2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.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>
43 #include <linux/interrupt.h>
45 #include <linux/inet.h>
46 #include <linux/slab.h>
47 #include <linux/netdevice.h>
48 #ifdef CONFIG_NET_CLS_ACT
49 #include <net/pkt_sched.h>
51 #include <linux/string.h>
52 #include <linux/skbuff.h>
53 #include <linux/splice.h>
54 #include <linux/cache.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/init.h>
57 #include <linux/scatterlist.h>
59 #include <net/protocol.h>
62 #include <net/checksum.h>
65 #include <asm/uaccess.h>
66 #include <asm/system.h>
70 static struct kmem_cache *skbuff_head_cache __read_mostly;
71 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
73 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
74 struct pipe_buffer *buf)
76 struct sk_buff *skb = (struct sk_buff *) buf->private;
81 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
82 struct pipe_buffer *buf)
84 struct sk_buff *skb = (struct sk_buff *) buf->private;
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 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 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 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 void skb_truesize_bug(struct sk_buff *skb)
152 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
153 "len=%u, sizeof(sk_buff)=%Zd\n",
154 skb->truesize, skb->len, sizeof(struct sk_buff));
156 EXPORT_SYMBOL(skb_truesize_bug);
158 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
159 * 'private' fields and also do memory statistics to find all the
165 * __alloc_skb - allocate a network buffer
166 * @size: size to allocate
167 * @gfp_mask: allocation mask
168 * @fclone: allocate from fclone cache instead of head cache
169 * and allocate a cloned (child) skb
170 * @node: numa node to allocate memory on
172 * Allocate a new &sk_buff. The returned buffer has no headroom and a
173 * tail room of size bytes. The object has a reference count of one.
174 * The return is the buffer. On a failure the return is %NULL.
176 * Buffers may only be allocated from interrupts using a @gfp_mask of
179 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
180 int fclone, int node)
182 struct kmem_cache *cache;
183 struct skb_shared_info *shinfo;
187 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
190 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
194 size = SKB_DATA_ALIGN(size);
195 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
201 * Only clear those fields we need to clear, not those that we will
202 * actually initialise below. Hence, don't put any more fields after
203 * the tail pointer in struct sk_buff!
205 memset(skb, 0, offsetof(struct sk_buff, tail));
206 skb->truesize = size + sizeof(struct sk_buff);
207 atomic_set(&skb->users, 1);
210 skb_reset_tail_pointer(skb);
211 skb->end = skb->tail + size;
212 /* make sure we initialize shinfo sequentially */
213 shinfo = skb_shinfo(skb);
214 atomic_set(&shinfo->dataref, 1);
215 shinfo->nr_frags = 0;
216 shinfo->gso_size = 0;
217 shinfo->gso_segs = 0;
218 shinfo->gso_type = 0;
219 shinfo->ip6_frag_id = 0;
220 shinfo->frag_list = NULL;
223 struct sk_buff *child = skb + 1;
224 atomic_t *fclone_ref = (atomic_t *) (child + 1);
226 skb->fclone = SKB_FCLONE_ORIG;
227 atomic_set(fclone_ref, 1);
229 child->fclone = SKB_FCLONE_UNAVAILABLE;
234 kmem_cache_free(cache, skb);
240 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
241 * @dev: network device to receive on
242 * @length: length to allocate
243 * @gfp_mask: get_free_pages mask, passed to alloc_skb
245 * Allocate a new &sk_buff and assign it a usage count of one. The
246 * buffer has unspecified headroom built in. Users should allocate
247 * the headroom they think they need without accounting for the
248 * built in space. The built in space is used for optimisations.
250 * %NULL is returned if there is no free memory.
252 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
253 unsigned int length, gfp_t gfp_mask)
255 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
258 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
260 skb_reserve(skb, NET_SKB_PAD);
267 * dev_alloc_skb - allocate an skbuff for receiving
268 * @length: length to allocate
270 * Allocate a new &sk_buff and assign it a usage count of one. The
271 * buffer has unspecified headroom built in. Users should allocate
272 * the headroom they think they need without accounting for the
273 * built in space. The built in space is used for optimisations.
275 * %NULL is returned if there is no free memory. Although this function
276 * allocates memory it can be called from an interrupt.
278 struct sk_buff *dev_alloc_skb(unsigned int length)
281 * There is more code here than it seems:
282 * __dev_alloc_skb is an inline
284 return __dev_alloc_skb(length, GFP_ATOMIC);
286 EXPORT_SYMBOL(dev_alloc_skb);
288 static void skb_drop_list(struct sk_buff **listp)
290 struct sk_buff *list = *listp;
295 struct sk_buff *this = list;
301 static inline void skb_drop_fraglist(struct sk_buff *skb)
303 skb_drop_list(&skb_shinfo(skb)->frag_list);
306 static void skb_clone_fraglist(struct sk_buff *skb)
308 struct sk_buff *list;
310 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
314 static void skb_release_data(struct sk_buff *skb)
317 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
318 &skb_shinfo(skb)->dataref)) {
319 if (skb_shinfo(skb)->nr_frags) {
321 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
322 put_page(skb_shinfo(skb)->frags[i].page);
325 if (skb_shinfo(skb)->frag_list)
326 skb_drop_fraglist(skb);
333 * Free an skbuff by memory without cleaning the state.
335 static void kfree_skbmem(struct sk_buff *skb)
337 struct sk_buff *other;
338 atomic_t *fclone_ref;
340 switch (skb->fclone) {
341 case SKB_FCLONE_UNAVAILABLE:
342 kmem_cache_free(skbuff_head_cache, skb);
345 case SKB_FCLONE_ORIG:
346 fclone_ref = (atomic_t *) (skb + 2);
347 if (atomic_dec_and_test(fclone_ref))
348 kmem_cache_free(skbuff_fclone_cache, skb);
351 case SKB_FCLONE_CLONE:
352 fclone_ref = (atomic_t *) (skb + 1);
355 /* The clone portion is available for
356 * fast-cloning again.
358 skb->fclone = SKB_FCLONE_UNAVAILABLE;
360 if (atomic_dec_and_test(fclone_ref))
361 kmem_cache_free(skbuff_fclone_cache, other);
366 static void skb_release_head_state(struct sk_buff *skb)
368 dst_release(skb->dst);
370 secpath_put(skb->sp);
372 if (skb->destructor) {
374 skb->destructor(skb);
376 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
377 nf_conntrack_put(skb->nfct);
378 nf_conntrack_put_reasm(skb->nfct_reasm);
380 #ifdef CONFIG_BRIDGE_NETFILTER
381 nf_bridge_put(skb->nf_bridge);
383 /* XXX: IS this still necessary? - JHS */
384 #ifdef CONFIG_NET_SCHED
386 #ifdef CONFIG_NET_CLS_ACT
392 /* Free everything but the sk_buff shell. */
393 static void skb_release_all(struct sk_buff *skb)
395 skb_release_head_state(skb);
396 skb_release_data(skb);
400 * __kfree_skb - private function
403 * Free an sk_buff. Release anything attached to the buffer.
