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>
61 #include <net/protocol.h>
64 #include <net/checksum.h>
67 #include <asm/uaccess.h>
68 #include <asm/system.h>
69 #include <trace/events/skb.h>
73 static struct kmem_cache *skbuff_head_cache __read_mostly;
74 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
76 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
77 struct pipe_buffer *buf)
82 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
83 struct pipe_buffer *buf)
88 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
89 struct pipe_buffer *buf)
95 /* Pipe buffer operations for a socket. */
96 static const struct pipe_buf_operations sock_pipe_buf_ops = {
98 .map = generic_pipe_buf_map,
99 .unmap = generic_pipe_buf_unmap,
100 .confirm = generic_pipe_buf_confirm,
101 .release = sock_pipe_buf_release,
102 .steal = sock_pipe_buf_steal,
103 .get = sock_pipe_buf_get,
107 * Keep out-of-line to prevent kernel bloat.
108 * __builtin_return_address is not used because it is not always
113 * skb_over_panic - private function
118 * Out of line support code for skb_put(). Not user callable.
120 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
122 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
123 "data:%p tail:%#lx end:%#lx dev:%s\n",
124 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 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here, skb->len, sz, skb->head, skb->data,
144 (unsigned long)skb->tail, (unsigned long)skb->end,
145 skb->dev ? skb->dev->name : "<NULL>");
149 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
156 * __alloc_skb - allocate a network buffer
157 * @size: size to allocate
158 * @gfp_mask: allocation mask
159 * @fclone: allocate from fclone cache instead of head cache
160 * and allocate a cloned (child) skb
161 * @node: numa node to allocate memory on
163 * Allocate a new &sk_buff. The returned buffer has no headroom and a
164 * tail room of size bytes. The object has a reference count of one.
165 * The return is the buffer. On a failure the return is %NULL.
167 * Buffers may only be allocated from interrupts using a @gfp_mask of
170 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
171 int fclone, int node)
173 struct kmem_cache *cache;
174 struct skb_shared_info *shinfo;
178 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
181 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
186 size = SKB_DATA_ALIGN(size);
187 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
191 prefetchw(data + size);
194 * Only clear those fields we need to clear, not those that we will
195 * actually initialise below. Hence, don't put any more fields after
196 * the tail pointer in struct sk_buff!
198 memset(skb, 0, offsetof(struct sk_buff, tail));
199 skb->truesize = size + sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
203 skb_reset_tail_pointer(skb);
204 skb->end = skb->tail + size;
205 #ifdef NET_SKBUFF_DATA_USES_OFFSET
206 skb->mac_header = ~0U;
209 /* make sure we initialize shinfo sequentially */
210 shinfo = skb_shinfo(skb);
211 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
212 atomic_set(&shinfo->dataref, 1);
215 struct sk_buff *child = skb + 1;
216 atomic_t *fclone_ref = (atomic_t *) (child + 1);
218 kmemcheck_annotate_bitfield(child, flags1);
219 kmemcheck_annotate_bitfield(child, flags2);
220 skb->fclone = SKB_FCLONE_ORIG;
221 atomic_set(fclone_ref, 1);
223 child->fclone = SKB_FCLONE_UNAVAILABLE;
228 kmem_cache_free(cache, skb);
232 EXPORT_SYMBOL(__alloc_skb);
235 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
236 * @dev: network device to receive on
237 * @length: length to allocate
238 * @gfp_mask: get_free_pages mask, passed to alloc_skb
240 * Allocate a new &sk_buff and assign it a usage count of one. The
241 * buffer has unspecified headroom built in. Users should allocate
242 * the headroom they think they need without accounting for the
243 * built in space. The built in space is used for optimisations.
245 * %NULL is returned if there is no free memory.
247 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
248 unsigned int length, gfp_t gfp_mask)
252 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
254 skb_reserve(skb, NET_SKB_PAD);
259 EXPORT_SYMBOL(__netdev_alloc_skb);
261 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
264 skb_fill_page_desc(skb, i, page, off, size);
266 skb->data_len += size;
267 skb->truesize += size;
269 EXPORT_SYMBOL(skb_add_rx_frag);
272 * dev_alloc_skb - allocate an skbuff for receiving
273 * @length: length to allocate
275 * Allocate a new &sk_buff and assign it a usage count of one. The
276 * buffer has unspecified headroom built in. Users should allocate
277 * the headroom they think they need without accounting for the
278 * built in space. The built in space is used for optimisations.
280 * %NULL is returned if there is no free memory. Although this function
281 * allocates memory it can be called from an interrupt.
283 struct sk_buff *dev_alloc_skb(unsigned int length)
286 * There is more code here than it seems:
287 * __dev_alloc_skb is an inline
289 return __dev_alloc_skb(length, GFP_ATOMIC);
291 EXPORT_SYMBOL(dev_alloc_skb);
293 static void skb_drop_list(struct sk_buff **listp)
295 struct sk_buff *list = *listp;
300 struct sk_buff *this = list;
306 static inline void skb_drop_fraglist(struct sk_buff *skb)
308 skb_drop_list(&skb_shinfo(skb)->frag_list);
311 static void skb_clone_fraglist(struct sk_buff *skb)
313 struct sk_buff *list;
315 skb_walk_frags(skb, list)
319 static void skb_release_data(struct sk_buff *skb)
322 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
323 &skb_shinfo(skb)->dataref)) {
324 if (skb_shinfo(skb)->nr_frags) {
326 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
327 put_page(skb_shinfo(skb)->frags[i].page);
330 if (skb_has_frag_list(skb))
331 skb_drop_fraglist(skb);
338 * Free an skbuff by memory without cleaning the state.
340 static void kfree_skbmem(struct sk_buff *skb)
342 struct sk_buff *other;
343 atomic_t *fclone_ref;
345 switch (skb->fclone) {
346 case SKB_FCLONE_UNAVAILABLE:
347 kmem_cache_free(skbuff_head_cache, skb);
350 case SKB_FCLONE_ORIG:
351 fclone_ref = (atomic_t *) (skb + 2);
352 if (atomic_dec_and_test(fclone_ref))
353 kmem_cache_free(skbuff_fclone_cache, skb);
356 case SKB_FCLONE_CLONE:
357 fclone_ref = (atomic_t *) (skb + 1);
360 /* The clone portion is available for
361 * fast-cloning again.
363 skb->fclone = SKB_FCLONE_UNAVAILABLE;
365 if (atomic_dec_and_test(fclone_ref))
366 kmem_cache_free(skbuff_fclone_cache, other);
371 static void skb_release_head_state(struct sk_buff *skb)
375 secpath_put(skb->sp);
377 if (skb->destructor) {
379 skb->destructor(skb);
381 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
382 nf_conntrack_put(skb->nfct);
383 nf_conntrack_put_reasm(skb->nfct_reasm);
385 #ifdef CONFIG_BRIDGE_NETFILTER
386 nf_bridge_put(skb->nf_bridge);
388 /* XXX: IS this still necessary? - JHS */
389 #ifdef CONFIG_NET_SCHED
391 #ifdef CONFIG_NET_CLS_ACT
397 /* Free everything but the sk_buff shell. */
398 static void skb_release_all(struct sk_buff *skb)
400 skb_release_head_state(skb);
401 skb_release_data(skb);
405 * __kfree_skb - private function
408 * Free an sk_buff. Release anything attached to the buffer.
409 * Clean the state. This is an internal helper function. Users should
410 * always call kfree_skb
413 void __kfree_skb(struct sk_buff *skb)
415 skb_release_all(skb);
418 EXPORT_SYMBOL(__kfree_skb);
421 * kfree_skb - free an sk_buff
422 * @skb: buffer to free
424 * Drop a reference to the buffer and free it if the usage count has
427 void kfree_skb(struct sk_buff *skb)
431 if (likely(atomic_read(&skb->users) == 1))
433 else if (likely(!atomic_dec_and_test(&skb->users)))
435 trace_kfree_skb(skb, __builtin_return_address(0));
438 EXPORT_SYMBOL(kfree_skb);
441 * consume_skb - free an skbuff
442 * @skb: buffer to free
444 * Drop a ref to the buffer and free it if the usage count has hit zero
445 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
446 * is being dropped after a failure and notes that
448 void consume_skb(struct sk_buff *skb)
452 if (likely(atomic_read(&skb->users) == 1))
454 else if (likely(!atomic_dec_and_test(&skb->users)))
456 trace_consume_skb(skb);
459 EXPORT_SYMBOL(consume_skb);
462 * skb_recycle_check - check if skb can be reused for receive
464 * @skb_size: minimum receive buffer size
466 * Checks that the skb passed in is not shared or cloned, and
467 * that it is linear and its head portion at least as large as
468 * skb_size so that it can be recycled as a receive buffer.
