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 <trace/events/skb.h>
70 #include <linux/highmem.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 const 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 static 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>");
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
140 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
148 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
170 int fclone, int node)
172 struct kmem_cache *cache;
173 struct skb_shared_info *shinfo;
177 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
180 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
190 size = SKB_DATA_ALIGN(size);
191 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
192 data = kmalloc_node_track_caller(size, gfp_mask, node);
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
199 size = SKB_WITH_OVERHEAD(ksize(data));
200 prefetchw(data + size);
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
207 memset(skb, 0, offsetof(struct sk_buff, tail));
208 /* Account for allocated memory : skb + skb->head */
209 skb->truesize = SKB_TRUESIZE(size);
210 atomic_set(&skb->users, 1);
213 skb_reset_tail_pointer(skb);
214 skb->end = skb->tail + size;
215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb->mac_header = ~0U;
219 /* make sure we initialize shinfo sequentially */
220 shinfo = skb_shinfo(skb);
221 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
222 atomic_set(&shinfo->dataref, 1);
223 kmemcheck_annotate_variable(shinfo->destructor_arg);
226 struct sk_buff *child = skb + 1;
227 atomic_t *fclone_ref = (atomic_t *) (child + 1);
229 kmemcheck_annotate_bitfield(child, flags1);
230 kmemcheck_annotate_bitfield(child, flags2);
231 skb->fclone = SKB_FCLONE_ORIG;
232 atomic_set(fclone_ref, 1);
234 child->fclone = SKB_FCLONE_UNAVAILABLE;
239 kmem_cache_free(cache, skb);
243 EXPORT_SYMBOL(__alloc_skb);
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
248 * @frag_size: size of fragment, or 0 if head was kmalloced
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
262 struct sk_buff *build_skb(void *data, unsigned int frag_size)
264 struct skb_shared_info *shinfo;
266 unsigned int size = frag_size ? : ksize(data);
268 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
272 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
274 memset(skb, 0, offsetof(struct sk_buff, tail));
275 skb->truesize = SKB_TRUESIZE(size);
276 skb->head_frag = frag_size != 0;
277 atomic_set(&skb->users, 1);
280 skb_reset_tail_pointer(skb);
281 skb->end = skb->tail + size;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb->mac_header = ~0U;
286 /* make sure we initialize shinfo sequentially */
287 shinfo = skb_shinfo(skb);
288 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
289 atomic_set(&shinfo->dataref, 1);
290 kmemcheck_annotate_variable(shinfo->destructor_arg);
294 EXPORT_SYMBOL(build_skb);
297 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
298 * @dev: network device to receive on
299 * @length: length to allocate
300 * @gfp_mask: get_free_pages mask, passed to alloc_skb
302 * Allocate a new &sk_buff and assign it a usage count of one. The
303 * buffer has unspecified headroom built in. Users should allocate
304 * the headroom they think they need without accounting for the
305 * built in space. The built in space is used for optimisations.
307 * %NULL is returned if there is no free memory.
309 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
310 unsigned int length, gfp_t gfp_mask)
314 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
316 skb_reserve(skb, NET_SKB_PAD);
321 EXPORT_SYMBOL(__netdev_alloc_skb);
323 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
324 int size, unsigned int truesize)
326 skb_fill_page_desc(skb, i, page, off, size);
328 skb->data_len += size;
329 skb->truesize += truesize;
331 EXPORT_SYMBOL(skb_add_rx_frag);
334 * dev_alloc_skb - allocate an skbuff for receiving
335 * @length: length to allocate
337 * Allocate a new &sk_buff and assign it a usage count of one. The
338 * buffer has unspecified headroom built in. Users should allocate
339 * the headroom they think they need without accounting for the
340 * built in space. The built in space is used for optimisations.
342 * %NULL is returned if there is no free memory. Although this function
343 * allocates memory it can be called from an interrupt.
345 struct sk_buff *dev_alloc_skb(unsigned int length)
348 * There is more code here than it seems:
349 * __dev_alloc_skb is an inline
351 return __dev_alloc_skb(length, GFP_ATOMIC);
353 EXPORT_SYMBOL(dev_alloc_skb);
355 static void skb_drop_list(struct sk_buff **listp)
357 struct sk_buff *list = *listp;
362 struct sk_buff *this = list;
368 static inline void skb_drop_fraglist(struct sk_buff *skb)
370 skb_drop_list(&skb_shinfo(skb)->frag_list);
373 static void skb_clone_fraglist(struct sk_buff *skb)
375 struct sk_buff *list;
377 skb_walk_frags(skb, list)
381 static void skb_free_head(struct sk_buff *skb)
384 put_page(virt_to_head_page(skb->head));
389 static void skb_release_data(struct sk_buff *skb)
392 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
393 &skb_shinfo(skb)->dataref)) {
394 if (skb_shinfo(skb)->nr_frags) {
396 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
397 skb_frag_unref(skb, i);
401 * If skb buf is from userspace, we need to notify the caller
402 * the lower device DMA has done;
404 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
405 struct ubuf_info *uarg;
407 uarg = skb_shinfo(skb)->destructor_arg;
409 uarg->callback(uarg);
412 if (skb_has_frag_list(skb))
413 skb_drop_fraglist(skb);
420 * Free an skbuff by memory without cleaning the state.
422 static void kfree_skbmem(struct sk_buff *skb)
424 struct sk_buff *other;
425 atomic_t *fclone_ref;
427 switch (skb->fclone) {
428 case SKB_FCLONE_UNAVAILABLE:
429 kmem_cache_free(skbuff_head_cache, skb);
432 case SKB_FCLONE_ORIG:
433 fclone_ref = (atomic_t *) (skb + 2);
434 if (atomic_dec_and_test(fclone_ref))
435 kmem_cache_free(skbuff_fclone_cache, skb);
438 case SKB_FCLONE_CLONE:
439 fclone_ref = (atomic_t *) (skb + 1);
442 /* The clone portion is available for
443 * fast-cloning again.