404 * Clean the state. This is an internal helper function. Users should
405 * always call kfree_skb
408 void __kfree_skb(struct sk_buff *skb)
410 skb_release_all(skb);
415 * kfree_skb - free an sk_buff
416 * @skb: buffer to free
418 * Drop a reference to the buffer and free it if the usage count has
421 void kfree_skb(struct sk_buff *skb)
425 if (likely(atomic_read(&skb->users) == 1))
427 else if (likely(!atomic_dec_and_test(&skb->users)))
432 int skb_recycle_check(struct sk_buff *skb, int skb_size)
434 struct skb_shared_info *shinfo;
436 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
439 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
440 if (skb_end_pointer(skb) - skb->head < skb_size)
443 if (skb_shared(skb) || skb_cloned(skb))
446 skb_release_head_state(skb);
447 shinfo = skb_shinfo(skb);
448 atomic_set(&shinfo->dataref, 1);
449 shinfo->nr_frags = 0;
450 shinfo->gso_size = 0;
451 shinfo->gso_segs = 0;
452 shinfo->gso_type = 0;
453 shinfo->ip6_frag_id = 0;
454 shinfo->frag_list = NULL;
456 memset(skb, 0, offsetof(struct sk_buff, tail));
457 skb_reset_tail_pointer(skb);
458 skb->data = skb->head + NET_SKB_PAD;
462 EXPORT_SYMBOL(skb_recycle_check);
464 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
466 new->tstamp = old->tstamp;
468 new->transport_header = old->transport_header;
469 new->network_header = old->network_header;
470 new->mac_header = old->mac_header;
471 new->dst = dst_clone(old->dst);
473 new->sp = secpath_get(old->sp);
475 memcpy(new->cb, old->cb, sizeof(old->cb));
476 new->csum_start = old->csum_start;
477 new->csum_offset = old->csum_offset;
478 new->local_df = old->local_df;
479 new->pkt_type = old->pkt_type;
480 new->ip_summed = old->ip_summed;
481 skb_copy_queue_mapping(new, old);
482 new->priority = old->priority;
483 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
484 new->ipvs_property = old->ipvs_property;
486 new->protocol = old->protocol;
487 new->mark = old->mark;
489 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
490 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
491 new->nf_trace = old->nf_trace;
493 #ifdef CONFIG_NET_SCHED
494 new->tc_index = old->tc_index;
495 #ifdef CONFIG_NET_CLS_ACT
496 new->tc_verd = old->tc_verd;
499 new->vlan_tci = old->vlan_tci;
501 skb_copy_secmark(new, old);
504 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
506 #define C(x) n->x = skb->x
508 n->next = n->prev = NULL;
510 __copy_skb_header(n, skb);
515 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
518 n->destructor = NULL;
525 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
528 atomic_set(&n->users, 1);
530 atomic_inc(&(skb_shinfo(skb)->dataref));
538 * skb_morph - morph one skb into another
539 * @dst: the skb to receive the contents
540 * @src: the skb to supply the contents
542 * This is identical to skb_clone except that the target skb is
543 * supplied by the user.
545 * The target skb is returned upon exit.
547 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
549 skb_release_all(dst);
550 return __skb_clone(dst, src);
552 EXPORT_SYMBOL_GPL(skb_morph);
555 * skb_clone - duplicate an sk_buff
556 * @skb: buffer to clone
557 * @gfp_mask: allocation priority
559 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
560 * copies share the same packet data but not structure. The new
561 * buffer has a reference count of 1. If the allocation fails the
562 * function returns %NULL otherwise the new buffer is returned.
564 * If this function is called from an interrupt gfp_mask() must be
568 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
573 if (skb->fclone == SKB_FCLONE_ORIG &&
574 n->fclone == SKB_FCLONE_UNAVAILABLE) {
575 atomic_t *fclone_ref = (atomic_t *) (n + 1);
576 n->fclone = SKB_FCLONE_CLONE;
577 atomic_inc(fclone_ref);
579 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
582 n->fclone = SKB_FCLONE_UNAVAILABLE;
585 return __skb_clone(n, skb);
588 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
590 #ifndef NET_SKBUFF_DATA_USES_OFFSET
592 * Shift between the two data areas in bytes
594 unsigned long offset = new->data - old->data;
597 __copy_skb_header(new, old);
599 #ifndef NET_SKBUFF_DATA_USES_OFFSET
600 /* {transport,network,mac}_header are relative to skb->head */
601 new->transport_header += offset;
602 new->network_header += offset;
603 new->mac_header += offset;
605 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
606 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
607 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
611 * skb_copy - create private copy of an sk_buff
612 * @skb: buffer to copy
613 * @gfp_mask: allocation priority
615 * Make a copy of both an &sk_buff and its data. This is used when the
616 * caller wishes to modify the data and needs a private copy of the
617 * data to alter. Returns %NULL on failure or the pointer to the buffer
618 * on success. The returned buffer has a reference count of 1.
620 * As by-product this function converts non-linear &sk_buff to linear
621 * one, so that &sk_buff becomes completely private and caller is allowed
622 * to modify all the data of returned buffer. This means that this
623 * function is not recommended for use in circumstances when only
624 * header is going to be modified. Use pskb_copy() instead.
627 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
629 int headerlen = skb->data - skb->head;
631 * Allocate the copy buffer
634 #ifdef NET_SKBUFF_DATA_USES_OFFSET
635 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
637 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
642 /* Set the data pointer */
643 skb_reserve(n, headerlen);
644 /* Set the tail pointer and length */
645 skb_put(n, skb->len);
647 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
650 copy_skb_header(n, skb);
656 * pskb_copy - create copy of an sk_buff with private head.
657 * @skb: buffer to copy
658 * @gfp_mask: allocation priority
660 * Make a copy of both an &sk_buff and part of its data, located
661 * in header. Fragmented data remain shared. This is used when
662 * the caller wishes to modify only header of &sk_buff and needs
663 * private copy of the header to alter. Returns %NULL on failure
664 * or the pointer to the buffer on success.
665 * The returned buffer has a reference count of 1.
668 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
671 * Allocate the copy buffer
674 #ifdef NET_SKBUFF_DATA_USES_OFFSET
675 n = alloc_skb(skb->end, gfp_mask);
677 n = alloc_skb(skb->end - skb->head, gfp_mask);
682 /* Set the data pointer */
683 skb_reserve(n, skb->data - skb->head);
684 /* Set the tail pointer and length */
685 skb_put(n, skb_headlen(skb));
687 skb_copy_from_linear_data(skb, n->data, n->len);
689 n->truesize += skb->data_len;
690 n->data_len = skb->data_len;
693 if (skb_shinfo(skb)->nr_frags) {
696 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
697 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
698 get_page(skb_shinfo(n)->frags[i].page);
700 skb_shinfo(n)->nr_frags = i;
703 if (skb_shinfo(skb)->frag_list) {
704 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
705 skb_clone_fraglist(n);
708 copy_skb_header(n, skb);
714 * pskb_expand_head - reallocate header of &sk_buff
715 * @skb: buffer to reallocate
716 * @nhead: room to add at head
717 * @ntail: room to add at tail
718 * @gfp_mask: allocation priority
720 * Expands (or creates identical copy, if &nhead and &ntail are zero)
721 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
722 * reference count of 1. Returns zero in the case of success or error,
723 * if expansion failed. In the last case, &sk_buff is not changed.
