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>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/netdevice.h>
50 #ifdef CONFIG_NET_CLS_ACT
51 #include <net/pkt_sched.h>
53 #include <linux/string.h>
54 #include <linux/skbuff.h>
55 #include <linux/splice.h>
56 #include <linux/cache.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/init.h>
59 #include <linux/scatterlist.h>
61 #include <net/protocol.h>
64 #include <net/checksum.h>
67 #include <asm/uaccess.h>
68 #include <asm/system.h>
72 static struct kmem_cache *skbuff_head_cache __read_mostly;
73 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
75 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
76 struct pipe_buffer *buf)
78 struct sk_buff *skb = (struct sk_buff *) buf->private;
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
86 struct sk_buff *skb = (struct sk_buff *) buf->private;
91 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
92 struct pipe_buffer *buf)
98 /* Pipe buffer operations for a socket. */
99 static struct pipe_buf_operations sock_pipe_buf_ops = {
101 .map = generic_pipe_buf_map,
102 .unmap = generic_pipe_buf_unmap,
103 .confirm = generic_pipe_buf_confirm,
104 .release = sock_pipe_buf_release,
105 .steal = sock_pipe_buf_steal,
106 .get = sock_pipe_buf_get,
110 * Keep out-of-line to prevent kernel bloat.
111 * __builtin_return_address is not used because it is not always
116 * skb_over_panic - private function
121 * Out of line support code for skb_put(). Not user callable.
123 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
125 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
126 "data:%p tail:%#lx end:%#lx dev:%s\n",
127 here, skb->len, sz, skb->head, skb->data,
128 (unsigned long)skb->tail, (unsigned long)skb->end,
129 skb->dev ? skb->dev->name : "<NULL>");
134 * skb_under_panic - private function
139 * Out of line support code for skb_push(). Not user callable.
142 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
144 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
145 "data:%p tail:%#lx end:%#lx dev:%s\n",
146 here, skb->len, sz, skb->head, skb->data,
147 (unsigned long)skb->tail, (unsigned long)skb->end,
148 skb->dev ? skb->dev->name : "<NULL>");
152 void skb_truesize_bug(struct sk_buff *skb)
154 printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
155 "len=%u, sizeof(sk_buff)=%Zd\n",
156 skb->truesize, skb->len, sizeof(struct sk_buff));
158 EXPORT_SYMBOL(skb_truesize_bug);
160 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
161 * 'private' fields and also do memory statistics to find all the
167 * __alloc_skb - allocate a network buffer
168 * @size: size to allocate
169 * @gfp_mask: allocation mask
170 * @fclone: allocate from fclone cache instead of head cache
171 * and allocate a cloned (child) skb
172 * @node: numa node to allocate memory on
174 * Allocate a new &sk_buff. The returned buffer has no headroom and a
175 * tail room of size bytes. The object has a reference count of one.
176 * The return is the buffer. On a failure the return is %NULL.
178 * Buffers may only be allocated from interrupts using a @gfp_mask of
181 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
182 int fclone, int node)
184 struct kmem_cache *cache;
185 struct skb_shared_info *shinfo;
189 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
192 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
196 size = SKB_DATA_ALIGN(size);
197 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
203 * See comment in sk_buff definition, just before the 'tail' member
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)
280 return __dev_alloc_skb(length, GFP_ATOMIC);
282 EXPORT_SYMBOL(dev_alloc_skb);
284 static void skb_drop_list(struct sk_buff **listp)
286 struct sk_buff *list = *listp;
291 struct sk_buff *this = list;
297 static inline void skb_drop_fraglist(struct sk_buff *skb)
299 skb_drop_list(&skb_shinfo(skb)->frag_list);
302 static void skb_clone_fraglist(struct sk_buff *skb)
304 struct sk_buff *list;
306 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
310 static void skb_release_data(struct sk_buff *skb)
313 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
314 &skb_shinfo(skb)->dataref)) {
315 if (skb_shinfo(skb)->nr_frags) {
317 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
318 put_page(skb_shinfo(skb)->frags[i].page);
321 if (skb_shinfo(skb)->frag_list)
322 skb_drop_fraglist(skb);
329 * Free an skbuff by memory without cleaning the state.
331 static void kfree_skbmem(struct sk_buff *skb)
333 struct sk_buff *other;
334 atomic_t *fclone_ref;
336 switch (skb->fclone) {
337 case SKB_FCLONE_UNAVAILABLE:
338 kmem_cache_free(skbuff_head_cache, skb);
341 case SKB_FCLONE_ORIG:
342 fclone_ref = (atomic_t *) (skb + 2);
343 if (atomic_dec_and_test(fclone_ref))
344 kmem_cache_free(skbuff_fclone_cache, skb);
347 case SKB_FCLONE_CLONE:
348 fclone_ref = (atomic_t *) (skb + 1);
351 /* The clone portion is available for
352 * fast-cloning again.
354 skb->fclone = SKB_FCLONE_UNAVAILABLE;
356 if (atomic_dec_and_test(fclone_ref))
357 kmem_cache_free(skbuff_fclone_cache, other);
362 /* Free everything but the sk_buff shell. */
363 static void skb_release_all(struct sk_buff *skb)
365 dst_release(skb->dst);
367 secpath_put(skb->sp);
369 if (skb->destructor) {
371 skb->destructor(skb);
373 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
374 nf_conntrack_put(skb->nfct);
375 nf_conntrack_put_reasm(skb->nfct_reasm);
377 #ifdef CONFIG_BRIDGE_NETFILTER
378 nf_bridge_put(skb->nf_bridge);
380 /* XXX: IS this still necessary? - JHS */
381 #ifdef CONFIG_NET_SCHED
383 #ifdef CONFIG_NET_CLS_ACT
387 skb_release_data(skb);
391 * __kfree_skb - private function
394 * Free an sk_buff. Release anything attached to the buffer.