445 skb->fclone = SKB_FCLONE_UNAVAILABLE;
447 if (atomic_dec_and_test(fclone_ref))
448 kmem_cache_free(skbuff_fclone_cache, other);
453 static void skb_release_head_state(struct sk_buff *skb)
457 secpath_put(skb->sp);
459 if (skb->destructor) {
461 skb->destructor(skb);
463 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
464 nf_conntrack_put(skb->nfct);
466 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
467 nf_conntrack_put_reasm(skb->nfct_reasm);
469 #ifdef CONFIG_BRIDGE_NETFILTER
470 nf_bridge_put(skb->nf_bridge);
472 /* XXX: IS this still necessary? - JHS */
473 #ifdef CONFIG_NET_SCHED
475 #ifdef CONFIG_NET_CLS_ACT
481 /* Free everything but the sk_buff shell. */
482 static void skb_release_all(struct sk_buff *skb)
484 skb_release_head_state(skb);
485 skb_release_data(skb);
489 * __kfree_skb - private function
492 * Free an sk_buff. Release anything attached to the buffer.
493 * Clean the state. This is an internal helper function. Users should
494 * always call kfree_skb
497 void __kfree_skb(struct sk_buff *skb)
499 skb_release_all(skb);
502 EXPORT_SYMBOL(__kfree_skb);
505 * kfree_skb - free an sk_buff
506 * @skb: buffer to free
508 * Drop a reference to the buffer and free it if the usage count has
511 void kfree_skb(struct sk_buff *skb)
515 if (likely(atomic_read(&skb->users) == 1))
517 else if (likely(!atomic_dec_and_test(&skb->users)))
519 trace_kfree_skb(skb, __builtin_return_address(0));
522 EXPORT_SYMBOL(kfree_skb);
525 * consume_skb - free an skbuff
526 * @skb: buffer to free
528 * Drop a ref to the buffer and free it if the usage count has hit zero
529 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
530 * is being dropped after a failure and notes that
532 void consume_skb(struct sk_buff *skb)
536 if (likely(atomic_read(&skb->users) == 1))
538 else if (likely(!atomic_dec_and_test(&skb->users)))
540 trace_consume_skb(skb);
543 EXPORT_SYMBOL(consume_skb);
546 * skb_recycle - clean up an skb for reuse
549 * Recycles the skb to be reused as a receive buffer. This
550 * function does any necessary reference count dropping, and
551 * cleans up the skbuff as if it just came from __alloc_skb().
553 void skb_recycle(struct sk_buff *skb)
555 struct skb_shared_info *shinfo;
557 skb_release_head_state(skb);
559 shinfo = skb_shinfo(skb);
560 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
561 atomic_set(&shinfo->dataref, 1);
563 memset(skb, 0, offsetof(struct sk_buff, tail));
564 skb->data = skb->head + NET_SKB_PAD;
565 skb_reset_tail_pointer(skb);
567 EXPORT_SYMBOL(skb_recycle);
570 * skb_recycle_check - check if skb can be reused for receive
572 * @skb_size: minimum receive buffer size
574 * Checks that the skb passed in is not shared or cloned, and
575 * that it is linear and its head portion at least as large as
576 * skb_size so that it can be recycled as a receive buffer.
577 * If these conditions are met, this function does any necessary
578 * reference count dropping and cleans up the skbuff as if it
579 * just came from __alloc_skb().
581 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
583 if (!skb_is_recycleable(skb, skb_size))
590 EXPORT_SYMBOL(skb_recycle_check);
592 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
594 new->tstamp = old->tstamp;
596 new->transport_header = old->transport_header;
597 new->network_header = old->network_header;
598 new->mac_header = old->mac_header;
599 skb_dst_copy(new, old);
600 new->rxhash = old->rxhash;
601 new->ooo_okay = old->ooo_okay;
602 new->l4_rxhash = old->l4_rxhash;
603 new->no_fcs = old->no_fcs;
605 new->sp = secpath_get(old->sp);
607 memcpy(new->cb, old->cb, sizeof(old->cb));
608 new->csum = old->csum;
609 new->local_df = old->local_df;
610 new->pkt_type = old->pkt_type;
611 new->ip_summed = old->ip_summed;
612 skb_copy_queue_mapping(new, old);
613 new->priority = old->priority;
614 #if IS_ENABLED(CONFIG_IP_VS)
615 new->ipvs_property = old->ipvs_property;
617 new->protocol = old->protocol;
618 new->mark = old->mark;
619 new->skb_iif = old->skb_iif;
621 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
622 new->nf_trace = old->nf_trace;
624 #ifdef CONFIG_NET_SCHED
625 new->tc_index = old->tc_index;
626 #ifdef CONFIG_NET_CLS_ACT
627 new->tc_verd = old->tc_verd;
630 new->vlan_tci = old->vlan_tci;
632 skb_copy_secmark(new, old);
636 * You should not add any new code to this function. Add it to
637 * __copy_skb_header above instead.
639 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
641 #define C(x) n->x = skb->x
643 n->next = n->prev = NULL;
645 __copy_skb_header(n, skb);
650 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
653 n->destructor = NULL;
660 atomic_set(&n->users, 1);
662 atomic_inc(&(skb_shinfo(skb)->dataref));
670 * skb_morph - morph one skb into another
671 * @dst: the skb to receive the contents
672 * @src: the skb to supply the contents
674 * This is identical to skb_clone except that the target skb is
675 * supplied by the user.
677 * The target skb is returned upon exit.
679 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
681 skb_release_all(dst);
682 return __skb_clone(dst, src);
684 EXPORT_SYMBOL_GPL(skb_morph);
686 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
687 * @skb: the skb to modify
688 * @gfp_mask: allocation priority
690 * This must be called on SKBTX_DEV_ZEROCOPY skb.
691 * It will copy all frags into kernel and drop the reference
692 * to userspace pages.
694 * If this function is called from an interrupt gfp_mask() must be
697 * Returns 0 on success or a negative error code on failure
698 * to allocate kernel memory to copy to.
700 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
703 int num_frags = skb_shinfo(skb)->nr_frags;
704 struct page *page, *head = NULL;
705 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
707 for (i = 0; i < num_frags; i++) {
709 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
711 page = alloc_page(GFP_ATOMIC);
714 struct page *next = (struct page *)head->private;
720 vaddr = kmap_atomic(skb_frag_page(f));
721 memcpy(page_address(page),
722 vaddr + f->page_offset, skb_frag_size(f));
723 kunmap_atomic(vaddr);
724 page->private = (unsigned long)head;
728 /* skb frags release userspace buffers */
729 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
730 skb_frag_unref(skb, i);
732 uarg->callback(uarg);
734 /* skb frags point to kernel buffers */
735 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
736 __skb_fill_page_desc(skb, i-1, head, 0,
737 skb_shinfo(skb)->frags[i - 1].size);
738 head = (struct page *)head->private;
741 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
747 * skb_clone - duplicate an sk_buff
748 * @skb: buffer to clone
749 * @gfp_mask: allocation priority
751 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
752 * copies share the same packet data but not structure. The new
753 * buffer has a reference count of 1. If the allocation fails the
754 * function returns %NULL otherwise the new buffer is returned.