725 * All the pointers pointing into skb header may change and must be
726 * reloaded after call to this function.
729 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
734 #ifdef NET_SKBUFF_DATA_USES_OFFSET
735 int size = nhead + skb->end + ntail;
737 int size = nhead + (skb->end - skb->head) + ntail;
744 size = SKB_DATA_ALIGN(size);
746 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
750 /* Copy only real data... and, alas, header. This should be
751 * optimized for the cases when header is void. */
752 #ifdef NET_SKBUFF_DATA_USES_OFFSET
753 memcpy(data + nhead, skb->head, skb->tail);
755 memcpy(data + nhead, skb->head, skb->tail - skb->head);
757 memcpy(data + size, skb_end_pointer(skb),
758 sizeof(struct skb_shared_info));
760 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
761 get_page(skb_shinfo(skb)->frags[i].page);
763 if (skb_shinfo(skb)->frag_list)
764 skb_clone_fraglist(skb);
766 skb_release_data(skb);
768 off = (data + nhead) - skb->head;
772 #ifdef NET_SKBUFF_DATA_USES_OFFSET
776 skb->end = skb->head + size;
778 /* {transport,network,mac}_header and tail are relative to skb->head */
780 skb->transport_header += off;
781 skb->network_header += off;
782 skb->mac_header += off;
783 skb->csum_start += nhead;
787 atomic_set(&skb_shinfo(skb)->dataref, 1);
794 /* Make private copy of skb with writable head and some headroom */
796 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
798 struct sk_buff *skb2;
799 int delta = headroom - skb_headroom(skb);
802 skb2 = pskb_copy(skb, GFP_ATOMIC);
804 skb2 = skb_clone(skb, GFP_ATOMIC);
805 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
816 * skb_copy_expand - copy and expand sk_buff
817 * @skb: buffer to copy
818 * @newheadroom: new free bytes at head
819 * @newtailroom: new free bytes at tail
820 * @gfp_mask: allocation priority
822 * Make a copy of both an &sk_buff and its data and while doing so
823 * allocate additional space.
825 * This is used when the caller wishes to modify the data and needs a
826 * private copy of the data to alter as well as more space for new fields.
827 * Returns %NULL on failure or the pointer to the buffer
828 * on success. The returned buffer has a reference count of 1.
830 * You must pass %GFP_ATOMIC as the allocation priority if this function
831 * is called from an interrupt.
833 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
834 int newheadroom, int newtailroom,
838 * Allocate the copy buffer
840 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
842 int oldheadroom = skb_headroom(skb);
843 int head_copy_len, head_copy_off;
849 skb_reserve(n, newheadroom);
851 /* Set the tail pointer and length */
852 skb_put(n, skb->len);
854 head_copy_len = oldheadroom;
856 if (newheadroom <= head_copy_len)
857 head_copy_len = newheadroom;
859 head_copy_off = newheadroom - head_copy_len;
861 /* Copy the linear header and data. */
862 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
863 skb->len + head_copy_len))
866 copy_skb_header(n, skb);
868 off = newheadroom - oldheadroom;
869 n->csum_start += off;
870 #ifdef NET_SKBUFF_DATA_USES_OFFSET
871 n->transport_header += off;
872 n->network_header += off;
873 n->mac_header += off;
880 * skb_pad - zero pad the tail of an skb
881 * @skb: buffer to pad
884 * Ensure that a buffer is followed by a padding area that is zero
885 * filled. Used by network drivers which may DMA or transfer data
886 * beyond the buffer end onto the wire.
888 * May return error in out of memory cases. The skb is freed on error.
891 int skb_pad(struct sk_buff *skb, int pad)
896 /* If the skbuff is non linear tailroom is always zero.. */
897 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
898 memset(skb->data+skb->len, 0, pad);
902 ntail = skb->data_len + pad - (skb->end - skb->tail);
903 if (likely(skb_cloned(skb) || ntail > 0)) {
904 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
909 /* FIXME: The use of this function with non-linear skb's really needs
912 err = skb_linearize(skb);
916 memset(skb->data + skb->len, 0, pad);
925 * skb_put - add data to a buffer
926 * @skb: buffer to use
927 * @len: amount of data to add
929 * This function extends the used data area of the buffer. If this would
930 * exceed the total buffer size the kernel will panic. A pointer to the
931 * first byte of the extra data is returned.
933 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
935 unsigned char *tmp = skb_tail_pointer(skb);
936 SKB_LINEAR_ASSERT(skb);
939 if (unlikely(skb->tail > skb->end))
940 skb_over_panic(skb, len, __builtin_return_address(0));
943 EXPORT_SYMBOL(skb_put);
946 * skb_push - add data to the start of a buffer
947 * @skb: buffer to use
948 * @len: amount of data to add
950 * This function extends the used data area of the buffer at the buffer
951 * start. If this would exceed the total buffer headroom the kernel will
952 * panic. A pointer to the first byte of the extra data is returned.
954 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
958 if (unlikely(skb->data<skb->head))
959 skb_under_panic(skb, len, __builtin_return_address(0));
962 EXPORT_SYMBOL(skb_push);
965 * skb_pull - remove data from the start of a buffer
966 * @skb: buffer to use
967 * @len: amount of data to remove
969 * This function removes data from the start of a buffer, returning
970 * the memory to the headroom. A pointer to the next data in the buffer
971 * is returned. Once the data has been pulled future pushes will overwrite
974 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
976 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
978 EXPORT_SYMBOL(skb_pull);
981 * skb_trim - remove end from a buffer
982 * @skb: buffer to alter
985 * Cut the length of a buffer down by removing data from the tail. If
986 * the buffer is already under the length specified it is not modified.
987 * The skb must be linear.
989 void skb_trim(struct sk_buff *skb, unsigned int len)
992 __skb_trim(skb, len);
994 EXPORT_SYMBOL(skb_trim);
996 /* Trims skb to length len. It can change skb pointers.
