2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
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>
58 #include <linux/errqueue.h>
60 #include <net/protocol.h>
63 #include <net/checksum.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <trace/skb.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)
81 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
82 struct pipe_buffer *buf)
87 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
88 struct pipe_buffer *buf)
94 /* Pipe buffer operations for a socket. */
95 static struct pipe_buf_operations sock_pipe_buf_ops = {
97 .map = generic_pipe_buf_map,
98 .unmap = generic_pipe_buf_unmap,
99 .confirm = generic_pipe_buf_confirm,
100 .release = sock_pipe_buf_release,
101 .steal = sock_pipe_buf_steal,
102 .get = sock_pipe_buf_get,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
121 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here, skb->len, sz, skb->head, skb->data,
124 (unsigned long)skb->tail, (unsigned long)skb->end,
125 skb->dev ? skb->dev->name : "<NULL>");
128 EXPORT_SYMBOL(skb_over_panic);
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
142 "data:%p tail:%#lx end:%#lx dev:%s\n",
143 here, skb->len, sz, skb->head, skb->data,
144 (unsigned long)skb->tail, (unsigned long)skb->end,
145 skb->dev ? skb->dev->name : "<NULL>");
148 EXPORT_SYMBOL(skb_under_panic);
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
186 size = SKB_DATA_ALIGN(size);
187 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb, 0, offsetof(struct sk_buff, tail));
198 skb->truesize = size + sizeof(struct sk_buff);
199 atomic_set(&skb->users, 1);
202 skb_reset_tail_pointer(skb);
203 skb->end = skb->tail + size;
204 /* make sure we initialize shinfo sequentially */
205 shinfo = skb_shinfo(skb);
206 atomic_set(&shinfo->dataref, 1);
207 shinfo->nr_frags = 0;
208 shinfo->gso_size = 0;
209 shinfo->gso_segs = 0;
210 shinfo->gso_type = 0;
211 shinfo->ip6_frag_id = 0;
212 shinfo->tx_flags.flags = 0;
213 skb_frag_list_init(skb);
214 memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
217 struct sk_buff *child = skb + 1;
218 atomic_t *fclone_ref = (atomic_t *) (child + 1);
220 skb->fclone = SKB_FCLONE_ORIG;
221 atomic_set(fclone_ref, 1);
223 child->fclone = SKB_FCLONE_UNAVAILABLE;
228 kmem_cache_free(cache, skb);
232 EXPORT_SYMBOL(__alloc_skb);
235 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
236 * @dev: network device to receive on
237 * @length: length to allocate
238 * @gfp_mask: get_free_pages mask, passed to alloc_skb
240 * Allocate a new &sk_buff and assign it a usage count of one. The
241 * buffer has unspecified headroom built in. Users should allocate
242 * the headroom they think they need without accounting for the
243 * built in space. The built in space is used for optimisations.
245 * %NULL is returned if there is no free memory.
247 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
248 unsigned int length, gfp_t gfp_mask)
250 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
253 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
255 skb_reserve(skb, NET_SKB_PAD);
260 EXPORT_SYMBOL(__netdev_alloc_skb);
262 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
264 int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
267 page = alloc_pages_node(node, gfp_mask, 0);
270 EXPORT_SYMBOL(__netdev_alloc_page);
272 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
275 skb_fill_page_desc(skb, i, page, off, size);
277 skb->data_len += size;
278 skb->truesize += size;
280 EXPORT_SYMBOL(skb_add_rx_frag);
283 * dev_alloc_skb - allocate an skbuff for receiving
284 * @length: length to allocate
286 * Allocate a new &sk_buff and assign it a usage count of one. The
287 * buffer has unspecified headroom built in. Users should allocate
288 * the headroom they think they need without accounting for the
289 * built in space. The built in space is used for optimisations.
291 * %NULL is returned if there is no free memory. Although this function
292 * allocates memory it can be called from an interrupt.
294 struct sk_buff *dev_alloc_skb(unsigned int length)
297 * There is more code here than it seems:
298 * __dev_alloc_skb is an inline
300 return __dev_alloc_skb(length, GFP_ATOMIC);
302 EXPORT_SYMBOL(dev_alloc_skb);
304 static void skb_drop_list(struct sk_buff **listp)
306 struct sk_buff *list = *listp;
311 struct sk_buff *this = list;
317 static inline void skb_drop_fraglist(struct sk_buff *skb)
319 skb_drop_list(&skb_shinfo(skb)->frag_list);
322 static void skb_clone_fraglist(struct sk_buff *skb)
324 struct sk_buff *list;
326 skb_walk_frags(skb, list)
330 static void skb_release_data(struct sk_buff *skb)
333 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
334 &skb_shinfo(skb)->dataref)) {
335 if (skb_shinfo(skb)->nr_frags) {
337 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
338 put_page(skb_shinfo(skb)->frags[i].page);
341 if (skb_has_frags(skb))
342 skb_drop_fraglist(skb);
349 * Free an skbuff by memory without cleaning the state.
351 static void kfree_skbmem(struct sk_buff *skb)
353 struct sk_buff *other;
354 atomic_t *fclone_ref;
356 switch (skb->fclone) {
357 case SKB_FCLONE_UNAVAILABLE:
358 kmem_cache_free(skbuff_head_cache, skb);
361 case SKB_FCLONE_ORIG:
362 fclone_ref = (atomic_t *) (skb + 2);
363 if (atomic_dec_and_test(fclone_ref))
364 kmem_cache_free(skbuff_fclone_cache, skb);
367 case SKB_FCLONE_CLONE:
368 fclone_ref = (atomic_t *) (skb + 1);
371 /* The clone portion is available for
372 * fast-cloning again.
374 skb->fclone = SKB_FCLONE_UNAVAILABLE;
376 if (atomic_dec_and_test(fclone_ref))
377 kmem_cache_free(skbuff_fclone_cache, other);
382 static void skb_release_head_state(struct sk_buff *skb)
386 secpath_put(skb->sp);
388 if (skb->destructor) {
390 skb->destructor(skb);
392 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
393 nf_conntrack_put(skb->nfct);
394 nf_conntrack_put_reasm(skb->nfct_reasm);
396 #ifdef CONFIG_BRIDGE_NETFILTER
397 nf_bridge_put(skb->nf_bridge);
399 /* XXX: IS this still necessary? - JHS */
400 #ifdef CONFIG_NET_SCHED
402 #ifdef CONFIG_NET_CLS_ACT
408 /* Free everything but the sk_buff shell. */
409 static void skb_release_all(struct sk_buff *skb)
411 skb_release_head_state(skb);
412 skb_release_data(skb);
416 * __kfree_skb - private function
419 * Free an sk_buff. Release anything attached to the buffer.
420 * Clean the state. This is an internal helper function. Users should
421 * always call kfree_skb
424 void __kfree_skb(struct sk_buff *skb)
426 skb_release_all(skb);
429 EXPORT_SYMBOL(__kfree_skb);
432 * kfree_skb - free an sk_buff
433 * @skb: buffer to free
435 * Drop a reference to the buffer and free it if the usage count has
438 void kfree_skb(struct sk_buff *skb)
442 if (likely(atomic_read(&skb->users) == 1))
444 else if (likely(!atomic_dec_and_test(&skb->users)))
446 trace_kfree_skb(skb, __builtin_return_address(0));
449 EXPORT_SYMBOL(kfree_skb);
452 * consume_skb - free an skbuff
453 * @skb: buffer to free
455 * Drop a ref to the buffer and free it if the usage count has hit zero
456 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
457 * is being dropped after a failure and notes that
459 void consume_skb(struct sk_buff *skb)
463 if (likely(atomic_read(&skb->users) == 1))
465 else if (likely(!atomic_dec_and_test(&skb->users)))
469 EXPORT_SYMBOL(consume_skb);
472 * skb_recycle_check - check if skb can be reused for receive
474 * @skb_size: minimum receive buffer size
476 * Checks that the skb passed in is not shared or cloned, and
477 * that it is linear and its head portion at least as large as
478 * skb_size so that it can be recycled as a receive buffer.