756 * If this function is called from an interrupt gfp_mask() must be
760 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
764 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
765 if (skb_copy_ubufs(skb, gfp_mask))
770 if (skb->fclone == SKB_FCLONE_ORIG &&
771 n->fclone == SKB_FCLONE_UNAVAILABLE) {
772 atomic_t *fclone_ref = (atomic_t *) (n + 1);
773 n->fclone = SKB_FCLONE_CLONE;
774 atomic_inc(fclone_ref);
776 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
780 kmemcheck_annotate_bitfield(n, flags1);
781 kmemcheck_annotate_bitfield(n, flags2);
782 n->fclone = SKB_FCLONE_UNAVAILABLE;
785 return __skb_clone(n, skb);
787 EXPORT_SYMBOL(skb_clone);
789 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
791 #ifndef NET_SKBUFF_DATA_USES_OFFSET
793 * Shift between the two data areas in bytes
795 unsigned long offset = new->data - old->data;
798 __copy_skb_header(new, old);
800 #ifndef NET_SKBUFF_DATA_USES_OFFSET
801 /* {transport,network,mac}_header are relative to skb->head */
802 new->transport_header += offset;
803 new->network_header += offset;
804 if (skb_mac_header_was_set(new))
805 new->mac_header += offset;
807 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
808 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
809 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
813 * skb_copy - create private copy of an sk_buff
814 * @skb: buffer to copy
815 * @gfp_mask: allocation priority
817 * Make a copy of both an &sk_buff and its data. This is used when the
818 * caller wishes to modify the data and needs a private copy of the
819 * data to alter. Returns %NULL on failure or the pointer to the buffer
820 * on success. The returned buffer has a reference count of 1.
822 * As by-product this function converts non-linear &sk_buff to linear
823 * one, so that &sk_buff becomes completely private and caller is allowed
824 * to modify all the data of returned buffer. This means that this
825 * function is not recommended for use in circumstances when only
826 * header is going to be modified. Use pskb_copy() instead.
829 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
831 int headerlen = skb_headroom(skb);
832 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
833 struct sk_buff *n = alloc_skb(size, gfp_mask);
838 /* Set the data pointer */
839 skb_reserve(n, headerlen);
840 /* Set the tail pointer and length */
841 skb_put(n, skb->len);
843 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
846 copy_skb_header(n, skb);
849 EXPORT_SYMBOL(skb_copy);
852 * __pskb_copy - create copy of an sk_buff with private head.
853 * @skb: buffer to copy
854 * @headroom: headroom of new skb
855 * @gfp_mask: allocation priority
857 * Make a copy of both an &sk_buff and part of its data, located
858 * in header. Fragmented data remain shared. This is used when
859 * the caller wishes to modify only header of &sk_buff and needs
860 * private copy of the header to alter. Returns %NULL on failure
861 * or the pointer to the buffer on success.
862 * The returned buffer has a reference count of 1.
865 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
867 unsigned int size = skb_headlen(skb) + headroom;
868 struct sk_buff *n = alloc_skb(size, gfp_mask);
873 /* Set the data pointer */
874 skb_reserve(n, headroom);
875 /* Set the tail pointer and length */
876 skb_put(n, skb_headlen(skb));
878 skb_copy_from_linear_data(skb, n->data, n->len);
880 n->truesize += skb->data_len;
881 n->data_len = skb->data_len;
884 if (skb_shinfo(skb)->nr_frags) {
887 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
888 if (skb_copy_ubufs(skb, gfp_mask)) {
894 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
895 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
896 skb_frag_ref(skb, i);
898 skb_shinfo(n)->nr_frags = i;
901 if (skb_has_frag_list(skb)) {
902 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
903 skb_clone_fraglist(n);
906 copy_skb_header(n, skb);
910 EXPORT_SYMBOL(__pskb_copy);
913 * pskb_expand_head - reallocate header of &sk_buff
914 * @skb: buffer to reallocate
915 * @nhead: room to add at head
916 * @ntail: room to add at tail
917 * @gfp_mask: allocation priority
919 * Expands (or creates identical copy, if &nhead and &ntail are zero)
920 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
921 * reference count of 1. Returns zero in the case of success or error,
922 * if expansion failed. In the last case, &sk_buff is not changed.
924 * All the pointers pointing into skb header may change and must be
925 * reloaded after call to this function.
928 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
933 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
942 size = SKB_DATA_ALIGN(size);
944 /* Check if we can avoid taking references on fragments if we own
945 * the last reference on skb->head. (see skb_release_data())
950 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
951 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
954 if (fastpath && !skb->head_frag &&
955 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
956 memmove(skb->head + size, skb_shinfo(skb),
957 offsetof(struct skb_shared_info,
958 frags[skb_shinfo(skb)->nr_frags]));
959 memmove(skb->head + nhead, skb->head,
960 skb_tail_pointer(skb) - skb->head);
965 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
969 size = SKB_WITH_OVERHEAD(ksize(data));
971 /* Copy only real data... and, alas, header. This should be
972 * optimized for the cases when header is void.
974 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
976 memcpy((struct skb_shared_info *)(data + size),
978 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
983 /* copy this zero copy skb frags */
984 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
985 if (skb_copy_ubufs(skb, gfp_mask))
988 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
989 skb_frag_ref(skb, i);
991 if (skb_has_frag_list(skb))
992 skb_clone_fraglist(skb);
994 skb_release_data(skb);
996 off = (data + nhead) - skb->head;
1002 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1006 skb->end = skb->head + size;
1008 /* {transport,network,mac}_header and tail are relative to skb->head */
1010 skb->transport_header += off;
1011 skb->network_header += off;
1012 if (skb_mac_header_was_set(skb))
1013 skb->mac_header += off;
1014 /* Only adjust this if it actually is csum_start rather than csum */
1015 if (skb->ip_summed == CHECKSUM_PARTIAL)
1016 skb->csum_start += nhead;
1020 atomic_set(&skb_shinfo(skb)->dataref, 1);
1028 EXPORT_SYMBOL(pskb_expand_head);
1030 /* Make private copy of skb with writable head and some headroom */
1032 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1034 struct sk_buff *skb2;
1035 int delta = headroom - skb_headroom(skb);
1038 skb2 = pskb_copy(skb, GFP_ATOMIC);
1040 skb2 = skb_clone(skb, GFP_ATOMIC);
1041 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1049 EXPORT_SYMBOL(skb_realloc_headroom);
1052 * skb_copy_expand - copy and expand sk_buff
1053 * @skb: buffer to copy
1054 * @newheadroom: new free bytes at head
1055 * @newtailroom: new free bytes at tail
1056 * @gfp_mask: allocation priority
1058 * Make a copy of both an &sk_buff and its data and while doing so
1059 * allocate additional space.