999 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1001 struct sk_buff **fragp;
1002 struct sk_buff *frag;
1003 int offset = skb_headlen(skb);
1004 int nfrags = skb_shinfo(skb)->nr_frags;
1008 if (skb_cloned(skb) &&
1009 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1016 for (; i < nfrags; i++) {
1017 int end = offset + skb_shinfo(skb)->frags[i].size;
1024 skb_shinfo(skb)->frags[i++].size = len - offset;
1027 skb_shinfo(skb)->nr_frags = i;
1029 for (; i < nfrags; i++)
1030 put_page(skb_shinfo(skb)->frags[i].page);
1032 if (skb_shinfo(skb)->frag_list)
1033 skb_drop_fraglist(skb);
1037 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1038 fragp = &frag->next) {
1039 int end = offset + frag->len;
1041 if (skb_shared(frag)) {
1042 struct sk_buff *nfrag;
1044 nfrag = skb_clone(frag, GFP_ATOMIC);
1045 if (unlikely(!nfrag))
1048 nfrag->next = frag->next;
1060 unlikely((err = pskb_trim(frag, len - offset))))
1064 skb_drop_list(&frag->next);
1069 if (len > skb_headlen(skb)) {
1070 skb->data_len -= skb->len - len;
1075 skb_set_tail_pointer(skb, len);
1082 * __pskb_pull_tail - advance tail of skb header
1083 * @skb: buffer to reallocate
1084 * @delta: number of bytes to advance tail
1086 * The function makes a sense only on a fragmented &sk_buff,
1087 * it expands header moving its tail forward and copying necessary
1088 * data from fragmented part.
1090 * &sk_buff MUST have reference count of 1.
1092 * Returns %NULL (and &sk_buff does not change) if pull failed
1093 * or value of new tail of skb in the case of success.
1095 * All the pointers pointing into skb header may change and must be
1096 * reloaded after call to this function.
1099 /* Moves tail of skb head forward, copying data from fragmented part,
1100 * when it is necessary.
1101 * 1. It may fail due to malloc failure.
1102 * 2. It may change skb pointers.
1104 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1106 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1108 /* If skb has not enough free space at tail, get new one
1109 * plus 128 bytes for future expansions. If we have enough
1110 * room at tail, reallocate without expansion only if skb is cloned.
1112 int i, k, eat = (skb->tail + delta) - skb->end;
1114 if (eat > 0 || skb_cloned(skb)) {
1115 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1120 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1123 /* Optimization: no fragments, no reasons to preestimate
1124 * size of pulled pages. Superb.
1126 if (!skb_shinfo(skb)->frag_list)
1129 /* Estimate size of pulled pages. */
1131 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1132 if (skb_shinfo(skb)->frags[i].size >= eat)
1134 eat -= skb_shinfo(skb)->frags[i].size;
1137 /* If we need update frag list, we are in troubles.
1138 * Certainly, it possible to add an offset to skb data,
1139 * but taking into account that pulling is expected to
1140 * be very rare operation, it is worth to fight against
1141 * further bloating skb head and crucify ourselves here instead.
1142 * Pure masohism, indeed. 8)8)
1145 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1146 struct sk_buff *clone = NULL;
1147 struct sk_buff *insp = NULL;
1152 if (list->len <= eat) {
1153 /* Eaten as whole. */
1158 /* Eaten partially. */
1160 if (skb_shared(list)) {
1161 /* Sucks! We need to fork list. :-( */
1162 clone = skb_clone(list, GFP_ATOMIC);
1168 /* This may be pulled without
1172 if (!pskb_pull(list, eat)) {
1181 /* Free pulled out fragments. */
1182 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1183 skb_shinfo(skb)->frag_list = list->next;
1186 /* And insert new clone at head. */
1189 skb_shinfo(skb)->frag_list = clone;
1192 /* Success! Now we may commit changes to skb data. */
1197 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1198 if (skb_shinfo(skb)->frags[i].size <= eat) {
1199 put_page(skb_shinfo(skb)->frags[i].page);
1200 eat -= skb_shinfo(skb)->frags[i].size;
1202 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1204 skb_shinfo(skb)->frags[k].page_offset += eat;
1205 skb_shinfo(skb)->frags[k].size -= eat;
1211 skb_shinfo(skb)->nr_frags = k;
1214 skb->data_len -= delta;
1216 return skb_tail_pointer(skb);
1219 /* Copy some data bits from skb to kernel buffer. */
1221 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1224 int start = skb_headlen(skb);
1226 if (offset > (int)skb->len - len)
1230 if ((copy = start - offset) > 0) {
1233 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1234 if ((len -= copy) == 0)
1240 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1243 WARN_ON(start > offset + len);
1245 end = start + skb_shinfo(skb)->frags[i].size;
1246 if ((copy = end - offset) > 0) {
1252 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1254 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1255 offset - start, copy);
1256 kunmap_skb_frag(vaddr);
1258 if ((len -= copy) == 0)
1266 if (skb_shinfo(skb)->frag_list) {
1267 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1269 for (; list; list = list->next) {
1272 WARN_ON(start > offset + len);
1274 end = start + list->len;
1275 if ((copy = end - offset) > 0) {
1278 if (skb_copy_bits(list, offset - start,
1281 if ((len -= copy) == 0)
1297 * Callback from splice_to_pipe(), if we need to release some pages
1298 * at the end of the spd in case we error'ed out in filling the pipe.
1300 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1302 struct sk_buff *skb = (struct sk_buff *) spd->partial[i].private;
1308 * Fill page/offset/length into spd, if it can hold more pages.
1310 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1311 unsigned int len, unsigned int offset,
1312 struct sk_buff *skb)
1314 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1317 spd->pages[spd->nr_pages] = page;
1318 spd->partial[spd->nr_pages].len = len;
1319 spd->partial[spd->nr_pages].offset = offset;
1320 spd->partial[spd->nr_pages].private = (unsigned long) skb_get(skb);
1325 static inline void __segment_seek(struct page **page, unsigned int *poff,
1326 unsigned int *plen, unsigned int off)
1329 *page += *poff / PAGE_SIZE;
1330 *poff = *poff % PAGE_SIZE;
1334 static inline int __splice_segment(struct page *page, unsigned int poff,
1335 unsigned int plen, unsigned int *off,
1336 unsigned int *len, struct sk_buff *skb,
1337 struct splice_pipe_desc *spd)
1342 /* skip this segment if already processed */
1348 /* ignore any bits we already processed */
1350 __segment_seek(&page, &poff, &plen, *off);
1355 unsigned int flen = min(*len, plen);
1357 /* the linear region may spread across several pages */
1358 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1360 if (spd_fill_page(spd, page, flen, poff, skb))
1363 __segment_seek(&page, &poff, &plen, flen);
1366 } while (*len && plen);
1372 * Map linear and fragment data from the skb to spd. It reports failure if the
1373 * pipe is full or if we already spliced the requested length.
1375 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1377 struct splice_pipe_desc *spd)
1382 * map the linear part
1384 if (__splice_segment(virt_to_page(skb->data),
1385 (unsigned long) skb->data & (PAGE_SIZE - 1),
1387 offset, len, skb, spd))
1391 * then map the fragments
1393 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1394 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1396 if (__splice_segment(f->page, f->page_offset, f->size,
1397 offset, len, skb, spd))
1405 * Map data from the skb to a pipe. Should handle both the linear part,
1406 * the fragments, and the frag list. It does NOT handle frag lists within
1407 * the frag list, if such a thing exists. We'd probably need to recurse to
1408 * handle that cleanly.