395 * Clean the state. This is an internal helper function. Users should
396 * always call kfree_skb
399 void __kfree_skb(struct sk_buff *skb)
401 skb_release_all(skb);
406 * kfree_skb - free an sk_buff
407 * @skb: buffer to free
409 * Drop a reference to the buffer and free it if the usage count has
412 void kfree_skb(struct sk_buff *skb)
416 if (likely(atomic_read(&skb->users) == 1))
418 else if (likely(!atomic_dec_and_test(&skb->users)))
423 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
425 new->tstamp = old->tstamp;
427 new->transport_header = old->transport_header;
428 new->network_header = old->network_header;
429 new->mac_header = old->mac_header;
430 new->dst = dst_clone(old->dst);
432 new->sp = secpath_get(old->sp);
434 memcpy(new->cb, old->cb, sizeof(old->cb));
435 new->csum_start = old->csum_start;
436 new->csum_offset = old->csum_offset;
437 new->local_df = old->local_df;
438 new->pkt_type = old->pkt_type;
439 new->ip_summed = old->ip_summed;
440 skb_copy_queue_mapping(new, old);
441 new->priority = old->priority;
442 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
443 new->ipvs_property = old->ipvs_property;
445 new->protocol = old->protocol;
446 new->mark = old->mark;
448 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
449 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
450 new->nf_trace = old->nf_trace;
452 #ifdef CONFIG_NET_SCHED
453 new->tc_index = old->tc_index;
454 #ifdef CONFIG_NET_CLS_ACT
455 new->tc_verd = old->tc_verd;
458 skb_copy_secmark(new, old);
461 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
463 #define C(x) n->x = skb->x
465 n->next = n->prev = NULL;
467 __copy_skb_header(n, skb);
472 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
475 n->destructor = NULL;
482 atomic_set(&n->users, 1);
484 atomic_inc(&(skb_shinfo(skb)->dataref));
492 * skb_morph - morph one skb into another
493 * @dst: the skb to receive the contents
494 * @src: the skb to supply the contents
496 * This is identical to skb_clone except that the target skb is
497 * supplied by the user.
499 * The target skb is returned upon exit.
501 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
503 skb_release_all(dst);
504 return __skb_clone(dst, src);
506 EXPORT_SYMBOL_GPL(skb_morph);
509 * skb_clone - duplicate an sk_buff
510 * @skb: buffer to clone
511 * @gfp_mask: allocation priority
513 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
514 * copies share the same packet data but not structure. The new
515 * buffer has a reference count of 1. If the allocation fails the
516 * function returns %NULL otherwise the new buffer is returned.
518 * If this function is called from an interrupt gfp_mask() must be
522 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
527 if (skb->fclone == SKB_FCLONE_ORIG &&
528 n->fclone == SKB_FCLONE_UNAVAILABLE) {
529 atomic_t *fclone_ref = (atomic_t *) (n + 1);
530 n->fclone = SKB_FCLONE_CLONE;
531 atomic_inc(fclone_ref);
533 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
536 n->fclone = SKB_FCLONE_UNAVAILABLE;
539 return __skb_clone(n, skb);
542 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
544 #ifndef NET_SKBUFF_DATA_USES_OFFSET
546 * Shift between the two data areas in bytes
548 unsigned long offset = new->data - old->data;
551 __copy_skb_header(new, old);
553 #ifndef NET_SKBUFF_DATA_USES_OFFSET
554 /* {transport,network,mac}_header are relative to skb->head */
555 new->transport_header += offset;
556 new->network_header += offset;
557 new->mac_header += offset;
559 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
560 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
561 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
565 * skb_copy - create private copy of an sk_buff
566 * @skb: buffer to copy
567 * @gfp_mask: allocation priority
569 * Make a copy of both an &sk_buff and its data. This is used when the
570 * caller wishes to modify the data and needs a private copy of the
571 * data to alter. Returns %NULL on failure or the pointer to the buffer
572 * on success. The returned buffer has a reference count of 1.
574 * As by-product this function converts non-linear &sk_buff to linear
575 * one, so that &sk_buff becomes completely private and caller is allowed
576 * to modify all the data of returned buffer. This means that this
577 * function is not recommended for use in circumstances when only
578 * header is going to be modified. Use pskb_copy() instead.
581 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
583 int headerlen = skb->data - skb->head;
585 * Allocate the copy buffer
588 #ifdef NET_SKBUFF_DATA_USES_OFFSET
589 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
591 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
596 /* Set the data pointer */
597 skb_reserve(n, headerlen);
598 /* Set the tail pointer and length */
599 skb_put(n, skb->len);
601 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
604 copy_skb_header(n, skb);
610 * pskb_copy - create copy of an sk_buff with private head.
611 * @skb: buffer to copy
612 * @gfp_mask: allocation priority
614 * Make a copy of both an &sk_buff and part of its data, located
615 * in header. Fragmented data remain shared. This is used when
616 * the caller wishes to modify only header of &sk_buff and needs
617 * private copy of the header to alter. Returns %NULL on failure
618 * or the pointer to the buffer on success.
619 * The returned buffer has a reference count of 1.
622 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
625 * Allocate the copy buffer
628 #ifdef NET_SKBUFF_DATA_USES_OFFSET
629 n = alloc_skb(skb->end, gfp_mask);
631 n = alloc_skb(skb->end - skb->head, gfp_mask);
636 /* Set the data pointer */
637 skb_reserve(n, skb->data - skb->head);
638 /* Set the tail pointer and length */
639 skb_put(n, skb_headlen(skb));
641 skb_copy_from_linear_data(skb, n->data, n->len);
643 n->truesize += skb->data_len;
644 n->data_len = skb->data_len;
647 if (skb_shinfo(skb)->nr_frags) {
650 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
651 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
652 get_page(skb_shinfo(n)->frags[i].page);
654 skb_shinfo(n)->nr_frags = i;
657 if (skb_shinfo(skb)->frag_list) {
658 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
659 skb_clone_fraglist(n);
662 copy_skb_header(n, skb);
668 * pskb_expand_head - reallocate header of &sk_buff
669 * @skb: buffer to reallocate
670 * @nhead: room to add at head
671 * @ntail: room to add at tail
672 * @gfp_mask: allocation priority
674 * Expands (or creates identical copy, if &nhead and &ntail are zero)
675 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
676 * reference count of 1. Returns zero in the case of success or error,
677 * if expansion failed. In the last case, &sk_buff is not changed.
679 * All the pointers pointing into skb header may change and must be
680 * reloaded after call to this function.