1061 * This is used when the caller wishes to modify the data and needs a
1062 * private copy of the data to alter as well as more space for new fields.
1063 * Returns %NULL on failure or the pointer to the buffer
1064 * on success. The returned buffer has a reference count of 1.
1066 * You must pass %GFP_ATOMIC as the allocation priority if this function
1067 * is called from an interrupt.
1069 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1070 int newheadroom, int newtailroom,
1074 * Allocate the copy buffer
1076 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1078 int oldheadroom = skb_headroom(skb);
1079 int head_copy_len, head_copy_off;
1085 skb_reserve(n, newheadroom);
1087 /* Set the tail pointer and length */
1088 skb_put(n, skb->len);
1090 head_copy_len = oldheadroom;
1092 if (newheadroom <= head_copy_len)
1093 head_copy_len = newheadroom;
1095 head_copy_off = newheadroom - head_copy_len;
1097 /* Copy the linear header and data. */
1098 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1099 skb->len + head_copy_len))
1102 copy_skb_header(n, skb);
1104 off = newheadroom - oldheadroom;
1105 if (n->ip_summed == CHECKSUM_PARTIAL)
1106 n->csum_start += off;
1107 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1108 n->transport_header += off;
1109 n->network_header += off;
1110 if (skb_mac_header_was_set(skb))
1111 n->mac_header += off;
1116 EXPORT_SYMBOL(skb_copy_expand);
1119 * skb_pad - zero pad the tail of an skb
1120 * @skb: buffer to pad
1121 * @pad: space to pad
1123 * Ensure that a buffer is followed by a padding area that is zero
1124 * filled. Used by network drivers which may DMA or transfer data
1125 * beyond the buffer end onto the wire.
1127 * May return error in out of memory cases. The skb is freed on error.
1130 int skb_pad(struct sk_buff *skb, int pad)
1135 /* If the skbuff is non linear tailroom is always zero.. */
1136 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1137 memset(skb->data+skb->len, 0, pad);
1141 ntail = skb->data_len + pad - (skb->end - skb->tail);
1142 if (likely(skb_cloned(skb) || ntail > 0)) {
1143 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1148 /* FIXME: The use of this function with non-linear skb's really needs
1151 err = skb_linearize(skb);
1155 memset(skb->data + skb->len, 0, pad);
1162 EXPORT_SYMBOL(skb_pad);
1165 * skb_put - add data to a buffer
1166 * @skb: buffer to use
1167 * @len: amount of data to add
1169 * This function extends the used data area of the buffer. If this would
1170 * exceed the total buffer size the kernel will panic. A pointer to the
1171 * first byte of the extra data is returned.
1173 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1175 unsigned char *tmp = skb_tail_pointer(skb);
1176 SKB_LINEAR_ASSERT(skb);
1179 if (unlikely(skb->tail > skb->end))
1180 skb_over_panic(skb, len, __builtin_return_address(0));
1183 EXPORT_SYMBOL(skb_put);
1186 * skb_push - add data to the start of a buffer
1187 * @skb: buffer to use
1188 * @len: amount of data to add
1190 * This function extends the used data area of the buffer at the buffer
1191 * start. If this would exceed the total buffer headroom the kernel will
1192 * panic. A pointer to the first byte of the extra data is returned.
1194 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1198 if (unlikely(skb->data<skb->head))
1199 skb_under_panic(skb, len, __builtin_return_address(0));
1202 EXPORT_SYMBOL(skb_push);
1205 * skb_pull - remove data from the start of a buffer
1206 * @skb: buffer to use
1207 * @len: amount of data to remove
1209 * This function removes data from the start of a buffer, returning
1210 * the memory to the headroom. A pointer to the next data in the buffer
1211 * is returned. Once the data has been pulled future pushes will overwrite
1214 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1216 return skb_pull_inline(skb, len);
1218 EXPORT_SYMBOL(skb_pull);
1221 * skb_trim - remove end from a buffer
1222 * @skb: buffer to alter
1225 * Cut the length of a buffer down by removing data from the tail. If
1226 * the buffer is already under the length specified it is not modified.
1227 * The skb must be linear.
1229 void skb_trim(struct sk_buff *skb, unsigned int len)
1232 __skb_trim(skb, len);
1234 EXPORT_SYMBOL(skb_trim);
1236 /* Trims skb to length len. It can change skb pointers.
1239 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1241 struct sk_buff **fragp;
1242 struct sk_buff *frag;
1243 int offset = skb_headlen(skb);
1244 int nfrags = skb_shinfo(skb)->nr_frags;
1248 if (skb_cloned(skb) &&
1249 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1256 for (; i < nfrags; i++) {
1257 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1264 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1267 skb_shinfo(skb)->nr_frags = i;
1269 for (; i < nfrags; i++)
1270 skb_frag_unref(skb, i);
1272 if (skb_has_frag_list(skb))
1273 skb_drop_fraglist(skb);
1277 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1278 fragp = &frag->next) {
1279 int end = offset + frag->len;
1281 if (skb_shared(frag)) {
1282 struct sk_buff *nfrag;
1284 nfrag = skb_clone(frag, GFP_ATOMIC);
1285 if (unlikely(!nfrag))
1288 nfrag->next = frag->next;
1300 unlikely((err = pskb_trim(frag, len - offset))))
1304 skb_drop_list(&frag->next);
1309 if (len > skb_headlen(skb)) {
1310 skb->data_len -= skb->len - len;
1315 skb_set_tail_pointer(skb, len);
1320 EXPORT_SYMBOL(___pskb_trim);
1323 * __pskb_pull_tail - advance tail of skb header
1324 * @skb: buffer to reallocate
1325 * @delta: number of bytes to advance tail
1327 * The function makes a sense only on a fragmented &sk_buff,
1328 * it expands header moving its tail forward and copying necessary
1329 * data from fragmented part.