1410 int skb_splice_bits(struct sk_buff *__skb, unsigned int offset,
1411 struct pipe_inode_info *pipe, unsigned int tlen,
1414 struct partial_page partial[PIPE_BUFFERS];
1415 struct page *pages[PIPE_BUFFERS];
1416 struct splice_pipe_desc spd = {
1420 .ops = &sock_pipe_buf_ops,
1421 .spd_release = sock_spd_release,
1423 struct sk_buff *skb;
1426 * I'd love to avoid the clone here, but tcp_read_sock()
1427 * ignores reference counts and unconditonally kills the sk_buff
1428 * on return from the actor.
1430 skb = skb_clone(__skb, GFP_KERNEL);
1435 * __skb_splice_bits() only fails if the output has no room left,
1436 * so no point in going over the frag_list for the error case.
1438 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1444 * now see if we have a frag_list to map
1446 if (skb_shinfo(skb)->frag_list) {
1447 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1449 for (; list && tlen; list = list->next) {
1450 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1457 * drop our reference to the clone, the pipe consumption will
1464 struct sock *sk = __skb->sk;
1467 * Drop the socket lock, otherwise we have reverse
1468 * locking dependencies between sk_lock and i_mutex
1469 * here as compared to sendfile(). We enter here
1470 * with the socket lock held, and splice_to_pipe() will
1471 * grab the pipe inode lock. For sendfile() emulation,
1472 * we call into ->sendpage() with the i_mutex lock held
1473 * and networking will grab the socket lock.
1476 ret = splice_to_pipe(pipe, &spd);
1485 * skb_store_bits - store bits from kernel buffer to skb
1486 * @skb: destination buffer
1487 * @offset: offset in destination
1488 * @from: source buffer
1489 * @len: number of bytes to copy
1491 * Copy the specified number of bytes from the source buffer to the
1492 * destination skb. This function handles all the messy bits of
1493 * traversing fragment lists and such.
1496 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1499 int start = skb_headlen(skb);
1501 if (offset > (int)skb->len - len)
1504 if ((copy = start - offset) > 0) {
1507 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1508 if ((len -= copy) == 0)
1514 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1518 WARN_ON(start > offset + len);
1520 end = start + frag->size;
1521 if ((copy = end - offset) > 0) {
1527 vaddr = kmap_skb_frag(frag);
1528 memcpy(vaddr + frag->page_offset + offset - start,
1530 kunmap_skb_frag(vaddr);
1532 if ((len -= copy) == 0)
1540 if (skb_shinfo(skb)->frag_list) {
1541 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1543 for (; list; list = list->next) {
1546 WARN_ON(start > offset + len);
1548 end = start + list->len;
1549 if ((copy = end - offset) > 0) {
1552 if (skb_store_bits(list, offset - start,
1555 if ((len -= copy) == 0)
1570 EXPORT_SYMBOL(skb_store_bits);
1572 /* Checksum skb data. */
1574 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1575 int len, __wsum csum)
1577 int start = skb_headlen(skb);
1578 int i, copy = start - offset;
1581 /* Checksum header. */
1585 csum = csum_partial(skb->data + offset, copy, csum);
1586 if ((len -= copy) == 0)
1592 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1595 WARN_ON(start > offset + len);
1597 end = start + skb_shinfo(skb)->frags[i].size;
1598 if ((copy = end - offset) > 0) {
1601 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1605 vaddr = kmap_skb_frag(frag);
1606 csum2 = csum_partial(vaddr + frag->page_offset +
1607 offset - start, copy, 0);
1608 kunmap_skb_frag(vaddr);
1609 csum = csum_block_add(csum, csum2, pos);
1618 if (skb_shinfo(skb)->frag_list) {
1619 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1621 for (; list; list = list->next) {
1624 WARN_ON(start > offset + len);
1626 end = start + list->len;
1627 if ((copy = end - offset) > 0) {
1631 csum2 = skb_checksum(list, offset - start,
1633 csum = csum_block_add(csum, csum2, pos);
1634 if ((len -= copy) == 0)
1647 /* Both of above in one bottle. */
1649 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1650 u8 *to, int len, __wsum csum)
1652 int start = skb_headlen(skb);
1653 int i, copy = start - offset;
1660 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1662 if ((len -= copy) == 0)
1669 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1672 WARN_ON(start > offset + len);
1674 end = start + skb_shinfo(skb)->frags[i].size;
1675 if ((copy = end - offset) > 0) {
1678 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1682 vaddr = kmap_skb_frag(frag);
1683 csum2 = csum_partial_copy_nocheck(vaddr +
1687 kunmap_skb_frag(vaddr);
1688 csum = csum_block_add(csum, csum2, pos);
1698 if (skb_shinfo(skb)->frag_list) {
1699 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1701 for (; list; list = list->next) {
1705 WARN_ON(start > offset + len);
1707 end = start + list->len;
1708 if ((copy = end - offset) > 0) {
1711 csum2 = skb_copy_and_csum_bits(list,
1714 csum = csum_block_add(csum, csum2, pos);
1715 if ((len -= copy) == 0)
1728 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1733 if (skb->ip_summed == CHECKSUM_PARTIAL)
1734 csstart = skb->csum_start - skb_headroom(skb);
1736 csstart = skb_headlen(skb);
1738 BUG_ON(csstart > skb_headlen(skb));
1740 skb_copy_from_linear_data(skb, to, csstart);
1743 if (csstart != skb->len)
1744 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1745 skb->len - csstart, 0);
1747 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1748 long csstuff = csstart + skb->csum_offset;
1750 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1755 * skb_dequeue - remove from the head of the queue
1756 * @list: list to dequeue from
1758 * Remove the head of the list. The list lock is taken so the function
1759 * may be used safely with other locking list functions. The head item is
1760 * returned or %NULL if the list is empty.
1763 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1765 unsigned long flags;
1766 struct sk_buff *result;
1768 spin_lock_irqsave(&list->lock, flags);
1769 result = __skb_dequeue(list);
1770 spin_unlock_irqrestore(&list->lock, flags);
1775 * skb_dequeue_tail - remove from the tail of the queue
1776 * @list: list to dequeue from
1778 * Remove the tail of the list. The list lock is taken so the function
1779 * may be used safely with other locking list functions. The tail item is
1780 * returned or %NULL if the list is empty.
1782 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1784 unsigned long flags;
1785 struct sk_buff *result;
1787 spin_lock_irqsave(&list->lock, flags);
1788 result = __skb_dequeue_tail(list);
1789 spin_unlock_irqrestore(&list->lock, flags);
1794 * skb_queue_purge - empty a list
1795 * @list: list to empty
1797 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1798 * the list and one reference dropped. This function takes the list
1799 * lock and is atomic with respect to other list locking functions.
1801 void skb_queue_purge(struct sk_buff_head *list)
1803 struct sk_buff *skb;
1804 while ((skb = skb_dequeue(list)) != NULL)
1809 * skb_queue_head - queue a buffer at the list head
1810 * @list: list to use
1811 * @newsk: buffer to queue
1813 * Queue a buffer at the start of the list. This function takes the
1814 * list lock and can be used safely with other locking &sk_buff functions
1817 * A buffer cannot be placed on two lists at the same time.
