1322 * or value of new tail of skb in the case of success.
1324 * All the pointers pointing into skb header may change and must be
1325 * reloaded after call to this function.
1328 /* Moves tail of skb head forward, copying data from fragmented part,
1329 * when it is necessary.
1330 * 1. It may fail due to malloc failure.
1331 * 2. It may change skb pointers.
1333 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1335 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1337 /* If skb has not enough free space at tail, get new one
1338 * plus 128 bytes for future expansions. If we have enough
1339 * room at tail, reallocate without expansion only if skb is cloned.
1341 int i, k, eat = (skb->tail + delta) - skb->end;
1343 if (eat > 0 || skb_cloned(skb)) {
1344 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1349 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1352 /* Optimization: no fragments, no reasons to preestimate
1353 * size of pulled pages. Superb.
1355 if (!skb_has_frag_list(skb))
1358 /* Estimate size of pulled pages. */
1360 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1361 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1368 /* If we need update frag list, we are in troubles.
1369 * Certainly, it possible to add an offset to skb data,
1370 * but taking into account that pulling is expected to
1371 * be very rare operation, it is worth to fight against
1372 * further bloating skb head and crucify ourselves here instead.
1373 * Pure masohism, indeed. 8)8)
1376 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1377 struct sk_buff *clone = NULL;
1378 struct sk_buff *insp = NULL;
1383 if (list->len <= eat) {
1384 /* Eaten as whole. */
1389 /* Eaten partially. */
1391 if (skb_shared(list)) {
1392 /* Sucks! We need to fork list. :-( */
1393 clone = skb_clone(list, GFP_ATOMIC);
1399 /* This may be pulled without
1403 if (!pskb_pull(list, eat)) {
1411 /* Free pulled out fragments. */
1412 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1413 skb_shinfo(skb)->frag_list = list->next;
1416 /* And insert new clone at head. */
1419 skb_shinfo(skb)->frag_list = clone;
1422 /* Success! Now we may commit changes to skb data. */
1427 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1428 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1431 skb_frag_unref(skb, i);
1434 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1436 skb_shinfo(skb)->frags[k].page_offset += eat;
1437 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1443 skb_shinfo(skb)->nr_frags = k;
1446 skb->data_len -= delta;
1448 return skb_tail_pointer(skb);
1450 EXPORT_SYMBOL(__pskb_pull_tail);
1453 * skb_copy_bits - copy bits from skb to kernel buffer
1455 * @offset: offset in source
1456 * @to: destination buffer
1457 * @len: number of bytes to copy
1459 * Copy the specified number of bytes from the source skb to the
1460 * destination buffer.
1463 * If its prototype is ever changed,
1464 * check arch/{*}/net/{*}.S files,
1465 * since it is called from BPF assembly code.
1467 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1469 int start = skb_headlen(skb);
1470 struct sk_buff *frag_iter;
1473 if (offset > (int)skb->len - len)
1477 if ((copy = start - offset) > 0) {
1480 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1481 if ((len -= copy) == 0)
1487 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1490 WARN_ON(start > offset + len);
1492 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1493 if ((copy = end - offset) > 0) {
1499 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1501 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1502 offset - start, copy);
1503 kunmap_skb_frag(vaddr);
1505 if ((len -= copy) == 0)
1513 skb_walk_frags(skb, frag_iter) {
1516 WARN_ON(start > offset + len);
1518 end = start + frag_iter->len;
1519 if ((copy = end - offset) > 0) {
1522 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1524 if ((len -= copy) == 0)
1538 EXPORT_SYMBOL(skb_copy_bits);
1541 * Callback from splice_to_pipe(), if we need to release some pages
1542 * at the end of the spd in case we error'ed out in filling the pipe.
1544 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1546 put_page(spd->pages[i]);
1549 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1550 unsigned int *offset,
1551 struct sk_buff *skb, struct sock *sk)
1553 struct page *p = sk->sk_sndmsg_page;
1558 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1562 off = sk->sk_sndmsg_off = 0;
1563 /* hold one ref to this page until it's full */
1567 off = sk->sk_sndmsg_off;
1568 mlen = PAGE_SIZE - off;
1569 if (mlen < 64 && mlen < *len) {
1574 *len = min_t(unsigned int, *len, mlen);
1577 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1578 sk->sk_sndmsg_off += *len;
1586 * Fill page/offset/length into spd, if it can hold more pages.
1588 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1589 struct pipe_inode_info *pipe, struct page *page,
1590 unsigned int *len, unsigned int offset,
1591 struct sk_buff *skb, int linear,
1594 if (unlikely(spd->nr_pages == pipe->buffers))
1598 page = linear_to_page(page, len, &offset, skb, sk);
1604 spd->pages[spd->nr_pages] = page;
1605 spd->partial[spd->nr_pages].len = *len;
1606 spd->partial[spd->nr_pages].offset = offset;
1612 static inline void __segment_seek(struct page **page, unsigned int *poff,
1613 unsigned int *plen, unsigned int off)
1618 n = *poff / PAGE_SIZE;
1620 *page = nth_page(*page, n);
1622 *poff = *poff % PAGE_SIZE;
1626 static inline int __splice_segment(struct page *page, unsigned int poff,
1627 unsigned int plen, unsigned int *off,
1628 unsigned int *len, struct sk_buff *skb,
1629 struct splice_pipe_desc *spd, int linear,
1631 struct pipe_inode_info *pipe)
1636 /* skip this segment if already processed */
1642 /* ignore any bits we already processed */
1644 __segment_seek(&page, &poff, &plen, *off);
1649 unsigned int flen = min(*len, plen);
1651 /* the linear region may spread across several pages */
1652 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1654 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1657 __segment_seek(&page, &poff, &plen, flen);
1660 } while (*len && plen);
1666 * Map linear and fragment data from the skb to spd. It reports failure if the
1667 * pipe is full or if we already spliced the requested length.