1331 * &sk_buff MUST have reference count of 1.
1333 * Returns %NULL (and &sk_buff does not change) if pull failed
1334 * or value of new tail of skb in the case of success.
1336 * All the pointers pointing into skb header may change and must be
1337 * reloaded after call to this function.
1340 /* Moves tail of skb head forward, copying data from fragmented part,
1341 * when it is necessary.
1342 * 1. It may fail due to malloc failure.
1343 * 2. It may change skb pointers.
1345 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1347 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1349 /* If skb has not enough free space at tail, get new one
1350 * plus 128 bytes for future expansions. If we have enough
1351 * room at tail, reallocate without expansion only if skb is cloned.
1353 int i, k, eat = (skb->tail + delta) - skb->end;
1355 if (eat > 0 || skb_cloned(skb)) {
1356 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1361 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1364 /* Optimization: no fragments, no reasons to preestimate
1365 * size of pulled pages. Superb.
1367 if (!skb_has_frag_list(skb))
1370 /* Estimate size of pulled pages. */
1372 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1373 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1380 /* If we need update frag list, we are in troubles.
1381 * Certainly, it possible to add an offset to skb data,
1382 * but taking into account that pulling is expected to
1383 * be very rare operation, it is worth to fight against
1384 * further bloating skb head and crucify ourselves here instead.
1385 * Pure masohism, indeed. 8)8)
1388 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1389 struct sk_buff *clone = NULL;
1390 struct sk_buff *insp = NULL;
1395 if (list->len <= eat) {
1396 /* Eaten as whole. */
1401 /* Eaten partially. */
1403 if (skb_shared(list)) {
1404 /* Sucks! We need to fork list. :-( */
1405 clone = skb_clone(list, GFP_ATOMIC);
1411 /* This may be pulled without
1415 if (!pskb_pull(list, eat)) {
1423 /* Free pulled out fragments. */
1424 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1425 skb_shinfo(skb)->frag_list = list->next;
1428 /* And insert new clone at head. */
1431 skb_shinfo(skb)->frag_list = clone;
1434 /* Success! Now we may commit changes to skb data. */
1439 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1440 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1443 skb_frag_unref(skb, i);
1446 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1448 skb_shinfo(skb)->frags[k].page_offset += eat;
1449 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1455 skb_shinfo(skb)->nr_frags = k;
1458 skb->data_len -= delta;
1460 return skb_tail_pointer(skb);
1462 EXPORT_SYMBOL(__pskb_pull_tail);
1465 * skb_copy_bits - copy bits from skb to kernel buffer
1467 * @offset: offset in source
1468 * @to: destination buffer
1469 * @len: number of bytes to copy
1471 * Copy the specified number of bytes from the source skb to the
1472 * destination buffer.
1475 * If its prototype is ever changed,
1476 * check arch/{*}/net/{*}.S files,
1477 * since it is called from BPF assembly code.
1479 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1481 int start = skb_headlen(skb);
1482 struct sk_buff *frag_iter;
1485 if (offset > (int)skb->len - len)
1489 if ((copy = start - offset) > 0) {
1492 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1493 if ((len -= copy) == 0)
1499 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1501 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1503 WARN_ON(start > offset + len);
1505 end = start + skb_frag_size(f);
1506 if ((copy = end - offset) > 0) {
1512 vaddr = kmap_atomic(skb_frag_page(f));
1514 vaddr + f->page_offset + offset - start,
1516 kunmap_atomic(vaddr);
1518 if ((len -= copy) == 0)
1526 skb_walk_frags(skb, frag_iter) {
1529 WARN_ON(start > offset + len);
1531 end = start + frag_iter->len;
1532 if ((copy = end - offset) > 0) {
1535 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1537 if ((len -= copy) == 0)
1551 EXPORT_SYMBOL(skb_copy_bits);
1554 * Callback from splice_to_pipe(), if we need to release some pages
1555 * at the end of the spd in case we error'ed out in filling the pipe.
1557 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1559 put_page(spd->pages[i]);
1562 static struct page *linear_to_page(struct page *page, unsigned int *len,
1563 unsigned int *offset,
1564 struct sk_buff *skb, struct sock *sk)
1566 struct page *p = sk->sk_sndmsg_page;
1571 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1575 off = sk->sk_sndmsg_off = 0;
1576 /* hold one ref to this page until it's full */
1580 /* If we are the only user of the page, we can reset offset */
1581 if (page_count(p) == 1)
1582 sk->sk_sndmsg_off = 0;
1583 off = sk->sk_sndmsg_off;
1584 mlen = PAGE_SIZE - off;
1585 if (mlen < 64 && mlen < *len) {
1590 *len = min_t(unsigned int, *len, mlen);
1593 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1594 sk->sk_sndmsg_off += *len;
1600 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1602 unsigned int offset)
1604 return spd->nr_pages &&
1605 spd->pages[spd->nr_pages - 1] == page &&
1606 (spd->partial[spd->nr_pages - 1].offset +
1607 spd->partial[spd->nr_pages - 1].len == offset);
1611 * Fill page/offset/length into spd, if it can hold more pages.
1613 static bool spd_fill_page(struct splice_pipe_desc *spd,
1614 struct pipe_inode_info *pipe, struct page *page,
1615 unsigned int *len, unsigned int offset,
1616 struct sk_buff *skb, bool linear,
1619 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1623 page = linear_to_page(page, len, &offset, skb, sk);
1627 if (spd_can_coalesce(spd, page, offset)) {
1628 spd->partial[spd->nr_pages - 1].len += *len;
1632 spd->pages[spd->nr_pages] = page;
1633 spd->partial[spd->nr_pages].len = *len;
1634 spd->partial[spd->nr_pages].offset = offset;
1640 static inline void __segment_seek(struct page **page, unsigned int *poff,
1641 unsigned int *plen, unsigned int off)
1646 n = *poff / PAGE_SIZE;
1648 *page = nth_page(*page, n);
1650 *poff = *poff % PAGE_SIZE;
1654 static bool __splice_segment(struct page *page, unsigned int poff,
1655 unsigned int plen, unsigned int *off,
1656 unsigned int *len, struct sk_buff *skb,
1657 struct splice_pipe_desc *spd, bool linear,
1659 struct pipe_inode_info *pipe)
1664 /* skip this segment if already processed */
1670 /* ignore any bits we already processed */
1672 __segment_seek(&page, &poff, &plen, *off);
1677 unsigned int flen = min(*len, plen);
1679 /* the linear region may spread across several pages */
1680 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1682 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1685 __segment_seek(&page, &poff, &plen, flen);
1688 } while (*len && plen);
1694 * Map linear and fragment data from the skb to spd. It reports true if the
1695 * pipe is full or if we already spliced the requested length.