479 * If these conditions are met, this function does any necessary
480 * reference count dropping and cleans up the skbuff as if it
481 * just came from __alloc_skb().
483 int skb_recycle_check(struct sk_buff *skb, int skb_size)
485 struct skb_shared_info *shinfo;
487 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
490 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
491 if (skb_end_pointer(skb) - skb->head < skb_size)
494 if (skb_shared(skb) || skb_cloned(skb))
497 skb_release_head_state(skb);
498 shinfo = skb_shinfo(skb);
499 atomic_set(&shinfo->dataref, 1);
500 shinfo->nr_frags = 0;
501 shinfo->gso_size = 0;
502 shinfo->gso_segs = 0;
503 shinfo->gso_type = 0;
504 shinfo->ip6_frag_id = 0;
505 shinfo->tx_flags.flags = 0;
506 skb_frag_list_init(skb);
507 memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
509 memset(skb, 0, offsetof(struct sk_buff, tail));
510 skb->data = skb->head + NET_SKB_PAD;
511 skb_reset_tail_pointer(skb);
515 EXPORT_SYMBOL(skb_recycle_check);
517 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
519 new->tstamp = old->tstamp;
521 new->transport_header = old->transport_header;
522 new->network_header = old->network_header;
523 new->mac_header = old->mac_header;
524 skb_dst_set(new, dst_clone(skb_dst(old)));
526 new->sp = secpath_get(old->sp);
528 memcpy(new->cb, old->cb, sizeof(old->cb));
529 new->csum = old->csum;
530 new->local_df = old->local_df;
531 new->pkt_type = old->pkt_type;
532 new->ip_summed = old->ip_summed;
533 skb_copy_queue_mapping(new, old);
534 new->priority = old->priority;
535 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
536 new->ipvs_property = old->ipvs_property;
538 new->protocol = old->protocol;
539 new->mark = old->mark;
542 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
543 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
544 new->nf_trace = old->nf_trace;
546 #ifdef CONFIG_NET_SCHED
547 new->tc_index = old->tc_index;
548 #ifdef CONFIG_NET_CLS_ACT
549 new->tc_verd = old->tc_verd;
552 new->vlan_tci = old->vlan_tci;
553 #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE)
554 new->do_not_encrypt = old->do_not_encrypt;
555 new->requeue = old->requeue;
558 skb_copy_secmark(new, old);
562 * You should not add any new code to this function. Add it to
563 * __copy_skb_header above instead.
565 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
567 #define C(x) n->x = skb->x
569 n->next = n->prev = NULL;
571 __copy_skb_header(n, skb);
576 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
579 n->destructor = NULL;
585 atomic_set(&n->users, 1);
587 atomic_inc(&(skb_shinfo(skb)->dataref));
595 * skb_morph - morph one skb into another
596 * @dst: the skb to receive the contents
597 * @src: the skb to supply the contents
599 * This is identical to skb_clone except that the target skb is
600 * supplied by the user.
602 * The target skb is returned upon exit.
604 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
606 skb_release_all(dst);
607 return __skb_clone(dst, src);
609 EXPORT_SYMBOL_GPL(skb_morph);
612 * skb_clone - duplicate an sk_buff
613 * @skb: buffer to clone
614 * @gfp_mask: allocation priority
616 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
617 * copies share the same packet data but not structure. The new
618 * buffer has a reference count of 1. If the allocation fails the
619 * function returns %NULL otherwise the new buffer is returned.
621 * If this function is called from an interrupt gfp_mask() must be
625 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
630 if (skb->fclone == SKB_FCLONE_ORIG &&
631 n->fclone == SKB_FCLONE_UNAVAILABLE) {
632 atomic_t *fclone_ref = (atomic_t *) (n + 1);
633 n->fclone = SKB_FCLONE_CLONE;
634 atomic_inc(fclone_ref);
636 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
639 n->fclone = SKB_FCLONE_UNAVAILABLE;
642 return __skb_clone(n, skb);
644 EXPORT_SYMBOL(skb_clone);
646 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
648 #ifndef NET_SKBUFF_DATA_USES_OFFSET
650 * Shift between the two data areas in bytes
652 unsigned long offset = new->data - old->data;
655 __copy_skb_header(new, old);
657 #ifndef NET_SKBUFF_DATA_USES_OFFSET
658 /* {transport,network,mac}_header are relative to skb->head */
659 new->transport_header += offset;
660 new->network_header += offset;
661 new->mac_header += offset;
663 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
664 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
665 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
669 * skb_copy - create private copy of an sk_buff
670 * @skb: buffer to copy
671 * @gfp_mask: allocation priority
673 * Make a copy of both an &sk_buff and its data. This is used when the
674 * caller wishes to modify the data and needs a private copy of the
675 * data to alter. Returns %NULL on failure or the pointer to the buffer
676 * on success. The returned buffer has a reference count of 1.
678 * As by-product this function converts non-linear &sk_buff to linear
679 * one, so that &sk_buff becomes completely private and caller is allowed
680 * to modify all the data of returned buffer. This means that this
681 * function is not recommended for use in circumstances when only
682 * header is going to be modified. Use pskb_copy() instead.
685 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
687 int headerlen = skb->data - skb->head;
689 * Allocate the copy buffer
692 #ifdef NET_SKBUFF_DATA_USES_OFFSET
693 n = alloc_skb(skb->end + skb->data_len, gfp_mask);
695 n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
700 /* Set the data pointer */
701 skb_reserve(n, headerlen);
702 /* Set the tail pointer and length */
703 skb_put(n, skb->len);
705 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
708 copy_skb_header(n, skb);
711 EXPORT_SYMBOL(skb_copy);
714 * pskb_copy - create copy of an sk_buff with private head.
715 * @skb: buffer to copy
716 * @gfp_mask: allocation priority
718 * Make a copy of both an &sk_buff and part of its data, located
719 * in header. Fragmented data remain shared. This is used when
720 * the caller wishes to modify only header of &sk_buff and needs
721 * private copy of the header to alter. Returns %NULL on failure
722 * or the pointer to the buffer on success.
723 * The returned buffer has a reference count of 1.
726 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
729 * Allocate the copy buffer
732 #ifdef NET_SKBUFF_DATA_USES_OFFSET
733 n = alloc_skb(skb->end, gfp_mask);
735 n = alloc_skb(skb->end - skb->head, gfp_mask);
740 /* Set the data pointer */
741 skb_reserve(n, skb->data - skb->head);
742 /* Set the tail pointer and length */
743 skb_put(n, skb_headlen(skb));
745 skb_copy_from_linear_data(skb, n->data, n->len);
747 n->truesize += skb->data_len;
748 n->data_len = skb->data_len;
751 if (skb_shinfo(skb)->nr_frags) {
754 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
755 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
756 get_page(skb_shinfo(n)->frags[i].page);
758 skb_shinfo(n)->nr_frags = i;
761 if (skb_has_frags(skb)) {
762 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
763 skb_clone_fraglist(n);
766 copy_skb_header(n, skb);
770 EXPORT_SYMBOL(pskb_copy);
773 * pskb_expand_head - reallocate header of &sk_buff
774 * @skb: buffer to reallocate
775 * @nhead: room to add at head
776 * @ntail: room to add at tail
777 * @gfp_mask: allocation priority
779 * Expands (or creates identical copy, if &nhead and &ntail are zero)
780 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
781 * reference count of 1. Returns zero in the case of success or error,
782 * if expansion failed. In the last case, &sk_buff is not changed.
784 * All the pointers pointing into skb header may change and must be
785 * reloaded after call to this function.