469 * If these conditions are met, this function does any necessary
470 * reference count dropping and cleans up the skbuff as if it
471 * just came from __alloc_skb().
473 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
475 struct skb_shared_info *shinfo;
480 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
483 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
484 if (skb_end_pointer(skb) - skb->head < skb_size)
487 if (skb_shared(skb) || skb_cloned(skb))
490 skb_release_head_state(skb);
492 shinfo = skb_shinfo(skb);
493 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
494 atomic_set(&shinfo->dataref, 1);
496 memset(skb, 0, offsetof(struct sk_buff, tail));
497 skb->data = skb->head + NET_SKB_PAD;
498 skb_reset_tail_pointer(skb);
502 EXPORT_SYMBOL(skb_recycle_check);
504 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
506 new->tstamp = old->tstamp;
508 new->transport_header = old->transport_header;
509 new->network_header = old->network_header;
510 new->mac_header = old->mac_header;
511 skb_dst_copy(new, old);
512 new->rxhash = old->rxhash;
514 new->sp = secpath_get(old->sp);
516 memcpy(new->cb, old->cb, sizeof(old->cb));
517 new->csum = old->csum;
518 new->local_df = old->local_df;
519 new->pkt_type = old->pkt_type;
520 new->ip_summed = old->ip_summed;
521 skb_copy_queue_mapping(new, old);
522 new->priority = old->priority;
523 new->deliver_no_wcard = old->deliver_no_wcard;
524 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
525 new->ipvs_property = old->ipvs_property;
527 new->protocol = old->protocol;
528 new->mark = old->mark;
529 new->skb_iif = old->skb_iif;
531 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
532 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
533 new->nf_trace = old->nf_trace;
535 #ifdef CONFIG_NET_SCHED
536 new->tc_index = old->tc_index;
537 #ifdef CONFIG_NET_CLS_ACT
538 new->tc_verd = old->tc_verd;
541 new->vlan_tci = old->vlan_tci;
543 skb_copy_secmark(new, old);
547 * You should not add any new code to this function. Add it to
548 * __copy_skb_header above instead.
550 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
552 #define C(x) n->x = skb->x
554 n->next = n->prev = NULL;
556 __copy_skb_header(n, skb);
561 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
564 n->destructor = NULL;
570 atomic_set(&n->users, 1);
572 atomic_inc(&(skb_shinfo(skb)->dataref));
580 * skb_morph - morph one skb into another
581 * @dst: the skb to receive the contents
582 * @src: the skb to supply the contents
584 * This is identical to skb_clone except that the target skb is
585 * supplied by the user.
587 * The target skb is returned upon exit.
589 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
591 skb_release_all(dst);
592 return __skb_clone(dst, src);
594 EXPORT_SYMBOL_GPL(skb_morph);
597 * skb_clone - duplicate an sk_buff
598 * @skb: buffer to clone
599 * @gfp_mask: allocation priority
601 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
602 * copies share the same packet data but not structure. The new
603 * buffer has a reference count of 1. If the allocation fails the
604 * function returns %NULL otherwise the new buffer is returned.
606 * If this function is called from an interrupt gfp_mask() must be
610 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
615 if (skb->fclone == SKB_FCLONE_ORIG &&
616 n->fclone == SKB_FCLONE_UNAVAILABLE) {
617 atomic_t *fclone_ref = (atomic_t *) (n + 1);
618 n->fclone = SKB_FCLONE_CLONE;
619 atomic_inc(fclone_ref);
621 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
625 kmemcheck_annotate_bitfield(n, flags1);
626 kmemcheck_annotate_bitfield(n, flags2);
627 n->fclone = SKB_FCLONE_UNAVAILABLE;
630 return __skb_clone(n, skb);
632 EXPORT_SYMBOL(skb_clone);
634 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
636 #ifndef NET_SKBUFF_DATA_USES_OFFSET
638 * Shift between the two data areas in bytes
640 unsigned long offset = new->data - old->data;
643 __copy_skb_header(new, old);
645 #ifndef NET_SKBUFF_DATA_USES_OFFSET
646 /* {transport,network,mac}_header are relative to skb->head */
647 new->transport_header += offset;
648 new->network_header += offset;
649 if (skb_mac_header_was_set(new))
650 new->mac_header += offset;
652 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
653 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
654 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
658 * skb_copy - create private copy of an sk_buff
659 * @skb: buffer to copy
660 * @gfp_mask: allocation priority
662 * Make a copy of both an &sk_buff and its data. This is used when the
663 * caller wishes to modify the data and needs a private copy of the
664 * data to alter. Returns %NULL on failure or the pointer to the buffer
665 * on success. The returned buffer has a reference count of 1.
667 * As by-product this function converts non-linear &sk_buff to linear
668 * one, so that &sk_buff becomes completely private and caller is allowed
669 * to modify all the data of returned buffer. This means that this
670 * function is not recommended for use in circumstances when only
671 * header is going to be modified. Use pskb_copy() instead.
674 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
676 int headerlen = skb_headroom(skb);
677 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
678 struct sk_buff *n = alloc_skb(size, gfp_mask);
683 /* Set the data pointer */
684 skb_reserve(n, headerlen);
685 /* Set the tail pointer and length */
686 skb_put(n, skb->len);
688 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
691 copy_skb_header(n, skb);
694 EXPORT_SYMBOL(skb_copy);
697 * pskb_copy - create copy of an sk_buff with private head.
698 * @skb: buffer to copy
699 * @gfp_mask: allocation priority
701 * Make a copy of both an &sk_buff and part of its data, located
702 * in header. Fragmented data remain shared. This is used when
703 * the caller wishes to modify only header of &sk_buff and needs
704 * private copy of the header to alter. Returns %NULL on failure
705 * or the pointer to the buffer on success.
706 * The returned buffer has a reference count of 1.
709 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
711 unsigned int size = skb_end_pointer(skb) - skb->head;
712 struct sk_buff *n = alloc_skb(size, gfp_mask);
717 /* Set the data pointer */
718 skb_reserve(n, skb_headroom(skb));
719 /* Set the tail pointer and length */
720 skb_put(n, skb_headlen(skb));
722 skb_copy_from_linear_data(skb, n->data, n->len);
724 n->truesize += skb->data_len;
725 n->data_len = skb->data_len;
728 if (skb_shinfo(skb)->nr_frags) {
731 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
732 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
733 get_page(skb_shinfo(n)->frags[i].page);
735 skb_shinfo(n)->nr_frags = i;
738 if (skb_has_frag_list(skb)) {
739 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
740 skb_clone_fraglist(n);
743 copy_skb_header(n, skb);
747 EXPORT_SYMBOL(pskb_copy);
750 * pskb_expand_head - reallocate header of &sk_buff
751 * @skb: buffer to reallocate
752 * @nhead: room to add at head
753 * @ntail: room to add at tail
754 * @gfp_mask: allocation priority
756 * Expands (or creates identical copy, if &nhead and &ntail are zero)
757 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
758 * reference count of 1. Returns zero in the case of success or error,
759 * if expansion failed. In the last case, &sk_buff is not changed.
761 * All the pointers pointing into skb header may change and must be
762 * reloaded after call to this function.
765 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
770 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
779 size = SKB_DATA_ALIGN(size);
781 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
785 /* Copy only real data... and, alas, header. This should be
786 * optimized for the cases when header is void.