1697 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1698 unsigned int *offset, unsigned int *len,
1699 struct splice_pipe_desc *spd, struct sock *sk)
1704 * map the linear part
1706 if (__splice_segment(virt_to_page(skb->data),
1707 (unsigned long) skb->data & (PAGE_SIZE - 1),
1709 offset, len, skb, spd, true, sk, pipe))
1713 * then map the fragments
1715 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1716 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1718 if (__splice_segment(skb_frag_page(f),
1719 f->page_offset, skb_frag_size(f),
1720 offset, len, skb, spd, false, sk, pipe))
1728 * Map data from the skb to a pipe. Should handle both the linear part,
1729 * the fragments, and the frag list. It does NOT handle frag lists within
1730 * the frag list, if such a thing exists. We'd probably need to recurse to
1731 * handle that cleanly.
1733 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1734 struct pipe_inode_info *pipe, unsigned int tlen,
1737 struct partial_page partial[MAX_SKB_FRAGS];
1738 struct page *pages[MAX_SKB_FRAGS];
1739 struct splice_pipe_desc spd = {
1743 .ops = &sock_pipe_buf_ops,
1744 .spd_release = sock_spd_release,
1746 struct sk_buff *frag_iter;
1747 struct sock *sk = skb->sk;
1751 * __skb_splice_bits() only fails if the output has no room left,
1752 * so no point in going over the frag_list for the error case.
1754 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1760 * now see if we have a frag_list to map
1762 skb_walk_frags(skb, frag_iter) {
1765 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1772 * Drop the socket lock, otherwise we have reverse
1773 * locking dependencies between sk_lock and i_mutex
1774 * here as compared to sendfile(). We enter here
1775 * with the socket lock held, and splice_to_pipe() will
1776 * grab the pipe inode lock. For sendfile() emulation,
1777 * we call into ->sendpage() with the i_mutex lock held
1778 * and networking will grab the socket lock.
1781 ret = splice_to_pipe(pipe, &spd);
1789 * skb_store_bits - store bits from kernel buffer to skb
1790 * @skb: destination buffer
1791 * @offset: offset in destination
1792 * @from: source buffer
1793 * @len: number of bytes to copy
1795 * Copy the specified number of bytes from the source buffer to the
1796 * destination skb. This function handles all the messy bits of
1797 * traversing fragment lists and such.
1800 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1802 int start = skb_headlen(skb);
1803 struct sk_buff *frag_iter;
1806 if (offset > (int)skb->len - len)
1809 if ((copy = start - offset) > 0) {
1812 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1813 if ((len -= copy) == 0)
1819 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1820 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1823 WARN_ON(start > offset + len);
1825 end = start + skb_frag_size(frag);
1826 if ((copy = end - offset) > 0) {
1832 vaddr = kmap_atomic(skb_frag_page(frag));
1833 memcpy(vaddr + frag->page_offset + offset - start,
1835 kunmap_atomic(vaddr);
1837 if ((len -= copy) == 0)
1845 skb_walk_frags(skb, frag_iter) {
1848 WARN_ON(start > offset + len);
1850 end = start + frag_iter->len;
1851 if ((copy = end - offset) > 0) {
1854 if (skb_store_bits(frag_iter, offset - start,
1857 if ((len -= copy) == 0)
1870 EXPORT_SYMBOL(skb_store_bits);
1872 /* Checksum skb data. */
1874 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1875 int len, __wsum csum)
1877 int start = skb_headlen(skb);
1878 int i, copy = start - offset;
1879 struct sk_buff *frag_iter;
1882 /* Checksum header. */
1886 csum = csum_partial(skb->data + offset, copy, csum);
1887 if ((len -= copy) == 0)
1893 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1895 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1897 WARN_ON(start > offset + len);
1899 end = start + skb_frag_size(frag);
1900 if ((copy = end - offset) > 0) {
1906 vaddr = kmap_atomic(skb_frag_page(frag));
1907 csum2 = csum_partial(vaddr + frag->page_offset +
1908 offset - start, copy, 0);
1909 kunmap_atomic(vaddr);
1910 csum = csum_block_add(csum, csum2, pos);
1919 skb_walk_frags(skb, frag_iter) {
1922 WARN_ON(start > offset + len);
1924 end = start + frag_iter->len;
1925 if ((copy = end - offset) > 0) {
1929 csum2 = skb_checksum(frag_iter, offset - start,
1931 csum = csum_block_add(csum, csum2, pos);
1932 if ((len -= copy) == 0)
1943 EXPORT_SYMBOL(skb_checksum);
1945 /* Both of above in one bottle. */
1947 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1948 u8 *to, int len, __wsum csum)
1950 int start = skb_headlen(skb);
1951 int i, copy = start - offset;
1952 struct sk_buff *frag_iter;
1959 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1961 if ((len -= copy) == 0)
1968 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1971 WARN_ON(start > offset + len);
1973 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1974 if ((copy = end - offset) > 0) {
1977 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1981 vaddr = kmap_atomic(skb_frag_page(frag));
1982 csum2 = csum_partial_copy_nocheck(vaddr +
1986 kunmap_atomic(vaddr);
1987 csum = csum_block_add(csum, csum2, pos);
1997 skb_walk_frags(skb, frag_iter) {
2001 WARN_ON(start > offset + len);
2003 end = start + frag_iter->len;
2004 if ((copy = end - offset) > 0) {
2007 csum2 = skb_copy_and_csum_bits(frag_iter,
2010 csum = csum_block_add(csum, csum2, pos);
2011 if ((len -= copy) == 0)
2022 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2024 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2029 if (skb->ip_summed == CHECKSUM_PARTIAL)
2030 csstart = skb_checksum_start_offset(skb);
2032 csstart = skb_headlen(skb);
2034 BUG_ON(csstart > skb_headlen(skb));
2036 skb_copy_from_linear_data(skb, to, csstart);
2039 if (csstart != skb->len)
2040 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2041 skb->len - csstart, 0);
2043 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2044 long csstuff = csstart + skb->csum_offset;
2046 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2049 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2052 * skb_dequeue - remove from the head of the queue
2053 * @list: list to dequeue from
2055 * Remove the head of the list. The list lock is taken so the function
2056 * may be used safely with other locking list functions. The head item is
2057 * returned or %NULL if the list is empty.