1819 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1821 unsigned long flags;
1823 spin_lock_irqsave(&list->lock, flags);
1824 __skb_queue_head(list, newsk);
1825 spin_unlock_irqrestore(&list->lock, flags);
1829 * skb_queue_tail - queue a buffer at the list tail
1830 * @list: list to use
1831 * @newsk: buffer to queue
1833 * Queue a buffer at the tail of the list. This function takes the
1834 * list lock and can be used safely with other locking &sk_buff functions
1837 * A buffer cannot be placed on two lists at the same time.
1839 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1841 unsigned long flags;
1843 spin_lock_irqsave(&list->lock, flags);
1844 __skb_queue_tail(list, newsk);
1845 spin_unlock_irqrestore(&list->lock, flags);
1849 * skb_unlink - remove a buffer from a list
1850 * @skb: buffer to remove
1851 * @list: list to use
1853 * Remove a packet from a list. The list locks are taken and this
1854 * function is atomic with respect to other list locked calls
1856 * You must know what list the SKB is on.
1858 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1860 unsigned long flags;
1862 spin_lock_irqsave(&list->lock, flags);
1863 __skb_unlink(skb, list);
1864 spin_unlock_irqrestore(&list->lock, flags);
1868 * skb_append - append a buffer
1869 * @old: buffer to insert after
1870 * @newsk: buffer to insert
1871 * @list: list to use
1873 * Place a packet after a given packet in a list. The list locks are taken
1874 * and this function is atomic with respect to other list locked calls.
1875 * A buffer cannot be placed on two lists at the same time.
1877 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1879 unsigned long flags;
1881 spin_lock_irqsave(&list->lock, flags);
1882 __skb_queue_after(list, old, newsk);
1883 spin_unlock_irqrestore(&list->lock, flags);
1888 * skb_insert - insert a buffer
1889 * @old: buffer to insert before
1890 * @newsk: buffer to insert
1891 * @list: list to use
1893 * Place a packet before a given packet in a list. The list locks are
1894 * taken and this function is atomic with respect to other list locked
1897 * A buffer cannot be placed on two lists at the same time.
1899 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1901 unsigned long flags;
1903 spin_lock_irqsave(&list->lock, flags);
1904 __skb_insert(newsk, old->prev, old, list);
1905 spin_unlock_irqrestore(&list->lock, flags);
1908 static inline void skb_split_inside_header(struct sk_buff *skb,
1909 struct sk_buff* skb1,
1910 const u32 len, const int pos)
1914 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1916 /* And move data appendix as is. */
1917 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1918 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1920 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1921 skb_shinfo(skb)->nr_frags = 0;
1922 skb1->data_len = skb->data_len;
1923 skb1->len += skb1->data_len;
1926 skb_set_tail_pointer(skb, len);
1929 static inline void skb_split_no_header(struct sk_buff *skb,
1930 struct sk_buff* skb1,
1931 const u32 len, int pos)
1934 const int nfrags = skb_shinfo(skb)->nr_frags;
1936 skb_shinfo(skb)->nr_frags = 0;
1937 skb1->len = skb1->data_len = skb->len - len;
1939 skb->data_len = len - pos;
1941 for (i = 0; i < nfrags; i++) {
1942 int size = skb_shinfo(skb)->frags[i].size;
1944 if (pos + size > len) {
1945 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1949 * We have two variants in this case:
1950 * 1. Move all the frag to the second
1951 * part, if it is possible. F.e.
1952 * this approach is mandatory for TUX,
1953 * where splitting is expensive.
1954 * 2. Split is accurately. We make this.
1956 get_page(skb_shinfo(skb)->frags[i].page);
1957 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1958 skb_shinfo(skb1)->frags[0].size -= len - pos;
1959 skb_shinfo(skb)->frags[i].size = len - pos;
1960 skb_shinfo(skb)->nr_frags++;
1964 skb_shinfo(skb)->nr_frags++;
1967 skb_shinfo(skb1)->nr_frags = k;
1971 * skb_split - Split fragmented skb to two parts at length len.
1972 * @skb: the buffer to split
1973 * @skb1: the buffer to receive the second part
1974 * @len: new length for skb
1976 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1978 int pos = skb_headlen(skb);
1980 if (len < pos) /* Split line is inside header. */
1981 skb_split_inside_header(skb, skb1, len, pos);
1982 else /* Second chunk has no header, nothing to copy. */
1983 skb_split_no_header(skb, skb1, len, pos);
1987 * skb_prepare_seq_read - Prepare a sequential read of skb data
1988 * @skb: the buffer to read
1989 * @from: lower offset of data to be read
1990 * @to: upper offset of data to be read
1991 * @st: state variable
1993 * Initializes the specified state variable. Must be called before
1994 * invoking skb_seq_read() for the first time.
1996 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1997 unsigned int to, struct skb_seq_state *st)
1999 st->lower_offset = from;
2000 st->upper_offset = to;
2001 st->root_skb = st->cur_skb = skb;
2002 st->frag_idx = st->stepped_offset = 0;
2003 st->frag_data = NULL;
2007 * skb_seq_read - Sequentially read skb data
2008 * @consumed: number of bytes consumed by the caller so far
2009 * @data: destination pointer for data to be returned
2010 * @st: state variable
2012 * Reads a block of skb data at &consumed relative to the
2013 * lower offset specified to skb_prepare_seq_read(). Assigns
2014 * the head of the data block to &data and returns the length
2015 * of the block or 0 if the end of the skb data or the upper
2016 * offset has been reached.
2018 * The caller is not required to consume all of the data
2019 * returned, i.e. &consumed is typically set to the number
2020 * of bytes already consumed and the next call to
2021 * skb_seq_read() will return the remaining part of the block.
2023 * Note 1: The size of each block of data returned can be arbitary,
2024 * this limitation is the cost for zerocopy seqeuental
2025 * reads of potentially non linear data.
2027 * Note 2: Fragment lists within fragments are not implemented
2028 * at the moment, state->root_skb could be replaced with
2029 * a stack for this purpose.