683 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
688 #ifdef NET_SKBUFF_DATA_USES_OFFSET
689 int size = nhead + skb->end + ntail;
691 int size = nhead + (skb->end - skb->head) + ntail;
698 size = SKB_DATA_ALIGN(size);
700 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
704 /* Copy only real data... and, alas, header. This should be
705 * optimized for the cases when header is void. */
706 #ifdef NET_SKBUFF_DATA_USES_OFFSET
707 memcpy(data + nhead, skb->head, skb->tail);
709 memcpy(data + nhead, skb->head, skb->tail - skb->head);
711 memcpy(data + size, skb_end_pointer(skb),
712 sizeof(struct skb_shared_info));
714 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
715 get_page(skb_shinfo(skb)->frags[i].page);
717 if (skb_shinfo(skb)->frag_list)
718 skb_clone_fraglist(skb);
720 skb_release_data(skb);
722 off = (data + nhead) - skb->head;
726 #ifdef NET_SKBUFF_DATA_USES_OFFSET
730 skb->end = skb->head + size;
732 /* {transport,network,mac}_header and tail are relative to skb->head */
734 skb->transport_header += off;
735 skb->network_header += off;
736 skb->mac_header += off;
737 skb->csum_start += nhead;
741 atomic_set(&skb_shinfo(skb)->dataref, 1);
748 /* Make private copy of skb with writable head and some headroom */
750 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
752 struct sk_buff *skb2;
753 int delta = headroom - skb_headroom(skb);
756 skb2 = pskb_copy(skb, GFP_ATOMIC);
758 skb2 = skb_clone(skb, GFP_ATOMIC);
759 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
770 * skb_copy_expand - copy and expand sk_buff
771 * @skb: buffer to copy
772 * @newheadroom: new free bytes at head
773 * @newtailroom: new free bytes at tail
774 * @gfp_mask: allocation priority
776 * Make a copy of both an &sk_buff and its data and while doing so
777 * allocate additional space.
779 * This is used when the caller wishes to modify the data and needs a
780 * private copy of the data to alter as well as more space for new fields.
781 * Returns %NULL on failure or the pointer to the buffer
782 * on success. The returned buffer has a reference count of 1.
784 * You must pass %GFP_ATOMIC as the allocation priority if this function
785 * is called from an interrupt.
787 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
788 int newheadroom, int newtailroom,
792 * Allocate the copy buffer
794 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
796 int oldheadroom = skb_headroom(skb);
797 int head_copy_len, head_copy_off;
803 skb_reserve(n, newheadroom);
805 /* Set the tail pointer and length */
806 skb_put(n, skb->len);
808 head_copy_len = oldheadroom;
810 if (newheadroom <= head_copy_len)
811 head_copy_len = newheadroom;
813 head_copy_off = newheadroom - head_copy_len;
815 /* Copy the linear header and data. */
816 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
817 skb->len + head_copy_len))
820 copy_skb_header(n, skb);
822 off = newheadroom - oldheadroom;
823 n->csum_start += off;
824 #ifdef NET_SKBUFF_DATA_USES_OFFSET
825 n->transport_header += off;
826 n->network_header += off;
827 n->mac_header += off;
834 * skb_pad - zero pad the tail of an skb
835 * @skb: buffer to pad
838 * Ensure that a buffer is followed by a padding area that is zero
839 * filled. Used by network drivers which may DMA or transfer data
840 * beyond the buffer end onto the wire.
842 * May return error in out of memory cases. The skb is freed on error.
845 int skb_pad(struct sk_buff *skb, int pad)
850 /* If the skbuff is non linear tailroom is always zero.. */
851 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
852 memset(skb->data+skb->len, 0, pad);
856 ntail = skb->data_len + pad - (skb->end - skb->tail);
857 if (likely(skb_cloned(skb) || ntail > 0)) {
858 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
863 /* FIXME: The use of this function with non-linear skb's really needs
866 err = skb_linearize(skb);
870 memset(skb->data + skb->len, 0, pad);
879 * skb_put - add data to a buffer
880 * @skb: buffer to use
881 * @len: amount of data to add
883 * This function extends the used data area of the buffer. If this would
884 * exceed the total buffer size the kernel will panic. A pointer to the
885 * first byte of the extra data is returned.
887 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
889 unsigned char *tmp = skb_tail_pointer(skb);
890 SKB_LINEAR_ASSERT(skb);
893 if (unlikely(skb->tail > skb->end))
894 skb_over_panic(skb, len, __builtin_return_address(0));
897 EXPORT_SYMBOL(skb_put);
900 * skb_pull - remove data from the start of a buffer
901 * @skb: buffer to use
902 * @len: amount of data to remove
904 * This function removes data from the start of a buffer, returning
905 * the memory to the headroom. A pointer to the next data in the buffer
906 * is returned. Once the data has been pulled future pushes will overwrite
909 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
911 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
913 EXPORT_SYMBOL(skb_pull);
915 /* Trims skb to length len. It can change skb pointers.
918 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
920 struct sk_buff **fragp;
921 struct sk_buff *frag;
922 int offset = skb_headlen(skb);
923 int nfrags = skb_shinfo(skb)->nr_frags;
927 if (skb_cloned(skb) &&
928 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
935 for (; i < nfrags; i++) {
936 int end = offset + skb_shinfo(skb)->frags[i].size;
943 skb_shinfo(skb)->frags[i++].size = len - offset;
946 skb_shinfo(skb)->nr_frags = i;
948 for (; i < nfrags; i++)
949 put_page(skb_shinfo(skb)->frags[i].page);
951 if (skb_shinfo(skb)->frag_list)
952 skb_drop_fraglist(skb);
956 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
957 fragp = &frag->next) {
958 int end = offset + frag->len;
960 if (skb_shared(frag)) {
961 struct sk_buff *nfrag;
963 nfrag = skb_clone(frag, GFP_ATOMIC);
964 if (unlikely(!nfrag))
967 nfrag->next = frag->next;
979 unlikely((err = pskb_trim(frag, len - offset))))
983 skb_drop_list(&frag->next);
988 if (len > skb_headlen(skb)) {
989 skb->data_len -= skb->len - len;
994 skb_set_tail_pointer(skb, len);
1001 * __pskb_pull_tail - advance tail of skb header
1002 * @skb: buffer to reallocate
1003 * @delta: number of bytes to advance tail
1005 * The function makes a sense only on a fragmented &sk_buff,
1006 * it expands header moving its tail forward and copying necessary
1007 * data from fragmented part.
1009 * &sk_buff MUST have reference count of 1.
1011 * Returns %NULL (and &sk_buff does not change) if pull failed
1012 * or value of new tail of skb in the case of success.
1014 * All the pointers pointing into skb header may change and must be
1015 * reloaded after call to this function.
1018 /* Moves tail of skb head forward, copying data from fragmented part,
1019 * when it is necessary.
1020 * 1. It may fail due to malloc failure.
1021 * 2. It may change skb pointers.
1023 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1025 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1027 /* If skb has not enough free space at tail, get new one
1028 * plus 128 bytes for future expansions. If we have enough
1029 * room at tail, reallocate without expansion only if skb is cloned.
1031 int i, k, eat = (skb->tail + delta) - skb->end;
1033 if (eat > 0 || skb_cloned(skb)) {
1034 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1039 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1042 /* Optimization: no fragments, no reasons to preestimate
1043 * size of pulled pages. Superb.