1669 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1670 unsigned int *offset, unsigned int *len,
1671 struct splice_pipe_desc *spd, struct sock *sk)
1676 * map the linear part
1678 if (__splice_segment(virt_to_page(skb->data),
1679 (unsigned long) skb->data & (PAGE_SIZE - 1),
1681 offset, len, skb, spd, 1, sk, pipe))
1685 * then map the fragments
1687 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1688 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1690 if (__splice_segment(skb_frag_page(f),
1691 f->page_offset, skb_frag_size(f),
1692 offset, len, skb, spd, 0, sk, pipe))
1700 * Map data from the skb to a pipe. Should handle both the linear part,
1701 * the fragments, and the frag list. It does NOT handle frag lists within
1702 * the frag list, if such a thing exists. We'd probably need to recurse to
1703 * handle that cleanly.
1705 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1706 struct pipe_inode_info *pipe, unsigned int tlen,
1709 struct partial_page partial[PIPE_DEF_BUFFERS];
1710 struct page *pages[PIPE_DEF_BUFFERS];
1711 struct splice_pipe_desc spd = {
1715 .ops = &sock_pipe_buf_ops,
1716 .spd_release = sock_spd_release,
1718 struct sk_buff *frag_iter;
1719 struct sock *sk = skb->sk;
1722 if (splice_grow_spd(pipe, &spd))
1726 * __skb_splice_bits() only fails if the output has no room left,
1727 * so no point in going over the frag_list for the error case.
1729 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1735 * now see if we have a frag_list to map
1737 skb_walk_frags(skb, frag_iter) {
1740 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1747 * Drop the socket lock, otherwise we have reverse
1748 * locking dependencies between sk_lock and i_mutex
1749 * here as compared to sendfile(). We enter here
1750 * with the socket lock held, and splice_to_pipe() will
1751 * grab the pipe inode lock. For sendfile() emulation,
1752 * we call into ->sendpage() with the i_mutex lock held
1753 * and networking will grab the socket lock.
1756 ret = splice_to_pipe(pipe, &spd);
1760 splice_shrink_spd(pipe, &spd);
1765 * skb_store_bits - store bits from kernel buffer to skb
1766 * @skb: destination buffer
1767 * @offset: offset in destination
1768 * @from: source buffer
1769 * @len: number of bytes to copy
1771 * Copy the specified number of bytes from the source buffer to the
1772 * destination skb. This function handles all the messy bits of
1773 * traversing fragment lists and such.
1776 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1778 int start = skb_headlen(skb);
1779 struct sk_buff *frag_iter;
1782 if (offset > (int)skb->len - len)
1785 if ((copy = start - offset) > 0) {
1788 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1789 if ((len -= copy) == 0)
1795 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1796 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1799 WARN_ON(start > offset + len);
1801 end = start + skb_frag_size(frag);
1802 if ((copy = end - offset) > 0) {
1808 vaddr = kmap_skb_frag(frag);
1809 memcpy(vaddr + frag->page_offset + offset - start,
1811 kunmap_skb_frag(vaddr);
1813 if ((len -= copy) == 0)
1821 skb_walk_frags(skb, frag_iter) {
1824 WARN_ON(start > offset + len);
1826 end = start + frag_iter->len;
1827 if ((copy = end - offset) > 0) {
1830 if (skb_store_bits(frag_iter, offset - start,
1833 if ((len -= copy) == 0)
1846 EXPORT_SYMBOL(skb_store_bits);
1848 /* Checksum skb data. */
1850 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1851 int len, __wsum csum)
1853 int start = skb_headlen(skb);
1854 int i, copy = start - offset;
1855 struct sk_buff *frag_iter;
1858 /* Checksum header. */
1862 csum = csum_partial(skb->data + offset, copy, csum);
1863 if ((len -= copy) == 0)
1869 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1872 WARN_ON(start > offset + len);
1874 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1875 if ((copy = end - offset) > 0) {
1878 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1882 vaddr = kmap_skb_frag(frag);
1883 csum2 = csum_partial(vaddr + frag->page_offset +
1884 offset - start, copy, 0);
1885 kunmap_skb_frag(vaddr);
1886 csum = csum_block_add(csum, csum2, pos);
1895 skb_walk_frags(skb, frag_iter) {
1898 WARN_ON(start > offset + len);
1900 end = start + frag_iter->len;
1901 if ((copy = end - offset) > 0) {
1905 csum2 = skb_checksum(frag_iter, offset - start,
1907 csum = csum_block_add(csum, csum2, pos);
1908 if ((len -= copy) == 0)
1919 EXPORT_SYMBOL(skb_checksum);
1921 /* Both of above in one bottle. */
1923 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1924 u8 *to, int len, __wsum csum)
1926 int start = skb_headlen(skb);
1927 int i, copy = start - offset;
1928 struct sk_buff *frag_iter;
1935 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1937 if ((len -= copy) == 0)
1944 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1947 WARN_ON(start > offset + len);