788 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
793 #ifdef NET_SKBUFF_DATA_USES_OFFSET
794 int size = nhead + skb->end + ntail;
796 int size = nhead + (skb->end - skb->head) + ntail;
805 size = SKB_DATA_ALIGN(size);
807 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
811 /* Copy only real data... and, alas, header. This should be
812 * optimized for the cases when header is void. */
813 #ifdef NET_SKBUFF_DATA_USES_OFFSET
814 memcpy(data + nhead, skb->head, skb->tail);
816 memcpy(data + nhead, skb->head, skb->tail - skb->head);
818 memcpy(data + size, skb_end_pointer(skb),
819 sizeof(struct skb_shared_info));
821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
822 get_page(skb_shinfo(skb)->frags[i].page);
824 if (skb_has_frags(skb))
825 skb_clone_fraglist(skb);
827 skb_release_data(skb);
829 off = (data + nhead) - skb->head;
833 #ifdef NET_SKBUFF_DATA_USES_OFFSET
837 skb->end = skb->head + size;
839 /* {transport,network,mac}_header and tail are relative to skb->head */
841 skb->transport_header += off;
842 skb->network_header += off;
843 skb->mac_header += off;
844 skb->csum_start += nhead;
848 atomic_set(&skb_shinfo(skb)->dataref, 1);
854 EXPORT_SYMBOL(pskb_expand_head);
856 /* Make private copy of skb with writable head and some headroom */
858 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
860 struct sk_buff *skb2;
861 int delta = headroom - skb_headroom(skb);
864 skb2 = pskb_copy(skb, GFP_ATOMIC);
866 skb2 = skb_clone(skb, GFP_ATOMIC);
867 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
875 EXPORT_SYMBOL(skb_realloc_headroom);
878 * skb_copy_expand - copy and expand sk_buff
879 * @skb: buffer to copy
880 * @newheadroom: new free bytes at head
881 * @newtailroom: new free bytes at tail
882 * @gfp_mask: allocation priority
884 * Make a copy of both an &sk_buff and its data and while doing so
885 * allocate additional space.
887 * This is used when the caller wishes to modify the data and needs a
888 * private copy of the data to alter as well as more space for new fields.
889 * Returns %NULL on failure or the pointer to the buffer
890 * on success. The returned buffer has a reference count of 1.
892 * You must pass %GFP_ATOMIC as the allocation priority if this function
893 * is called from an interrupt.
895 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
896 int newheadroom, int newtailroom,
900 * Allocate the copy buffer
902 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
904 int oldheadroom = skb_headroom(skb);
905 int head_copy_len, head_copy_off;
911 skb_reserve(n, newheadroom);
913 /* Set the tail pointer and length */
914 skb_put(n, skb->len);
916 head_copy_len = oldheadroom;
918 if (newheadroom <= head_copy_len)
919 head_copy_len = newheadroom;
921 head_copy_off = newheadroom - head_copy_len;
923 /* Copy the linear header and data. */
924 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
925 skb->len + head_copy_len))
928 copy_skb_header(n, skb);
930 off = newheadroom - oldheadroom;
931 n->csum_start += off;
932 #ifdef NET_SKBUFF_DATA_USES_OFFSET
933 n->transport_header += off;
934 n->network_header += off;
935 n->mac_header += off;
940 EXPORT_SYMBOL(skb_copy_expand);
943 * skb_pad - zero pad the tail of an skb
944 * @skb: buffer to pad
947 * Ensure that a buffer is followed by a padding area that is zero
948 * filled. Used by network drivers which may DMA or transfer data
949 * beyond the buffer end onto the wire.
951 * May return error in out of memory cases. The skb is freed on error.
954 int skb_pad(struct sk_buff *skb, int pad)
959 /* If the skbuff is non linear tailroom is always zero.. */
960 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
961 memset(skb->data+skb->len, 0, pad);
965 ntail = skb->data_len + pad - (skb->end - skb->tail);
966 if (likely(skb_cloned(skb) || ntail > 0)) {
967 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
972 /* FIXME: The use of this function with non-linear skb's really needs
975 err = skb_linearize(skb);
979 memset(skb->data + skb->len, 0, pad);
986 EXPORT_SYMBOL(skb_pad);
989 * skb_put - add data to a buffer
990 * @skb: buffer to use
991 * @len: amount of data to add
993 * This function extends the used data area of the buffer. If this would
994 * exceed the total buffer size the kernel will panic. A pointer to the
995 * first byte of the extra data is returned.
997 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
999 unsigned char *tmp = skb_tail_pointer(skb);
1000 SKB_LINEAR_ASSERT(skb);
1003 if (unlikely(skb->tail > skb->end))
1004 skb_over_panic(skb, len, __builtin_return_address(0));
1007 EXPORT_SYMBOL(skb_put);
1010 * skb_push - add data to the start of a buffer
1011 * @skb: buffer to use
1012 * @len: amount of data to add
1014 * This function extends the used data area of the buffer at the buffer
1015 * start. If this would exceed the total buffer headroom the kernel will
1016 * panic. A pointer to the first byte of the extra data is returned.
1018 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1022 if (unlikely(skb->data<skb->head))
1023 skb_under_panic(skb, len, __builtin_return_address(0));
1026 EXPORT_SYMBOL(skb_push);
1029 * skb_pull - remove data from the start of a buffer
1030 * @skb: buffer to use
1031 * @len: amount of data to remove
1033 * This function removes data from the start of a buffer, returning
1034 * the memory to the headroom. A pointer to the next data in the buffer
1035 * is returned. Once the data has been pulled future pushes will overwrite
1038 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1040 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1042 EXPORT_SYMBOL(skb_pull);
1045 * skb_trim - remove end from a buffer
1046 * @skb: buffer to alter
1049 * Cut the length of a buffer down by removing data from the tail. If
1050 * the buffer is already under the length specified it is not modified.
1051 * The skb must be linear.
1053 void skb_trim(struct sk_buff *skb, unsigned int len)
1056 __skb_trim(skb, len);
1058 EXPORT_SYMBOL(skb_trim);
1060 /* Trims skb to length len. It can change skb pointers.
1063 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1065 struct sk_buff **fragp;
1066 struct sk_buff *frag;
1067 int offset = skb_headlen(skb);
1068 int nfrags = skb_shinfo(skb)->nr_frags;
1072 if (skb_cloned(skb) &&
1073 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1080 for (; i < nfrags; i++) {
1081 int end = offset + skb_shinfo(skb)->frags[i].size;
1088 skb_shinfo(skb)->frags[i++].size = len - offset;
1091 skb_shinfo(skb)->nr_frags = i;
1093 for (; i < nfrags; i++)
1094 put_page(skb_shinfo(skb)->frags[i].page);
1096 if (skb_has_frags(skb))
1097 skb_drop_fraglist(skb);
1101 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1102 fragp = &frag->next) {
1103 int end = offset + frag->len;
1105 if (skb_shared(frag)) {
1106 struct sk_buff *nfrag;
1108 nfrag = skb_clone(frag, GFP_ATOMIC);
1109 if (unlikely(!nfrag))
1112 nfrag->next = frag->next;
1124 unlikely((err = pskb_trim(frag, len - offset))))
1128 skb_drop_list(&frag->next);
1133 if (len > skb_headlen(skb)) {
1134 skb->data_len -= skb->len - len;
1139 skb_set_tail_pointer(skb, len);
1144 EXPORT_SYMBOL(___pskb_trim);
1147 * __pskb_pull_tail - advance tail of skb header
1148 * @skb: buffer to reallocate
1149 * @delta: number of bytes to advance tail
1151 * The function makes a sense only on a fragmented &sk_buff,
1152 * it expands header moving its tail forward and copying necessary
1153 * data from fragmented part.
1155 * &sk_buff MUST have reference count of 1.
1157 * Returns %NULL (and &sk_buff does not change) if pull failed
1158 * or value of new tail of skb in the case of success.
1160 * All the pointers pointing into skb header may change and must be
1161 * reloaded after call to this function.
1164 /* Moves tail of skb head forward, copying data from fragmented part,
1165 * when it is necessary.
1166 * 1. It may fail due to malloc failure.
1167 * 2. It may change skb pointers.
1169 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1171 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1173 /* If skb has not enough free space at tail, get new one
1174 * plus 128 bytes for future expansions. If we have enough
1175 * room at tail, reallocate without expansion only if skb is cloned.