2031 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2032 struct skb_seq_state *st)
2034 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2037 if (unlikely(abs_offset >= st->upper_offset))
2041 block_limit = skb_headlen(st->cur_skb);
2043 if (abs_offset < block_limit) {
2044 *data = st->cur_skb->data + abs_offset;
2045 return block_limit - abs_offset;
2048 if (st->frag_idx == 0 && !st->frag_data)
2049 st->stepped_offset += skb_headlen(st->cur_skb);
2051 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2052 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2053 block_limit = frag->size + st->stepped_offset;
2055 if (abs_offset < block_limit) {
2057 st->frag_data = kmap_skb_frag(frag);
2059 *data = (u8 *) st->frag_data + frag->page_offset +
2060 (abs_offset - st->stepped_offset);
2062 return block_limit - abs_offset;
2065 if (st->frag_data) {
2066 kunmap_skb_frag(st->frag_data);
2067 st->frag_data = NULL;
2071 st->stepped_offset += frag->size;
2074 if (st->frag_data) {
2075 kunmap_skb_frag(st->frag_data);
2076 st->frag_data = NULL;
2079 if (st->cur_skb->next) {
2080 st->cur_skb = st->cur_skb->next;
2083 } else if (st->root_skb == st->cur_skb &&
2084 skb_shinfo(st->root_skb)->frag_list) {
2085 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2093 * skb_abort_seq_read - Abort a sequential read of skb data
2094 * @st: state variable
2096 * Must be called if skb_seq_read() was not called until it
2099 void skb_abort_seq_read(struct skb_seq_state *st)
2102 kunmap_skb_frag(st->frag_data);
2105 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2107 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2108 struct ts_config *conf,
2109 struct ts_state *state)
2111 return skb_seq_read(offset, text, TS_SKB_CB(state));
2114 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2116 skb_abort_seq_read(TS_SKB_CB(state));
2120 * skb_find_text - Find a text pattern in skb data
2121 * @skb: the buffer to look in
2122 * @from: search offset
2124 * @config: textsearch configuration
2125 * @state: uninitialized textsearch state variable
2127 * Finds a pattern in the skb data according to the specified
2128 * textsearch configuration. Use textsearch_next() to retrieve
2129 * subsequent occurrences of the pattern. Returns the offset
2130 * to the first occurrence or UINT_MAX if no match was found.
2132 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2133 unsigned int to, struct ts_config *config,
2134 struct ts_state *state)
2138 config->get_next_block = skb_ts_get_next_block;
2139 config->finish = skb_ts_finish;
2141 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2143 ret = textsearch_find(config, state);
2144 return (ret <= to - from ? ret : UINT_MAX);
2148 * skb_append_datato_frags: - append the user data to a skb
2149 * @sk: sock structure
2150 * @skb: skb structure to be appened with user data.
2151 * @getfrag: call back function to be used for getting the user data
2152 * @from: pointer to user message iov
2153 * @length: length of the iov message
2155 * Description: This procedure append the user data in the fragment part
2156 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2158 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2159 int (*getfrag)(void *from, char *to, int offset,
2160 int len, int odd, struct sk_buff *skb),
2161 void *from, int length)
2164 skb_frag_t *frag = NULL;
2165 struct page *page = NULL;
2171 /* Return error if we don't have space for new frag */
2172 frg_cnt = skb_shinfo(skb)->nr_frags;
2173 if (frg_cnt >= MAX_SKB_FRAGS)
2176 /* allocate a new page for next frag */
2177 page = alloc_pages(sk->sk_allocation, 0);
2179 /* If alloc_page fails just return failure and caller will
2180 * free previous allocated pages by doing kfree_skb()
2185 /* initialize the next frag */
2186 sk->sk_sndmsg_page = page;
2187 sk->sk_sndmsg_off = 0;
2188 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2189 skb->truesize += PAGE_SIZE;
2190 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2192 /* get the new initialized frag */
2193 frg_cnt = skb_shinfo(skb)->nr_frags;
2194 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2196 /* copy the user data to page */
2197 left = PAGE_SIZE - frag->page_offset;
2198 copy = (length > left)? left : length;
2200 ret = getfrag(from, (page_address(frag->page) +
2201 frag->page_offset + frag->size),
2202 offset, copy, 0, skb);
2206 /* copy was successful so update the size parameters */
2207 sk->sk_sndmsg_off += copy;
2210 skb->data_len += copy;
2214 } while (length > 0);
2220 * skb_pull_rcsum - pull skb and update receive checksum
2221 * @skb: buffer to update
2222 * @len: length of data pulled
2224 * This function performs an skb_pull on the packet and updates
2225 * the CHECKSUM_COMPLETE checksum. It should be used on
2226 * receive path processing instead of skb_pull unless you know
2227 * that the checksum difference is zero (e.g., a valid IP header)
2228 * or you are setting ip_summed to CHECKSUM_NONE.
2230 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2232 BUG_ON(len > skb->len);
2234 BUG_ON(skb->len < skb->data_len);
2235 skb_postpull_rcsum(skb, skb->data, len);
2236 return skb->data += len;
2239 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2242 * skb_segment - Perform protocol segmentation on skb.
2243 * @skb: buffer to segment
2244 * @features: features for the output path (see dev->features)
2246 * This function performs segmentation on the given skb. It returns
2247 * a pointer to the first in a list of new skbs for the segments.
2248 * In case of error it returns ERR_PTR(err).
2250 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2252 struct sk_buff *segs = NULL;
2253 struct sk_buff *tail = NULL;
2254 unsigned int mss = skb_shinfo(skb)->gso_size;
2255 unsigned int doffset = skb->data - skb_mac_header(skb);
2256 unsigned int offset = doffset;
2257 unsigned int headroom;
2259 int sg = features & NETIF_F_SG;
2260 int nfrags = skb_shinfo(skb)->nr_frags;
2265 __skb_push(skb, doffset);
2266 headroom = skb_headroom(skb);
2267 pos = skb_headlen(skb);
2270 struct sk_buff *nskb;
2276 len = skb->len - offset;
2280 hsize = skb_headlen(skb) - offset;
2283 if (hsize > len || !sg)
2286 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
2287 if (unlikely(!nskb))
2296 __copy_skb_header(nskb, skb);
2297 nskb->mac_len = skb->mac_len;
2299 skb_reserve(nskb, headroom);
2300 skb_reset_mac_header(nskb);
2301 skb_set_network_header(nskb, skb->mac_len);
2302 nskb->transport_header = (nskb->network_header +
2303 skb_network_header_len(skb));
2304 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
2307 nskb->ip_summed = CHECKSUM_NONE;
2308 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2314 frag = skb_shinfo(nskb)->frags;
2317 skb_copy_from_linear_data_offset(skb, offset,
2318 skb_put(nskb, hsize), hsize);
2320 while (pos < offset + len) {
2321 BUG_ON(i >= nfrags);
2323 *frag = skb_shinfo(skb)->frags[i];
2324 get_page(frag->page);
2328 frag->page_offset += offset - pos;
2329 frag->size -= offset - pos;
2334 if (pos + size <= offset + len) {
2338 frag->size -= pos + size - (offset + len);
2345 skb_shinfo(nskb)->nr_frags = k;
2346 nskb->data_len = len - hsize;
2347 nskb->len += nskb->data_len;
2348 nskb->truesize += nskb->data_len;
2349 } while ((offset += len) < skb->len);
2354 while ((skb = segs)) {
2358 return ERR_PTR(err);
2361 EXPORT_SYMBOL_GPL(skb_segment);
2363 void __init skb_init(void)
2365 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2366 sizeof(struct sk_buff),
2368 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2370 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2371 (2*sizeof(struct sk_buff)) +
2374 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2379 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2380 * @skb: Socket buffer containing the buffers to be mapped
2381 * @sg: The scatter-gather list to map into
2382 * @offset: The offset into the buffer's contents to start mapping
2383 * @len: Length of buffer space to be mapped
2385 * Fill the specified scatter-gather list with mappings/pointers into a
2386 * region of the buffer space attached to a socket buffer.
