2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
74 static struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here, skb->len, sz, skb->head, skb->data,
126 (unsigned long)skb->tail, (unsigned long)skb->end,
127 skb->dev ? skb->dev->name : "<NULL>");
132 * skb_under_panic - private function
137 * Out of line support code for skb_push(). Not user callable.
140 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here, skb->len, sz, skb->head, skb->data,
145 (unsigned long)skb->tail, (unsigned long)skb->end,
146 skb->dev ? skb->dev->name : "<NULL>");
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
171 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
187 /* We do our best to align skb_shared_info on a separate cache
188 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
189 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
190 * Both skb->head and skb_shared_info are cache line aligned.
192 size = SKB_DATA_ALIGN(size);
193 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
194 data = kmalloc_node_track_caller(size, gfp_mask, node);
197 /* kmalloc(size) might give us more room than requested.
198 * Put skb_shared_info exactly at the end of allocated zone,
199 * to allow max possible filling before reallocation.
201 size = SKB_WITH_OVERHEAD(ksize(data));
202 prefetchw(data + size);
205 * Only clear those fields we need to clear, not those that we will
206 * actually initialise below. Hence, don't put any more fields after
207 * the tail pointer in struct sk_buff!
209 memset(skb, 0, offsetof(struct sk_buff, tail));
210 /* Account for allocated memory : skb + skb->head */
211 skb->truesize = SKB_TRUESIZE(size);
212 atomic_set(&skb->users, 1);
215 skb_reset_tail_pointer(skb);
216 skb->end = skb->tail + size;
217 #ifdef NET_SKBUFF_DATA_USES_OFFSET
218 skb->mac_header = ~0U;
221 /* make sure we initialize shinfo sequentially */
222 shinfo = skb_shinfo(skb);
223 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
224 atomic_set(&shinfo->dataref, 1);
225 kmemcheck_annotate_variable(shinfo->destructor_arg);
228 struct sk_buff *child = skb + 1;
229 atomic_t *fclone_ref = (atomic_t *) (child + 1);
231 kmemcheck_annotate_bitfield(child, flags1);
232 kmemcheck_annotate_bitfield(child, flags2);
233 skb->fclone = SKB_FCLONE_ORIG;
234 atomic_set(fclone_ref, 1);
236 child->fclone = SKB_FCLONE_UNAVAILABLE;
241 kmem_cache_free(cache, skb);
245 EXPORT_SYMBOL(__alloc_skb);
248 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
249 * @dev: network device to receive on
250 * @length: length to allocate
251 * @gfp_mask: get_free_pages mask, passed to alloc_skb
253 * Allocate a new &sk_buff and assign it a usage count of one. The
254 * buffer has unspecified headroom built in. Users should allocate
255 * the headroom they think they need without accounting for the
256 * built in space. The built in space is used for optimisations.
258 * %NULL is returned if there is no free memory.
260 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
261 unsigned int length, gfp_t gfp_mask)
265 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
267 skb_reserve(skb, NET_SKB_PAD);
272 EXPORT_SYMBOL(__netdev_alloc_skb);
274 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
277 skb_fill_page_desc(skb, i, page, off, size);
279 skb->data_len += size;
280 skb->truesize += size;
282 EXPORT_SYMBOL(skb_add_rx_frag);
285 * dev_alloc_skb - allocate an skbuff for receiving
286 * @length: length to allocate
288 * Allocate a new &sk_buff and assign it a usage count of one. The
289 * buffer has unspecified headroom built in. Users should allocate
290 * the headroom they think they need without accounting for the
291 * built in space. The built in space is used for optimisations.
293 * %NULL is returned if there is no free memory. Although this function
294 * allocates memory it can be called from an interrupt.
296 struct sk_buff *dev_alloc_skb(unsigned int length)
299 * There is more code here than it seems:
300 * __dev_alloc_skb is an inline
302 return __dev_alloc_skb(length, GFP_ATOMIC);
304 EXPORT_SYMBOL(dev_alloc_skb);
306 static void skb_drop_list(struct sk_buff **listp)
308 struct sk_buff *list = *listp;
313 struct sk_buff *this = list;
319 static inline void skb_drop_fraglist(struct sk_buff *skb)
321 skb_drop_list(&skb_shinfo(skb)->frag_list);
324 static void skb_clone_fraglist(struct sk_buff *skb)
326 struct sk_buff *list;
328 skb_walk_frags(skb, list)
332 static void skb_release_data(struct sk_buff *skb)
335 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
336 &skb_shinfo(skb)->dataref)) {
337 if (skb_shinfo(skb)->nr_frags) {
339 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
340 skb_frag_unref(skb, i);
344 * If skb buf is from userspace, we need to notify the caller
345 * the lower device DMA has done;
347 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
348 struct ubuf_info *uarg;
350 uarg = skb_shinfo(skb)->destructor_arg;
352 uarg->callback(uarg);
355 if (skb_has_frag_list(skb))
356 skb_drop_fraglist(skb);
363 * Free an skbuff by memory without cleaning the state.
365 static void kfree_skbmem(struct sk_buff *skb)
367 struct sk_buff *other;
368 atomic_t *fclone_ref;
370 switch (skb->fclone) {
371 case SKB_FCLONE_UNAVAILABLE:
372 kmem_cache_free(skbuff_head_cache, skb);
375 case SKB_FCLONE_ORIG:
376 fclone_ref = (atomic_t *) (skb + 2);
377 if (atomic_dec_and_test(fclone_ref))
378 kmem_cache_free(skbuff_fclone_cache, skb);
381 case SKB_FCLONE_CLONE:
382 fclone_ref = (atomic_t *) (skb + 1);
385 /* The clone portion is available for
386 * fast-cloning again.
388 skb->fclone = SKB_FCLONE_UNAVAILABLE;
390 if (atomic_dec_and_test(fclone_ref))
391 kmem_cache_free(skbuff_fclone_cache, other);
396 static void skb_release_head_state(struct sk_buff *skb)
400 secpath_put(skb->sp);
402 if (skb->destructor) {
404 skb->destructor(skb);
406 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
407 nf_conntrack_put(skb->nfct);
409 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
410 nf_conntrack_put_reasm(skb->nfct_reasm);
412 #ifdef CONFIG_BRIDGE_NETFILTER
413 nf_bridge_put(skb->nf_bridge);
415 /* XXX: IS this still necessary? - JHS */
416 #ifdef CONFIG_NET_SCHED
418 #ifdef CONFIG_NET_CLS_ACT
424 /* Free everything but the sk_buff shell. */
425 static void skb_release_all(struct sk_buff *skb)
427 skb_release_head_state(skb);
428 skb_release_data(skb);
432 * __kfree_skb - private function
435 * Free an sk_buff. Release anything attached to the buffer.
436 * Clean the state. This is an internal helper function. Users should
437 * always call kfree_skb
440 void __kfree_skb(struct sk_buff *skb)
442 skb_release_all(skb);
445 EXPORT_SYMBOL(__kfree_skb);
448 * kfree_skb - free an sk_buff
449 * @skb: buffer to free
451 * Drop a reference to the buffer and free it if the usage count has
454 void kfree_skb(struct sk_buff *skb)
458 if (likely(atomic_read(&skb->users) == 1))
460 else if (likely(!atomic_dec_and_test(&skb->users)))
462 trace_kfree_skb(skb, __builtin_return_address(0));
465 EXPORT_SYMBOL(kfree_skb);
468 * consume_skb - free an skbuff
469 * @skb: buffer to free
471 * Drop a ref to the buffer and free it if the usage count has hit zero
472 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
473 * is being dropped after a failure and notes that
475 void consume_skb(struct sk_buff *skb)
479 if (likely(atomic_read(&skb->users) == 1))
481 else if (likely(!atomic_dec_and_test(&skb->users)))
483 trace_consume_skb(skb);
486 EXPORT_SYMBOL(consume_skb);
489 * skb_recycle - clean up an skb for reuse
492 * Recycles the skb to be reused as a receive buffer. This
493 * function does any necessary reference count dropping, and
494 * cleans up the skbuff as if it just came from __alloc_skb().