1177 int i, k, eat = (skb->tail + delta) - skb->end;
1179 if (eat > 0 || skb_cloned(skb)) {
1180 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1185 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1188 /* Optimization: no fragments, no reasons to preestimate
1189 * size of pulled pages. Superb.
1191 if (!skb_has_frags(skb))
1194 /* Estimate size of pulled pages. */
1196 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1197 if (skb_shinfo(skb)->frags[i].size >= eat)
1199 eat -= skb_shinfo(skb)->frags[i].size;
1202 /* If we need update frag list, we are in troubles.
1203 * Certainly, it possible to add an offset to skb data,
1204 * but taking into account that pulling is expected to
1205 * be very rare operation, it is worth to fight against
1206 * further bloating skb head and crucify ourselves here instead.
1207 * Pure masohism, indeed. 8)8)
1210 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1211 struct sk_buff *clone = NULL;
1212 struct sk_buff *insp = NULL;
1217 if (list->len <= eat) {
1218 /* Eaten as whole. */
1223 /* Eaten partially. */
1225 if (skb_shared(list)) {
1226 /* Sucks! We need to fork list. :-( */
1227 clone = skb_clone(list, GFP_ATOMIC);
1233 /* This may be pulled without
1237 if (!pskb_pull(list, eat)) {
1245 /* Free pulled out fragments. */
1246 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1247 skb_shinfo(skb)->frag_list = list->next;
1250 /* And insert new clone at head. */
1253 skb_shinfo(skb)->frag_list = clone;
1256 /* Success! Now we may commit changes to skb data. */
1261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1262 if (skb_shinfo(skb)->frags[i].size <= eat) {
1263 put_page(skb_shinfo(skb)->frags[i].page);
1264 eat -= skb_shinfo(skb)->frags[i].size;
1266 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1268 skb_shinfo(skb)->frags[k].page_offset += eat;
1269 skb_shinfo(skb)->frags[k].size -= eat;
1275 skb_shinfo(skb)->nr_frags = k;
1278 skb->data_len -= delta;
1280 return skb_tail_pointer(skb);
1282 EXPORT_SYMBOL(__pskb_pull_tail);
1284 /* Copy some data bits from skb to kernel buffer. */
1286 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1288 int start = skb_headlen(skb);
1289 struct sk_buff *frag_iter;
1292 if (offset > (int)skb->len - len)
1296 if ((copy = start - offset) > 0) {
1299 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1300 if ((len -= copy) == 0)
1306 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1309 WARN_ON(start > offset + len);
1311 end = start + skb_shinfo(skb)->frags[i].size;
1312 if ((copy = end - offset) > 0) {
1318 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1320 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1321 offset - start, copy);
1322 kunmap_skb_frag(vaddr);
1324 if ((len -= copy) == 0)
1332 skb_walk_frags(skb, frag_iter) {
1335 WARN_ON(start > offset + len);
1337 end = start + frag_iter->len;
1338 if ((copy = end - offset) > 0) {
1341 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1343 if ((len -= copy) == 0)
1356 EXPORT_SYMBOL(skb_copy_bits);
1359 * Callback from splice_to_pipe(), if we need to release some pages
1360 * at the end of the spd in case we error'ed out in filling the pipe.
1362 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1364 put_page(spd->pages[i]);
1367 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1368 unsigned int *offset,
1369 struct sk_buff *skb, struct sock *sk)
1371 struct page *p = sk->sk_sndmsg_page;
1376 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1380 off = sk->sk_sndmsg_off = 0;
1381 /* hold one ref to this page until it's full */
1385 off = sk->sk_sndmsg_off;
1386 mlen = PAGE_SIZE - off;
1387 if (mlen < 64 && mlen < *len) {
1392 *len = min_t(unsigned int, *len, mlen);
1395 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1396 sk->sk_sndmsg_off += *len;
1404 * Fill page/offset/length into spd, if it can hold more pages.
1406 static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
1407 unsigned int *len, unsigned int offset,
1408 struct sk_buff *skb, int linear,
1411 if (unlikely(spd->nr_pages == PIPE_BUFFERS))
1415 page = linear_to_page(page, len, &offset, skb, sk);
1421 spd->pages[spd->nr_pages] = page;
1422 spd->partial[spd->nr_pages].len = *len;
1423 spd->partial[spd->nr_pages].offset = offset;
1429 static inline void __segment_seek(struct page **page, unsigned int *poff,
1430 unsigned int *plen, unsigned int off)
1435 n = *poff / PAGE_SIZE;
1437 *page = nth_page(*page, n);
1439 *poff = *poff % PAGE_SIZE;
1443 static inline int __splice_segment(struct page *page, unsigned int poff,
1444 unsigned int plen, unsigned int *off,
1445 unsigned int *len, struct sk_buff *skb,
1446 struct splice_pipe_desc *spd, int linear,
1452 /* skip this segment if already processed */
1458 /* ignore any bits we already processed */
1460 __segment_seek(&page, &poff, &plen, *off);
1465 unsigned int flen = min(*len, plen);
1467 /* the linear region may spread across several pages */
1468 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1470 if (spd_fill_page(spd, page, &flen, poff, skb, linear, sk))
1473 __segment_seek(&page, &poff, &plen, flen);
1476 } while (*len && plen);
1482 * Map linear and fragment data from the skb to spd. It reports failure if the
1483 * pipe is full or if we already spliced the requested length.
1485 static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
1486 unsigned int *len, struct splice_pipe_desc *spd,
1492 * map the linear part
1494 if (__splice_segment(virt_to_page(skb->data),
1495 (unsigned long) skb->data & (PAGE_SIZE - 1),
1497 offset, len, skb, spd, 1, sk))
1501 * then map the fragments
1503 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1504 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1506 if (__splice_segment(f->page, f->page_offset, f->size,
1507 offset, len, skb, spd, 0, sk))
1515 * Map data from the skb to a pipe. Should handle both the linear part,
1516 * the fragments, and the frag list. It does NOT handle frag lists within
1517 * the frag list, if such a thing exists. We'd probably need to recurse to
1518 * handle that cleanly.
1520 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1521 struct pipe_inode_info *pipe, unsigned int tlen,
1524 struct partial_page partial[PIPE_BUFFERS];
1525 struct page *pages[PIPE_BUFFERS];
1526 struct splice_pipe_desc spd = {
1530 .ops = &sock_pipe_buf_ops,
1531 .spd_release = sock_spd_release,
1533 struct sk_buff *frag_iter;
1534 struct sock *sk = skb->sk;
1537 * __skb_splice_bits() only fails if the output has no room left,
1538 * so no point in going over the frag_list for the error case.
1540 if (__skb_splice_bits(skb, &offset, &tlen, &spd, sk))
1546 * now see if we have a frag_list to map
1548 skb_walk_frags(skb, frag_iter) {
1551 if (__skb_splice_bits(frag_iter, &offset, &tlen, &spd, sk))
1560 * Drop the socket lock, otherwise we have reverse
1561 * locking dependencies between sk_lock and i_mutex
1562 * here as compared to sendfile(). We enter here
1563 * with the socket lock held, and splice_to_pipe() will
1564 * grab the pipe inode lock. For sendfile() emulation,
1565 * we call into ->sendpage() with the i_mutex lock held
1566 * and networking will grab the socket lock.
1569 ret = splice_to_pipe(pipe, &spd);
1578 * skb_store_bits - store bits from kernel buffer to skb
1579 * @skb: destination buffer
1580 * @offset: offset in destination
1581 * @from: source buffer
1582 * @len: number of bytes to copy
1584 * Copy the specified number of bytes from the source buffer to the
1585 * destination skb. This function handles all the messy bits of
1586 * traversing fragment lists and such.