788 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
790 memcpy((struct skb_shared_info *)(data + size),
792 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
794 /* Check if we can avoid taking references on fragments if we own
795 * the last reference on skb->head. (see skb_release_data())
800 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
802 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
808 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
809 get_page(skb_shinfo(skb)->frags[i].page);
811 if (skb_has_frag_list(skb))
812 skb_clone_fraglist(skb);
814 skb_release_data(skb);
816 off = (data + nhead) - skb->head;
820 #ifdef NET_SKBUFF_DATA_USES_OFFSET
824 skb->end = skb->head + size;
826 /* {transport,network,mac}_header and tail are relative to skb->head */
828 skb->transport_header += off;
829 skb->network_header += off;
830 if (skb_mac_header_was_set(skb))
831 skb->mac_header += off;
832 /* Only adjust this if it actually is csum_start rather than csum */
833 if (skb->ip_summed == CHECKSUM_PARTIAL)
834 skb->csum_start += nhead;
838 atomic_set(&skb_shinfo(skb)->dataref, 1);
844 EXPORT_SYMBOL(pskb_expand_head);
846 /* Make private copy of skb with writable head and some headroom */
848 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
850 struct sk_buff *skb2;
851 int delta = headroom - skb_headroom(skb);
854 skb2 = pskb_copy(skb, GFP_ATOMIC);
856 skb2 = skb_clone(skb, GFP_ATOMIC);
857 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
865 EXPORT_SYMBOL(skb_realloc_headroom);
868 * skb_copy_expand - copy and expand sk_buff
869 * @skb: buffer to copy
870 * @newheadroom: new free bytes at head
871 * @newtailroom: new free bytes at tail
872 * @gfp_mask: allocation priority
874 * Make a copy of both an &sk_buff and its data and while doing so
875 * allocate additional space.
877 * This is used when the caller wishes to modify the data and needs a
878 * private copy of the data to alter as well as more space for new fields.
879 * Returns %NULL on failure or the pointer to the buffer
880 * on success. The returned buffer has a reference count of 1.
882 * You must pass %GFP_ATOMIC as the allocation priority if this function
883 * is called from an interrupt.
885 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
886 int newheadroom, int newtailroom,
890 * Allocate the copy buffer
892 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
894 int oldheadroom = skb_headroom(skb);
895 int head_copy_len, head_copy_off;
901 skb_reserve(n, newheadroom);
903 /* Set the tail pointer and length */
904 skb_put(n, skb->len);
906 head_copy_len = oldheadroom;
908 if (newheadroom <= head_copy_len)
909 head_copy_len = newheadroom;
911 head_copy_off = newheadroom - head_copy_len;
913 /* Copy the linear header and data. */
914 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
915 skb->len + head_copy_len))
918 copy_skb_header(n, skb);
920 off = newheadroom - oldheadroom;
921 if (n->ip_summed == CHECKSUM_PARTIAL)
922 n->csum_start += off;
923 #ifdef NET_SKBUFF_DATA_USES_OFFSET
924 n->transport_header += off;
925 n->network_header += off;
926 if (skb_mac_header_was_set(skb))
927 n->mac_header += off;
932 EXPORT_SYMBOL(skb_copy_expand);
935 * skb_pad - zero pad the tail of an skb
936 * @skb: buffer to pad
939 * Ensure that a buffer is followed by a padding area that is zero
940 * filled. Used by network drivers which may DMA or transfer data
941 * beyond the buffer end onto the wire.
943 * May return error in out of memory cases. The skb is freed on error.
946 int skb_pad(struct sk_buff *skb, int pad)
951 /* If the skbuff is non linear tailroom is always zero.. */
952 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
953 memset(skb->data+skb->len, 0, pad);
957 ntail = skb->data_len + pad - (skb->end - skb->tail);
958 if (likely(skb_cloned(skb) || ntail > 0)) {
959 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
964 /* FIXME: The use of this function with non-linear skb's really needs
967 err = skb_linearize(skb);
971 memset(skb->data + skb->len, 0, pad);
978 EXPORT_SYMBOL(skb_pad);
981 * skb_put - add data to a buffer
982 * @skb: buffer to use
983 * @len: amount of data to add
985 * This function extends the used data area of the buffer. If this would
986 * exceed the total buffer size the kernel will panic. A pointer to the
987 * first byte of the extra data is returned.
989 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
991 unsigned char *tmp = skb_tail_pointer(skb);
992 SKB_LINEAR_ASSERT(skb);
995 if (unlikely(skb->tail > skb->end))
996 skb_over_panic(skb, len, __builtin_return_address(0));
999 EXPORT_SYMBOL(skb_put);
1002 * skb_push - add data to the start of a buffer
1003 * @skb: buffer to use
1004 * @len: amount of data to add
1006 * This function extends the used data area of the buffer at the buffer
1007 * start. If this would exceed the total buffer headroom the kernel will
1008 * panic. A pointer to the first byte of the extra data is returned.
1010 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1014 if (unlikely(skb->data<skb->head))
1015 skb_under_panic(skb, len, __builtin_return_address(0));
1018 EXPORT_SYMBOL(skb_push);
1021 * skb_pull - remove data from the start of a buffer
1022 * @skb: buffer to use
1023 * @len: amount of data to remove
1025 * This function removes data from the start of a buffer, returning
1026 * the memory to the headroom. A pointer to the next data in the buffer
1027 * is returned. Once the data has been pulled future pushes will overwrite
1030 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1032 return skb_pull_inline(skb, len);
1034 EXPORT_SYMBOL(skb_pull);
1037 * skb_trim - remove end from a buffer
1038 * @skb: buffer to alter
1041 * Cut the length of a buffer down by removing data from the tail. If
1042 * the buffer is already under the length specified it is not modified.
1043 * The skb must be linear.
1045 void skb_trim(struct sk_buff *skb, unsigned int len)
1048 __skb_trim(skb, len);
1050 EXPORT_SYMBOL(skb_trim);
1052 /* Trims skb to length len. It can change skb pointers.
1055 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1057 struct sk_buff **fragp;
1058 struct sk_buff *frag;
1059 int offset = skb_headlen(skb);
1060 int nfrags = skb_shinfo(skb)->nr_frags;
1064 if (skb_cloned(skb) &&
1065 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1072 for (; i < nfrags; i++) {
1073 int end = offset + skb_shinfo(skb)->frags[i].size;
1080 skb_shinfo(skb)->frags[i++].size = len - offset;
1083 skb_shinfo(skb)->nr_frags = i;
1085 for (; i < nfrags; i++)
1086 put_page(skb_shinfo(skb)->frags[i].page);
1088 if (skb_has_frag_list(skb))
1089 skb_drop_fraglist(skb);
1093 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1094 fragp = &frag->next) {
1095 int end = offset + frag->len;
1097 if (skb_shared(frag)) {
1098 struct sk_buff *nfrag;
1100 nfrag = skb_clone(frag, GFP_ATOMIC);
1101 if (unlikely(!nfrag))
1104 nfrag->next = frag->next;
1116 unlikely((err = pskb_trim(frag, len - offset))))
1120 skb_drop_list(&frag->next);
1125 if (len > skb_headlen(skb)) {
1126 skb->data_len -= skb->len - len;
1131 skb_set_tail_pointer(skb, len);
1136 EXPORT_SYMBOL(___pskb_trim);
1139 * __pskb_pull_tail - advance tail of skb header
1140 * @skb: buffer to reallocate
1141 * @delta: number of bytes to advance tail
1143 * The function makes a sense only on a fragmented &sk_buff,
1144 * it expands header moving its tail forward and copying necessary
1145 * data from fragmented part.
1147 * &sk_buff MUST have reference count of 1.
1149 * Returns %NULL (and &sk_buff does not change) if pull failed
1150 * or value of new tail of skb in the case of success.
1152 * All the pointers pointing into skb header may change and must be
1153 * reloaded after call to this function.
1156 /* Moves tail of skb head forward, copying data from fragmented part,
1157 * when it is necessary.
1158 * 1. It may fail due to malloc failure.
1159 * 2. It may change skb pointers.
1161 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1163 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1165 /* If skb has not enough free space at tail, get new one
1166 * plus 128 bytes for future expansions. If we have enough
1167 * room at tail, reallocate without expansion only if skb is cloned.
1169 int i, k, eat = (skb->tail + delta) - skb->end;
1171 if (eat > 0 || skb_cloned(skb)) {
1172 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1177 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1180 /* Optimization: no fragments, no reasons to preestimate
1181 * size of pulled pages. Superb.