496 void skb_recycle(struct sk_buff *skb)
498 struct skb_shared_info *shinfo;
500 skb_release_head_state(skb);
502 shinfo = skb_shinfo(skb);
503 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
504 atomic_set(&shinfo->dataref, 1);
506 memset(skb, 0, offsetof(struct sk_buff, tail));
507 skb->data = skb->head + NET_SKB_PAD;
508 skb_reset_tail_pointer(skb);
510 EXPORT_SYMBOL(skb_recycle);
513 * skb_recycle_check - check if skb can be reused for receive
515 * @skb_size: minimum receive buffer size
517 * Checks that the skb passed in is not shared or cloned, and
518 * that it is linear and its head portion at least as large as
519 * skb_size so that it can be recycled as a receive buffer.
520 * If these conditions are met, this function does any necessary
521 * reference count dropping and cleans up the skbuff as if it
522 * just came from __alloc_skb().
524 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
526 if (!skb_is_recycleable(skb, skb_size))
533 EXPORT_SYMBOL(skb_recycle_check);
535 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
537 new->tstamp = old->tstamp;
539 new->transport_header = old->transport_header;
540 new->network_header = old->network_header;
541 new->mac_header = old->mac_header;
542 skb_dst_copy(new, old);
543 new->rxhash = old->rxhash;
544 new->ooo_okay = old->ooo_okay;
545 new->l4_rxhash = old->l4_rxhash;
547 new->sp = secpath_get(old->sp);
549 memcpy(new->cb, old->cb, sizeof(old->cb));
550 new->csum = old->csum;
551 new->local_df = old->local_df;
552 new->pkt_type = old->pkt_type;
553 new->ip_summed = old->ip_summed;
554 skb_copy_queue_mapping(new, old);
555 new->priority = old->priority;
556 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
557 new->ipvs_property = old->ipvs_property;
559 new->protocol = old->protocol;
560 new->mark = old->mark;
561 new->skb_iif = old->skb_iif;
563 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
564 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
565 new->nf_trace = old->nf_trace;
567 #ifdef CONFIG_NET_SCHED
568 new->tc_index = old->tc_index;
569 #ifdef CONFIG_NET_CLS_ACT
570 new->tc_verd = old->tc_verd;
573 new->vlan_tci = old->vlan_tci;
575 skb_copy_secmark(new, old);
579 * You should not add any new code to this function. Add it to
580 * __copy_skb_header above instead.
582 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
584 #define C(x) n->x = skb->x
586 n->next = n->prev = NULL;
588 __copy_skb_header(n, skb);
593 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
596 n->destructor = NULL;
602 atomic_set(&n->users, 1);
604 atomic_inc(&(skb_shinfo(skb)->dataref));
612 * skb_morph - morph one skb into another
613 * @dst: the skb to receive the contents
614 * @src: the skb to supply the contents
616 * This is identical to skb_clone except that the target skb is
617 * supplied by the user.
619 * The target skb is returned upon exit.
621 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
623 skb_release_all(dst);
624 return __skb_clone(dst, src);
626 EXPORT_SYMBOL_GPL(skb_morph);
628 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
629 * @skb: the skb to modify
630 * @gfp_mask: allocation priority
632 * This must be called on SKBTX_DEV_ZEROCOPY skb.
633 * It will copy all frags into kernel and drop the reference
634 * to userspace pages.
636 * If this function is called from an interrupt gfp_mask() must be
639 * Returns 0 on success or a negative error code on failure
640 * to allocate kernel memory to copy to.
642 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
645 int num_frags = skb_shinfo(skb)->nr_frags;
646 struct page *page, *head = NULL;
647 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
649 for (i = 0; i < num_frags; i++) {
651 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
653 page = alloc_page(GFP_ATOMIC);
656 struct page *next = (struct page *)head->private;
662 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
663 memcpy(page_address(page),
664 vaddr + f->page_offset, skb_frag_size(f));
665 kunmap_skb_frag(vaddr);
666 page->private = (unsigned long)head;
670 /* skb frags release userspace buffers */
671 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
672 skb_frag_unref(skb, i);
674 uarg->callback(uarg);
676 /* skb frags point to kernel buffers */
677 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
678 __skb_fill_page_desc(skb, i-1, head, 0,
679 skb_shinfo(skb)->frags[i - 1].size);
680 head = (struct page *)head->private;
683 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
689 * skb_clone - duplicate an sk_buff
690 * @skb: buffer to clone
691 * @gfp_mask: allocation priority
693 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
694 * copies share the same packet data but not structure. The new
695 * buffer has a reference count of 1. If the allocation fails the
696 * function returns %NULL otherwise the new buffer is returned.
698 * If this function is called from an interrupt gfp_mask() must be
702 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
706 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
707 if (skb_copy_ubufs(skb, gfp_mask))
712 if (skb->fclone == SKB_FCLONE_ORIG &&
713 n->fclone == SKB_FCLONE_UNAVAILABLE) {
714 atomic_t *fclone_ref = (atomic_t *) (n + 1);
715 n->fclone = SKB_FCLONE_CLONE;
716 atomic_inc(fclone_ref);
718 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
722 kmemcheck_annotate_bitfield(n, flags1);
723 kmemcheck_annotate_bitfield(n, flags2);
724 n->fclone = SKB_FCLONE_UNAVAILABLE;
727 return __skb_clone(n, skb);
729 EXPORT_SYMBOL(skb_clone);
731 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
733 #ifndef NET_SKBUFF_DATA_USES_OFFSET
735 * Shift between the two data areas in bytes
737 unsigned long offset = new->data - old->data;
740 __copy_skb_header(new, old);
742 #ifndef NET_SKBUFF_DATA_USES_OFFSET
743 /* {transport,network,mac}_header are relative to skb->head */
744 new->transport_header += offset;
745 new->network_header += offset;
746 if (skb_mac_header_was_set(new))
747 new->mac_header += offset;
749 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
750 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
751 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
755 * skb_copy - create private copy of an sk_buff
756 * @skb: buffer to copy
757 * @gfp_mask: allocation priority
759 * Make a copy of both an &sk_buff and its data. This is used when the
760 * caller wishes to modify the data and needs a private copy of the
761 * data to alter. Returns %NULL on failure or the pointer to the buffer
762 * on success. The returned buffer has a reference count of 1.
764 * As by-product this function converts non-linear &sk_buff to linear
765 * one, so that &sk_buff becomes completely private and caller is allowed
766 * to modify all the data of returned buffer. This means that this
767 * function is not recommended for use in circumstances when only
768 * header is going to be modified. Use pskb_copy() instead.
771 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
773 int headerlen = skb_headroom(skb);
774 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
775 struct sk_buff *n = alloc_skb(size, gfp_mask);
780 /* Set the data pointer */
781 skb_reserve(n, headerlen);
782 /* Set the tail pointer and length */
783 skb_put(n, skb->len);
785 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
788 copy_skb_header(n, skb);
791 EXPORT_SYMBOL(skb_copy);
794 * pskb_copy - create copy of an sk_buff with private head.
795 * @skb: buffer to copy
796 * @gfp_mask: allocation priority
798 * Make a copy of both an &sk_buff and part of its data, located
799 * in header. Fragmented data remain shared. This is used when
800 * the caller wishes to modify only header of &sk_buff and needs
801 * private copy of the header to alter. Returns %NULL on failure
802 * or the pointer to the buffer on success.
803 * The returned buffer has a reference count of 1.
806 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
808 unsigned int size = skb_end_pointer(skb) - skb->head;
809 struct sk_buff *n = alloc_skb(size, gfp_mask);
814 /* Set the data pointer */
815 skb_reserve(n, skb_headroom(skb));
816 /* Set the tail pointer and length */
817 skb_put(n, skb_headlen(skb));
819 skb_copy_from_linear_data(skb, n->data, n->len);
821 n->truesize += skb->data_len;
822 n->data_len = skb->data_len;
825 if (skb_shinfo(skb)->nr_frags) {
828 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
829 if (skb_copy_ubufs(skb, gfp_mask)) {
835 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
836 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
837 skb_frag_ref(skb, i);
839 skb_shinfo(n)->nr_frags = i;
842 if (skb_has_frag_list(skb)) {
843 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
844 skb_clone_fraglist(n);
847 copy_skb_header(n, skb);
851 EXPORT_SYMBOL(pskb_copy);
854 * pskb_expand_head - reallocate header of &sk_buff
855 * @skb: buffer to reallocate
856 * @nhead: room to add at head
857 * @ntail: room to add at tail
858 * @gfp_mask: allocation priority
860 * Expands (or creates identical copy, if &nhead and &ntail are zero)
861 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
862 * reference count of 1. Returns zero in the case of success or error,
863 * if expansion failed. In the last case, &sk_buff is not changed.