1949 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1950 if ((copy = end - offset) > 0) {
1953 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1957 vaddr = kmap_skb_frag(frag);
1958 csum2 = csum_partial_copy_nocheck(vaddr +
1962 kunmap_skb_frag(vaddr);
1963 csum = csum_block_add(csum, csum2, pos);
1973 skb_walk_frags(skb, frag_iter) {
1977 WARN_ON(start > offset + len);
1979 end = start + frag_iter->len;
1980 if ((copy = end - offset) > 0) {
1983 csum2 = skb_copy_and_csum_bits(frag_iter,
1986 csum = csum_block_add(csum, csum2, pos);
1987 if ((len -= copy) == 0)
1998 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2000 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2005 if (skb->ip_summed == CHECKSUM_PARTIAL)
2006 csstart = skb_checksum_start_offset(skb);
2008 csstart = skb_headlen(skb);
2010 BUG_ON(csstart > skb_headlen(skb));
2012 skb_copy_from_linear_data(skb, to, csstart);
2015 if (csstart != skb->len)
2016 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2017 skb->len - csstart, 0);
2019 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2020 long csstuff = csstart + skb->csum_offset;
2022 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2025 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2028 * skb_dequeue - remove from the head of the queue
2029 * @list: list to dequeue from
2031 * Remove the head of the list. The list lock is taken so the function
2032 * may be used safely with other locking list functions. The head item is
2033 * returned or %NULL if the list is empty.
2036 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2038 unsigned long flags;
2039 struct sk_buff *result;
2041 spin_lock_irqsave(&list->lock, flags);
2042 result = __skb_dequeue(list);
2043 spin_unlock_irqrestore(&list->lock, flags);
2046 EXPORT_SYMBOL(skb_dequeue);
2049 * skb_dequeue_tail - remove from the tail of the queue
2050 * @list: list to dequeue from
2052 * Remove the tail of the list. The list lock is taken so the function
2053 * may be used safely with other locking list functions. The tail item is
2054 * returned or %NULL if the list is empty.
2056 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2058 unsigned long flags;
2059 struct sk_buff *result;
2061 spin_lock_irqsave(&list->lock, flags);
2062 result = __skb_dequeue_tail(list);
2063 spin_unlock_irqrestore(&list->lock, flags);
2066 EXPORT_SYMBOL(skb_dequeue_tail);
2069 * skb_queue_purge - empty a list
2070 * @list: list to empty
2072 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2073 * the list and one reference dropped. This function takes the list
2074 * lock and is atomic with respect to other list locking functions.
2076 void skb_queue_purge(struct sk_buff_head *list)
2078 struct sk_buff *skb;
2079 while ((skb = skb_dequeue(list)) != NULL)
2082 EXPORT_SYMBOL(skb_queue_purge);
2085 * skb_queue_head - queue a buffer at the list head
2086 * @list: list to use
2087 * @newsk: buffer to queue
2089 * Queue a buffer at the start of the list. This function takes the
2090 * list lock and can be used safely with other locking &sk_buff functions
2093 * A buffer cannot be placed on two lists at the same time.
2095 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2097 unsigned long flags;
2099 spin_lock_irqsave(&list->lock, flags);
2100 __skb_queue_head(list, newsk);
2101 spin_unlock_irqrestore(&list->lock, flags);
2103 EXPORT_SYMBOL(skb_queue_head);
2106 * skb_queue_tail - queue a buffer at the list tail
2107 * @list: list to use
2108 * @newsk: buffer to queue
2110 * Queue a buffer at the tail of the list. This function takes the
2111 * list lock and can be used safely with other locking &sk_buff functions
2114 * A buffer cannot be placed on two lists at the same time.
2116 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2118 unsigned long flags;
2120 spin_lock_irqsave(&list->lock, flags);
2121 __skb_queue_tail(list, newsk);
2122 spin_unlock_irqrestore(&list->lock, flags);
2124 EXPORT_SYMBOL(skb_queue_tail);
2127 * skb_unlink - remove a buffer from a list
2128 * @skb: buffer to remove
2129 * @list: list to use
2131 * Remove a packet from a list. The list locks are taken and this
2132 * function is atomic with respect to other list locked calls
2134 * You must know what list the SKB is on.
2136 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2138 unsigned long flags;
2140 spin_lock_irqsave(&list->lock, flags);
2141 __skb_unlink(skb, list);
2142 spin_unlock_irqrestore(&list->lock, flags);
2144 EXPORT_SYMBOL(skb_unlink);
2147 * skb_append - append a buffer
2148 * @old: buffer to insert after
2149 * @newsk: buffer to insert
2150 * @list: list to use
2152 * Place a packet after a given packet in a list. The list locks are taken
2153 * and this function is atomic with respect to other list locked calls.
2154 * A buffer cannot be placed on two lists at the same time.