1045 if (!skb_shinfo(skb)->frag_list)
1048 /* Estimate size of pulled pages. */
1050 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1051 if (skb_shinfo(skb)->frags[i].size >= eat)
1053 eat -= skb_shinfo(skb)->frags[i].size;
1056 /* If we need update frag list, we are in troubles.
1057 * Certainly, it possible to add an offset to skb data,
1058 * but taking into account that pulling is expected to
1059 * be very rare operation, it is worth to fight against
1060 * further bloating skb head and crucify ourselves here instead.
1061 * Pure masohism, indeed. 8)8)
1064 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1065 struct sk_buff *clone = NULL;
1066 struct sk_buff *insp = NULL;
1071 if (list->len <= eat) {
1072 /* Eaten as whole. */
1077 /* Eaten partially. */
1079 if (skb_shared(list)) {
1080 /* Sucks! We need to fork list. :-( */
1081 clone = skb_clone(list, GFP_ATOMIC);
1087 /* This may be pulled without
1091 if (!pskb_pull(list, eat)) {
1100 /* Free pulled out fragments. */
1101 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1102 skb_shinfo(skb)->frag_list = list->next;
1105 /* And insert new clone at head. */
1108 skb_shinfo(skb)->frag_list = clone;
1111 /* Success! Now we may commit changes to skb data. */
1116 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1117 if (skb_shinfo(skb)->frags[i].size <= eat) {
1118 put_page(skb_shinfo(skb)->frags[i].page);
1119 eat -= skb_shinfo(skb)->frags[i].size;
1121 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1123 skb_shinfo(skb)->frags[k].page_offset += eat;
1124 skb_shinfo(skb)->frags[k].size -= eat;
1130 skb_shinfo(skb)->nr_frags = k;
1133 skb->data_len -= delta;
1135 return skb_tail_pointer(skb);
1138 /* Copy some data bits from skb to kernel buffer. */
1140 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1143 int start = skb_headlen(skb);
1145 if (offset > (int)skb->len - len)
1149 if ((copy = start - offset) > 0) {
1152 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1153 if ((len -= copy) == 0)
1159 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1162 BUG_TRAP(start <= offset + len);
1164 end = start + skb_shinfo(skb)->frags[i].size;
1165 if ((copy = end - offset) > 0) {
1171 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1173 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1174 offset - start, copy);
1175 kunmap_skb_frag(vaddr);
1177 if ((len -= copy) == 0)
1185 if (skb_shinfo(skb)->frag_list) {
1186 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1188 for (; list; list = list->next) {
1191 BUG_TRAP(start <= offset + len);
1193 end = start + list->len;
1194 if ((copy = end - offset) > 0) {
1197 if (skb_copy_bits(list, offset - start,
1200 if ((len -= copy) == 0)
1216 * Callback from splice_to_pipe(), if we need to release some pages
1217 * at the end of the spd in case we error'ed out in filling the pipe.
1219 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1221 struct sk_buff *skb = (struct sk_buff *) spd->partial[i].private;
1227 * Fill page/offset/length into spd, if it can hold more pages.
1229 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1230 unsigned int len, unsigned int offset,
1231 struct sk_buff *skb)
1233 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1236 spd->pages[spd->nr_pages] = page;
1237 spd->partial[spd->nr_pages].len = len;
1238 spd->partial[spd->nr_pages].offset = offset;
1239 spd->partial[spd->nr_pages].private = (unsigned long) skb_get(skb);
1245 * Map linear and fragment data from the skb to spd. Returns number of
1248 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1249 unsigned int *total_len,
1250 struct splice_pipe_desc *spd)
1252 unsigned int nr_pages = spd->nr_pages;
1253 unsigned int poff, plen, len, toff, tlen;
1262 * if the offset is greater than the linear part, go directly to
1265 headlen = skb_headlen(skb);
1266 if (toff >= headlen) {
1272 * first map the linear region into the pages/partial map, skipping
1273 * any potential initial offset.
1276 while (len < headlen) {
1277 void *p = skb->data + len;
1279 poff = (unsigned long) p & (PAGE_SIZE - 1);
1280 plen = min_t(unsigned int, headlen - len, PAGE_SIZE - poff);
1293 plen = min(plen, tlen);
1298 * just jump directly to update and return, no point
1299 * in going over fragments when the output is full.
1301 if (spd_fill_page(spd, virt_to_page(p), plen, poff, skb))
1308 * then map the fragments
1311 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1312 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1315 poff = f->page_offset;
1327 plen = min(plen, tlen);
1331 if (spd_fill_page(spd, f->page, plen, poff, skb))
1338 if (spd->nr_pages - nr_pages) {
1348 * Map data from the skb to a pipe. Should handle both the linear part,
1349 * the fragments, and the frag list. It does NOT handle frag lists within
1350 * the frag list, if such a thing exists. We'd probably need to recurse to
1351 * handle that cleanly.
1353 int skb_splice_bits(struct sk_buff *__skb, unsigned int offset,
1354 struct pipe_inode_info *pipe, unsigned int tlen,
1357 struct partial_page partial[PIPE_BUFFERS];
1358 struct page *pages[PIPE_BUFFERS];
1359 struct splice_pipe_desc spd = {
1363 .ops = &sock_pipe_buf_ops,
1364 .spd_release = sock_spd_release,
1366 struct sk_buff *skb;
1369 * I'd love to avoid the clone here, but tcp_read_sock()
1370 * ignores reference counts and unconditonally kills the sk_buff
1371 * on return from the actor.
1373 skb = skb_clone(__skb, GFP_KERNEL);
1378 * __skb_splice_bits() only fails if the output has no room left,
1379 * so no point in going over the frag_list for the error case.
1381 if (__skb_splice_bits(skb, &offset, &tlen, &spd))
1387 * now see if we have a frag_list to map
1389 if (skb_shinfo(skb)->frag_list) {
1390 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1392 for (; list && tlen; list = list->next) {
1393 if (__skb_splice_bits(list, &offset, &tlen, &spd))
1400 * drop our reference to the clone, the pipe consumption will
1409 * Drop the socket lock, otherwise we have reverse
1410 * locking dependencies between sk_lock and i_mutex
1411 * here as compared to sendfile(). We enter here
1412 * with the socket lock held, and splice_to_pipe() will
1413 * grab the pipe inode lock. For sendfile() emulation,
1414 * we call into ->sendpage() with the i_mutex lock held
1415 * and networking will grab the socket lock.
1417 release_sock(__skb->sk);
1418 ret = splice_to_pipe(pipe, &spd);
1419 lock_sock(__skb->sk);
1427 * skb_store_bits - store bits from kernel buffer to skb
1428 * @skb: destination buffer
1429 * @offset: offset in destination
1430 * @from: source buffer
1431 * @len: number of bytes to copy
1433 * Copy the specified number of bytes from the source buffer to the
1434 * destination skb. This function handles all the messy bits of
1435 * traversing fragment lists and such.