865 * All the pointers pointing into skb header may change and must be
866 * reloaded after call to this function.
869 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
874 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
883 size = SKB_DATA_ALIGN(size);
885 /* Check if we can avoid taking references on fragments if we own
886 * the last reference on skb->head. (see skb_release_data())
891 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
892 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
896 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
897 memmove(skb->head + size, skb_shinfo(skb),
898 offsetof(struct skb_shared_info,
899 frags[skb_shinfo(skb)->nr_frags]));
900 memmove(skb->head + nhead, skb->head,
901 skb_tail_pointer(skb) - skb->head);
906 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
910 size = SKB_WITH_OVERHEAD(ksize(data));
912 /* Copy only real data... and, alas, header. This should be
913 * optimized for the cases when header is void.
915 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
917 memcpy((struct skb_shared_info *)(data + size),
919 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
924 /* copy this zero copy skb frags */
925 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
926 if (skb_copy_ubufs(skb, gfp_mask))
929 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
930 skb_frag_ref(skb, i);
932 if (skb_has_frag_list(skb))
933 skb_clone_fraglist(skb);
935 skb_release_data(skb);
937 off = (data + nhead) - skb->head;
942 #ifdef NET_SKBUFF_DATA_USES_OFFSET
946 skb->end = skb->head + size;
948 /* {transport,network,mac}_header and tail are relative to skb->head */
950 skb->transport_header += off;
951 skb->network_header += off;
952 if (skb_mac_header_was_set(skb))
953 skb->mac_header += off;
954 /* Only adjust this if it actually is csum_start rather than csum */
955 if (skb->ip_summed == CHECKSUM_PARTIAL)
956 skb->csum_start += nhead;
960 atomic_set(&skb_shinfo(skb)->dataref, 1);
968 EXPORT_SYMBOL(pskb_expand_head);
970 /* Make private copy of skb with writable head and some headroom */
972 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
974 struct sk_buff *skb2;
975 int delta = headroom - skb_headroom(skb);
978 skb2 = pskb_copy(skb, GFP_ATOMIC);
980 skb2 = skb_clone(skb, GFP_ATOMIC);
981 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
989 EXPORT_SYMBOL(skb_realloc_headroom);
992 * skb_copy_expand - copy and expand sk_buff
993 * @skb: buffer to copy
994 * @newheadroom: new free bytes at head
995 * @newtailroom: new free bytes at tail
996 * @gfp_mask: allocation priority
998 * Make a copy of both an &sk_buff and its data and while doing so
999 * allocate additional space.
1001 * This is used when the caller wishes to modify the data and needs a
1002 * private copy of the data to alter as well as more space for new fields.
1003 * Returns %NULL on failure or the pointer to the buffer
1004 * on success. The returned buffer has a reference count of 1.
1006 * You must pass %GFP_ATOMIC as the allocation priority if this function
1007 * is called from an interrupt.
1009 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1010 int newheadroom, int newtailroom,
1014 * Allocate the copy buffer
1016 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1018 int oldheadroom = skb_headroom(skb);
1019 int head_copy_len, head_copy_off;
1025 skb_reserve(n, newheadroom);
1027 /* Set the tail pointer and length */
1028 skb_put(n, skb->len);
1030 head_copy_len = oldheadroom;
1032 if (newheadroom <= head_copy_len)
1033 head_copy_len = newheadroom;
1035 head_copy_off = newheadroom - head_copy_len;
1037 /* Copy the linear header and data. */
1038 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1039 skb->len + head_copy_len))
1042 copy_skb_header(n, skb);
1044 off = newheadroom - oldheadroom;
1045 if (n->ip_summed == CHECKSUM_PARTIAL)
1046 n->csum_start += off;
1047 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1048 n->transport_header += off;
1049 n->network_header += off;
1050 if (skb_mac_header_was_set(skb))
1051 n->mac_header += off;
1056 EXPORT_SYMBOL(skb_copy_expand);
1059 * skb_pad - zero pad the tail of an skb
1060 * @skb: buffer to pad
1061 * @pad: space to pad
1063 * Ensure that a buffer is followed by a padding area that is zero
1064 * filled. Used by network drivers which may DMA or transfer data
1065 * beyond the buffer end onto the wire.
1067 * May return error in out of memory cases. The skb is freed on error.
1070 int skb_pad(struct sk_buff *skb, int pad)
1075 /* If the skbuff is non linear tailroom is always zero.. */
1076 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1077 memset(skb->data+skb->len, 0, pad);
1081 ntail = skb->data_len + pad - (skb->end - skb->tail);
1082 if (likely(skb_cloned(skb) || ntail > 0)) {
1083 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1088 /* FIXME: The use of this function with non-linear skb's really needs
1091 err = skb_linearize(skb);
1095 memset(skb->data + skb->len, 0, pad);
1102 EXPORT_SYMBOL(skb_pad);
1105 * skb_put - add data to a buffer
1106 * @skb: buffer to use
1107 * @len: amount of data to add
1109 * This function extends the used data area of the buffer. If this would
1110 * exceed the total buffer size the kernel will panic. A pointer to the
1111 * first byte of the extra data is returned.
1113 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1115 unsigned char *tmp = skb_tail_pointer(skb);
1116 SKB_LINEAR_ASSERT(skb);
1119 if (unlikely(skb->tail > skb->end))
1120 skb_over_panic(skb, len, __builtin_return_address(0));
1123 EXPORT_SYMBOL(skb_put);
1126 * skb_push - add data to the start of a buffer
1127 * @skb: buffer to use
1128 * @len: amount of data to add
1130 * This function extends the used data area of the buffer at the buffer
1131 * start. If this would exceed the total buffer headroom the kernel will
1132 * panic. A pointer to the first byte of the extra data is returned.
1134 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1138 if (unlikely(skb->data<skb->head))
1139 skb_under_panic(skb, len, __builtin_return_address(0));
1142 EXPORT_SYMBOL(skb_push);
1145 * skb_pull - remove data from the start of a buffer
1146 * @skb: buffer to use
1147 * @len: amount of data to remove
1149 * This function removes data from the start of a buffer, returning
1150 * the memory to the headroom. A pointer to the next data in the buffer
1151 * is returned. Once the data has been pulled future pushes will overwrite
1154 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1156 return skb_pull_inline(skb, len);
1158 EXPORT_SYMBOL(skb_pull);
1161 * skb_trim - remove end from a buffer
1162 * @skb: buffer to alter
1165 * Cut the length of a buffer down by removing data from the tail. If
1166 * the buffer is already under the length specified it is not modified.
1167 * The skb must be linear.
1169 void skb_trim(struct sk_buff *skb, unsigned int len)
1172 __skb_trim(skb, len);
1174 EXPORT_SYMBOL(skb_trim);
1176 /* Trims skb to length len. It can change skb pointers.
1179 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1181 struct sk_buff **fragp;
1182 struct sk_buff *frag;
1183 int offset = skb_headlen(skb);
1184 int nfrags = skb_shinfo(skb)->nr_frags;
1188 if (skb_cloned(skb) &&
1189 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1196 for (; i < nfrags; i++) {
1197 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1204 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1207 skb_shinfo(skb)->nr_frags = i;
1209 for (; i < nfrags; i++)
1210 skb_frag_unref(skb, i);
1212 if (skb_has_frag_list(skb))
1213 skb_drop_fraglist(skb);
1217 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1218 fragp = &frag->next) {
1219 int end = offset + frag->len;
1221 if (skb_shared(frag)) {
1222 struct sk_buff *nfrag;
1224 nfrag = skb_clone(frag, GFP_ATOMIC);
1225 if (unlikely(!nfrag))
1228 nfrag->next = frag->next;
1240 unlikely((err = pskb_trim(frag, len - offset))))
1244 skb_drop_list(&frag->next);
1249 if (len > skb_headlen(skb)) {
1250 skb->data_len -= skb->len - len;
1255 skb_set_tail_pointer(skb, len);
1260 EXPORT_SYMBOL(___pskb_trim);
1263 * __pskb_pull_tail - advance tail of skb header
1264 * @skb: buffer to reallocate
1265 * @delta: number of bytes to advance tail
1267 * The function makes a sense only on a fragmented &sk_buff,
1268 * it expands header moving its tail forward and copying necessary
1269 * data from fragmented part.