2060 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2062 unsigned long flags;
2063 struct sk_buff *result;
2065 spin_lock_irqsave(&list->lock, flags);
2066 result = __skb_dequeue(list);
2067 spin_unlock_irqrestore(&list->lock, flags);
2070 EXPORT_SYMBOL(skb_dequeue);
2073 * skb_dequeue_tail - remove from the tail of the queue
2074 * @list: list to dequeue from
2076 * Remove the tail of the list. The list lock is taken so the function
2077 * may be used safely with other locking list functions. The tail item is
2078 * returned or %NULL if the list is empty.
2080 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2082 unsigned long flags;
2083 struct sk_buff *result;
2085 spin_lock_irqsave(&list->lock, flags);
2086 result = __skb_dequeue_tail(list);
2087 spin_unlock_irqrestore(&list->lock, flags);
2090 EXPORT_SYMBOL(skb_dequeue_tail);
2093 * skb_queue_purge - empty a list
2094 * @list: list to empty
2096 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2097 * the list and one reference dropped. This function takes the list
2098 * lock and is atomic with respect to other list locking functions.
2100 void skb_queue_purge(struct sk_buff_head *list)
2102 struct sk_buff *skb;
2103 while ((skb = skb_dequeue(list)) != NULL)
2106 EXPORT_SYMBOL(skb_queue_purge);
2109 * skb_queue_head - queue a buffer at the list head
2110 * @list: list to use
2111 * @newsk: buffer to queue
2113 * Queue a buffer at the start of the list. This function takes the
2114 * list lock and can be used safely with other locking &sk_buff functions
2117 * A buffer cannot be placed on two lists at the same time.
2119 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2121 unsigned long flags;
2123 spin_lock_irqsave(&list->lock, flags);
2124 __skb_queue_head(list, newsk);
2125 spin_unlock_irqrestore(&list->lock, flags);
2127 EXPORT_SYMBOL(skb_queue_head);
2130 * skb_queue_tail - queue a buffer at the list tail
2131 * @list: list to use
2132 * @newsk: buffer to queue
2134 * Queue a buffer at the tail of the list. This function takes the
2135 * list lock and can be used safely with other locking &sk_buff functions
2138 * A buffer cannot be placed on two lists at the same time.
2140 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2142 unsigned long flags;
2144 spin_lock_irqsave(&list->lock, flags);
2145 __skb_queue_tail(list, newsk);
2146 spin_unlock_irqrestore(&list->lock, flags);
2148 EXPORT_SYMBOL(skb_queue_tail);
2151 * skb_unlink - remove a buffer from a list
2152 * @skb: buffer to remove
2153 * @list: list to use
2155 * Remove a packet from a list. The list locks are taken and this
2156 * function is atomic with respect to other list locked calls
2158 * You must know what list the SKB is on.
2160 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2162 unsigned long flags;
2164 spin_lock_irqsave(&list->lock, flags);
2165 __skb_unlink(skb, list);
2166 spin_unlock_irqrestore(&list->lock, flags);
2168 EXPORT_SYMBOL(skb_unlink);
2171 * skb_append - append a buffer
2172 * @old: buffer to insert after
2173 * @newsk: buffer to insert
2174 * @list: list to use
2176 * Place a packet after a given packet in a list. The list locks are taken
2177 * and this function is atomic with respect to other list locked calls.
2178 * A buffer cannot be placed on two lists at the same time.
2180 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2182 unsigned long flags;
2184 spin_lock_irqsave(&list->lock, flags);
2185 __skb_queue_after(list, old, newsk);
2186 spin_unlock_irqrestore(&list->lock, flags);
2188 EXPORT_SYMBOL(skb_append);
2191 * skb_insert - insert a buffer
2192 * @old: buffer to insert before
2193 * @newsk: buffer to insert
2194 * @list: list to use
2196 * Place a packet before a given packet in a list. The list locks are
2197 * taken and this function is atomic with respect to other list locked
2200 * A buffer cannot be placed on two lists at the same time.
2202 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2204 unsigned long flags;
2206 spin_lock_irqsave(&list->lock, flags);
2207 __skb_insert(newsk, old->prev, old, list);
2208 spin_unlock_irqrestore(&list->lock, flags);
2210 EXPORT_SYMBOL(skb_insert);
2212 static inline void skb_split_inside_header(struct sk_buff *skb,
2213 struct sk_buff* skb1,
2214 const u32 len, const int pos)
2218 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2220 /* And move data appendix as is. */
2221 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2222 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2224 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2225 skb_shinfo(skb)->nr_frags = 0;
2226 skb1->data_len = skb->data_len;
2227 skb1->len += skb1->data_len;
2230 skb_set_tail_pointer(skb, len);
2233 static inline void skb_split_no_header(struct sk_buff *skb,
2234 struct sk_buff* skb1,
2235 const u32 len, int pos)
2238 const int nfrags = skb_shinfo(skb)->nr_frags;
2240 skb_shinfo(skb)->nr_frags = 0;
2241 skb1->len = skb1->data_len = skb->len - len;
2243 skb->data_len = len - pos;
2245 for (i = 0; i < nfrags; i++) {
2246 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2248 if (pos + size > len) {
2249 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2253 * We have two variants in this case:
2254 * 1. Move all the frag to the second
2255 * part, if it is possible. F.e.
2256 * this approach is mandatory for TUX,
2257 * where splitting is expensive.
2258 * 2. Split is accurately. We make this.
2260 skb_frag_ref(skb, i);
2261 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2262 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2263 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2264 skb_shinfo(skb)->nr_frags++;
2268 skb_shinfo(skb)->nr_frags++;
2271 skb_shinfo(skb1)->nr_frags = k;
2275 * skb_split - Split fragmented skb to two parts at length len.