1183 if (!skb_has_frag_list(skb))
1186 /* Estimate size of pulled pages. */
1188 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1189 if (skb_shinfo(skb)->frags[i].size >= eat)
1191 eat -= skb_shinfo(skb)->frags[i].size;
1194 /* If we need update frag list, we are in troubles.
1195 * Certainly, it possible to add an offset to skb data,
1196 * but taking into account that pulling is expected to
1197 * be very rare operation, it is worth to fight against
1198 * further bloating skb head and crucify ourselves here instead.
1199 * Pure masohism, indeed. 8)8)
1202 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1203 struct sk_buff *clone = NULL;
1204 struct sk_buff *insp = NULL;
1209 if (list->len <= eat) {
1210 /* Eaten as whole. */
1215 /* Eaten partially. */
1217 if (skb_shared(list)) {
1218 /* Sucks! We need to fork list. :-( */
1219 clone = skb_clone(list, GFP_ATOMIC);
1225 /* This may be pulled without
1229 if (!pskb_pull(list, eat)) {
1237 /* Free pulled out fragments. */
1238 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1239 skb_shinfo(skb)->frag_list = list->next;
1242 /* And insert new clone at head. */
1245 skb_shinfo(skb)->frag_list = clone;
1248 /* Success! Now we may commit changes to skb data. */
1253 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1254 if (skb_shinfo(skb)->frags[i].size <= eat) {
1255 put_page(skb_shinfo(skb)->frags[i].page);
1256 eat -= skb_shinfo(skb)->frags[i].size;
1258 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1260 skb_shinfo(skb)->frags[k].page_offset += eat;
1261 skb_shinfo(skb)->frags[k].size -= eat;
1267 skb_shinfo(skb)->nr_frags = k;
1270 skb->data_len -= delta;
1272 return skb_tail_pointer(skb);
1274 EXPORT_SYMBOL(__pskb_pull_tail);
1276 /* Copy some data bits from skb to kernel buffer. */
1278 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1280 int start = skb_headlen(skb);
1281 struct sk_buff *frag_iter;
1284 if (offset > (int)skb->len - len)
1288 if ((copy = start - offset) > 0) {
1291 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1292 if ((len -= copy) == 0)
1298 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1301 WARN_ON(start > offset + len);
1303 end = start + skb_shinfo(skb)->frags[i].size;
1304 if ((copy = end - offset) > 0) {
1310 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1312 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1313 offset - start, copy);
1314 kunmap_skb_frag(vaddr);
1316 if ((len -= copy) == 0)
1324 skb_walk_frags(skb, frag_iter) {
1327 WARN_ON(start > offset + len);
1329 end = start + frag_iter->len;
1330 if ((copy = end - offset) > 0) {
1333 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1335 if ((len -= copy) == 0)
1348 EXPORT_SYMBOL(skb_copy_bits);
1351 * Callback from splice_to_pipe(), if we need to release some pages
1352 * at the end of the spd in case we error'ed out in filling the pipe.
1354 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1356 put_page(spd->pages[i]);
1359 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1360 unsigned int *offset,
1361 struct sk_buff *skb, struct sock *sk)
1363 struct page *p = sk->sk_sndmsg_page;
1368 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1372 off = sk->sk_sndmsg_off = 0;
1373 /* hold one ref to this page until it's full */
1377 off = sk->sk_sndmsg_off;
1378 mlen = PAGE_SIZE - off;
1379 if (mlen < 64 && mlen < *len) {
1384 *len = min_t(unsigned int, *len, mlen);
1387 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1388 sk->sk_sndmsg_off += *len;
1396 * Fill page/offset/length into spd, if it can hold more pages.
1398 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1399 struct pipe_inode_info *pipe, struct page *page,
1400 unsigned int *len, unsigned int offset,
1401 struct sk_buff *skb, int linear,
1404 if (unlikely(spd->nr_pages == pipe->buffers))
1408 page = linear_to_page(page, len, &offset, skb, sk);
1414 spd->pages[spd->nr_pages] = page;
1415 spd->partial[spd->nr_pages].len = *len;
1416 spd->partial[spd->nr_pages].offset = offset;
1422 static inline void __segment_seek(struct page **page, unsigned int *poff,
1423 unsigned int *plen, unsigned int off)
1428 n = *poff / PAGE_SIZE;
1430 *page = nth_page(*page, n);
1432 *poff = *poff % PAGE_SIZE;
1436 static inline int __splice_segment(struct page *page, unsigned int poff,
1437 unsigned int plen, unsigned int *off,
1438 unsigned int *len, struct sk_buff *skb,
1439 struct splice_pipe_desc *spd, int linear,
1441 struct pipe_inode_info *pipe)
1446 /* skip this segment if already processed */
1452 /* ignore any bits we already processed */
1454 __segment_seek(&page, &poff, &plen, *off);
1459 unsigned int flen = min(*len, plen);
1461 /* the linear region may spread across several pages */
1462 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1464 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1467 __segment_seek(&page, &poff, &plen, flen);
1470 } while (*len && plen);
1476 * Map linear and fragment data from the skb to spd. It reports failure if the
1477 * pipe is full or if we already spliced the requested length.
1479 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1480 unsigned int *offset, unsigned int *len,
1481 struct splice_pipe_desc *spd, struct sock *sk)
1486 * map the linear part
1488 if (__splice_segment(virt_to_page(skb->data),
1489 (unsigned long) skb->data & (PAGE_SIZE - 1),
1491 offset, len, skb, spd, 1, sk, pipe))
1495 * then map the fragments
1497 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1498 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1500 if (__splice_segment(f->page, f->page_offset, f->size,
1501 offset, len, skb, spd, 0, sk, pipe))
1509 * Map data from the skb to a pipe. Should handle both the linear part,
1510 * the fragments, and the frag list. It does NOT handle frag lists within
1511 * the frag list, if such a thing exists. We'd probably need to recurse to
1512 * handle that cleanly.
1514 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1515 struct pipe_inode_info *pipe, unsigned int tlen,
1518 struct partial_page partial[PIPE_DEF_BUFFERS];
1519 struct page *pages[PIPE_DEF_BUFFERS];
1520 struct splice_pipe_desc spd = {
1524 .ops = &sock_pipe_buf_ops,
1525 .spd_release = sock_spd_release,
1527 struct sk_buff *frag_iter;
1528 struct sock *sk = skb->sk;
1531 if (splice_grow_spd(pipe, &spd))
1535 * __skb_splice_bits() only fails if the output has no room left,
1536 * so no point in going over the frag_list for the error case.
1538 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1544 * now see if we have a frag_list to map
1546 skb_walk_frags(skb, frag_iter) {
1549 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1556 * Drop the socket lock, otherwise we have reverse
1557 * locking dependencies between sk_lock and i_mutex
1558 * here as compared to sendfile(). We enter here
1559 * with the socket lock held, and splice_to_pipe() will
1560 * grab the pipe inode lock. For sendfile() emulation,
1561 * we call into ->sendpage() with the i_mutex lock held
1562 * and networking will grab the socket lock.
1565 ret = splice_to_pipe(pipe, &spd);
1569 splice_shrink_spd(pipe, &spd);
1574 * skb_store_bits - store bits from kernel buffer to skb
1575 * @skb: destination buffer
1576 * @offset: offset in destination
1577 * @from: source buffer
1578 * @len: number of bytes to copy
1580 * Copy the specified number of bytes from the source buffer to the
1581 * destination skb. This function handles all the messy bits of
1582 * traversing fragment lists and such.