1271 * &sk_buff MUST have reference count of 1.
1273 * Returns %NULL (and &sk_buff does not change) if pull failed
1274 * or value of new tail of skb in the case of success.
1276 * All the pointers pointing into skb header may change and must be
1277 * reloaded after call to this function.
1280 /* Moves tail of skb head forward, copying data from fragmented part,
1281 * when it is necessary.
1282 * 1. It may fail due to malloc failure.
1283 * 2. It may change skb pointers.
1285 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1287 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1289 /* If skb has not enough free space at tail, get new one
1290 * plus 128 bytes for future expansions. If we have enough
1291 * room at tail, reallocate without expansion only if skb is cloned.
1293 int i, k, eat = (skb->tail + delta) - skb->end;
1295 if (eat > 0 || skb_cloned(skb)) {
1296 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1301 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1304 /* Optimization: no fragments, no reasons to preestimate
1305 * size of pulled pages. Superb.
1307 if (!skb_has_frag_list(skb))
1310 /* Estimate size of pulled pages. */
1312 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1313 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1320 /* If we need update frag list, we are in troubles.
1321 * Certainly, it possible to add an offset to skb data,
1322 * but taking into account that pulling is expected to
1323 * be very rare operation, it is worth to fight against
1324 * further bloating skb head and crucify ourselves here instead.
1325 * Pure masohism, indeed. 8)8)
1328 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1329 struct sk_buff *clone = NULL;
1330 struct sk_buff *insp = NULL;
1335 if (list->len <= eat) {
1336 /* Eaten as whole. */
1341 /* Eaten partially. */
1343 if (skb_shared(list)) {
1344 /* Sucks! We need to fork list. :-( */
1345 clone = skb_clone(list, GFP_ATOMIC);
1351 /* This may be pulled without
1355 if (!pskb_pull(list, eat)) {
1363 /* Free pulled out fragments. */
1364 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1365 skb_shinfo(skb)->frag_list = list->next;
1368 /* And insert new clone at head. */
1371 skb_shinfo(skb)->frag_list = clone;
1374 /* Success! Now we may commit changes to skb data. */
1379 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1380 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1383 skb_frag_unref(skb, i);
1386 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1388 skb_shinfo(skb)->frags[k].page_offset += eat;
1389 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1395 skb_shinfo(skb)->nr_frags = k;
1398 skb->data_len -= delta;
1400 return skb_tail_pointer(skb);
1402 EXPORT_SYMBOL(__pskb_pull_tail);
1405 * skb_copy_bits - copy bits from skb to kernel buffer
1407 * @offset: offset in source
1408 * @to: destination buffer
1409 * @len: number of bytes to copy
1411 * Copy the specified number of bytes from the source skb to the
1412 * destination buffer.
1415 * If its prototype is ever changed,
1416 * check arch/{*}/net/{*}.S files,
1417 * since it is called from BPF assembly code.
1419 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1421 int start = skb_headlen(skb);
1422 struct sk_buff *frag_iter;
1425 if (offset > (int)skb->len - len)
1429 if ((copy = start - offset) > 0) {
1432 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1433 if ((len -= copy) == 0)
1439 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1442 WARN_ON(start > offset + len);
1444 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1445 if ((copy = end - offset) > 0) {
1451 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1453 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1454 offset - start, copy);
1455 kunmap_skb_frag(vaddr);
1457 if ((len -= copy) == 0)
1465 skb_walk_frags(skb, frag_iter) {
1468 WARN_ON(start > offset + len);
1470 end = start + frag_iter->len;
1471 if ((copy = end - offset) > 0) {
1474 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1476 if ((len -= copy) == 0)
1490 EXPORT_SYMBOL(skb_copy_bits);
1493 * Callback from splice_to_pipe(), if we need to release some pages
1494 * at the end of the spd in case we error'ed out in filling the pipe.
1496 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1498 put_page(spd->pages[i]);
1501 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1502 unsigned int *offset,
1503 struct sk_buff *skb, struct sock *sk)
1505 struct page *p = sk->sk_sndmsg_page;
1510 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1514 off = sk->sk_sndmsg_off = 0;
1515 /* hold one ref to this page until it's full */
1519 off = sk->sk_sndmsg_off;
1520 mlen = PAGE_SIZE - off;
1521 if (mlen < 64 && mlen < *len) {
1526 *len = min_t(unsigned int, *len, mlen);
1529 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1530 sk->sk_sndmsg_off += *len;
1538 * Fill page/offset/length into spd, if it can hold more pages.
1540 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1541 struct pipe_inode_info *pipe, struct page *page,
1542 unsigned int *len, unsigned int offset,
1543 struct sk_buff *skb, int linear,
1546 if (unlikely(spd->nr_pages == pipe->buffers))
1550 page = linear_to_page(page, len, &offset, skb, sk);
1556 spd->pages[spd->nr_pages] = page;
1557 spd->partial[spd->nr_pages].len = *len;
1558 spd->partial[spd->nr_pages].offset = offset;
1564 static inline void __segment_seek(struct page **page, unsigned int *poff,
1565 unsigned int *plen, unsigned int off)
1570 n = *poff / PAGE_SIZE;
1572 *page = nth_page(*page, n);
1574 *poff = *poff % PAGE_SIZE;
1578 static inline int __splice_segment(struct page *page, unsigned int poff,
1579 unsigned int plen, unsigned int *off,
1580 unsigned int *len, struct sk_buff *skb,
1581 struct splice_pipe_desc *spd, int linear,
1583 struct pipe_inode_info *pipe)
1588 /* skip this segment if already processed */
1594 /* ignore any bits we already processed */
1596 __segment_seek(&page, &poff, &plen, *off);
1601 unsigned int flen = min(*len, plen);
1603 /* the linear region may spread across several pages */
1604 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1606 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1609 __segment_seek(&page, &poff, &plen, flen);
1612 } while (*len && plen);
1618 * Map linear and fragment data from the skb to spd. It reports failure if the
1619 * pipe is full or if we already spliced the requested length.
1621 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1622 unsigned int *offset, unsigned int *len,
1623 struct splice_pipe_desc *spd, struct sock *sk)
1628 * map the linear part
1630 if (__splice_segment(virt_to_page(skb->data),
1631 (unsigned long) skb->data & (PAGE_SIZE - 1),
1633 offset, len, skb, spd, 1, sk, pipe))
1637 * then map the fragments
1639 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1640 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1642 if (__splice_segment(skb_frag_page(f),
1643 f->page_offset, skb_frag_size(f),
1644 offset, len, skb, spd, 0, sk, pipe))
1652 * Map data from the skb to a pipe. Should handle both the linear part,
1653 * the fragments, and the frag list. It does NOT handle frag lists within
1654 * the frag list, if such a thing exists. We'd probably need to recurse to
1655 * handle that cleanly.
1657 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1658 struct pipe_inode_info *pipe, unsigned int tlen,
1661 struct partial_page partial[PIPE_DEF_BUFFERS];
1662 struct page *pages[PIPE_DEF_BUFFERS];
1663 struct splice_pipe_desc spd = {
1667 .ops = &sock_pipe_buf_ops,
1668 .spd_release = sock_spd_release,
1670 struct sk_buff *frag_iter;
1671 struct sock *sk = skb->sk;
1674 if (splice_grow_spd(pipe, &spd))
1678 * __skb_splice_bits() only fails if the output has no room left,
1679 * so no point in going over the frag_list for the error case.
1681 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1687 * now see if we have a frag_list to map
1689 skb_walk_frags(skb, frag_iter) {
1692 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1699 * Drop the socket lock, otherwise we have reverse
1700 * locking dependencies between sk_lock and i_mutex
1701 * here as compared to sendfile(). We enter here
1702 * with the socket lock held, and splice_to_pipe() will
1703 * grab the pipe inode lock. For sendfile() emulation,
1704 * we call into ->sendpage() with the i_mutex lock held
1705 * and networking will grab the socket lock.