1438 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1441 int start = skb_headlen(skb);
1443 if (offset > (int)skb->len - len)
1446 if ((copy = start - offset) > 0) {
1449 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1450 if ((len -= copy) == 0)
1456 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1457 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1460 BUG_TRAP(start <= offset + len);
1462 end = start + frag->size;
1463 if ((copy = end - offset) > 0) {
1469 vaddr = kmap_skb_frag(frag);
1470 memcpy(vaddr + frag->page_offset + offset - start,
1472 kunmap_skb_frag(vaddr);
1474 if ((len -= copy) == 0)
1482 if (skb_shinfo(skb)->frag_list) {
1483 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1485 for (; list; list = list->next) {
1488 BUG_TRAP(start <= offset + len);
1490 end = start + list->len;
1491 if ((copy = end - offset) > 0) {
1494 if (skb_store_bits(list, offset - start,
1497 if ((len -= copy) == 0)
1512 EXPORT_SYMBOL(skb_store_bits);
1514 /* Checksum skb data. */
1516 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1517 int len, __wsum csum)
1519 int start = skb_headlen(skb);
1520 int i, copy = start - offset;
1523 /* Checksum header. */
1527 csum = csum_partial(skb->data + offset, copy, csum);
1528 if ((len -= copy) == 0)
1534 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1537 BUG_TRAP(start <= offset + len);
1539 end = start + skb_shinfo(skb)->frags[i].size;
1540 if ((copy = end - offset) > 0) {
1543 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1547 vaddr = kmap_skb_frag(frag);
1548 csum2 = csum_partial(vaddr + frag->page_offset +
1549 offset - start, copy, 0);
1550 kunmap_skb_frag(vaddr);
1551 csum = csum_block_add(csum, csum2, pos);
1560 if (skb_shinfo(skb)->frag_list) {
1561 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1563 for (; list; list = list->next) {
1566 BUG_TRAP(start <= offset + len);
1568 end = start + list->len;
1569 if ((copy = end - offset) > 0) {
1573 csum2 = skb_checksum(list, offset - start,
1575 csum = csum_block_add(csum, csum2, pos);
1576 if ((len -= copy) == 0)
1589 /* Both of above in one bottle. */
1591 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1592 u8 *to, int len, __wsum csum)
1594 int start = skb_headlen(skb);
1595 int i, copy = start - offset;
1602 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1604 if ((len -= copy) == 0)
1611 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1614 BUG_TRAP(start <= offset + len);
1616 end = start + skb_shinfo(skb)->frags[i].size;
1617 if ((copy = end - offset) > 0) {
1620 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1624 vaddr = kmap_skb_frag(frag);
1625 csum2 = csum_partial_copy_nocheck(vaddr +
1629 kunmap_skb_frag(vaddr);
1630 csum = csum_block_add(csum, csum2, pos);
1640 if (skb_shinfo(skb)->frag_list) {
1641 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1643 for (; list; list = list->next) {
1647 BUG_TRAP(start <= offset + len);
1649 end = start + list->len;
1650 if ((copy = end - offset) > 0) {
1653 csum2 = skb_copy_and_csum_bits(list,
1656 csum = csum_block_add(csum, csum2, pos);
1657 if ((len -= copy) == 0)
1670 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1675 if (skb->ip_summed == CHECKSUM_PARTIAL)
1676 csstart = skb->csum_start - skb_headroom(skb);
1678 csstart = skb_headlen(skb);
1680 BUG_ON(csstart > skb_headlen(skb));
1682 skb_copy_from_linear_data(skb, to, csstart);
1685 if (csstart != skb->len)
1686 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1687 skb->len - csstart, 0);
1689 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1690 long csstuff = csstart + skb->csum_offset;
1692 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1697 * skb_dequeue - remove from the head of the queue
1698 * @list: list to dequeue from
1700 * Remove the head of the list. The list lock is taken so the function
1701 * may be used safely with other locking list functions. The head item is
1702 * returned or %NULL if the list is empty.
1705 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1707 unsigned long flags;
1708 struct sk_buff *result;
1710 spin_lock_irqsave(&list->lock, flags);
1711 result = __skb_dequeue(list);
1712 spin_unlock_irqrestore(&list->lock, flags);
1717 * skb_dequeue_tail - remove from the tail of the queue
1718 * @list: list to dequeue from
1720 * Remove the tail of the list. The list lock is taken so the function
1721 * may be used safely with other locking list functions. The tail item is
1722 * returned or %NULL if the list is empty.
1724 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1726 unsigned long flags;
1727 struct sk_buff *result;
1729 spin_lock_irqsave(&list->lock, flags);
1730 result = __skb_dequeue_tail(list);
1731 spin_unlock_irqrestore(&list->lock, flags);
1736 * skb_queue_purge - empty a list
1737 * @list: list to empty
1739 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1740 * the list and one reference dropped. This function takes the list
1741 * lock and is atomic with respect to other list locking functions.
1743 void skb_queue_purge(struct sk_buff_head *list)
1745 struct sk_buff *skb;
1746 while ((skb = skb_dequeue(list)) != NULL)
1751 * skb_queue_head - queue a buffer at the list head
1752 * @list: list to use
1753 * @newsk: buffer to queue
1755 * Queue a buffer at the start of the list. This function takes the
1756 * list lock and can be used safely with other locking &sk_buff functions
1759 * A buffer cannot be placed on two lists at the same time.
1761 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1763 unsigned long flags;
1765 spin_lock_irqsave(&list->lock, flags);
1766 __skb_queue_head(list, newsk);
1767 spin_unlock_irqrestore(&list->lock, flags);
1771 * skb_queue_tail - queue a buffer at the list tail
1772 * @list: list to use
1773 * @newsk: buffer to queue
1775 * Queue a buffer at the tail of the list. This function takes the
1776 * list lock and can be used safely with other locking &sk_buff functions
1779 * A buffer cannot be placed on two lists at the same time.
1781 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1783 unsigned long flags;
1785 spin_lock_irqsave(&list->lock, flags);
1786 __skb_queue_tail(list, newsk);
1787 spin_unlock_irqrestore(&list->lock, flags);
1791 * skb_unlink - remove a buffer from a list
1792 * @skb: buffer to remove
1793 * @list: list to use
1795 * Remove a packet from a list. The list locks are taken and this
1796 * function is atomic with respect to other list locked calls
1798 * You must know what list the SKB is on.