2156 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2158 unsigned long flags;
2160 spin_lock_irqsave(&list->lock, flags);
2161 __skb_queue_after(list, old, newsk);
2162 spin_unlock_irqrestore(&list->lock, flags);
2164 EXPORT_SYMBOL(skb_append);
2167 * skb_insert - insert a buffer
2168 * @old: buffer to insert before
2169 * @newsk: buffer to insert
2170 * @list: list to use
2172 * Place a packet before a given packet in a list. The list locks are
2173 * taken and this function is atomic with respect to other list locked
2176 * A buffer cannot be placed on two lists at the same time.
2178 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2180 unsigned long flags;
2182 spin_lock_irqsave(&list->lock, flags);
2183 __skb_insert(newsk, old->prev, old, list);
2184 spin_unlock_irqrestore(&list->lock, flags);
2186 EXPORT_SYMBOL(skb_insert);
2188 static inline void skb_split_inside_header(struct sk_buff *skb,
2189 struct sk_buff* skb1,
2190 const u32 len, const int pos)
2194 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2196 /* And move data appendix as is. */
2197 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2198 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2200 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2201 skb_shinfo(skb)->nr_frags = 0;
2202 skb1->data_len = skb->data_len;
2203 skb1->len += skb1->data_len;
2206 skb_set_tail_pointer(skb, len);
2209 static inline void skb_split_no_header(struct sk_buff *skb,
2210 struct sk_buff* skb1,
2211 const u32 len, int pos)
2214 const int nfrags = skb_shinfo(skb)->nr_frags;
2216 skb_shinfo(skb)->nr_frags = 0;
2217 skb1->len = skb1->data_len = skb->len - len;
2219 skb->data_len = len - pos;
2221 for (i = 0; i < nfrags; i++) {
2222 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2224 if (pos + size > len) {
2225 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2229 * We have two variants in this case:
2230 * 1. Move all the frag to the second
2231 * part, if it is possible. F.e.
2232 * this approach is mandatory for TUX,
2233 * where splitting is expensive.
2234 * 2. Split is accurately. We make this.
2236 skb_frag_ref(skb, i);
2237 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2238 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2239 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2240 skb_shinfo(skb)->nr_frags++;
2244 skb_shinfo(skb)->nr_frags++;
2247 skb_shinfo(skb1)->nr_frags = k;
2251 * skb_split - Split fragmented skb to two parts at length len.
2252 * @skb: the buffer to split
2253 * @skb1: the buffer to receive the second part
2254 * @len: new length for skb
2256 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2258 int pos = skb_headlen(skb);
2260 if (len < pos) /* Split line is inside header. */
2261 skb_split_inside_header(skb, skb1, len, pos);
2262 else /* Second chunk has no header, nothing to copy. */
2263 skb_split_no_header(skb, skb1, len, pos);
2265 EXPORT_SYMBOL(skb_split);
2267 /* Shifting from/to a cloned skb is a no-go.
2269 * Caller cannot keep skb_shinfo related pointers past calling here!
2271 static int skb_prepare_for_shift(struct sk_buff *skb)
2273 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2277 * skb_shift - Shifts paged data partially from skb to another
2278 * @tgt: buffer into which tail data gets added
2279 * @skb: buffer from which the paged data comes from
2280 * @shiftlen: shift up to this many bytes
2282 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2283 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2284 * It's up to caller to free skb if everything was shifted.
2286 * If @tgt runs out of frags, the whole operation is aborted.
2288 * Skb cannot include anything else but paged data while tgt is allowed
2289 * to have non-paged data as well.
2291 * TODO: full sized shift could be optimized but that would need
2292 * specialized skb free'er to handle frags without up-to-date nr_frags.
2294 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2296 int from, to, merge, todo;
2297 struct skb_frag_struct *fragfrom, *fragto;
2299 BUG_ON(shiftlen > skb->len);
2300 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2304 to = skb_shinfo(tgt)->nr_frags;
2305 fragfrom = &skb_shinfo(skb)->frags[from];
2307 /* Actual merge is delayed until the point when we know we can
2308 * commit all, so that we don't have to undo partial changes
2311 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2312 fragfrom->page_offset)) {
2317 todo -= skb_frag_size(fragfrom);
2319 if (skb_prepare_for_shift(skb) ||
2320 skb_prepare_for_shift(tgt))
2323 /* All previous frag pointers might be stale! */
2324 fragfrom = &skb_shinfo(skb)->frags[from];
2325 fragto = &skb_shinfo(tgt)->frags[merge];
2327 skb_frag_size_add(fragto, shiftlen);
2328 skb_frag_size_sub(fragfrom, shiftlen);
2329 fragfrom->page_offset += shiftlen;
2337 /* Skip full, not-fitting skb to avoid expensive operations */
2338 if ((shiftlen == skb->len) &&
2339 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2342 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2345 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2346 if (to == MAX_SKB_FRAGS)
2349 fragfrom = &skb_shinfo(skb)->frags[from];
2350 fragto = &skb_shinfo(tgt)->frags[to];
2352 if (todo >= skb_frag_size(fragfrom)) {
2353 *fragto = *fragfrom;
2354 todo -= skb_frag_size(fragfrom);
2359 __skb_frag_ref(fragfrom);
2360 fragto->page = fragfrom->page;
2361 fragto->page_offset = fragfrom->page_offset;
2362 skb_frag_size_set(fragto, todo);
2364 fragfrom->page_offset += todo;
2365 skb_frag_size_sub(fragfrom, todo);
2373 /* Ready to "commit" this state change to tgt */
2374 skb_shinfo(tgt)->nr_frags = to;
2377 fragfrom = &skb_shinfo(skb)->frags[0];
2378 fragto = &skb_shinfo(tgt)->frags[merge];
2380 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2381 __skb_frag_unref(fragfrom);
2384 /* Reposition in the original skb */
2386 while (from < skb_shinfo(skb)->nr_frags)
2387 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2388 skb_shinfo(skb)->nr_frags = to;
2390 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2393 /* Most likely the tgt won't ever need its checksum anymore, skb on
2394 * the other hand might need it if it needs to be resent
2396 tgt->ip_summed = CHECKSUM_PARTIAL;
2397 skb->ip_summed = CHECKSUM_PARTIAL;
2399 /* Yak, is it really working this way? Some helper please? */
2400 skb->len -= shiftlen;
2401 skb->data_len -= shiftlen;
2402 skb->truesize -= shiftlen;
2403 tgt->len += shiftlen;
2404 tgt->data_len += shiftlen;
2405 tgt->truesize += shiftlen;
2411 * skb_prepare_seq_read - Prepare a sequential read of skb data
2412 * @skb: the buffer to read
2413 * @from: lower offset of data to be read
2414 * @to: upper offset of data to be read
2415 * @st: state variable
2417 * Initializes the specified state variable. Must be called before
2418 * invoking skb_seq_read() for the first time.