1708 ret = splice_to_pipe(pipe, &spd);
1712 splice_shrink_spd(pipe, &spd);
1717 * skb_store_bits - store bits from kernel buffer to skb
1718 * @skb: destination buffer
1719 * @offset: offset in destination
1720 * @from: source buffer
1721 * @len: number of bytes to copy
1723 * Copy the specified number of bytes from the source buffer to the
1724 * destination skb. This function handles all the messy bits of
1725 * traversing fragment lists and such.
1728 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1730 int start = skb_headlen(skb);
1731 struct sk_buff *frag_iter;
1734 if (offset > (int)skb->len - len)
1737 if ((copy = start - offset) > 0) {
1740 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1741 if ((len -= copy) == 0)
1747 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1748 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1751 WARN_ON(start > offset + len);
1753 end = start + skb_frag_size(frag);
1754 if ((copy = end - offset) > 0) {
1760 vaddr = kmap_skb_frag(frag);
1761 memcpy(vaddr + frag->page_offset + offset - start,
1763 kunmap_skb_frag(vaddr);
1765 if ((len -= copy) == 0)
1773 skb_walk_frags(skb, frag_iter) {
1776 WARN_ON(start > offset + len);
1778 end = start + frag_iter->len;
1779 if ((copy = end - offset) > 0) {
1782 if (skb_store_bits(frag_iter, offset - start,
1785 if ((len -= copy) == 0)
1798 EXPORT_SYMBOL(skb_store_bits);
1800 /* Checksum skb data. */
1802 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1803 int len, __wsum csum)
1805 int start = skb_headlen(skb);
1806 int i, copy = start - offset;
1807 struct sk_buff *frag_iter;
1810 /* Checksum header. */
1814 csum = csum_partial(skb->data + offset, copy, csum);
1815 if ((len -= copy) == 0)
1821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1824 WARN_ON(start > offset + len);
1826 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1827 if ((copy = end - offset) > 0) {
1830 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1834 vaddr = kmap_skb_frag(frag);
1835 csum2 = csum_partial(vaddr + frag->page_offset +
1836 offset - start, copy, 0);
1837 kunmap_skb_frag(vaddr);
1838 csum = csum_block_add(csum, csum2, pos);
1847 skb_walk_frags(skb, frag_iter) {
1850 WARN_ON(start > offset + len);
1852 end = start + frag_iter->len;
1853 if ((copy = end - offset) > 0) {
1857 csum2 = skb_checksum(frag_iter, offset - start,
1859 csum = csum_block_add(csum, csum2, pos);
1860 if ((len -= copy) == 0)
1871 EXPORT_SYMBOL(skb_checksum);
1873 /* Both of above in one bottle. */
1875 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1876 u8 *to, int len, __wsum csum)
1878 int start = skb_headlen(skb);
1879 int i, copy = start - offset;
1880 struct sk_buff *frag_iter;
1887 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1889 if ((len -= copy) == 0)
1896 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1899 WARN_ON(start > offset + len);
1901 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1902 if ((copy = end - offset) > 0) {
1905 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1909 vaddr = kmap_skb_frag(frag);
1910 csum2 = csum_partial_copy_nocheck(vaddr +
1914 kunmap_skb_frag(vaddr);
1915 csum = csum_block_add(csum, csum2, pos);
1925 skb_walk_frags(skb, frag_iter) {
1929 WARN_ON(start > offset + len);
1931 end = start + frag_iter->len;
1932 if ((copy = end - offset) > 0) {
1935 csum2 = skb_copy_and_csum_bits(frag_iter,
1938 csum = csum_block_add(csum, csum2, pos);
1939 if ((len -= copy) == 0)
1950 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1952 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1957 if (skb->ip_summed == CHECKSUM_PARTIAL)
1958 csstart = skb_checksum_start_offset(skb);
1960 csstart = skb_headlen(skb);
1962 BUG_ON(csstart > skb_headlen(skb));
1964 skb_copy_from_linear_data(skb, to, csstart);
1967 if (csstart != skb->len)
1968 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1969 skb->len - csstart, 0);
1971 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1972 long csstuff = csstart + skb->csum_offset;
1974 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1977 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1980 * skb_dequeue - remove from the head of the queue
1981 * @list: list to dequeue from
1983 * Remove the head of the list. The list lock is taken so the function
1984 * may be used safely with other locking list functions. The head item is
1985 * returned or %NULL if the list is empty.
1988 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1990 unsigned long flags;
1991 struct sk_buff *result;
1993 spin_lock_irqsave(&list->lock, flags);
1994 result = __skb_dequeue(list);
1995 spin_unlock_irqrestore(&list->lock, flags);
1998 EXPORT_SYMBOL(skb_dequeue);
2001 * skb_dequeue_tail - remove from the tail of the queue
2002 * @list: list to dequeue from
2004 * Remove the tail of the list. The list lock is taken so the function
2005 * may be used safely with other locking list functions. The tail item is
2006 * returned or %NULL if the list is empty.
2008 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2010 unsigned long flags;
2011 struct sk_buff *result;
2013 spin_lock_irqsave(&list->lock, flags);
2014 result = __skb_dequeue_tail(list);
2015 spin_unlock_irqrestore(&list->lock, flags);
2018 EXPORT_SYMBOL(skb_dequeue_tail);
2021 * skb_queue_purge - empty a list
2022 * @list: list to empty
2024 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2025 * the list and one reference dropped. This function takes the list
2026 * lock and is atomic with respect to other list locking functions.
2028 void skb_queue_purge(struct sk_buff_head *list)
2030 struct sk_buff *skb;
2031 while ((skb = skb_dequeue(list)) != NULL)
2034 EXPORT_SYMBOL(skb_queue_purge);
2037 * skb_queue_head - queue a buffer at the list head
2038 * @list: list to use
2039 * @newsk: buffer to queue
2041 * Queue a buffer at the start of the list. This function takes the
2042 * list lock and can be used safely with other locking &sk_buff functions
2045 * A buffer cannot be placed on two lists at the same time.
2047 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2049 unsigned long flags;
2051 spin_lock_irqsave(&list->lock, flags);
2052 __skb_queue_head(list, newsk);
2053 spin_unlock_irqrestore(&list->lock, flags);
2055 EXPORT_SYMBOL(skb_queue_head);
2058 * skb_queue_tail - queue a buffer at the list tail
2059 * @list: list to use
2060 * @newsk: buffer to queue
2062 * Queue a buffer at the tail of the list. This function takes the
2063 * list lock and can be used safely with other locking &sk_buff functions
2066 * A buffer cannot be placed on two lists at the same time.
2068 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2070 unsigned long flags;
2072 spin_lock_irqsave(&list->lock, flags);
2073 __skb_queue_tail(list, newsk);
2074 spin_unlock_irqrestore(&list->lock, flags);
2076 EXPORT_SYMBOL(skb_queue_tail);
2079 * skb_unlink - remove a buffer from a list
2080 * @skb: buffer to remove
2081 * @list: list to use
2083 * Remove a packet from a list. The list locks are taken and this
2084 * function is atomic with respect to other list locked calls
2086 * You must know what list the SKB is on.
2088 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2090 unsigned long flags;
2092 spin_lock_irqsave(&list->lock, flags);
2093 __skb_unlink(skb, list);
2094 spin_unlock_irqrestore(&list->lock, flags);
2096 EXPORT_SYMBOL(skb_unlink);
2099 * skb_append - append a buffer
2100 * @old: buffer to insert after
2101 * @newsk: buffer to insert
2102 * @list: list to use
2104 * Place a packet after a given packet in a list. The list locks are taken
2105 * and this function is atomic with respect to other list locked calls.
2106 * A buffer cannot be placed on two lists at the same time.
2108 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2110 unsigned long flags;
2112 spin_lock_irqsave(&list->lock, flags);
2113 __skb_queue_after(list, old, newsk);
2114 spin_unlock_irqrestore(&list->lock, flags);
2116 EXPORT_SYMBOL(skb_append);
2119 * skb_insert - insert a buffer
2120 * @old: buffer to insert before
2121 * @newsk: buffer to insert
2122 * @list: list to use
2124 * Place a packet before a given packet in a list. The list locks are
2125 * taken and this function is atomic with respect to other list locked
2128 * A buffer cannot be placed on two lists at the same time.