1585 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1587 int start = skb_headlen(skb);
1588 struct sk_buff *frag_iter;
1591 if (offset > (int)skb->len - len)
1594 if ((copy = start - offset) > 0) {
1597 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1598 if ((len -= copy) == 0)
1604 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1605 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1608 WARN_ON(start > offset + len);
1610 end = start + frag->size;
1611 if ((copy = end - offset) > 0) {
1617 vaddr = kmap_skb_frag(frag);
1618 memcpy(vaddr + frag->page_offset + offset - start,
1620 kunmap_skb_frag(vaddr);
1622 if ((len -= copy) == 0)
1630 skb_walk_frags(skb, frag_iter) {
1633 WARN_ON(start > offset + len);
1635 end = start + frag_iter->len;
1636 if ((copy = end - offset) > 0) {
1639 if (skb_store_bits(frag_iter, offset - start,
1642 if ((len -= copy) == 0)
1655 EXPORT_SYMBOL(skb_store_bits);
1657 /* Checksum skb data. */
1659 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1660 int len, __wsum csum)
1662 int start = skb_headlen(skb);
1663 int i, copy = start - offset;
1664 struct sk_buff *frag_iter;
1667 /* Checksum header. */
1671 csum = csum_partial(skb->data + offset, copy, csum);
1672 if ((len -= copy) == 0)
1678 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1681 WARN_ON(start > offset + len);
1683 end = start + skb_shinfo(skb)->frags[i].size;
1684 if ((copy = end - offset) > 0) {
1687 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1691 vaddr = kmap_skb_frag(frag);
1692 csum2 = csum_partial(vaddr + frag->page_offset +
1693 offset - start, copy, 0);
1694 kunmap_skb_frag(vaddr);
1695 csum = csum_block_add(csum, csum2, pos);
1704 skb_walk_frags(skb, frag_iter) {
1707 WARN_ON(start > offset + len);
1709 end = start + frag_iter->len;
1710 if ((copy = end - offset) > 0) {
1714 csum2 = skb_checksum(frag_iter, offset - start,
1716 csum = csum_block_add(csum, csum2, pos);
1717 if ((len -= copy) == 0)
1728 EXPORT_SYMBOL(skb_checksum);
1730 /* Both of above in one bottle. */
1732 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1733 u8 *to, int len, __wsum csum)
1735 int start = skb_headlen(skb);
1736 int i, copy = start - offset;
1737 struct sk_buff *frag_iter;
1744 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1746 if ((len -= copy) == 0)
1753 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1756 WARN_ON(start > offset + len);
1758 end = start + skb_shinfo(skb)->frags[i].size;
1759 if ((copy = end - offset) > 0) {
1762 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1766 vaddr = kmap_skb_frag(frag);
1767 csum2 = csum_partial_copy_nocheck(vaddr +
1771 kunmap_skb_frag(vaddr);
1772 csum = csum_block_add(csum, csum2, pos);
1782 skb_walk_frags(skb, frag_iter) {
1786 WARN_ON(start > offset + len);
1788 end = start + frag_iter->len;
1789 if ((copy = end - offset) > 0) {
1792 csum2 = skb_copy_and_csum_bits(frag_iter,
1795 csum = csum_block_add(csum, csum2, pos);
1796 if ((len -= copy) == 0)
1807 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1809 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1814 if (skb->ip_summed == CHECKSUM_PARTIAL)
1815 csstart = skb->csum_start - skb_headroom(skb);
1817 csstart = skb_headlen(skb);
1819 BUG_ON(csstart > skb_headlen(skb));
1821 skb_copy_from_linear_data(skb, to, csstart);
1824 if (csstart != skb->len)
1825 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1826 skb->len - csstart, 0);
1828 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1829 long csstuff = csstart + skb->csum_offset;
1831 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1834 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1837 * skb_dequeue - remove from the head of the queue
1838 * @list: list to dequeue from
1840 * Remove the head of the list. The list lock is taken so the function
1841 * may be used safely with other locking list functions. The head item is
1842 * returned or %NULL if the list is empty.
1845 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1847 unsigned long flags;
1848 struct sk_buff *result;
1850 spin_lock_irqsave(&list->lock, flags);
1851 result = __skb_dequeue(list);
1852 spin_unlock_irqrestore(&list->lock, flags);
1855 EXPORT_SYMBOL(skb_dequeue);
1858 * skb_dequeue_tail - remove from the tail of the queue
1859 * @list: list to dequeue from
1861 * Remove the tail of the list. The list lock is taken so the function
1862 * may be used safely with other locking list functions. The tail item is
1863 * returned or %NULL if the list is empty.
1865 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1867 unsigned long flags;
1868 struct sk_buff *result;
1870 spin_lock_irqsave(&list->lock, flags);
1871 result = __skb_dequeue_tail(list);
1872 spin_unlock_irqrestore(&list->lock, flags);
1875 EXPORT_SYMBOL(skb_dequeue_tail);
1878 * skb_queue_purge - empty a list
1879 * @list: list to empty
1881 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1882 * the list and one reference dropped. This function takes the list
1883 * lock and is atomic with respect to other list locking functions.
1885 void skb_queue_purge(struct sk_buff_head *list)
1887 struct sk_buff *skb;
1888 while ((skb = skb_dequeue(list)) != NULL)
1891 EXPORT_SYMBOL(skb_queue_purge);
1894 * skb_queue_head - queue a buffer at the list head
1895 * @list: list to use
1896 * @newsk: buffer to queue
1898 * Queue a buffer at the start of the list. This function takes the
1899 * list lock and can be used safely with other locking &sk_buff functions
1902 * A buffer cannot be placed on two lists at the same time.
1904 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1906 unsigned long flags;
1908 spin_lock_irqsave(&list->lock, flags);
1909 __skb_queue_head(list, newsk);
1910 spin_unlock_irqrestore(&list->lock, flags);
1912 EXPORT_SYMBOL(skb_queue_head);
1915 * skb_queue_tail - queue a buffer at the list tail
1916 * @list: list to use
1917 * @newsk: buffer to queue
1919 * Queue a buffer at the tail of the list. This function takes the
1920 * list lock and can be used safely with other locking &sk_buff functions
1923 * A buffer cannot be placed on two lists at the same time.
1925 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1927 unsigned long flags;
1929 spin_lock_irqsave(&list->lock, flags);
1930 __skb_queue_tail(list, newsk);
1931 spin_unlock_irqrestore(&list->lock, flags);
1933 EXPORT_SYMBOL(skb_queue_tail);
1936 * skb_unlink - remove a buffer from a list
1937 * @skb: buffer to remove
1938 * @list: list to use
1940 * Remove a packet from a list. The list locks are taken and this
1941 * function is atomic with respect to other list locked calls
1943 * You must know what list the SKB is on.
1945 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1947 unsigned long flags;
1949 spin_lock_irqsave(&list->lock, flags);
1950 __skb_unlink(skb, list);
1951 spin_unlock_irqrestore(&list->lock, flags);
1953 EXPORT_SYMBOL(skb_unlink);
1956 * skb_append - append a buffer
1957 * @old: buffer to insert after
1958 * @newsk: buffer to insert
1959 * @list: list to use
1961 * Place a packet after a given packet in a list. The list locks are taken
1962 * and this function is atomic with respect to other list locked calls.
1963 * A buffer cannot be placed on two lists at the same time.
1965 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1967 unsigned long flags;
1969 spin_lock_irqsave(&list->lock, flags);
1970 __skb_queue_after(list, old, newsk);
1971 spin_unlock_irqrestore(&list->lock, flags);
1973 EXPORT_SYMBOL(skb_append);
1976 * skb_insert - insert a buffer
1977 * @old: buffer to insert before
1978 * @newsk: buffer to insert
1979 * @list: list to use
1981 * Place a packet before a given packet in a list. The list locks are
1982 * taken and this function is atomic with respect to other list locked
1985 * A buffer cannot be placed on two lists at the same time.
1987 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1989 unsigned long flags;
1991 spin_lock_irqsave(&list->lock, flags);
1992 __skb_insert(newsk, old->prev, old, list);
1993 spin_unlock_irqrestore(&list->lock, flags);
1995 EXPORT_SYMBOL(skb_insert);
1997 static inline void skb_split_inside_header(struct sk_buff *skb,
1998 struct sk_buff* skb1,
1999 const u32 len, const int pos)
2003 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2005 /* And move data appendix as is. */
2006 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2007 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2009 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2010 skb_shinfo(skb)->nr_frags = 0;
2011 skb1->data_len = skb->data_len;
2012 skb1->len += skb1->data_len;
2015 skb_set_tail_pointer(skb, len);
2018 static inline void skb_split_no_header(struct sk_buff *skb,
2019 struct sk_buff* skb1,
2020 const u32 len, int pos)
2023 const int nfrags = skb_shinfo(skb)->nr_frags;
2025 skb_shinfo(skb)->nr_frags = 0;
2026 skb1->len = skb1->data_len = skb->len - len;
2028 skb->data_len = len - pos;
2030 for (i = 0; i < nfrags; i++) {
2031 int size = skb_shinfo(skb)->frags[i].size;
2033 if (pos + size > len) {
2034 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2038 * We have two variants in this case:
2039 * 1. Move all the frag to the second
2040 * part, if it is possible. F.e.