2420 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2421 unsigned int to, struct skb_seq_state *st)
2423 st->lower_offset = from;
2424 st->upper_offset = to;
2425 st->root_skb = st->cur_skb = skb;
2426 st->frag_idx = st->stepped_offset = 0;
2427 st->frag_data = NULL;
2429 EXPORT_SYMBOL(skb_prepare_seq_read);
2432 * skb_seq_read - Sequentially read skb data
2433 * @consumed: number of bytes consumed by the caller so far
2434 * @data: destination pointer for data to be returned
2435 * @st: state variable
2437 * Reads a block of skb data at &consumed relative to the
2438 * lower offset specified to skb_prepare_seq_read(). Assigns
2439 * the head of the data block to &data and returns the length
2440 * of the block or 0 if the end of the skb data or the upper
2441 * offset has been reached.
2443 * The caller is not required to consume all of the data
2444 * returned, i.e. &consumed is typically set to the number
2445 * of bytes already consumed and the next call to
2446 * skb_seq_read() will return the remaining part of the block.
2448 * Note 1: The size of each block of data returned can be arbitrary,
2449 * this limitation is the cost for zerocopy seqeuental
2450 * reads of potentially non linear data.
2452 * Note 2: Fragment lists within fragments are not implemented
2453 * at the moment, state->root_skb could be replaced with
2454 * a stack for this purpose.
2456 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2457 struct skb_seq_state *st)
2459 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2462 if (unlikely(abs_offset >= st->upper_offset))
2466 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2468 if (abs_offset < block_limit && !st->frag_data) {
2469 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2470 return block_limit - abs_offset;
2473 if (st->frag_idx == 0 && !st->frag_data)
2474 st->stepped_offset += skb_headlen(st->cur_skb);
2476 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2477 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2478 block_limit = skb_frag_size(frag) + st->stepped_offset;
2480 if (abs_offset < block_limit) {
2482 st->frag_data = kmap_skb_frag(frag);
2484 *data = (u8 *) st->frag_data + frag->page_offset +
2485 (abs_offset - st->stepped_offset);
2487 return block_limit - abs_offset;
2490 if (st->frag_data) {
2491 kunmap_skb_frag(st->frag_data);
2492 st->frag_data = NULL;
2496 st->stepped_offset += skb_frag_size(frag);
2499 if (st->frag_data) {
2500 kunmap_skb_frag(st->frag_data);
2501 st->frag_data = NULL;
2504 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2505 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2508 } else if (st->cur_skb->next) {
2509 st->cur_skb = st->cur_skb->next;
2516 EXPORT_SYMBOL(skb_seq_read);
2519 * skb_abort_seq_read - Abort a sequential read of skb data
2520 * @st: state variable
2522 * Must be called if skb_seq_read() was not called until it
2525 void skb_abort_seq_read(struct skb_seq_state *st)
2528 kunmap_skb_frag(st->frag_data);
2530 EXPORT_SYMBOL(skb_abort_seq_read);
2532 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2534 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2535 struct ts_config *conf,
2536 struct ts_state *state)
2538 return skb_seq_read(offset, text, TS_SKB_CB(state));
2541 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2543 skb_abort_seq_read(TS_SKB_CB(state));
2547 * skb_find_text - Find a text pattern in skb data
2548 * @skb: the buffer to look in
2549 * @from: search offset
2551 * @config: textsearch configuration
2552 * @state: uninitialized textsearch state variable
2554 * Finds a pattern in the skb data according to the specified
2555 * textsearch configuration. Use textsearch_next() to retrieve
2556 * subsequent occurrences of the pattern. Returns the offset
2557 * to the first occurrence or UINT_MAX if no match was found.