2130 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2132 unsigned long flags;
2134 spin_lock_irqsave(&list->lock, flags);
2135 __skb_insert(newsk, old->prev, old, list);
2136 spin_unlock_irqrestore(&list->lock, flags);
2138 EXPORT_SYMBOL(skb_insert);
2140 static inline void skb_split_inside_header(struct sk_buff *skb,
2141 struct sk_buff* skb1,
2142 const u32 len, const int pos)
2146 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2148 /* And move data appendix as is. */
2149 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2150 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2152 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2153 skb_shinfo(skb)->nr_frags = 0;
2154 skb1->data_len = skb->data_len;
2155 skb1->len += skb1->data_len;
2158 skb_set_tail_pointer(skb, len);
2161 static inline void skb_split_no_header(struct sk_buff *skb,
2162 struct sk_buff* skb1,
2163 const u32 len, int pos)
2166 const int nfrags = skb_shinfo(skb)->nr_frags;
2168 skb_shinfo(skb)->nr_frags = 0;
2169 skb1->len = skb1->data_len = skb->len - len;
2171 skb->data_len = len - pos;
2173 for (i = 0; i < nfrags; i++) {
2174 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2176 if (pos + size > len) {
2177 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2181 * We have two variants in this case:
2182 * 1. Move all the frag to the second
2183 * part, if it is possible. F.e.
2184 * this approach is mandatory for TUX,
2185 * where splitting is expensive.
2186 * 2. Split is accurately. We make this.
2188 skb_frag_ref(skb, i);
2189 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2190 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2191 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2192 skb_shinfo(skb)->nr_frags++;
2196 skb_shinfo(skb)->nr_frags++;
2199 skb_shinfo(skb1)->nr_frags = k;
2203 * skb_split - Split fragmented skb to two parts at length len.
2204 * @skb: the buffer to split
2205 * @skb1: the buffer to receive the second part
2206 * @len: new length for skb
2208 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2210 int pos = skb_headlen(skb);
2212 if (len < pos) /* Split line is inside header. */
2213 skb_split_inside_header(skb, skb1, len, pos);
2214 else /* Second chunk has no header, nothing to copy. */
2215 skb_split_no_header(skb, skb1, len, pos);
2217 EXPORT_SYMBOL(skb_split);
2219 /* Shifting from/to a cloned skb is a no-go.
2221 * Caller cannot keep skb_shinfo related pointers past calling here!
2223 static int skb_prepare_for_shift(struct sk_buff *skb)
2225 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2229 * skb_shift - Shifts paged data partially from skb to another
2230 * @tgt: buffer into which tail data gets added
2231 * @skb: buffer from which the paged data comes from
2232 * @shiftlen: shift up to this many bytes
2234 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2235 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2236 * It's up to caller to free skb if everything was shifted.
2238 * If @tgt runs out of frags, the whole operation is aborted.
2240 * Skb cannot include anything else but paged data while tgt is allowed
2241 * to have non-paged data as well.
2243 * TODO: full sized shift could be optimized but that would need
2244 * specialized skb free'er to handle frags without up-to-date nr_frags.
2246 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2248 int from, to, merge, todo;
2249 struct skb_frag_struct *fragfrom, *fragto;
2251 BUG_ON(shiftlen > skb->len);
2252 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2256 to = skb_shinfo(tgt)->nr_frags;
2257 fragfrom = &skb_shinfo(skb)->frags[from];
2259 /* Actual merge is delayed until the point when we know we can
2260 * commit all, so that we don't have to undo partial changes
2263 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2264 fragfrom->page_offset)) {
2269 todo -= skb_frag_size(fragfrom);
2271 if (skb_prepare_for_shift(skb) ||
2272 skb_prepare_for_shift(tgt))
2275 /* All previous frag pointers might be stale! */
2276 fragfrom = &skb_shinfo(skb)->frags[from];
2277 fragto = &skb_shinfo(tgt)->frags[merge];
2279 skb_frag_size_add(fragto, shiftlen);
2280 skb_frag_size_sub(fragfrom, shiftlen);
2281 fragfrom->page_offset += shiftlen;
2289 /* Skip full, not-fitting skb to avoid expensive operations */
2290 if ((shiftlen == skb->len) &&
2291 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2294 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2297 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2298 if (to == MAX_SKB_FRAGS)
2301 fragfrom = &skb_shinfo(skb)->frags[from];
2302 fragto = &skb_shinfo(tgt)->frags[to];
2304 if (todo >= skb_frag_size(fragfrom)) {
2305 *fragto = *fragfrom;
2306 todo -= skb_frag_size(fragfrom);
2311 __skb_frag_ref(fragfrom);
2312 fragto->page = fragfrom->page;
2313 fragto->page_offset = fragfrom->page_offset;
2314 skb_frag_size_set(fragto, todo);
2316 fragfrom->page_offset += todo;
2317 skb_frag_size_sub(fragfrom, todo);
2325 /* Ready to "commit" this state change to tgt */
2326 skb_shinfo(tgt)->nr_frags = to;
2329 fragfrom = &skb_shinfo(skb)->frags[0];
2330 fragto = &skb_shinfo(tgt)->frags[merge];
2332 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2333 __skb_frag_unref(fragfrom);
2336 /* Reposition in the original skb */
2338 while (from < skb_shinfo(skb)->nr_frags)
2339 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2340 skb_shinfo(skb)->nr_frags = to;
2342 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2345 /* Most likely the tgt won't ever need its checksum anymore, skb on
2346 * the other hand might need it if it needs to be resent
2348 tgt->ip_summed = CHECKSUM_PARTIAL;
2349 skb->ip_summed = CHECKSUM_PARTIAL;
2351 /* Yak, is it really working this way? Some helper please? */
2352 skb->len -= shiftlen;
2353 skb->data_len -= shiftlen;
2354 skb->truesize -= shiftlen;
2355 tgt->len += shiftlen;
2356 tgt->data_len += shiftlen;
2357 tgt->truesize += shiftlen;
2363 * skb_prepare_seq_read - Prepare a sequential read of skb data
2364 * @skb: the buffer to read
2365 * @from: lower offset of data to be read
2366 * @to: upper offset of data to be read
2367 * @st: state variable
2369 * Initializes the specified state variable. Must be called before
2370 * invoking skb_seq_read() for the first time.
2372 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2373 unsigned int to, struct skb_seq_state *st)
2375 st->lower_offset = from;
2376 st->upper_offset = to;
2377 st->root_skb = st->cur_skb = skb;
2378 st->frag_idx = st->stepped_offset = 0;
2379 st->frag_data = NULL;
2381 EXPORT_SYMBOL(skb_prepare_seq_read);
2384 * skb_seq_read - Sequentially read skb data
2385 * @consumed: number of bytes consumed by the caller so far
2386 * @data: destination pointer for data to be returned
2387 * @st: state variable
2389 * Reads a block of skb data at &consumed relative to the
2390 * lower offset specified to skb_prepare_seq_read(). Assigns
2391 * the head of the data block to &data and returns the length
2392 * of the block or 0 if the end of the skb data or the upper
2393 * offset has been reached.
2395 * The caller is not required to consume all of the data
2396 * returned, i.e. &consumed is typically set to the number
2397 * of bytes already consumed and the next call to
2398 * skb_seq_read() will return the remaining part of the block.
2400 * Note 1: The size of each block of data returned can be arbitrary,
2401 * this limitation is the cost for zerocopy seqeuental
2402 * reads of potentially non linear data.
2404 * Note 2: Fragment lists within fragments are not implemented
2405 * at the moment, state->root_skb could be replaced with
2406 * a stack for this purpose.