1589 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1591 int start = skb_headlen(skb);
1592 struct sk_buff *frag_iter;
1595 if (offset > (int)skb->len - len)
1598 if ((copy = start - offset) > 0) {
1601 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1602 if ((len -= copy) == 0)
1608 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1609 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1612 WARN_ON(start > offset + len);
1614 end = start + frag->size;
1615 if ((copy = end - offset) > 0) {
1621 vaddr = kmap_skb_frag(frag);
1622 memcpy(vaddr + frag->page_offset + offset - start,
1624 kunmap_skb_frag(vaddr);
1626 if ((len -= copy) == 0)
1634 skb_walk_frags(skb, frag_iter) {
1637 WARN_ON(start > offset + len);
1639 end = start + frag_iter->len;
1640 if ((copy = end - offset) > 0) {
1643 if (skb_store_bits(frag_iter, offset - start,
1646 if ((len -= copy) == 0)
1659 EXPORT_SYMBOL(skb_store_bits);
1661 /* Checksum skb data. */
1663 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1664 int len, __wsum csum)
1666 int start = skb_headlen(skb);
1667 int i, copy = start - offset;
1668 struct sk_buff *frag_iter;
1671 /* Checksum header. */
1675 csum = csum_partial(skb->data + offset, copy, csum);
1676 if ((len -= copy) == 0)
1682 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1685 WARN_ON(start > offset + len);
1687 end = start + skb_shinfo(skb)->frags[i].size;
1688 if ((copy = end - offset) > 0) {
1691 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1695 vaddr = kmap_skb_frag(frag);
1696 csum2 = csum_partial(vaddr + frag->page_offset +
1697 offset - start, copy, 0);
1698 kunmap_skb_frag(vaddr);
1699 csum = csum_block_add(csum, csum2, pos);
1708 skb_walk_frags(skb, frag_iter) {
1711 WARN_ON(start > offset + len);
1713 end = start + frag_iter->len;
1714 if ((copy = end - offset) > 0) {
1718 csum2 = skb_checksum(frag_iter, offset - start,
1720 csum = csum_block_add(csum, csum2, pos);
1721 if ((len -= copy) == 0)
1732 EXPORT_SYMBOL(skb_checksum);
1734 /* Both of above in one bottle. */
1736 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1737 u8 *to, int len, __wsum csum)
1739 int start = skb_headlen(skb);
1740 int i, copy = start - offset;
1741 struct sk_buff *frag_iter;
1748 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1750 if ((len -= copy) == 0)
1757 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1760 WARN_ON(start > offset + len);
1762 end = start + skb_shinfo(skb)->frags[i].size;
1763 if ((copy = end - offset) > 0) {
1766 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1770 vaddr = kmap_skb_frag(frag);
1771 csum2 = csum_partial_copy_nocheck(vaddr +
1775 kunmap_skb_frag(vaddr);
1776 csum = csum_block_add(csum, csum2, pos);
1786 skb_walk_frags(skb, frag_iter) {
1790 WARN_ON(start > offset + len);
1792 end = start + frag_iter->len;
1793 if ((copy = end - offset) > 0) {
1796 csum2 = skb_copy_and_csum_bits(frag_iter,
1799 csum = csum_block_add(csum, csum2, pos);
1800 if ((len -= copy) == 0)
1811 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1813 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1818 if (skb->ip_summed == CHECKSUM_PARTIAL)
1819 csstart = skb->csum_start - skb_headroom(skb);
1821 csstart = skb_headlen(skb);
1823 BUG_ON(csstart > skb_headlen(skb));
1825 skb_copy_from_linear_data(skb, to, csstart);
1828 if (csstart != skb->len)
1829 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1830 skb->len - csstart, 0);
1832 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1833 long csstuff = csstart + skb->csum_offset;
1835 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1838 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1841 * skb_dequeue - remove from the head of the queue
1842 * @list: list to dequeue from
1844 * Remove the head of the list. The list lock is taken so the function
1845 * may be used safely with other locking list functions. The head item is
1846 * returned or %NULL if the list is empty.
1849 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1851 unsigned long flags;
1852 struct sk_buff *result;
1854 spin_lock_irqsave(&list->lock, flags);
1855 result = __skb_dequeue(list);
1856 spin_unlock_irqrestore(&list->lock, flags);
1859 EXPORT_SYMBOL(skb_dequeue);
1862 * skb_dequeue_tail - remove from the tail of the queue
1863 * @list: list to dequeue from
1865 * Remove the tail of the list. The list lock is taken so the function
1866 * may be used safely with other locking list functions. The tail item is
1867 * returned or %NULL if the list is empty.
1869 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1871 unsigned long flags;
1872 struct sk_buff *result;
1874 spin_lock_irqsave(&list->lock, flags);
1875 result = __skb_dequeue_tail(list);
1876 spin_unlock_irqrestore(&list->lock, flags);
1879 EXPORT_SYMBOL(skb_dequeue_tail);
1882 * skb_queue_purge - empty a list
1883 * @list: list to empty
1885 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1886 * the list and one reference dropped. This function takes the list
1887 * lock and is atomic with respect to other list locking functions.
1889 void skb_queue_purge(struct sk_buff_head *list)
1891 struct sk_buff *skb;
1892 while ((skb = skb_dequeue(list)) != NULL)
1895 EXPORT_SYMBOL(skb_queue_purge);
1898 * skb_queue_head - queue a buffer at the list head
1899 * @list: list to use
1900 * @newsk: buffer to queue
1902 * Queue a buffer at the start of the list. This function takes the
1903 * list lock and can be used safely with other locking &sk_buff functions
1906 * A buffer cannot be placed on two lists at the same time.
1908 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1910 unsigned long flags;
1912 spin_lock_irqsave(&list->lock, flags);
1913 __skb_queue_head(list, newsk);
1914 spin_unlock_irqrestore(&list->lock, flags);
1916 EXPORT_SYMBOL(skb_queue_head);
1919 * skb_queue_tail - queue a buffer at the list tail
1920 * @list: list to use
1921 * @newsk: buffer to queue
1923 * Queue a buffer at the tail of the list. This function takes the
1924 * list lock and can be used safely with other locking &sk_buff functions
1927 * A buffer cannot be placed on two lists at the same time.
1929 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1931 unsigned long flags;
1933 spin_lock_irqsave(&list->lock, flags);
1934 __skb_queue_tail(list, newsk);
1935 spin_unlock_irqrestore(&list->lock, flags);
1937 EXPORT_SYMBOL(skb_queue_tail);
1940 * skb_unlink - remove a buffer from a list
1941 * @skb: buffer to remove
1942 * @list: list to use
1944 * Remove a packet from a list. The list locks are taken and this
1945 * function is atomic with respect to other list locked calls
1947 * You must know what list the SKB is on.
1949 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1951 unsigned long flags;
1953 spin_lock_irqsave(&list->lock, flags);
1954 __skb_unlink(skb, list);
1955 spin_unlock_irqrestore(&list->lock, flags);
1957 EXPORT_SYMBOL(skb_unlink);
1960 * skb_append - append a buffer
1961 * @old: buffer to insert after
1962 * @newsk: buffer to insert
1963 * @list: list to use
1965 * Place a packet after a given packet in a list. The list locks are taken
1966 * and this function is atomic with respect to other list locked calls.
1967 * A buffer cannot be placed on two lists at the same time.
1969 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1971 unsigned long flags;
1973 spin_lock_irqsave(&list->lock, flags);
1974 __skb_queue_after(list, old, newsk);
1975 spin_unlock_irqrestore(&list->lock, flags);
1977 EXPORT_SYMBOL(skb_append);
1980 * skb_insert - insert a buffer
1981 * @old: buffer to insert before
1982 * @newsk: buffer to insert
1983 * @list: list to use
1985 * Place a packet before a given packet in a list. The list locks are
1986 * taken and this function is atomic with respect to other list locked
1989 * A buffer cannot be placed on two lists at the same time.