2041 * this approach is mandatory for TUX,
2042 * where splitting is expensive.
2043 * 2. Split is accurately. We make this.
2045 get_page(skb_shinfo(skb)->frags[i].page);
2046 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2047 skb_shinfo(skb1)->frags[0].size -= len - pos;
2048 skb_shinfo(skb)->frags[i].size = len - pos;
2049 skb_shinfo(skb)->nr_frags++;
2053 skb_shinfo(skb)->nr_frags++;
2056 skb_shinfo(skb1)->nr_frags = k;
2060 * skb_split - Split fragmented skb to two parts at length len.
2061 * @skb: the buffer to split
2062 * @skb1: the buffer to receive the second part
2063 * @len: new length for skb
2065 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2067 int pos = skb_headlen(skb);
2069 if (len < pos) /* Split line is inside header. */
2070 skb_split_inside_header(skb, skb1, len, pos);
2071 else /* Second chunk has no header, nothing to copy. */
2072 skb_split_no_header(skb, skb1, len, pos);
2074 EXPORT_SYMBOL(skb_split);
2076 /* Shifting from/to a cloned skb is a no-go.
2078 * Caller cannot keep skb_shinfo related pointers past calling here!
2080 static int skb_prepare_for_shift(struct sk_buff *skb)
2082 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2086 * skb_shift - Shifts paged data partially from skb to another
2087 * @tgt: buffer into which tail data gets added
2088 * @skb: buffer from which the paged data comes from
2089 * @shiftlen: shift up to this many bytes
2091 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2092 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2093 * It's up to caller to free skb if everything was shifted.
2095 * If @tgt runs out of frags, the whole operation is aborted.
2097 * Skb cannot include anything else but paged data while tgt is allowed
2098 * to have non-paged data as well.
2100 * TODO: full sized shift could be optimized but that would need
2101 * specialized skb free'er to handle frags without up-to-date nr_frags.
2103 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2105 int from, to, merge, todo;
2106 struct skb_frag_struct *fragfrom, *fragto;
2108 BUG_ON(shiftlen > skb->len);
2109 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2113 to = skb_shinfo(tgt)->nr_frags;
2114 fragfrom = &skb_shinfo(skb)->frags[from];
2116 /* Actual merge is delayed until the point when we know we can
2117 * commit all, so that we don't have to undo partial changes
2120 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2125 todo -= fragfrom->size;
2127 if (skb_prepare_for_shift(skb) ||
2128 skb_prepare_for_shift(tgt))
2131 /* All previous frag pointers might be stale! */
2132 fragfrom = &skb_shinfo(skb)->frags[from];
2133 fragto = &skb_shinfo(tgt)->frags[merge];
2135 fragto->size += shiftlen;
2136 fragfrom->size -= shiftlen;
2137 fragfrom->page_offset += shiftlen;
2145 /* Skip full, not-fitting skb to avoid expensive operations */
2146 if ((shiftlen == skb->len) &&
2147 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2150 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2153 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2154 if (to == MAX_SKB_FRAGS)
2157 fragfrom = &skb_shinfo(skb)->frags[from];
2158 fragto = &skb_shinfo(tgt)->frags[to];
2160 if (todo >= fragfrom->size) {
2161 *fragto = *fragfrom;
2162 todo -= fragfrom->size;
2167 get_page(fragfrom->page);
2168 fragto->page = fragfrom->page;
2169 fragto->page_offset = fragfrom->page_offset;
2170 fragto->size = todo;
2172 fragfrom->page_offset += todo;
2173 fragfrom->size -= todo;
2181 /* Ready to "commit" this state change to tgt */
2182 skb_shinfo(tgt)->nr_frags = to;
2185 fragfrom = &skb_shinfo(skb)->frags[0];
2186 fragto = &skb_shinfo(tgt)->frags[merge];
2188 fragto->size += fragfrom->size;
2189 put_page(fragfrom->page);
2192 /* Reposition in the original skb */
2194 while (from < skb_shinfo(skb)->nr_frags)
2195 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2196 skb_shinfo(skb)->nr_frags = to;
2198 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2201 /* Most likely the tgt won't ever need its checksum anymore, skb on
2202 * the other hand might need it if it needs to be resent
2204 tgt->ip_summed = CHECKSUM_PARTIAL;
2205 skb->ip_summed = CHECKSUM_PARTIAL;
2207 /* Yak, is it really working this way? Some helper please? */
2208 skb->len -= shiftlen;
2209 skb->data_len -= shiftlen;
2210 skb->truesize -= shiftlen;
2211 tgt->len += shiftlen;
2212 tgt->data_len += shiftlen;
2213 tgt->truesize += shiftlen;
2219 * skb_prepare_seq_read - Prepare a sequential read of skb data
2220 * @skb: the buffer to read
2221 * @from: lower offset of data to be read
2222 * @to: upper offset of data to be read
2223 * @st: state variable
2225 * Initializes the specified state variable. Must be called before
2226 * invoking skb_seq_read() for the first time.
2228 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2229 unsigned int to, struct skb_seq_state *st)
2231 st->lower_offset = from;
2232 st->upper_offset = to;
2233 st->root_skb = st->cur_skb = skb;
2234 st->frag_idx = st->stepped_offset = 0;
2235 st->frag_data = NULL;
2237 EXPORT_SYMBOL(skb_prepare_seq_read);
2240 * skb_seq_read - Sequentially read skb data
2241 * @consumed: number of bytes consumed by the caller so far
2242 * @data: destination pointer for data to be returned
2243 * @st: state variable
2245 * Reads a block of skb data at &consumed relative to the
2246 * lower offset specified to skb_prepare_seq_read(). Assigns
2247 * the head of the data block to &data and returns the length
2248 * of the block or 0 if the end of the skb data or the upper
2249 * offset has been reached.
2251 * The caller is not required to consume all of the data
2252 * returned, i.e. &consumed is typically set to the number
2253 * of bytes already consumed and the next call to
2254 * skb_seq_read() will return the remaining part of the block.
2256 * Note 1: The size of each block of data returned can be arbitary,
2257 * this limitation is the cost for zerocopy seqeuental
2258 * reads of potentially non linear data.
2260 * Note 2: Fragment lists within fragments are not implemented
2261 * at the moment, state->root_skb could be replaced with
2262 * a stack for this purpose.
2264 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2265 struct skb_seq_state *st)
2267 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2270 if (unlikely(abs_offset >= st->upper_offset))
2274 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2276 if (abs_offset < block_limit && !st->frag_data) {
2277 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2278 return block_limit - abs_offset;
2281 if (st->frag_idx == 0 && !st->frag_data)
2282 st->stepped_offset += skb_headlen(st->cur_skb);
2284 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2285 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2286 block_limit = frag->size + st->stepped_offset;
2288 if (abs_offset < block_limit) {
2290 st->frag_data = kmap_skb_frag(frag);
2292 *data = (u8 *) st->frag_data + frag->page_offset +
2293 (abs_offset - st->stepped_offset);
2295 return block_limit - abs_offset;
2298 if (st->frag_data) {
2299 kunmap_skb_frag(st->frag_data);
2300 st->frag_data = NULL;
2304 st->stepped_offset += frag->size;
2307 if (st->frag_data) {
2308 kunmap_skb_frag(st->frag_data);
2309 st->frag_data = NULL;
2312 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2313 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2316 } else if (st->cur_skb->next) {
2317 st->cur_skb = st->cur_skb->next;
2324 EXPORT_SYMBOL(skb_seq_read);
2327 * skb_abort_seq_read - Abort a sequential read of skb data
2328 * @st: state variable
2330 * Must be called if skb_seq_read() was not called until it
2333 void skb_abort_seq_read(struct skb_seq_state *st)
2336 kunmap_skb_frag(st->frag_data);
2338 EXPORT_SYMBOL(skb_abort_seq_read);
2340 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2342 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2343 struct ts_config *conf,
2344 struct ts_state *state)
2346 return skb_seq_read(offset, text, TS_SKB_CB(state));
2349 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2351 skb_abort_seq_read(TS_SKB_CB(state));
2355 * skb_find_text - Find a text pattern in skb data
2356 * @skb: the buffer to look in
2357 * @from: search offset
2359 * @config: textsearch configuration
2360 * @state: uninitialized textsearch state variable
2362 * Finds a pattern in the skb data according to the specified
2363 * textsearch configuration. Use textsearch_next() to retrieve
2364 * subsequent occurrences of the pattern. Returns the offset
2365 * to the first occurrence or UINT_MAX if no match was found.