2408 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2409 struct skb_seq_state *st)
2411 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2414 if (unlikely(abs_offset >= st->upper_offset))
2418 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2420 if (abs_offset < block_limit && !st->frag_data) {
2421 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2422 return block_limit - abs_offset;
2425 if (st->frag_idx == 0 && !st->frag_data)
2426 st->stepped_offset += skb_headlen(st->cur_skb);
2428 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2429 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2430 block_limit = skb_frag_size(frag) + st->stepped_offset;
2432 if (abs_offset < block_limit) {
2434 st->frag_data = kmap_skb_frag(frag);
2436 *data = (u8 *) st->frag_data + frag->page_offset +
2437 (abs_offset - st->stepped_offset);
2439 return block_limit - abs_offset;
2442 if (st->frag_data) {
2443 kunmap_skb_frag(st->frag_data);
2444 st->frag_data = NULL;
2448 st->stepped_offset += skb_frag_size(frag);
2451 if (st->frag_data) {
2452 kunmap_skb_frag(st->frag_data);
2453 st->frag_data = NULL;
2456 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2457 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2460 } else if (st->cur_skb->next) {
2461 st->cur_skb = st->cur_skb->next;
2468 EXPORT_SYMBOL(skb_seq_read);
2471 * skb_abort_seq_read - Abort a sequential read of skb data
2472 * @st: state variable
2474 * Must be called if skb_seq_read() was not called until it
2477 void skb_abort_seq_read(struct skb_seq_state *st)
2480 kunmap_skb_frag(st->frag_data);
2482 EXPORT_SYMBOL(skb_abort_seq_read);
2484 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2486 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2487 struct ts_config *conf,
2488 struct ts_state *state)
2490 return skb_seq_read(offset, text, TS_SKB_CB(state));
2493 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2495 skb_abort_seq_read(TS_SKB_CB(state));
2499 * skb_find_text - Find a text pattern in skb data
2500 * @skb: the buffer to look in
2501 * @from: search offset
2503 * @config: textsearch configuration
2504 * @state: uninitialized textsearch state variable
2506 * Finds a pattern in the skb data according to the specified
2507 * textsearch configuration. Use textsearch_next() to retrieve
2508 * subsequent occurrences of the pattern. Returns the offset
2509 * to the first occurrence or UINT_MAX if no match was found.
2511 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2512 unsigned int to, struct ts_config *config,
2513 struct ts_state *state)
2517 config->get_next_block = skb_ts_get_next_block;
2518 config->finish = skb_ts_finish;
2520 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2522 ret = textsearch_find(config, state);
2523 return (ret <= to - from ? ret : UINT_MAX);
2525 EXPORT_SYMBOL(skb_find_text);
2528 * skb_append_datato_frags: - append the user data to a skb
2529 * @sk: sock structure
2530 * @skb: skb structure to be appened with user data.
2531 * @getfrag: call back function to be used for getting the user data
2532 * @from: pointer to user message iov
2533 * @length: length of the iov message
2535 * Description: This procedure append the user data in the fragment part
2536 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2538 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2539 int (*getfrag)(void *from, char *to, int offset,
2540 int len, int odd, struct sk_buff *skb),
2541 void *from, int length)
2544 skb_frag_t *frag = NULL;
2545 struct page *page = NULL;
2551 /* Return error if we don't have space for new frag */
2552 frg_cnt = skb_shinfo(skb)->nr_frags;
2553 if (frg_cnt >= MAX_SKB_FRAGS)
2556 /* allocate a new page for next frag */
2557 page = alloc_pages(sk->sk_allocation, 0);
2559 /* If alloc_page fails just return failure and caller will
2560 * free previous allocated pages by doing kfree_skb()
2565 /* initialize the next frag */
2566 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2567 skb->truesize += PAGE_SIZE;
2568 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2570 /* get the new initialized frag */
2571 frg_cnt = skb_shinfo(skb)->nr_frags;
2572 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2574 /* copy the user data to page */
2575 left = PAGE_SIZE - frag->page_offset;
2576 copy = (length > left)? left : length;
2578 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2579 offset, copy, 0, skb);
2583 /* copy was successful so update the size parameters */
2584 skb_frag_size_add(frag, copy);
2586 skb->data_len += copy;
2590 } while (length > 0);
2594 EXPORT_SYMBOL(skb_append_datato_frags);
2597 * skb_pull_rcsum - pull skb and update receive checksum
2598 * @skb: buffer to update
2599 * @len: length of data pulled
2601 * This function performs an skb_pull on the packet and updates
2602 * the CHECKSUM_COMPLETE checksum. It should be used on
2603 * receive path processing instead of skb_pull unless you know
2604 * that the checksum difference is zero (e.g., a valid IP header)
2605 * or you are setting ip_summed to CHECKSUM_NONE.
2607 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2609 BUG_ON(len > skb->len);
2611 BUG_ON(skb->len < skb->data_len);
2612 skb_postpull_rcsum(skb, skb->data, len);
2613 return skb->data += len;
2615 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2618 * skb_segment - Perform protocol segmentation on skb.
2619 * @skb: buffer to segment
2620 * @features: features for the output path (see dev->features)
2622 * This function performs segmentation on the given skb. It returns
2623 * a pointer to the first in a list of new skbs for the segments.
2624 * In case of error it returns ERR_PTR(err).
2626 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2628 struct sk_buff *segs = NULL;
2629 struct sk_buff *tail = NULL;
2630 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2631 unsigned int mss = skb_shinfo(skb)->gso_size;
2632 unsigned int doffset = skb->data - skb_mac_header(skb);
2633 unsigned int offset = doffset;
2634 unsigned int headroom;
2636 int sg = !!(features & NETIF_F_SG);
2637 int nfrags = skb_shinfo(skb)->nr_frags;
2642 __skb_push(skb, doffset);
2643 headroom = skb_headroom(skb);
2644 pos = skb_headlen(skb);
2647 struct sk_buff *nskb;
2652 len = skb->len - offset;
2656 hsize = skb_headlen(skb) - offset;
2659 if (hsize > len || !sg)
2662 if (!hsize && i >= nfrags) {
2663 BUG_ON(fskb->len != len);
2666 nskb = skb_clone(fskb, GFP_ATOMIC);
2669 if (unlikely(!nskb))
2672 hsize = skb_end_pointer(nskb) - nskb->head;
2673 if (skb_cow_head(nskb, doffset + headroom)) {
2678 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2680 skb_release_head_state(nskb);
2681 __skb_push(nskb, doffset);
2683 nskb = alloc_skb(hsize + doffset + headroom,
2686 if (unlikely(!nskb))
2689 skb_reserve(nskb, headroom);
2690 __skb_put(nskb, doffset);
2699 __copy_skb_header(nskb, skb);
2700 nskb->mac_len = skb->mac_len;
2702 /* nskb and skb might have different headroom */
2703 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2704 nskb->csum_start += skb_headroom(nskb) - headroom;
2706 skb_reset_mac_header(nskb);
2707 skb_set_network_header(nskb, skb->mac_len);
2708 nskb->transport_header = (nskb->network_header +
2709 skb_network_header_len(skb));
2710 skb_copy_from_linear_data(skb, nskb->data, doffset);
2712 if (fskb != skb_shinfo(skb)->frag_list)
2716 nskb->ip_summed = CHECKSUM_NONE;
2717 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2723 frag = skb_shinfo(nskb)->frags;
2725 skb_copy_from_linear_data_offset(skb, offset,
2726 skb_put(nskb, hsize), hsize);
2728 while (pos < offset + len && i < nfrags) {
2729 *frag = skb_shinfo(skb)->frags[i];
2730 __skb_frag_ref(frag);
2731 size = skb_frag_size(frag);
2734 frag->page_offset += offset - pos;
2735 skb_frag_size_sub(frag, offset - pos);
2738 skb_shinfo(nskb)->nr_frags++;
2740 if (pos + size <= offset + len) {
2744 skb_frag_size_sub(frag, pos + size - (offset + len));
2751 if (pos < offset + len) {
2752 struct sk_buff *fskb2 = fskb;
2754 BUG_ON(pos + fskb->len != offset + len);
2760 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2766 SKB_FRAG_ASSERT(nskb);
2767 skb_shinfo(nskb)->frag_list = fskb2;
2771 nskb->data_len = len - hsize;
2772 nskb->len += nskb->data_len;
2773 nskb->truesize += nskb->data_len;
2774 } while ((offset += len) < skb->len);
2779 while ((skb = segs)) {
2783 return ERR_PTR(err);
2785 EXPORT_SYMBOL_GPL(skb_segment);
2787 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2789 struct sk_buff *p = *head;
2790 struct sk_buff *nskb;
2791 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2792 struct skb_shared_info *pinfo = skb_shinfo(p);
2793 unsigned int headroom;
2794 unsigned int len = skb_gro_len(skb);
2795 unsigned int offset = skb_gro_offset(skb);
2796 unsigned int headlen = skb_headlen(skb);
2798 if (p->len + len >= 65536)
2801 if (pinfo->frag_list)
2803 else if (headlen <= offset) {
2806 int i = skbinfo->nr_frags;
2807 int nr_frags = pinfo->nr_frags + i;
2811 if (nr_frags > MAX_SKB_FRAGS)
2814 pinfo->nr_frags = nr_frags;
2815 skbinfo->nr_frags = 0;
2817 frag = pinfo->frags + nr_frags;
2818 frag2 = skbinfo->frags + i;
2823 frag->page_offset += offset;
2824 skb_frag_size_sub(frag, offset);
2826 skb->truesize -= skb->data_len;
2827 skb->len -= skb->data_len;
2830 NAPI_GRO_CB(skb)->free = 1;
2832 } else if (skb_gro_len(p) != pinfo->gso_size)
2835 headroom = skb_headroom(p);
2836 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2837 if (unlikely(!nskb))
2840 __copy_skb_header(nskb, p);
2841 nskb->mac_len = p->mac_len;
2843 skb_reserve(nskb, headroom);
2844 __skb_put(nskb, skb_gro_offset(p));
2846 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2847 skb_set_network_header(nskb, skb_network_offset(p));
2848 skb_set_transport_header(nskb, skb_transport_offset(p));
2850 __skb_pull(p, skb_gro_offset(p));
2851 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2852 p->data - skb_mac_header(p));
2854 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2855 skb_shinfo(nskb)->frag_list = p;
2856 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2857 pinfo->gso_size = 0;
2858 skb_header_release(p);
2861 nskb->data_len += p->len;
2862 nskb->truesize += p->len;
2863 nskb->len += p->len;
2866 nskb->next = p->next;
2872 if (offset > headlen) {
2873 unsigned int eat = offset - headlen;
2875 skbinfo->frags[0].page_offset += eat;
2876 skb_frag_size_sub(&skbinfo->frags[0], eat);
2877 skb->data_len -= eat;
2882 __skb_pull(skb, offset);
2884 p->prev->next = skb;
2886 skb_header_release(skb);
2889 NAPI_GRO_CB(p)->count++;
2894 NAPI_GRO_CB(skb)->same_flow = 1;
2897 EXPORT_SYMBOL_GPL(skb_gro_receive);
2899 void __init skb_init(void)
2901 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2902 sizeof(struct sk_buff),
2904 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2906 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2907 (2*sizeof(struct sk_buff)) +
2910 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2915 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2916 * @skb: Socket buffer containing the buffers to be mapped
2917 * @sg: The scatter-gather list to map into
2918 * @offset: The offset into the buffer's contents to start mapping
2919 * @len: Length of buffer space to be mapped
2921 * Fill the specified scatter-gather list with mappings/pointers into a
2922 * region of the buffer space attached to a socket buffer.