2559 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2560 unsigned int to, struct ts_config *config,
2561 struct ts_state *state)
2565 config->get_next_block = skb_ts_get_next_block;
2566 config->finish = skb_ts_finish;
2568 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2570 ret = textsearch_find(config, state);
2571 return (ret <= to - from ? ret : UINT_MAX);
2573 EXPORT_SYMBOL(skb_find_text);
2576 * skb_append_datato_frags: - append the user data to a skb
2577 * @sk: sock structure
2578 * @skb: skb structure to be appened with user data.
2579 * @getfrag: call back function to be used for getting the user data
2580 * @from: pointer to user message iov
2581 * @length: length of the iov message
2583 * Description: This procedure append the user data in the fragment part
2584 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2586 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2587 int (*getfrag)(void *from, char *to, int offset,
2588 int len, int odd, struct sk_buff *skb),
2589 void *from, int length)
2592 skb_frag_t *frag = NULL;
2593 struct page *page = NULL;
2599 /* Return error if we don't have space for new frag */
2600 frg_cnt = skb_shinfo(skb)->nr_frags;
2601 if (frg_cnt >= MAX_SKB_FRAGS)
2604 /* allocate a new page for next frag */
2605 page = alloc_pages(sk->sk_allocation, 0);
2607 /* If alloc_page fails just return failure and caller will
2608 * free previous allocated pages by doing kfree_skb()
2613 /* initialize the next frag */
2614 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2615 skb->truesize += PAGE_SIZE;
2616 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2618 /* get the new initialized frag */
2619 frg_cnt = skb_shinfo(skb)->nr_frags;
2620 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2622 /* copy the user data to page */
2623 left = PAGE_SIZE - frag->page_offset;
2624 copy = (length > left)? left : length;
2626 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2627 offset, copy, 0, skb);
2631 /* copy was successful so update the size parameters */
2632 skb_frag_size_add(frag, copy);
2634 skb->data_len += copy;
2638 } while (length > 0);
2642 EXPORT_SYMBOL(skb_append_datato_frags);
2645 * skb_pull_rcsum - pull skb and update receive checksum
2646 * @skb: buffer to update
2647 * @len: length of data pulled
2649 * This function performs an skb_pull on the packet and updates
2650 * the CHECKSUM_COMPLETE checksum. It should be used on
2651 * receive path processing instead of skb_pull unless you know
2652 * that the checksum difference is zero (e.g., a valid IP header)
2653 * or you are setting ip_summed to CHECKSUM_NONE.
2655 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2657 BUG_ON(len > skb->len);
2659 BUG_ON(skb->len < skb->data_len);
2660 skb_postpull_rcsum(skb, skb->data, len);
2661 return skb->data += len;
2663 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2666 * skb_segment - Perform protocol segmentation on skb.
2667 * @skb: buffer to segment
2668 * @features: features for the output path (see dev->features)
2670 * This function performs segmentation on the given skb. It returns
2671 * a pointer to the first in a list of new skbs for the segments.
2672 * In case of error it returns ERR_PTR(err).
2674 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2676 struct sk_buff *segs = NULL;
2677 struct sk_buff *tail = NULL;
2678 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2679 unsigned int mss = skb_shinfo(skb)->gso_size;
2680 unsigned int doffset = skb->data - skb_mac_header(skb);
2681 unsigned int offset = doffset;
2682 unsigned int headroom;
2684 int sg = !!(features & NETIF_F_SG);
2685 int nfrags = skb_shinfo(skb)->nr_frags;
2690 __skb_push(skb, doffset);
2691 headroom = skb_headroom(skb);
2692 pos = skb_headlen(skb);
2695 struct sk_buff *nskb;
2700 len = skb->len - offset;
2704 hsize = skb_headlen(skb) - offset;
2707 if (hsize > len || !sg)
2710 if (!hsize && i >= nfrags) {
2711 BUG_ON(fskb->len != len);
2714 nskb = skb_clone(fskb, GFP_ATOMIC);
2717 if (unlikely(!nskb))
2720 hsize = skb_end_pointer(nskb) - nskb->head;
2721 if (skb_cow_head(nskb, doffset + headroom)) {
2726 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2728 skb_release_head_state(nskb);
2729 __skb_push(nskb, doffset);
2731 nskb = alloc_skb(hsize + doffset + headroom,
2734 if (unlikely(!nskb))
2737 skb_reserve(nskb, headroom);
2738 __skb_put(nskb, doffset);
2747 __copy_skb_header(nskb, skb);
2748 nskb->mac_len = skb->mac_len;
2750 /* nskb and skb might have different headroom */
2751 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2752 nskb->csum_start += skb_headroom(nskb) - headroom;
2754 skb_reset_mac_header(nskb);
2755 skb_set_network_header(nskb, skb->mac_len);
2756 nskb->transport_header = (nskb->network_header +
2757 skb_network_header_len(skb));
2758 skb_copy_from_linear_data(skb, nskb->data, doffset);
2760 if (fskb != skb_shinfo(skb)->frag_list)
2764 nskb->ip_summed = CHECKSUM_NONE;
2765 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2771 frag = skb_shinfo(nskb)->frags;
2773 skb_copy_from_linear_data_offset(skb, offset,
2774 skb_put(nskb, hsize), hsize);
2776 while (pos < offset + len && i < nfrags) {
2777 *frag = skb_shinfo(skb)->frags[i];
2778 __skb_frag_ref(frag);