1991 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1993 unsigned long flags;
1995 spin_lock_irqsave(&list->lock, flags);
1996 __skb_insert(newsk, old->prev, old, list);
1997 spin_unlock_irqrestore(&list->lock, flags);
1999 EXPORT_SYMBOL(skb_insert);
2001 static inline void skb_split_inside_header(struct sk_buff *skb,
2002 struct sk_buff* skb1,
2003 const u32 len, const int pos)
2007 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2009 /* And move data appendix as is. */
2010 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2011 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2013 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2014 skb_shinfo(skb)->nr_frags = 0;
2015 skb1->data_len = skb->data_len;
2016 skb1->len += skb1->data_len;
2019 skb_set_tail_pointer(skb, len);
2022 static inline void skb_split_no_header(struct sk_buff *skb,
2023 struct sk_buff* skb1,
2024 const u32 len, int pos)
2027 const int nfrags = skb_shinfo(skb)->nr_frags;
2029 skb_shinfo(skb)->nr_frags = 0;
2030 skb1->len = skb1->data_len = skb->len - len;
2032 skb->data_len = len - pos;
2034 for (i = 0; i < nfrags; i++) {
2035 int size = skb_shinfo(skb)->frags[i].size;
2037 if (pos + size > len) {
2038 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2042 * We have two variants in this case:
2043 * 1. Move all the frag to the second
2044 * part, if it is possible. F.e.
2045 * this approach is mandatory for TUX,
2046 * where splitting is expensive.
2047 * 2. Split is accurately. We make this.
2049 get_page(skb_shinfo(skb)->frags[i].page);
2050 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2051 skb_shinfo(skb1)->frags[0].size -= len - pos;
2052 skb_shinfo(skb)->frags[i].size = len - pos;
2053 skb_shinfo(skb)->nr_frags++;
2057 skb_shinfo(skb)->nr_frags++;
2060 skb_shinfo(skb1)->nr_frags = k;
2064 * skb_split - Split fragmented skb to two parts at length len.
2065 * @skb: the buffer to split
2066 * @skb1: the buffer to receive the second part
2067 * @len: new length for skb
2069 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2071 int pos = skb_headlen(skb);
2073 if (len < pos) /* Split line is inside header. */
2074 skb_split_inside_header(skb, skb1, len, pos);
2075 else /* Second chunk has no header, nothing to copy. */
2076 skb_split_no_header(skb, skb1, len, pos);
2078 EXPORT_SYMBOL(skb_split);
2080 /* Shifting from/to a cloned skb is a no-go.
2082 * Caller cannot keep skb_shinfo related pointers past calling here!
2084 static int skb_prepare_for_shift(struct sk_buff *skb)
2086 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2090 * skb_shift - Shifts paged data partially from skb to another
2091 * @tgt: buffer into which tail data gets added
2092 * @skb: buffer from which the paged data comes from
2093 * @shiftlen: shift up to this many bytes
2095 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2096 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2097 * It's up to caller to free skb if everything was shifted.
2099 * If @tgt runs out of frags, the whole operation is aborted.
2101 * Skb cannot include anything else but paged data while tgt is allowed
2102 * to have non-paged data as well.
2104 * TODO: full sized shift could be optimized but that would need
2105 * specialized skb free'er to handle frags without up-to-date nr_frags.
2107 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2109 int from, to, merge, todo;
2110 struct skb_frag_struct *fragfrom, *fragto;
2112 BUG_ON(shiftlen > skb->len);
2113 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2117 to = skb_shinfo(tgt)->nr_frags;
2118 fragfrom = &skb_shinfo(skb)->frags[from];
2120 /* Actual merge is delayed until the point when we know we can
2121 * commit all, so that we don't have to undo partial changes
2124 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2129 todo -= fragfrom->size;
2131 if (skb_prepare_for_shift(skb) ||
2132 skb_prepare_for_shift(tgt))
2135 /* All previous frag pointers might be stale! */
2136 fragfrom = &skb_shinfo(skb)->frags[from];
2137 fragto = &skb_shinfo(tgt)->frags[merge];
2139 fragto->size += shiftlen;
2140 fragfrom->size -= shiftlen;
2141 fragfrom->page_offset += shiftlen;
2149 /* Skip full, not-fitting skb to avoid expensive operations */
2150 if ((shiftlen == skb->len) &&
2151 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2154 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2157 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2158 if (to == MAX_SKB_FRAGS)
2161 fragfrom = &skb_shinfo(skb)->frags[from];
2162 fragto = &skb_shinfo(tgt)->frags[to];
2164 if (todo >= fragfrom->size) {
2165 *fragto = *fragfrom;
2166 todo -= fragfrom->size;
2171 get_page(fragfrom->page);
2172 fragto->page = fragfrom->page;
2173 fragto->page_offset = fragfrom->page_offset;
2174 fragto->size = todo;
2176 fragfrom->page_offset += todo;
2177 fragfrom->size -= todo;
2185 /* Ready to "commit" this state change to tgt */
2186 skb_shinfo(tgt)->nr_frags = to;
2189 fragfrom = &skb_shinfo(skb)->frags[0];
2190 fragto = &skb_shinfo(tgt)->frags[merge];
2192 fragto->size += fragfrom->size;
2193 put_page(fragfrom->page);
2196 /* Reposition in the original skb */
2198 while (from < skb_shinfo(skb)->nr_frags)
2199 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2200 skb_shinfo(skb)->nr_frags = to;
2202 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2205 /* Most likely the tgt won't ever need its checksum anymore, skb on
2206 * the other hand might need it if it needs to be resent
2208 tgt->ip_summed = CHECKSUM_PARTIAL;
2209 skb->ip_summed = CHECKSUM_PARTIAL;
2211 /* Yak, is it really working this way? Some helper please? */
2212 skb->len -= shiftlen;
2213 skb->data_len -= shiftlen;
2214 skb->truesize -= shiftlen;
2215 tgt->len += shiftlen;
2216 tgt->data_len += shiftlen;
2217 tgt->truesize += shiftlen;
2223 * skb_prepare_seq_read - Prepare a sequential read of skb data
2224 * @skb: the buffer to read
2225 * @from: lower offset of data to be read
2226 * @to: upper offset of data to be read
2227 * @st: state variable
2229 * Initializes the specified state variable. Must be called before
2230 * invoking skb_seq_read() for the first time.
2232 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2233 unsigned int to, struct skb_seq_state *st)
2235 st->lower_offset = from;
2236 st->upper_offset = to;
2237 st->root_skb = st->cur_skb = skb;
2238 st->frag_idx = st->stepped_offset = 0;
2239 st->frag_data = NULL;
2241 EXPORT_SYMBOL(skb_prepare_seq_read);
2244 * skb_seq_read - Sequentially read skb data
2245 * @consumed: number of bytes consumed by the caller so far
2246 * @data: destination pointer for data to be returned
2247 * @st: state variable
2249 * Reads a block of skb data at &consumed relative to the
2250 * lower offset specified to skb_prepare_seq_read(). Assigns
2251 * the head of the data block to &data and returns the length
2252 * of the block or 0 if the end of the skb data or the upper
2253 * offset has been reached.
2255 * The caller is not required to consume all of the data
2256 * returned, i.e. &consumed is typically set to the number
2257 * of bytes already consumed and the next call to
2258 * skb_seq_read() will return the remaining part of the block.
2260 * Note 1: The size of each block of data returned can be arbitary,
2261 * this limitation is the cost for zerocopy seqeuental
2262 * reads of potentially non linear data.
2264 * Note 2: Fragment lists within fragments are not implemented
2265 * at the moment, state->root_skb could be replaced with
2266 * a stack for this purpose.