2925 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2927 int start = skb_headlen(skb);
2928 int i, copy = start - offset;
2929 struct sk_buff *frag_iter;
2935 sg_set_buf(sg, skb->data + offset, copy);
2937 if ((len -= copy) == 0)
2942 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2945 WARN_ON(start > offset + len);
2947 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2948 if ((copy = end - offset) > 0) {
2949 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2953 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
2954 frag->page_offset+offset-start);
2963 skb_walk_frags(skb, frag_iter) {
2966 WARN_ON(start > offset + len);
2968 end = start + frag_iter->len;
2969 if ((copy = end - offset) > 0) {
2972 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2974 if ((len -= copy) == 0)
2984 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2986 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2988 sg_mark_end(&sg[nsg - 1]);
2992 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2995 * skb_cow_data - Check that a socket buffer's data buffers are writable
2996 * @skb: The socket buffer to check.
2997 * @tailbits: Amount of trailing space to be added
2998 * @trailer: Returned pointer to the skb where the @tailbits space begins
3000 * Make sure that the data buffers attached to a socket buffer are
3001 * writable. If they are not, private copies are made of the data buffers
3002 * and the socket buffer is set to use these instead.
3004 * If @tailbits is given, make sure that there is space to write @tailbits
3005 * bytes of data beyond current end of socket buffer. @trailer will be
3006 * set to point to the skb in which this space begins.
3008 * The number of scatterlist elements required to completely map the
3009 * COW'd and extended socket buffer will be returned.
3011 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3015 struct sk_buff *skb1, **skb_p;
3017 /* If skb is cloned or its head is paged, reallocate
3018 * head pulling out all the pages (pages are considered not writable
3019 * at the moment even if they are anonymous).
3021 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3022 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3025 /* Easy case. Most of packets will go this way. */
3026 if (!skb_has_frag_list(skb)) {
3027 /* A little of trouble, not enough of space for trailer.
3028 * This should not happen, when stack is tuned to generate
3029 * good frames. OK, on miss we reallocate and reserve even more
3030 * space, 128 bytes is fair. */
3032 if (skb_tailroom(skb) < tailbits &&
3033 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3041 /* Misery. We are in troubles, going to mincer fragments... */
3044 skb_p = &skb_shinfo(skb)->frag_list;
3047 while ((skb1 = *skb_p) != NULL) {
3050 /* The fragment is partially pulled by someone,
3051 * this can happen on input. Copy it and everything
3054 if (skb_shared(skb1))
3057 /* If the skb is the last, worry about trailer. */
3059 if (skb1->next == NULL && tailbits) {
3060 if (skb_shinfo(skb1)->nr_frags ||
3061 skb_has_frag_list(skb1) ||
3062 skb_tailroom(skb1) < tailbits)
3063 ntail = tailbits + 128;
3069 skb_shinfo(skb1)->nr_frags ||
3070 skb_has_frag_list(skb1)) {
3071 struct sk_buff *skb2;
3073 /* Fuck, we are miserable poor guys... */
3075 skb2 = skb_copy(skb1, GFP_ATOMIC);
3077 skb2 = skb_copy_expand(skb1,
3081 if (unlikely(skb2 == NULL))
3085 skb_set_owner_w(skb2, skb1->sk);
3087 /* Looking around. Are we still alive?
3088 * OK, link new skb, drop old one */
3090 skb2->next = skb1->next;
3097 skb_p = &skb1->next;
3102 EXPORT_SYMBOL_GPL(skb_cow_data);
3104 static void sock_rmem_free(struct sk_buff *skb)
3106 struct sock *sk = skb->sk;
3108 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3112 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3114 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3118 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3119 (unsigned)sk->sk_rcvbuf)
3124 skb->destructor = sock_rmem_free;
3125 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3127 /* before exiting rcu section, make sure dst is refcounted */
3130 skb_queue_tail(&sk->sk_error_queue, skb);
3131 if (!sock_flag(sk, SOCK_DEAD))
3132 sk->sk_data_ready(sk, len);
3135 EXPORT_SYMBOL(sock_queue_err_skb);
3137 void skb_tstamp_tx(struct sk_buff *orig_skb,
3138 struct skb_shared_hwtstamps *hwtstamps)
3140 struct sock *sk = orig_skb->sk;
3141 struct sock_exterr_skb *serr;
3142 struct sk_buff *skb;
3148 skb = skb_clone(orig_skb, GFP_ATOMIC);
3153 *skb_hwtstamps(skb) =
3157 * no hardware time stamps available,
3158 * so keep the shared tx_flags and only
3159 * store software time stamp
3161 skb->tstamp = ktime_get_real();
3164 serr = SKB_EXT_ERR(skb);
3165 memset(serr, 0, sizeof(*serr));
3166 serr->ee.ee_errno = ENOMSG;
3167 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3169 err = sock_queue_err_skb(sk, skb);
3174 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3178 * skb_partial_csum_set - set up and verify partial csum values for packet
3179 * @skb: the skb to set
3180 * @start: the number of bytes after skb->data to start checksumming.
3181 * @off: the offset from start to place the checksum.
3183 * For untrusted partially-checksummed packets, we need to make sure the values
3184 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3186 * This function checks and sets those values and skb->ip_summed: if this
3187 * returns false you should drop the packet.
3189 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3191 if (unlikely(start > skb_headlen(skb)) ||
3192 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3193 if (net_ratelimit())
3195 "bad partial csum: csum=%u/%u len=%u\n",
3196 start, off, skb_headlen(skb));
3199 skb->ip_summed = CHECKSUM_PARTIAL;
3200 skb->csum_start = skb_headroom(skb) + start;
3201 skb->csum_offset = off;
3204 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3206 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3208 if (net_ratelimit())
3209 pr_warning("%s: received packets cannot be forwarded"
3210 " while LRO is enabled\n", skb->dev->name);
3212 EXPORT_SYMBOL(__skb_warn_lro_forwarding);