2367 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2368 unsigned int to, struct ts_config *config,
2369 struct ts_state *state)
2373 config->get_next_block = skb_ts_get_next_block;
2374 config->finish = skb_ts_finish;
2376 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2378 ret = textsearch_find(config, state);
2379 return (ret <= to - from ? ret : UINT_MAX);
2381 EXPORT_SYMBOL(skb_find_text);
2384 * skb_append_datato_frags: - append the user data to a skb
2385 * @sk: sock structure
2386 * @skb: skb structure to be appened with user data.
2387 * @getfrag: call back function to be used for getting the user data
2388 * @from: pointer to user message iov
2389 * @length: length of the iov message
2391 * Description: This procedure append the user data in the fragment part
2392 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2394 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2395 int (*getfrag)(void *from, char *to, int offset,
2396 int len, int odd, struct sk_buff *skb),
2397 void *from, int length)
2400 skb_frag_t *frag = NULL;
2401 struct page *page = NULL;
2407 /* Return error if we don't have space for new frag */
2408 frg_cnt = skb_shinfo(skb)->nr_frags;
2409 if (frg_cnt >= MAX_SKB_FRAGS)
2412 /* allocate a new page for next frag */
2413 page = alloc_pages(sk->sk_allocation, 0);
2415 /* If alloc_page fails just return failure and caller will
2416 * free previous allocated pages by doing kfree_skb()
2421 /* initialize the next frag */
2422 sk->sk_sndmsg_page = page;
2423 sk->sk_sndmsg_off = 0;
2424 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2425 skb->truesize += PAGE_SIZE;
2426 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2428 /* get the new initialized frag */
2429 frg_cnt = skb_shinfo(skb)->nr_frags;
2430 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2432 /* copy the user data to page */
2433 left = PAGE_SIZE - frag->page_offset;
2434 copy = (length > left)? left : length;
2436 ret = getfrag(from, (page_address(frag->page) +
2437 frag->page_offset + frag->size),
2438 offset, copy, 0, skb);
2442 /* copy was successful so update the size parameters */
2443 sk->sk_sndmsg_off += copy;
2446 skb->data_len += copy;
2450 } while (length > 0);
2454 EXPORT_SYMBOL(skb_append_datato_frags);
2457 * skb_pull_rcsum - pull skb and update receive checksum
2458 * @skb: buffer to update
2459 * @len: length of data pulled
2461 * This function performs an skb_pull on the packet and updates
2462 * the CHECKSUM_COMPLETE checksum. It should be used on
2463 * receive path processing instead of skb_pull unless you know
2464 * that the checksum difference is zero (e.g., a valid IP header)
2465 * or you are setting ip_summed to CHECKSUM_NONE.
2467 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2469 BUG_ON(len > skb->len);
2471 BUG_ON(skb->len < skb->data_len);
2472 skb_postpull_rcsum(skb, skb->data, len);
2473 return skb->data += len;
2475 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2478 * skb_segment - Perform protocol segmentation on skb.
2479 * @skb: buffer to segment
2480 * @features: features for the output path (see dev->features)
2482 * This function performs segmentation on the given skb. It returns
2483 * a pointer to the first in a list of new skbs for the segments.
2484 * In case of error it returns ERR_PTR(err).
2486 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2488 struct sk_buff *segs = NULL;
2489 struct sk_buff *tail = NULL;
2490 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2491 unsigned int mss = skb_shinfo(skb)->gso_size;
2492 unsigned int doffset = skb->data - skb_mac_header(skb);
2493 unsigned int offset = doffset;
2494 unsigned int headroom;
2496 int sg = features & NETIF_F_SG;
2497 int nfrags = skb_shinfo(skb)->nr_frags;
2502 __skb_push(skb, doffset);
2503 headroom = skb_headroom(skb);
2504 pos = skb_headlen(skb);
2507 struct sk_buff *nskb;
2512 len = skb->len - offset;
2516 hsize = skb_headlen(skb) - offset;
2519 if (hsize > len || !sg)
2522 if (!hsize && i >= nfrags) {
2523 BUG_ON(fskb->len != len);
2526 nskb = skb_clone(fskb, GFP_ATOMIC);
2529 if (unlikely(!nskb))
2532 hsize = skb_end_pointer(nskb) - nskb->head;
2533 if (skb_cow_head(nskb, doffset + headroom)) {
2538 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2540 skb_release_head_state(nskb);
2541 __skb_push(nskb, doffset);
2543 nskb = alloc_skb(hsize + doffset + headroom,
2546 if (unlikely(!nskb))
2549 skb_reserve(nskb, headroom);
2550 __skb_put(nskb, doffset);
2559 __copy_skb_header(nskb, skb);
2560 nskb->mac_len = skb->mac_len;
2562 /* nskb and skb might have different headroom */
2563 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2564 nskb->csum_start += skb_headroom(nskb) - headroom;
2566 skb_reset_mac_header(nskb);
2567 skb_set_network_header(nskb, skb->mac_len);
2568 nskb->transport_header = (nskb->network_header +
2569 skb_network_header_len(skb));
2570 skb_copy_from_linear_data(skb, nskb->data, doffset);
2572 if (fskb != skb_shinfo(skb)->frag_list)
2576 nskb->ip_summed = CHECKSUM_NONE;
2577 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2583 frag = skb_shinfo(nskb)->frags;
2585 skb_copy_from_linear_data_offset(skb, offset,
2586 skb_put(nskb, hsize), hsize);
2588 while (pos < offset + len && i < nfrags) {
2589 *frag = skb_shinfo(skb)->frags[i];
2590 get_page(frag->page);
2594 frag->page_offset += offset - pos;
2595 frag->size -= offset - pos;
2598 skb_shinfo(nskb)->nr_frags++;
2600 if (pos + size <= offset + len) {
2604 frag->size -= pos + size - (offset + len);
2611 if (pos < offset + len) {
2612 struct sk_buff *fskb2 = fskb;
2614 BUG_ON(pos + fskb->len != offset + len);
2620 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2626 SKB_FRAG_ASSERT(nskb);
2627 skb_shinfo(nskb)->frag_list = fskb2;
2631 nskb->data_len = len - hsize;
2632 nskb->len += nskb->data_len;
2633 nskb->truesize += nskb->data_len;
2634 } while ((offset += len) < skb->len);
2639 while ((skb = segs)) {
2643 return ERR_PTR(err);
2645 EXPORT_SYMBOL_GPL(skb_segment);
2647 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2649 struct sk_buff *p = *head;
2650 struct sk_buff *nskb;
2651 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2652 struct skb_shared_info *pinfo = skb_shinfo(p);
2653 unsigned int headroom;
2654 unsigned int len = skb_gro_len(skb);
2655 unsigned int offset = skb_gro_offset(skb);
2656 unsigned int headlen = skb_headlen(skb);
2658 if (p->len + len >= 65536)
2661 if (pinfo->frag_list)
2663 else if (headlen <= offset) {
2666 int i = skbinfo->nr_frags;
2667 int nr_frags = pinfo->nr_frags + i;
2671 if (nr_frags > MAX_SKB_FRAGS)
2674 pinfo->nr_frags = nr_frags;
2675 skbinfo->nr_frags = 0;
2677 frag = pinfo->frags + nr_frags;
2678 frag2 = skbinfo->frags + i;
2683 frag->page_offset += offset;
2684 frag->size -= offset;
2686 skb->truesize -= skb->data_len;
2687 skb->len -= skb->data_len;
2690 NAPI_GRO_CB(skb)->free = 1;
2692 } else if (skb_gro_len(p) != pinfo->gso_size)
2695 headroom = skb_headroom(p);
2696 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2697 if (unlikely(!nskb))
2700 __copy_skb_header(nskb, p);
2701 nskb->mac_len = p->mac_len;
2703 skb_reserve(nskb, headroom);
2704 __skb_put(nskb, skb_gro_offset(p));
2706 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2707 skb_set_network_header(nskb, skb_network_offset(p));
2708 skb_set_transport_header(nskb, skb_transport_offset(p));
2710 __skb_pull(p, skb_gro_offset(p));
2711 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2712 p->data - skb_mac_header(p));
2714 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2715 skb_shinfo(nskb)->frag_list = p;
2716 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2717 pinfo->gso_size = 0;
2718 skb_header_release(p);
2721 nskb->data_len += p->len;
2722 nskb->truesize += p->len;
2723 nskb->len += p->len;
2726 nskb->next = p->next;
2732 if (offset > headlen) {
2733 skbinfo->frags[0].page_offset += offset - headlen;
2734 skbinfo->frags[0].size -= offset - headlen;
2738 __skb_pull(skb, offset);
2740 p->prev->next = skb;
2742 skb_header_release(skb);
2745 NAPI_GRO_CB(p)->count++;
2750 NAPI_GRO_CB(skb)->same_flow = 1;
2753 EXPORT_SYMBOL_GPL(skb_gro_receive);
2755 void __init skb_init(void)
2757 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2758 sizeof(struct sk_buff),
2760 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2762 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2763 (2*sizeof(struct sk_buff)) +
2766 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2771 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2772 * @skb: Socket buffer containing the buffers to be mapped
2773 * @sg: The scatter-gather list to map into
2774 * @offset: The offset into the buffer's contents to start mapping
2775 * @len: Length of buffer space to be mapped
2777 * Fill the specified scatter-gather list with mappings/pointers into a
2778 * region of the buffer space attached to a socket buffer.