2779 size = skb_frag_size(frag);
2782 frag->page_offset += offset - pos;
2783 skb_frag_size_sub(frag, offset - pos);
2786 skb_shinfo(nskb)->nr_frags++;
2788 if (pos + size <= offset + len) {
2792 skb_frag_size_sub(frag, pos + size - (offset + len));
2799 if (pos < offset + len) {
2800 struct sk_buff *fskb2 = fskb;
2802 BUG_ON(pos + fskb->len != offset + len);
2808 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2814 SKB_FRAG_ASSERT(nskb);
2815 skb_shinfo(nskb)->frag_list = fskb2;
2819 nskb->data_len = len - hsize;
2820 nskb->len += nskb->data_len;
2821 nskb->truesize += nskb->data_len;
2822 } while ((offset += len) < skb->len);
2827 while ((skb = segs)) {
2831 return ERR_PTR(err);
2833 EXPORT_SYMBOL_GPL(skb_segment);
2835 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2837 struct sk_buff *p = *head;
2838 struct sk_buff *nskb;
2839 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2840 struct skb_shared_info *pinfo = skb_shinfo(p);
2841 unsigned int headroom;
2842 unsigned int len = skb_gro_len(skb);
2843 unsigned int offset = skb_gro_offset(skb);
2844 unsigned int headlen = skb_headlen(skb);
2846 if (p->len + len >= 65536)
2849 if (pinfo->frag_list)
2851 else if (headlen <= offset) {
2854 int i = skbinfo->nr_frags;
2855 int nr_frags = pinfo->nr_frags + i;
2859 if (nr_frags > MAX_SKB_FRAGS)
2862 pinfo->nr_frags = nr_frags;
2863 skbinfo->nr_frags = 0;
2865 frag = pinfo->frags + nr_frags;
2866 frag2 = skbinfo->frags + i;
2871 frag->page_offset += offset;
2872 skb_frag_size_sub(frag, offset);
2874 skb->truesize -= skb->data_len;
2875 skb->len -= skb->data_len;
2878 NAPI_GRO_CB(skb)->free = 1;
2880 } else if (skb_gro_len(p) != pinfo->gso_size)
2883 headroom = skb_headroom(p);
2884 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2885 if (unlikely(!nskb))
2888 __copy_skb_header(nskb, p);
2889 nskb->mac_len = p->mac_len;
2891 skb_reserve(nskb, headroom);
2892 __skb_put(nskb, skb_gro_offset(p));
2894 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2895 skb_set_network_header(nskb, skb_network_offset(p));
2896 skb_set_transport_header(nskb, skb_transport_offset(p));
2898 __skb_pull(p, skb_gro_offset(p));
2899 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2900 p->data - skb_mac_header(p));
2902 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2903 skb_shinfo(nskb)->frag_list = p;
2904 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2905 pinfo->gso_size = 0;
2906 skb_header_release(p);
2909 nskb->data_len += p->len;
2910 nskb->truesize += p->truesize;
2911 nskb->len += p->len;
2914 nskb->next = p->next;
2920 p->truesize += skb->truesize - len;
2921 if (offset > headlen) {
2922 unsigned int eat = offset - headlen;
2924 skbinfo->frags[0].page_offset += eat;
2925 skb_frag_size_sub(&skbinfo->frags[0], eat);
2926 skb->data_len -= eat;
2931 __skb_pull(skb, offset);
2933 p->prev->next = skb;
2935 skb_header_release(skb);
2938 NAPI_GRO_CB(p)->count++;
2943 NAPI_GRO_CB(skb)->same_flow = 1;
2946 EXPORT_SYMBOL_GPL(skb_gro_receive);
2948 void __init skb_init(void)
2950 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2951 sizeof(struct sk_buff),
2953 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2955 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2956 (2*sizeof(struct sk_buff)) +
2959 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2964 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2965 * @skb: Socket buffer containing the buffers to be mapped
2966 * @sg: The scatter-gather list to map into
2967 * @offset: The offset into the buffer's contents to start mapping
2968 * @len: Length of buffer space to be mapped
2970 * Fill the specified scatter-gather list with mappings/pointers into a
2971 * region of the buffer space attached to a socket buffer.
2974 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2976 int start = skb_headlen(skb);
2977 int i, copy = start - offset;
2978 struct sk_buff *frag_iter;
2984 sg_set_buf(sg, skb->data + offset, copy);
2986 if ((len -= copy) == 0)
2991 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2994 WARN_ON(start > offset + len);
2996 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2997 if ((copy = end - offset) > 0) {
2998 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3002 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3003 frag->page_offset+offset-start);
3012 skb_walk_frags(skb, frag_iter) {
3015 WARN_ON(start > offset + len);
3017 end = start + frag_iter->len;
3018 if ((copy = end - offset) > 0) {
3021 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3023 if ((len -= copy) == 0)
3033 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3035 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3037 sg_mark_end(&sg[nsg - 1]);
3041 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3044 * skb_cow_data - Check that a socket buffer's data buffers are writable
3045 * @skb: The socket buffer to check.
3046 * @tailbits: Amount of trailing space to be added
3047 * @trailer: Returned pointer to the skb where the @tailbits space begins
3049 * Make sure that the data buffers attached to a socket buffer are
3050 * writable. If they are not, private copies are made of the data buffers
3051 * and the socket buffer is set to use these instead.