1800 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1802 unsigned long flags;
1804 spin_lock_irqsave(&list->lock, flags);
1805 __skb_unlink(skb, list);
1806 spin_unlock_irqrestore(&list->lock, flags);
1810 * skb_append - append a buffer
1811 * @old: buffer to insert after
1812 * @newsk: buffer to insert
1813 * @list: list to use
1815 * Place a packet after a given packet in a list. The list locks are taken
1816 * and this function is atomic with respect to other list locked calls.
1817 * A buffer cannot be placed on two lists at the same time.
1819 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1821 unsigned long flags;
1823 spin_lock_irqsave(&list->lock, flags);
1824 __skb_append(old, newsk, list);
1825 spin_unlock_irqrestore(&list->lock, flags);
1830 * skb_insert - insert a buffer
1831 * @old: buffer to insert before
1832 * @newsk: buffer to insert
1833 * @list: list to use
1835 * Place a packet before a given packet in a list. The list locks are
1836 * taken and this function is atomic with respect to other list locked
1839 * A buffer cannot be placed on two lists at the same time.
1841 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1843 unsigned long flags;
1845 spin_lock_irqsave(&list->lock, flags);
1846 __skb_insert(newsk, old->prev, old, list);
1847 spin_unlock_irqrestore(&list->lock, flags);
1850 static inline void skb_split_inside_header(struct sk_buff *skb,
1851 struct sk_buff* skb1,
1852 const u32 len, const int pos)
1856 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1858 /* And move data appendix as is. */
1859 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1860 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1862 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1863 skb_shinfo(skb)->nr_frags = 0;
1864 skb1->data_len = skb->data_len;
1865 skb1->len += skb1->data_len;
1868 skb_set_tail_pointer(skb, len);
1871 static inline void skb_split_no_header(struct sk_buff *skb,
1872 struct sk_buff* skb1,
1873 const u32 len, int pos)
1876 const int nfrags = skb_shinfo(skb)->nr_frags;
1878 skb_shinfo(skb)->nr_frags = 0;
1879 skb1->len = skb1->data_len = skb->len - len;
1881 skb->data_len = len - pos;
1883 for (i = 0; i < nfrags; i++) {
1884 int size = skb_shinfo(skb)->frags[i].size;
1886 if (pos + size > len) {
1887 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1891 * We have two variants in this case:
1892 * 1. Move all the frag to the second
1893 * part, if it is possible. F.e.
1894 * this approach is mandatory for TUX,
1895 * where splitting is expensive.
1896 * 2. Split is accurately. We make this.
1898 get_page(skb_shinfo(skb)->frags[i].page);
1899 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1900 skb_shinfo(skb1)->frags[0].size -= len - pos;
1901 skb_shinfo(skb)->frags[i].size = len - pos;
1902 skb_shinfo(skb)->nr_frags++;
1906 skb_shinfo(skb)->nr_frags++;
1909 skb_shinfo(skb1)->nr_frags = k;
1913 * skb_split - Split fragmented skb to two parts at length len.
1914 * @skb: the buffer to split
1915 * @skb1: the buffer to receive the second part
1916 * @len: new length for skb
1918 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1920 int pos = skb_headlen(skb);
1922 if (len < pos) /* Split line is inside header. */
1923 skb_split_inside_header(skb, skb1, len, pos);
1924 else /* Second chunk has no header, nothing to copy. */
1925 skb_split_no_header(skb, skb1, len, pos);
1929 * skb_prepare_seq_read - Prepare a sequential read of skb data
1930 * @skb: the buffer to read
1931 * @from: lower offset of data to be read
1932 * @to: upper offset of data to be read
1933 * @st: state variable
1935 * Initializes the specified state variable. Must be called before
1936 * invoking skb_seq_read() for the first time.
1938 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1939 unsigned int to, struct skb_seq_state *st)
1941 st->lower_offset = from;
1942 st->upper_offset = to;
1943 st->root_skb = st->cur_skb = skb;
1944 st->frag_idx = st->stepped_offset = 0;
1945 st->frag_data = NULL;
1949 * skb_seq_read - Sequentially read skb data
1950 * @consumed: number of bytes consumed by the caller so far
1951 * @data: destination pointer for data to be returned
1952 * @st: state variable
1954 * Reads a block of skb data at &consumed relative to the
1955 * lower offset specified to skb_prepare_seq_read(). Assigns
1956 * the head of the data block to &data and returns the length
1957 * of the block or 0 if the end of the skb data or the upper
1958 * offset has been reached.
1960 * The caller is not required to consume all of the data
1961 * returned, i.e. &consumed is typically set to the number
1962 * of bytes already consumed and the next call to
1963 * skb_seq_read() will return the remaining part of the block.
1965 * Note 1: The size of each block of data returned can be arbitary,
1966 * this limitation is the cost for zerocopy seqeuental
1967 * reads of potentially non linear data.
1969 * Note 2: Fragment lists within fragments are not implemented
1970 * at the moment, state->root_skb could be replaced with
1971 * a stack for this purpose.
1973 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1974 struct skb_seq_state *st)
1976 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1979 if (unlikely(abs_offset >= st->upper_offset))
1983 block_limit = skb_headlen(st->cur_skb);
1985 if (abs_offset < block_limit) {
1986 *data = st->cur_skb->data + abs_offset;
1987 return block_limit - abs_offset;
1990 if (st->frag_idx == 0 && !st->frag_data)
1991 st->stepped_offset += skb_headlen(st->cur_skb);
1993 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1994 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1995 block_limit = frag->size + st->stepped_offset;
1997 if (abs_offset < block_limit) {
1999 st->frag_data = kmap_skb_frag(frag);
2001 *data = (u8 *) st->frag_data + frag->page_offset +
2002 (abs_offset - st->stepped_offset);
2004 return block_limit - abs_offset;
2007 if (st->frag_data) {
2008 kunmap_skb_frag(st->frag_data);
2009 st->frag_data = NULL;
2013 st->stepped_offset += frag->size;
2016 if (st->frag_data) {
2017 kunmap_skb_frag(st->frag_data);
2018 st->frag_data = NULL;
2021 if (st->cur_skb->next) {
2022 st->cur_skb = st->cur_skb->next;
2025 } else if (st->root_skb == st->cur_skb &&
2026 skb_shinfo(st->root_skb)->frag_list) {
2027 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2035 * skb_abort_seq_read - Abort a sequential read of skb data
2036 * @st: state variable
2038 * Must be called if skb_seq_read() was not called until it
2041 void skb_abort_seq_read(struct skb_seq_state *st)
2044 kunmap_skb_frag(st->frag_data);
2047 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2049 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2050 struct ts_config *conf,
2051 struct ts_state *state)
2053 return skb_seq_read(offset, text, TS_SKB_CB(state));
2056 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2058 skb_abort_seq_read(TS_SKB_CB(state));
2062 * skb_find_text - Find a text pattern in skb data
2063 * @skb: the buffer to look in
2064 * @from: search offset
2066 * @config: textsearch configuration
2067 * @state: uninitialized textsearch state variable
2069 * Finds a pattern in the skb data according to the specified
2070 * textsearch configuration. Use textsearch_next() to retrieve
2071 * subsequent occurrences of the pattern. Returns the offset
2072 * to the first occurrence or UINT_MAX if no match was found.