2268 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2269 struct skb_seq_state *st)
2271 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2274 if (unlikely(abs_offset >= st->upper_offset))
2278 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2280 if (abs_offset < block_limit && !st->frag_data) {
2281 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2282 return block_limit - abs_offset;
2285 if (st->frag_idx == 0 && !st->frag_data)
2286 st->stepped_offset += skb_headlen(st->cur_skb);
2288 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2289 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2290 block_limit = frag->size + st->stepped_offset;
2292 if (abs_offset < block_limit) {
2294 st->frag_data = kmap_skb_frag(frag);
2296 *data = (u8 *) st->frag_data + frag->page_offset +
2297 (abs_offset - st->stepped_offset);
2299 return block_limit - abs_offset;
2302 if (st->frag_data) {
2303 kunmap_skb_frag(st->frag_data);
2304 st->frag_data = NULL;
2308 st->stepped_offset += frag->size;
2311 if (st->frag_data) {
2312 kunmap_skb_frag(st->frag_data);
2313 st->frag_data = NULL;
2316 if (st->root_skb == st->cur_skb && skb_has_frags(st->root_skb)) {
2317 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2320 } else if (st->cur_skb->next) {
2321 st->cur_skb = st->cur_skb->next;
2328 EXPORT_SYMBOL(skb_seq_read);
2331 * skb_abort_seq_read - Abort a sequential read of skb data
2332 * @st: state variable
2334 * Must be called if skb_seq_read() was not called until it
2337 void skb_abort_seq_read(struct skb_seq_state *st)
2340 kunmap_skb_frag(st->frag_data);
2342 EXPORT_SYMBOL(skb_abort_seq_read);
2344 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2346 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2347 struct ts_config *conf,
2348 struct ts_state *state)
2350 return skb_seq_read(offset, text, TS_SKB_CB(state));
2353 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2355 skb_abort_seq_read(TS_SKB_CB(state));
2359 * skb_find_text - Find a text pattern in skb data
2360 * @skb: the buffer to look in
2361 * @from: search offset
2363 * @config: textsearch configuration
2364 * @state: uninitialized textsearch state variable
2366 * Finds a pattern in the skb data according to the specified
2367 * textsearch configuration. Use textsearch_next() to retrieve
2368 * subsequent occurrences of the pattern. Returns the offset
2369 * to the first occurrence or UINT_MAX if no match was found.
2371 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2372 unsigned int to, struct ts_config *config,
2373 struct ts_state *state)
2377 config->get_next_block = skb_ts_get_next_block;
2378 config->finish = skb_ts_finish;
2380 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2382 ret = textsearch_find(config, state);
2383 return (ret <= to - from ? ret : UINT_MAX);
2385 EXPORT_SYMBOL(skb_find_text);
2388 * skb_append_datato_frags: - append the user data to a skb
2389 * @sk: sock structure
2390 * @skb: skb structure to be appened with user data.
2391 * @getfrag: call back function to be used for getting the user data
2392 * @from: pointer to user message iov
2393 * @length: length of the iov message
2395 * Description: This procedure append the user data in the fragment part
2396 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2398 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2399 int (*getfrag)(void *from, char *to, int offset,
2400 int len, int odd, struct sk_buff *skb),
2401 void *from, int length)
2404 skb_frag_t *frag = NULL;
2405 struct page *page = NULL;
2411 /* Return error if we don't have space for new frag */
2412 frg_cnt = skb_shinfo(skb)->nr_frags;
2413 if (frg_cnt >= MAX_SKB_FRAGS)
2416 /* allocate a new page for next frag */
2417 page = alloc_pages(sk->sk_allocation, 0);
2419 /* If alloc_page fails just return failure and caller will
2420 * free previous allocated pages by doing kfree_skb()
2425 /* initialize the next frag */
2426 sk->sk_sndmsg_page = page;
2427 sk->sk_sndmsg_off = 0;
2428 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2429 skb->truesize += PAGE_SIZE;
2430 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2432 /* get the new initialized frag */
2433 frg_cnt = skb_shinfo(skb)->nr_frags;
2434 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2436 /* copy the user data to page */
2437 left = PAGE_SIZE - frag->page_offset;
2438 copy = (length > left)? left : length;
2440 ret = getfrag(from, (page_address(frag->page) +
2441 frag->page_offset + frag->size),
2442 offset, copy, 0, skb);
2446 /* copy was successful so update the size parameters */
2447 sk->sk_sndmsg_off += copy;
2450 skb->data_len += copy;
2454 } while (length > 0);
2458 EXPORT_SYMBOL(skb_append_datato_frags);
2461 * skb_pull_rcsum - pull skb and update receive checksum
2462 * @skb: buffer to update
2463 * @len: length of data pulled
2465 * This function performs an skb_pull on the packet and updates
2466 * the CHECKSUM_COMPLETE checksum. It should be used on
2467 * receive path processing instead of skb_pull unless you know
2468 * that the checksum difference is zero (e.g., a valid IP header)
2469 * or you are setting ip_summed to CHECKSUM_NONE.
2471 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2473 BUG_ON(len > skb->len);
2475 BUG_ON(skb->len < skb->data_len);
2476 skb_postpull_rcsum(skb, skb->data, len);
2477 return skb->data += len;
2480 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2483 * skb_segment - Perform protocol segmentation on skb.
2484 * @skb: buffer to segment
2485 * @features: features for the output path (see dev->features)
2487 * This function performs segmentation on the given skb. It returns
2488 * a pointer to the first in a list of new skbs for the segments.
2489 * In case of error it returns ERR_PTR(err).
2491 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2493 struct sk_buff *segs = NULL;
2494 struct sk_buff *tail = NULL;
2495 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2496 unsigned int mss = skb_shinfo(skb)->gso_size;
2497 unsigned int doffset = skb->data - skb_mac_header(skb);
2498 unsigned int offset = doffset;
2499 unsigned int headroom;
2501 int sg = features & NETIF_F_SG;
2502 int nfrags = skb_shinfo(skb)->nr_frags;
2507 __skb_push(skb, doffset);
2508 headroom = skb_headroom(skb);
2509 pos = skb_headlen(skb);
2512 struct sk_buff *nskb;
2517 len = skb->len - offset;
2521 hsize = skb_headlen(skb) - offset;
2524 if (hsize > len || !sg)
2527 if (!hsize && i >= nfrags) {
2528 BUG_ON(fskb->len != len);
2531 nskb = skb_clone(fskb, GFP_ATOMIC);
2534 if (unlikely(!nskb))
2537 hsize = skb_end_pointer(nskb) - nskb->head;
2538 if (skb_cow_head(nskb, doffset + headroom)) {
2543 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2545 skb_release_head_state(nskb);
2546 __skb_push(nskb, doffset);
2548 nskb = alloc_skb(hsize + doffset + headroom,
2551 if (unlikely(!nskb))
2554 skb_reserve(nskb, headroom);
2555 __skb_put(nskb, doffset);
2564 __copy_skb_header(nskb, skb);
2565 nskb->mac_len = skb->mac_len;
2567 skb_reset_mac_header(nskb);
2568 skb_set_network_header(nskb, skb->mac_len);
2569 nskb->transport_header = (nskb->network_header +
2570 skb_network_header_len(skb));
2571 skb_copy_from_linear_data(skb, nskb->data, doffset);
2573 if (fskb != skb_shinfo(skb)->frag_list)
2577 nskb->ip_summed = CHECKSUM_NONE;
2578 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2584 frag = skb_shinfo(nskb)->frags;
2586 skb_copy_from_linear_data_offset(skb, offset,
2587 skb_put(nskb, hsize), hsize);
2589 while (pos < offset + len && i < nfrags) {
2590 *frag = skb_shinfo(skb)->frags[i];
2591 get_page(frag->page);
2595 frag->page_offset += offset - pos;
2596 frag->size -= offset - pos;
2599 skb_shinfo(nskb)->nr_frags++;
2601 if (pos + size <= offset + len) {
2605 frag->size -= pos + size - (offset + len);
2612 if (pos < offset + len) {
2613 struct sk_buff *fskb2 = fskb;
2615 BUG_ON(pos + fskb->len != offset + len);
2621 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2627 SKB_FRAG_ASSERT(nskb);
2628 skb_shinfo(nskb)->frag_list = fskb2;
2632 nskb->data_len = len - hsize;
2633 nskb->len += nskb->data_len;
2634 nskb->truesize += nskb->data_len;
2635 } while ((offset += len) < skb->len);
2640 while ((skb = segs)) {
2644 return ERR_PTR(err);
2646 EXPORT_SYMBOL_GPL(skb_segment);