3053 * If @tailbits is given, make sure that there is space to write @tailbits
3054 * bytes of data beyond current end of socket buffer. @trailer will be
3055 * set to point to the skb in which this space begins.
3057 * The number of scatterlist elements required to completely map the
3058 * COW'd and extended socket buffer will be returned.
3060 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3064 struct sk_buff *skb1, **skb_p;
3066 /* If skb is cloned or its head is paged, reallocate
3067 * head pulling out all the pages (pages are considered not writable
3068 * at the moment even if they are anonymous).
3070 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3071 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3074 /* Easy case. Most of packets will go this way. */
3075 if (!skb_has_frag_list(skb)) {
3076 /* A little of trouble, not enough of space for trailer.
3077 * This should not happen, when stack is tuned to generate
3078 * good frames. OK, on miss we reallocate and reserve even more
3079 * space, 128 bytes is fair. */
3081 if (skb_tailroom(skb) < tailbits &&
3082 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3090 /* Misery. We are in troubles, going to mincer fragments... */
3093 skb_p = &skb_shinfo(skb)->frag_list;
3096 while ((skb1 = *skb_p) != NULL) {
3099 /* The fragment is partially pulled by someone,
3100 * this can happen on input. Copy it and everything
3103 if (skb_shared(skb1))
3106 /* If the skb is the last, worry about trailer. */
3108 if (skb1->next == NULL && tailbits) {
3109 if (skb_shinfo(skb1)->nr_frags ||
3110 skb_has_frag_list(skb1) ||
3111 skb_tailroom(skb1) < tailbits)
3112 ntail = tailbits + 128;
3118 skb_shinfo(skb1)->nr_frags ||
3119 skb_has_frag_list(skb1)) {
3120 struct sk_buff *skb2;
3122 /* Fuck, we are miserable poor guys... */
3124 skb2 = skb_copy(skb1, GFP_ATOMIC);
3126 skb2 = skb_copy_expand(skb1,
3130 if (unlikely(skb2 == NULL))
3134 skb_set_owner_w(skb2, skb1->sk);
3136 /* Looking around. Are we still alive?
3137 * OK, link new skb, drop old one */
3139 skb2->next = skb1->next;
3146 skb_p = &skb1->next;
3151 EXPORT_SYMBOL_GPL(skb_cow_data);
3153 static void sock_rmem_free(struct sk_buff *skb)
3155 struct sock *sk = skb->sk;
3157 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3161 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3163 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3165 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3166 (unsigned)sk->sk_rcvbuf)
3171 skb->destructor = sock_rmem_free;
3172 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3174 /* before exiting rcu section, make sure dst is refcounted */
3177 skb_queue_tail(&sk->sk_error_queue, skb);
3178 if (!sock_flag(sk, SOCK_DEAD))
3179 sk->sk_data_ready(sk, skb->len);
3182 EXPORT_SYMBOL(sock_queue_err_skb);
3184 void skb_tstamp_tx(struct sk_buff *orig_skb,
3185 struct skb_shared_hwtstamps *hwtstamps)
3187 struct sock *sk = orig_skb->sk;
3188 struct sock_exterr_skb *serr;
3189 struct sk_buff *skb;
3195 skb = skb_clone(orig_skb, GFP_ATOMIC);
3200 *skb_hwtstamps(skb) =
3204 * no hardware time stamps available,
3205 * so keep the shared tx_flags and only
3206 * store software time stamp
3208 skb->tstamp = ktime_get_real();
3211 serr = SKB_EXT_ERR(skb);
3212 memset(serr, 0, sizeof(*serr));
3213 serr->ee.ee_errno = ENOMSG;
3214 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3216 err = sock_queue_err_skb(sk, skb);
3221 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3223 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3225 struct sock *sk = skb->sk;
3226 struct sock_exterr_skb *serr;
3229 skb->wifi_acked_valid = 1;
3230 skb->wifi_acked = acked;
3232 serr = SKB_EXT_ERR(skb);
3233 memset(serr, 0, sizeof(*serr));
3234 serr->ee.ee_errno = ENOMSG;
3235 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3237 err = sock_queue_err_skb(sk, skb);
3241 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3245 * skb_partial_csum_set - set up and verify partial csum values for packet
3246 * @skb: the skb to set
3247 * @start: the number of bytes after skb->data to start checksumming.
3248 * @off: the offset from start to place the checksum.
3250 * For untrusted partially-checksummed packets, we need to make sure the values
3251 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3253 * This function checks and sets those values and skb->ip_summed: if this
3254 * returns false you should drop the packet.
3256 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3258 if (unlikely(start > skb_headlen(skb)) ||
3259 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3260 if (net_ratelimit())
3262 "bad partial csum: csum=%u/%u len=%u\n",
3263 start, off, skb_headlen(skb));
3266 skb->ip_summed = CHECKSUM_PARTIAL;
3267 skb->csum_start = skb_headroom(skb) + start;
3268 skb->csum_offset = off;
3271 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3273 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3275 if (net_ratelimit())
3276 pr_warning("%s: received packets cannot be forwarded"
3277 " while LRO is enabled\n", skb->dev->name);
3279 EXPORT_SYMBOL(__skb_warn_lro_forwarding);