2074 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2075 unsigned int to, struct ts_config *config,
2076 struct ts_state *state)
2080 config->get_next_block = skb_ts_get_next_block;
2081 config->finish = skb_ts_finish;
2083 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2085 ret = textsearch_find(config, state);
2086 return (ret <= to - from ? ret : UINT_MAX);
2090 * skb_append_datato_frags: - append the user data to a skb
2091 * @sk: sock structure
2092 * @skb: skb structure to be appened with user data.
2093 * @getfrag: call back function to be used for getting the user data
2094 * @from: pointer to user message iov
2095 * @length: length of the iov message
2097 * Description: This procedure append the user data in the fragment part
2098 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2100 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2101 int (*getfrag)(void *from, char *to, int offset,
2102 int len, int odd, struct sk_buff *skb),
2103 void *from, int length)
2106 skb_frag_t *frag = NULL;
2107 struct page *page = NULL;
2113 /* Return error if we don't have space for new frag */
2114 frg_cnt = skb_shinfo(skb)->nr_frags;
2115 if (frg_cnt >= MAX_SKB_FRAGS)
2118 /* allocate a new page for next frag */
2119 page = alloc_pages(sk->sk_allocation, 0);
2121 /* If alloc_page fails just return failure and caller will
2122 * free previous allocated pages by doing kfree_skb()
2127 /* initialize the next frag */
2128 sk->sk_sndmsg_page = page;
2129 sk->sk_sndmsg_off = 0;
2130 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2131 skb->truesize += PAGE_SIZE;
2132 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2134 /* get the new initialized frag */
2135 frg_cnt = skb_shinfo(skb)->nr_frags;
2136 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2138 /* copy the user data to page */
2139 left = PAGE_SIZE - frag->page_offset;
2140 copy = (length > left)? left : length;
2142 ret = getfrag(from, (page_address(frag->page) +
2143 frag->page_offset + frag->size),
2144 offset, copy, 0, skb);
2148 /* copy was successful so update the size parameters */
2149 sk->sk_sndmsg_off += copy;
2152 skb->data_len += copy;
2156 } while (length > 0);
2162 * skb_pull_rcsum - pull skb and update receive checksum
2163 * @skb: buffer to update
2164 * @len: length of data pulled
2166 * This function performs an skb_pull on the packet and updates
2167 * the CHECKSUM_COMPLETE checksum. It should be used on
2168 * receive path processing instead of skb_pull unless you know
2169 * that the checksum difference is zero (e.g., a valid IP header)
2170 * or you are setting ip_summed to CHECKSUM_NONE.
2172 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2174 BUG_ON(len > skb->len);
2176 BUG_ON(skb->len < skb->data_len);
2177 skb_postpull_rcsum(skb, skb->data, len);
2178 return skb->data += len;
2181 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2184 * skb_segment - Perform protocol segmentation on skb.
2185 * @skb: buffer to segment
2186 * @features: features for the output path (see dev->features)
2188 * This function performs segmentation on the given skb. It returns
2189 * the segment at the given position. It returns NULL if there are
2190 * no more segments to generate, or when an error is encountered.
2192 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2194 struct sk_buff *segs = NULL;
2195 struct sk_buff *tail = NULL;
2196 unsigned int mss = skb_shinfo(skb)->gso_size;
2197 unsigned int doffset = skb->data - skb_mac_header(skb);
2198 unsigned int offset = doffset;
2199 unsigned int headroom;
2201 int sg = features & NETIF_F_SG;
2202 int nfrags = skb_shinfo(skb)->nr_frags;
2207 __skb_push(skb, doffset);
2208 headroom = skb_headroom(skb);
2209 pos = skb_headlen(skb);
2212 struct sk_buff *nskb;
2218 len = skb->len - offset;
2222 hsize = skb_headlen(skb) - offset;
2225 if (hsize > len || !sg)
2228 nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
2229 if (unlikely(!nskb))
2238 nskb->dev = skb->dev;
2239 skb_copy_queue_mapping(nskb, skb);
2240 nskb->priority = skb->priority;
2241 nskb->protocol = skb->protocol;
2242 nskb->dst = dst_clone(skb->dst);
2243 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
2244 nskb->pkt_type = skb->pkt_type;
2245 nskb->mac_len = skb->mac_len;
2247 skb_reserve(nskb, headroom);
2248 skb_reset_mac_header(nskb);
2249 skb_set_network_header(nskb, skb->mac_len);
2250 nskb->transport_header = (nskb->network_header +
2251 skb_network_header_len(skb));
2252 skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
2255 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2261 frag = skb_shinfo(nskb)->frags;
2264 nskb->ip_summed = CHECKSUM_PARTIAL;
2265 nskb->csum = skb->csum;
2266 skb_copy_from_linear_data_offset(skb, offset,
2267 skb_put(nskb, hsize), hsize);
2269 while (pos < offset + len) {
2270 BUG_ON(i >= nfrags);
2272 *frag = skb_shinfo(skb)->frags[i];
2273 get_page(frag->page);
2277 frag->page_offset += offset - pos;
2278 frag->size -= offset - pos;
2283 if (pos + size <= offset + len) {
2287 frag->size -= pos + size - (offset + len);
2294 skb_shinfo(nskb)->nr_frags = k;
2295 nskb->data_len = len - hsize;
2296 nskb->len += nskb->data_len;
2297 nskb->truesize += nskb->data_len;
2298 } while ((offset += len) < skb->len);
2303 while ((skb = segs)) {
2307 return ERR_PTR(err);
2310 EXPORT_SYMBOL_GPL(skb_segment);
2312 void __init skb_init(void)
2314 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2315 sizeof(struct sk_buff),
2317 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2319 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2320 (2*sizeof(struct sk_buff)) +
2323 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2328 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2329 * @skb: Socket buffer containing the buffers to be mapped
2330 * @sg: The scatter-gather list to map into
2331 * @offset: The offset into the buffer's contents to start mapping
2332 * @len: Length of buffer space to be mapped
2334 * Fill the specified scatter-gather list with mappings/pointers into a
2335 * region of the buffer space attached to a socket buffer.