2648 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2650 struct sk_buff *p = *head;
2651 struct sk_buff *nskb;
2652 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2653 struct skb_shared_info *pinfo = skb_shinfo(p);
2654 unsigned int headroom;
2655 unsigned int len = skb_gro_len(skb);
2656 unsigned int offset = skb_gro_offset(skb);
2657 unsigned int headlen = skb_headlen(skb);
2659 if (p->len + len >= 65536)
2662 if (pinfo->frag_list)
2664 else if (headlen <= offset) {
2667 int i = skbinfo->nr_frags;
2668 int nr_frags = pinfo->nr_frags + i;
2672 if (nr_frags > MAX_SKB_FRAGS)
2675 pinfo->nr_frags = nr_frags;
2676 skbinfo->nr_frags = 0;
2678 frag = pinfo->frags + nr_frags;
2679 frag2 = skbinfo->frags + i;
2684 frag->page_offset += offset;
2685 frag->size -= offset;
2687 skb->truesize -= skb->data_len;
2688 skb->len -= skb->data_len;
2691 NAPI_GRO_CB(skb)->free = 1;
2695 headroom = skb_headroom(p);
2696 nskb = netdev_alloc_skb(p->dev, headroom + skb_gro_offset(p));
2697 if (unlikely(!nskb))
2700 __copy_skb_header(nskb, p);
2701 nskb->mac_len = p->mac_len;
2703 skb_reserve(nskb, headroom);
2704 __skb_put(nskb, skb_gro_offset(p));
2706 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2707 skb_set_network_header(nskb, skb_network_offset(p));
2708 skb_set_transport_header(nskb, skb_transport_offset(p));
2710 __skb_pull(p, skb_gro_offset(p));
2711 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2712 p->data - skb_mac_header(p));
2714 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2715 skb_shinfo(nskb)->frag_list = p;
2716 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2717 skb_header_release(p);
2720 nskb->data_len += p->len;
2721 nskb->truesize += p->len;
2722 nskb->len += p->len;
2725 nskb->next = p->next;
2731 if (offset > headlen) {
2732 skbinfo->frags[0].page_offset += offset - headlen;
2733 skbinfo->frags[0].size -= offset - headlen;
2737 __skb_pull(skb, offset);
2739 p->prev->next = skb;
2741 skb_header_release(skb);
2744 NAPI_GRO_CB(p)->count++;
2749 NAPI_GRO_CB(skb)->same_flow = 1;
2752 EXPORT_SYMBOL_GPL(skb_gro_receive);
2754 void __init skb_init(void)
2756 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2757 sizeof(struct sk_buff),
2759 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2761 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2762 (2*sizeof(struct sk_buff)) +
2765 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2770 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2771 * @skb: Socket buffer containing the buffers to be mapped
2772 * @sg: The scatter-gather list to map into
2773 * @offset: The offset into the buffer's contents to start mapping
2774 * @len: Length of buffer space to be mapped
2776 * Fill the specified scatter-gather list with mappings/pointers into a
2777 * region of the buffer space attached to a socket buffer.
2780 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2782 int start = skb_headlen(skb);
2783 int i, copy = start - offset;
2784 struct sk_buff *frag_iter;
2790 sg_set_buf(sg, skb->data + offset, copy);
2792 if ((len -= copy) == 0)
2797 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2800 WARN_ON(start > offset + len);
2802 end = start + skb_shinfo(skb)->frags[i].size;
2803 if ((copy = end - offset) > 0) {
2804 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2808 sg_set_page(&sg[elt], frag->page, copy,
2809 frag->page_offset+offset-start);
2818 skb_walk_frags(skb, frag_iter) {
2821 WARN_ON(start > offset + len);
2823 end = start + frag_iter->len;
2824 if ((copy = end - offset) > 0) {
2827 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2829 if ((len -= copy) == 0)
2839 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2841 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2843 sg_mark_end(&sg[nsg - 1]);
2847 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2850 * skb_cow_data - Check that a socket buffer's data buffers are writable
2851 * @skb: The socket buffer to check.
2852 * @tailbits: Amount of trailing space to be added
2853 * @trailer: Returned pointer to the skb where the @tailbits space begins
2855 * Make sure that the data buffers attached to a socket buffer are
2856 * writable. If they are not, private copies are made of the data buffers
2857 * and the socket buffer is set to use these instead.
2859 * If @tailbits is given, make sure that there is space to write @tailbits
2860 * bytes of data beyond current end of socket buffer. @trailer will be
2861 * set to point to the skb in which this space begins.
2863 * The number of scatterlist elements required to completely map the
2864 * COW'd and extended socket buffer will be returned.
2866 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2870 struct sk_buff *skb1, **skb_p;
2872 /* If skb is cloned or its head is paged, reallocate
2873 * head pulling out all the pages (pages are considered not writable
2874 * at the moment even if they are anonymous).
2876 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2877 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2880 /* Easy case. Most of packets will go this way. */
2881 if (!skb_has_frags(skb)) {
2882 /* A little of trouble, not enough of space for trailer.
2883 * This should not happen, when stack is tuned to generate
2884 * good frames. OK, on miss we reallocate and reserve even more
2885 * space, 128 bytes is fair. */
2887 if (skb_tailroom(skb) < tailbits &&
2888 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2896 /* Misery. We are in troubles, going to mincer fragments... */
2899 skb_p = &skb_shinfo(skb)->frag_list;
2902 while ((skb1 = *skb_p) != NULL) {
2905 /* The fragment is partially pulled by someone,
2906 * this can happen on input. Copy it and everything
2909 if (skb_shared(skb1))
2912 /* If the skb is the last, worry about trailer. */
2914 if (skb1->next == NULL && tailbits) {
2915 if (skb_shinfo(skb1)->nr_frags ||
2916 skb_has_frags(skb1) ||
2917 skb_tailroom(skb1) < tailbits)
2918 ntail = tailbits + 128;
2924 skb_shinfo(skb1)->nr_frags ||
2925 skb_has_frags(skb1)) {
2926 struct sk_buff *skb2;
2928 /* Fuck, we are miserable poor guys... */
2930 skb2 = skb_copy(skb1, GFP_ATOMIC);
2932 skb2 = skb_copy_expand(skb1,
2936 if (unlikely(skb2 == NULL))
2940 skb_set_owner_w(skb2, skb1->sk);
2942 /* Looking around. Are we still alive?
2943 * OK, link new skb, drop old one */
2945 skb2->next = skb1->next;
2952 skb_p = &skb1->next;
2957 EXPORT_SYMBOL_GPL(skb_cow_data);
2959 void skb_tstamp_tx(struct sk_buff *orig_skb,
2960 struct skb_shared_hwtstamps *hwtstamps)
2962 struct sock *sk = orig_skb->sk;
2963 struct sock_exterr_skb *serr;
2964 struct sk_buff *skb;
2970 skb = skb_clone(orig_skb, GFP_ATOMIC);
2975 *skb_hwtstamps(skb) =
2979 * no hardware time stamps available,
2980 * so keep the skb_shared_tx and only
2981 * store software time stamp
2983 skb->tstamp = ktime_get_real();
2986 serr = SKB_EXT_ERR(skb);
2987 memset(serr, 0, sizeof(*serr));
2988 serr->ee.ee_errno = ENOMSG;
2989 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
2990 err = sock_queue_err_skb(sk, skb);
2994 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
2998 * skb_partial_csum_set - set up and verify partial csum values for packet
2999 * @skb: the skb to set
3000 * @start: the number of bytes after skb->data to start checksumming.
3001 * @off: the offset from start to place the checksum.
3003 * For untrusted partially-checksummed packets, we need to make sure the values
3004 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3006 * This function checks and sets those values and skb->ip_summed: if this
3007 * returns false you should drop the packet.
3009 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3011 if (unlikely(start > skb_headlen(skb)) ||
3012 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3013 if (net_ratelimit())
3015 "bad partial csum: csum=%u/%u len=%u\n",
3016 start, off, skb_headlen(skb));
3019 skb->ip_summed = CHECKSUM_PARTIAL;
3020 skb->csum_start = skb_headroom(skb) + start;
3021 skb->csum_offset = off;
3024 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3026 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3028 if (net_ratelimit())
3029 pr_warning("%s: received packets cannot be forwarded"
3030 " while LRO is enabled\n", skb->dev->name);
3032 EXPORT_SYMBOL(__skb_warn_lro_forwarding);