2781 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2783 int start = skb_headlen(skb);
2784 int i, copy = start - offset;
2785 struct sk_buff *frag_iter;
2791 sg_set_buf(sg, skb->data + offset, copy);
2793 if ((len -= copy) == 0)
2798 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2801 WARN_ON(start > offset + len);
2803 end = start + skb_shinfo(skb)->frags[i].size;
2804 if ((copy = end - offset) > 0) {
2805 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2809 sg_set_page(&sg[elt], frag->page, copy,
2810 frag->page_offset+offset-start);
2819 skb_walk_frags(skb, frag_iter) {
2822 WARN_ON(start > offset + len);
2824 end = start + frag_iter->len;
2825 if ((copy = end - offset) > 0) {
2828 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2830 if ((len -= copy) == 0)
2840 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2842 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2844 sg_mark_end(&sg[nsg - 1]);
2848 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2851 * skb_cow_data - Check that a socket buffer's data buffers are writable
2852 * @skb: The socket buffer to check.
2853 * @tailbits: Amount of trailing space to be added
2854 * @trailer: Returned pointer to the skb where the @tailbits space begins
2856 * Make sure that the data buffers attached to a socket buffer are
2857 * writable. If they are not, private copies are made of the data buffers
2858 * and the socket buffer is set to use these instead.
2860 * If @tailbits is given, make sure that there is space to write @tailbits
2861 * bytes of data beyond current end of socket buffer. @trailer will be
2862 * set to point to the skb in which this space begins.
2864 * The number of scatterlist elements required to completely map the
2865 * COW'd and extended socket buffer will be returned.
2867 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2871 struct sk_buff *skb1, **skb_p;
2873 /* If skb is cloned or its head is paged, reallocate
2874 * head pulling out all the pages (pages are considered not writable
2875 * at the moment even if they are anonymous).
2877 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2878 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2881 /* Easy case. Most of packets will go this way. */
2882 if (!skb_has_frag_list(skb)) {
2883 /* A little of trouble, not enough of space for trailer.
2884 * This should not happen, when stack is tuned to generate
2885 * good frames. OK, on miss we reallocate and reserve even more
2886 * space, 128 bytes is fair. */
2888 if (skb_tailroom(skb) < tailbits &&
2889 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2897 /* Misery. We are in troubles, going to mincer fragments... */
2900 skb_p = &skb_shinfo(skb)->frag_list;
2903 while ((skb1 = *skb_p) != NULL) {
2906 /* The fragment is partially pulled by someone,
2907 * this can happen on input. Copy it and everything
2910 if (skb_shared(skb1))
2913 /* If the skb is the last, worry about trailer. */
2915 if (skb1->next == NULL && tailbits) {
2916 if (skb_shinfo(skb1)->nr_frags ||
2917 skb_has_frag_list(skb1) ||
2918 skb_tailroom(skb1) < tailbits)
2919 ntail = tailbits + 128;
2925 skb_shinfo(skb1)->nr_frags ||
2926 skb_has_frag_list(skb1)) {
2927 struct sk_buff *skb2;
2929 /* Fuck, we are miserable poor guys... */
2931 skb2 = skb_copy(skb1, GFP_ATOMIC);
2933 skb2 = skb_copy_expand(skb1,
2937 if (unlikely(skb2 == NULL))
2941 skb_set_owner_w(skb2, skb1->sk);
2943 /* Looking around. Are we still alive?
2944 * OK, link new skb, drop old one */
2946 skb2->next = skb1->next;
2953 skb_p = &skb1->next;
2958 EXPORT_SYMBOL_GPL(skb_cow_data);
2960 static void sock_rmem_free(struct sk_buff *skb)
2962 struct sock *sk = skb->sk;
2964 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2968 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
2970 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
2972 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
2973 (unsigned)sk->sk_rcvbuf)
2978 skb->destructor = sock_rmem_free;
2979 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2981 skb_queue_tail(&sk->sk_error_queue, skb);
2982 if (!sock_flag(sk, SOCK_DEAD))
2983 sk->sk_data_ready(sk, skb->len);
2986 EXPORT_SYMBOL(sock_queue_err_skb);
2988 void skb_tstamp_tx(struct sk_buff *orig_skb,
2989 struct skb_shared_hwtstamps *hwtstamps)
2991 struct sock *sk = orig_skb->sk;
2992 struct sock_exterr_skb *serr;
2993 struct sk_buff *skb;
2999 skb = skb_clone(orig_skb, GFP_ATOMIC);
3004 *skb_hwtstamps(skb) =
3008 * no hardware time stamps available,
3009 * so keep the shared tx_flags and only
3010 * store software time stamp
3012 skb->tstamp = ktime_get_real();
3015 serr = SKB_EXT_ERR(skb);
3016 memset(serr, 0, sizeof(*serr));
3017 serr->ee.ee_errno = ENOMSG;
3018 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3020 err = sock_queue_err_skb(sk, skb);
3025 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3029 * skb_partial_csum_set - set up and verify partial csum values for packet
3030 * @skb: the skb to set
3031 * @start: the number of bytes after skb->data to start checksumming.
3032 * @off: the offset from start to place the checksum.
3034 * For untrusted partially-checksummed packets, we need to make sure the values
3035 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3037 * This function checks and sets those values and skb->ip_summed: if this
3038 * returns false you should drop the packet.
3040 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3042 if (unlikely(start > skb_headlen(skb)) ||
3043 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3044 if (net_ratelimit())
3046 "bad partial csum: csum=%u/%u len=%u\n",
3047 start, off, skb_headlen(skb));
3050 skb->ip_summed = CHECKSUM_PARTIAL;
3051 skb->csum_start = skb_headroom(skb) + start;
3052 skb->csum_offset = off;
3055 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3057 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3059 if (net_ratelimit())
3060 pr_warning("%s: received packets cannot be forwarded"
3061 " while LRO is enabled\n", skb->dev->name);
3063 EXPORT_SYMBOL(__skb_warn_lro_forwarding);