2338 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2340 int start = skb_headlen(skb);
2341 int i, copy = start - offset;
2347 sg_set_buf(sg, skb->data + offset, copy);
2349 if ((len -= copy) == 0)
2354 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2357 BUG_TRAP(start <= offset + len);
2359 end = start + skb_shinfo(skb)->frags[i].size;
2360 if ((copy = end - offset) > 0) {
2361 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2365 sg_set_page(&sg[elt], frag->page, copy,
2366 frag->page_offset+offset-start);
2375 if (skb_shinfo(skb)->frag_list) {
2376 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2378 for (; list; list = list->next) {
2381 BUG_TRAP(start <= offset + len);
2383 end = start + list->len;
2384 if ((copy = end - offset) > 0) {
2387 elt += __skb_to_sgvec(list, sg+elt, offset - start,
2389 if ((len -= copy) == 0)
2400 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2402 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2404 sg_mark_end(&sg[nsg - 1]);
2410 * skb_cow_data - Check that a socket buffer's data buffers are writable
2411 * @skb: The socket buffer to check.
2412 * @tailbits: Amount of trailing space to be added
2413 * @trailer: Returned pointer to the skb where the @tailbits space begins
2415 * Make sure that the data buffers attached to a socket buffer are
2416 * writable. If they are not, private copies are made of the data buffers
2417 * and the socket buffer is set to use these instead.
2419 * If @tailbits is given, make sure that there is space to write @tailbits
2420 * bytes of data beyond current end of socket buffer. @trailer will be
2421 * set to point to the skb in which this space begins.
2423 * The number of scatterlist elements required to completely map the
2424 * COW'd and extended socket buffer will be returned.
2426 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2430 struct sk_buff *skb1, **skb_p;
2432 /* If skb is cloned or its head is paged, reallocate
2433 * head pulling out all the pages (pages are considered not writable
2434 * at the moment even if they are anonymous).
2436 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2437 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2440 /* Easy case. Most of packets will go this way. */
2441 if (!skb_shinfo(skb)->frag_list) {
2442 /* A little of trouble, not enough of space for trailer.
2443 * This should not happen, when stack is tuned to generate
2444 * good frames. OK, on miss we reallocate and reserve even more
2445 * space, 128 bytes is fair. */
2447 if (skb_tailroom(skb) < tailbits &&
2448 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2456 /* Misery. We are in troubles, going to mincer fragments... */
2459 skb_p = &skb_shinfo(skb)->frag_list;
2462 while ((skb1 = *skb_p) != NULL) {
2465 /* The fragment is partially pulled by someone,
2466 * this can happen on input. Copy it and everything
2469 if (skb_shared(skb1))
2472 /* If the skb is the last, worry about trailer. */
2474 if (skb1->next == NULL && tailbits) {
2475 if (skb_shinfo(skb1)->nr_frags ||
2476 skb_shinfo(skb1)->frag_list ||
2477 skb_tailroom(skb1) < tailbits)
2478 ntail = tailbits + 128;
2484 skb_shinfo(skb1)->nr_frags ||
2485 skb_shinfo(skb1)->frag_list) {
2486 struct sk_buff *skb2;
2488 /* Fuck, we are miserable poor guys... */
2490 skb2 = skb_copy(skb1, GFP_ATOMIC);
2492 skb2 = skb_copy_expand(skb1,
2496 if (unlikely(skb2 == NULL))
2500 skb_set_owner_w(skb2, skb1->sk);
2502 /* Looking around. Are we still alive?
2503 * OK, link new skb, drop old one */
2505 skb2->next = skb1->next;
2512 skb_p = &skb1->next;
2519 * skb_partial_csum_set - set up and verify partial csum values for packet
2520 * @skb: the skb to set
2521 * @start: the number of bytes after skb->data to start checksumming.
2522 * @off: the offset from start to place the checksum.
2524 * For untrusted partially-checksummed packets, we need to make sure the values
2525 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2527 * This function checks and sets those values and skb->ip_summed: if this
2528 * returns false you should drop the packet.
2530 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2532 if (unlikely(start > skb->len - 2) ||
2533 unlikely((int)start + off > skb->len - 2)) {
2534 if (net_ratelimit())
2536 "bad partial csum: csum=%u/%u len=%u\n",
2537 start, off, skb->len);
2540 skb->ip_summed = CHECKSUM_PARTIAL;
2541 skb->csum_start = skb_headroom(skb) + start;
2542 skb->csum_offset = off;
2546 EXPORT_SYMBOL(___pskb_trim);
2547 EXPORT_SYMBOL(__kfree_skb);
2548 EXPORT_SYMBOL(kfree_skb);
2549 EXPORT_SYMBOL(__pskb_pull_tail);
2550 EXPORT_SYMBOL(__alloc_skb);
2551 EXPORT_SYMBOL(__netdev_alloc_skb);
2552 EXPORT_SYMBOL(pskb_copy);
2553 EXPORT_SYMBOL(pskb_expand_head);
2554 EXPORT_SYMBOL(skb_checksum);
2555 EXPORT_SYMBOL(skb_clone);
2556 EXPORT_SYMBOL(skb_copy);
2557 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2558 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2559 EXPORT_SYMBOL(skb_copy_bits);
2560 EXPORT_SYMBOL(skb_copy_expand);
2561 EXPORT_SYMBOL(skb_over_panic);
2562 EXPORT_SYMBOL(skb_pad);
2563 EXPORT_SYMBOL(skb_realloc_headroom);
2564 EXPORT_SYMBOL(skb_under_panic);
2565 EXPORT_SYMBOL(skb_dequeue);
2566 EXPORT_SYMBOL(skb_dequeue_tail);
2567 EXPORT_SYMBOL(skb_insert);
2568 EXPORT_SYMBOL(skb_queue_purge);
2569 EXPORT_SYMBOL(skb_queue_head);
2570 EXPORT_SYMBOL(skb_queue_tail);
2571 EXPORT_SYMBOL(skb_unlink);
2572 EXPORT_SYMBOL(skb_append);
2573 EXPORT_SYMBOL(skb_split);
2574 EXPORT_SYMBOL(skb_prepare_seq_read);
2575 EXPORT_SYMBOL(skb_seq_read);
2576 EXPORT_SYMBOL(skb_abort_seq_read);
2577 EXPORT_SYMBOL(skb_find_text);
2578 EXPORT_SYMBOL(skb_append_datato_frags);
2580 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2581 EXPORT_SYMBOL_GPL(skb_cow_data);
2582 EXPORT_SYMBOL_GPL(skb_partial_csum_set);