2389 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2391 int start = skb_headlen(skb);
2392 int i, copy = start - offset;
2398 sg_set_buf(sg, skb->data + offset, copy);
2400 if ((len -= copy) == 0)
2405 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2408 WARN_ON(start > offset + len);
2410 end = start + skb_shinfo(skb)->frags[i].size;
2411 if ((copy = end - offset) > 0) {
2412 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2416 sg_set_page(&sg[elt], frag->page, copy,
2417 frag->page_offset+offset-start);
2426 if (skb_shinfo(skb)->frag_list) {
2427 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2429 for (; list; list = list->next) {
2432 WARN_ON(start > offset + len);
2434 end = start + list->len;
2435 if ((copy = end - offset) > 0) {
2438 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2440 if ((len -= copy) == 0)
2451 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2453 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2455 sg_mark_end(&sg[nsg - 1]);
2461 * skb_cow_data - Check that a socket buffer's data buffers are writable
2462 * @skb: The socket buffer to check.
2463 * @tailbits: Amount of trailing space to be added
2464 * @trailer: Returned pointer to the skb where the @tailbits space begins
2466 * Make sure that the data buffers attached to a socket buffer are
2467 * writable. If they are not, private copies are made of the data buffers
2468 * and the socket buffer is set to use these instead.
2470 * If @tailbits is given, make sure that there is space to write @tailbits
2471 * bytes of data beyond current end of socket buffer. @trailer will be
2472 * set to point to the skb in which this space begins.
2474 * The number of scatterlist elements required to completely map the
2475 * COW'd and extended socket buffer will be returned.
2477 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2481 struct sk_buff *skb1, **skb_p;
2483 /* If skb is cloned or its head is paged, reallocate
2484 * head pulling out all the pages (pages are considered not writable
2485 * at the moment even if they are anonymous).
2487 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2488 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2491 /* Easy case. Most of packets will go this way. */
2492 if (!skb_shinfo(skb)->frag_list) {
2493 /* A little of trouble, not enough of space for trailer.
2494 * This should not happen, when stack is tuned to generate
2495 * good frames. OK, on miss we reallocate and reserve even more
2496 * space, 128 bytes is fair. */
2498 if (skb_tailroom(skb) < tailbits &&
2499 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2507 /* Misery. We are in troubles, going to mincer fragments... */
2510 skb_p = &skb_shinfo(skb)->frag_list;
2513 while ((skb1 = *skb_p) != NULL) {
2516 /* The fragment is partially pulled by someone,
2517 * this can happen on input. Copy it and everything
2520 if (skb_shared(skb1))
2523 /* If the skb is the last, worry about trailer. */
2525 if (skb1->next == NULL && tailbits) {
2526 if (skb_shinfo(skb1)->nr_frags ||
2527 skb_shinfo(skb1)->frag_list ||
2528 skb_tailroom(skb1) < tailbits)
2529 ntail = tailbits + 128;
2535 skb_shinfo(skb1)->nr_frags ||
2536 skb_shinfo(skb1)->frag_list) {
2537 struct sk_buff *skb2;
2539 /* Fuck, we are miserable poor guys... */
2541 skb2 = skb_copy(skb1, GFP_ATOMIC);
2543 skb2 = skb_copy_expand(skb1,
2547 if (unlikely(skb2 == NULL))
2551 skb_set_owner_w(skb2, skb1->sk);
2553 /* Looking around. Are we still alive?
2554 * OK, link new skb, drop old one */
2556 skb2->next = skb1->next;
2563 skb_p = &skb1->next;
2570 * skb_partial_csum_set - set up and verify partial csum values for packet
2571 * @skb: the skb to set
2572 * @start: the number of bytes after skb->data to start checksumming.
2573 * @off: the offset from start to place the checksum.
2575 * For untrusted partially-checksummed packets, we need to make sure the values
2576 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2578 * This function checks and sets those values and skb->ip_summed: if this
2579 * returns false you should drop the packet.
2581 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2583 if (unlikely(start > skb->len - 2) ||
2584 unlikely((int)start + off > skb->len - 2)) {
2585 if (net_ratelimit())
2587 "bad partial csum: csum=%u/%u len=%u\n",
2588 start, off, skb->len);
2591 skb->ip_summed = CHECKSUM_PARTIAL;
2592 skb->csum_start = skb_headroom(skb) + start;
2593 skb->csum_offset = off;
2597 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
2599 if (net_ratelimit())
2600 pr_warning("%s: received packets cannot be forwarded"
2601 " while LRO is enabled\n", skb->dev->name);
2604 EXPORT_SYMBOL(___pskb_trim);
2605 EXPORT_SYMBOL(__kfree_skb);
2606 EXPORT_SYMBOL(kfree_skb);
2607 EXPORT_SYMBOL(__pskb_pull_tail);
2608 EXPORT_SYMBOL(__alloc_skb);
2609 EXPORT_SYMBOL(__netdev_alloc_skb);
2610 EXPORT_SYMBOL(pskb_copy);
2611 EXPORT_SYMBOL(pskb_expand_head);
2612 EXPORT_SYMBOL(skb_checksum);
2613 EXPORT_SYMBOL(skb_clone);
2614 EXPORT_SYMBOL(skb_copy);
2615 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2616 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2617 EXPORT_SYMBOL(skb_copy_bits);
2618 EXPORT_SYMBOL(skb_copy_expand);
2619 EXPORT_SYMBOL(skb_over_panic);
2620 EXPORT_SYMBOL(skb_pad);
2621 EXPORT_SYMBOL(skb_realloc_headroom);
2622 EXPORT_SYMBOL(skb_under_panic);
2623 EXPORT_SYMBOL(skb_dequeue);
2624 EXPORT_SYMBOL(skb_dequeue_tail);
2625 EXPORT_SYMBOL(skb_insert);
2626 EXPORT_SYMBOL(skb_queue_purge);
2627 EXPORT_SYMBOL(skb_queue_head);
2628 EXPORT_SYMBOL(skb_queue_tail);
2629 EXPORT_SYMBOL(skb_unlink);
2630 EXPORT_SYMBOL(skb_append);
2631 EXPORT_SYMBOL(skb_split);
2632 EXPORT_SYMBOL(skb_prepare_seq_read);
2633 EXPORT_SYMBOL(skb_seq_read);
2634 EXPORT_SYMBOL(skb_abort_seq_read);
2635 EXPORT_SYMBOL(skb_find_text);
2636 EXPORT_SYMBOL(skb_append_datato_frags);
2637 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
2639 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2640 EXPORT_SYMBOL_GPL(skb_cow_data);
2641 EXPORT_SYMBOL_GPL(skb_partial_csum_set);