2276 * @skb: the buffer to split
2277 * @skb1: the buffer to receive the second part
2278 * @len: new length for skb
2280 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2282 int pos = skb_headlen(skb);
2284 if (len < pos) /* Split line is inside header. */
2285 skb_split_inside_header(skb, skb1, len, pos);
2286 else /* Second chunk has no header, nothing to copy. */
2287 skb_split_no_header(skb, skb1, len, pos);
2289 EXPORT_SYMBOL(skb_split);
2291 /* Shifting from/to a cloned skb is a no-go.
2293 * Caller cannot keep skb_shinfo related pointers past calling here!
2295 static int skb_prepare_for_shift(struct sk_buff *skb)
2297 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2301 * skb_shift - Shifts paged data partially from skb to another
2302 * @tgt: buffer into which tail data gets added
2303 * @skb: buffer from which the paged data comes from
2304 * @shiftlen: shift up to this many bytes
2306 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2307 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2308 * It's up to caller to free skb if everything was shifted.
2310 * If @tgt runs out of frags, the whole operation is aborted.
2312 * Skb cannot include anything else but paged data while tgt is allowed
2313 * to have non-paged data as well.
2315 * TODO: full sized shift could be optimized but that would need
2316 * specialized skb free'er to handle frags without up-to-date nr_frags.
2318 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2320 int from, to, merge, todo;
2321 struct skb_frag_struct *fragfrom, *fragto;
2323 BUG_ON(shiftlen > skb->len);
2324 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2328 to = skb_shinfo(tgt)->nr_frags;
2329 fragfrom = &skb_shinfo(skb)->frags[from];
2331 /* Actual merge is delayed until the point when we know we can
2332 * commit all, so that we don't have to undo partial changes
2335 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2336 fragfrom->page_offset)) {
2341 todo -= skb_frag_size(fragfrom);
2343 if (skb_prepare_for_shift(skb) ||
2344 skb_prepare_for_shift(tgt))
2347 /* All previous frag pointers might be stale! */
2348 fragfrom = &skb_shinfo(skb)->frags[from];
2349 fragto = &skb_shinfo(tgt)->frags[merge];
2351 skb_frag_size_add(fragto, shiftlen);
2352 skb_frag_size_sub(fragfrom, shiftlen);
2353 fragfrom->page_offset += shiftlen;
2361 /* Skip full, not-fitting skb to avoid expensive operations */
2362 if ((shiftlen == skb->len) &&
2363 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2366 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2369 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2370 if (to == MAX_SKB_FRAGS)
2373 fragfrom = &skb_shinfo(skb)->frags[from];
2374 fragto = &skb_shinfo(tgt)->frags[to];
2376 if (todo >= skb_frag_size(fragfrom)) {
2377 *fragto = *fragfrom;
2378 todo -= skb_frag_size(fragfrom);
2383 __skb_frag_ref(fragfrom);
2384 fragto->page = fragfrom->page;
2385 fragto->page_offset = fragfrom->page_offset;
2386 skb_frag_size_set(fragto, todo);
2388 fragfrom->page_offset += todo;
2389 skb_frag_size_sub(fragfrom, todo);
2397 /* Ready to "commit" this state change to tgt */
2398 skb_shinfo(tgt)->nr_frags = to;
2401 fragfrom = &skb_shinfo(skb)->frags[0];
2402 fragto = &skb_shinfo(tgt)->frags[merge];
2404 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2405 __skb_frag_unref(fragfrom);
2408 /* Reposition in the original skb */
2410 while (from < skb_shinfo(skb)->nr_frags)
2411 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2412 skb_shinfo(skb)->nr_frags = to;
2414 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2417 /* Most likely the tgt won't ever need its checksum anymore, skb on
2418 * the other hand might need it if it needs to be resent
2420 tgt->ip_summed = CHECKSUM_PARTIAL;
2421 skb->ip_summed = CHECKSUM_PARTIAL;
2423 /* Yak, is it really working this way? Some helper please? */
2424 skb->len -= shiftlen;
2425 skb->data_len -= shiftlen;
2426 skb->truesize -= shiftlen;
2427 tgt->len += shiftlen;
2428 tgt->data_len += shiftlen;
2429 tgt->truesize += shiftlen;
2435 * skb_prepare_seq_read - Prepare a sequential read of skb data
2436 * @skb: the buffer to read
2437 * @from: lower offset of data to be read
2438 * @to: upper offset of data to be read
2439 * @st: state variable
2441 * Initializes the specified state variable. Must be called before
2442 * invoking skb_seq_read() for the first time.
2444 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2445 unsigned int to, struct skb_seq_state *st)
2447 st->lower_offset = from;
2448 st->upper_offset = to;
2449 st->root_skb = st->cur_skb = skb;
2450 st->frag_idx = st->stepped_offset = 0;
2451 st->frag_data = NULL;
2453 EXPORT_SYMBOL(skb_prepare_seq_read);
2456 * skb_seq_read - Sequentially read skb data
2457 * @consumed: number of bytes consumed by the caller so far
2458 * @data: destination pointer for data to be returned
2459 * @st: state variable
2461 * Reads a block of skb data at &consumed relative to the
2462 * lower offset specified to skb_prepare_seq_read(). Assigns
2463 * the head of the data block to &data and returns the length
2464 * of the block or 0 if the end of the skb data or the upper
2465 * offset has been reached.
2467 * The caller is not required to consume all of the data
2468 * returned, i.e. &consumed is typically set to the number
2469 * of bytes already consumed and the next call to
2470 * skb_seq_read() will return the remaining part of the block.
2472 * Note 1: The size of each block of data returned can be arbitrary,
2473 * this limitation is the cost for zerocopy seqeuental
2474 * reads of potentially non linear data.
2476 * Note 2: Fragment lists within fragments are not implemented
2477 * at the moment, state->root_skb could be replaced with
2478 * a stack for this purpose.
2480 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2481 struct skb_seq_state *st)
2483 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2486 if (unlikely(abs_offset >= st->upper_offset))
2490 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2492 if (abs_offset < block_limit && !st->frag_data) {
2493 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2494 return block_limit - abs_offset;
2497 if (st->frag_idx == 0 && !st->frag_data)
2498 st->stepped_offset += skb_headlen(st->cur_skb);
2500 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2501 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2502 block_limit = skb_frag_size(frag) + st->stepped_offset;
2504 if (abs_offset < block_limit) {
2506 st->frag_data = kmap_atomic(skb_frag_page(frag));
2508 *data = (u8 *) st->frag_data + frag->page_offset +
2509 (abs_offset - st->stepped_offset);
2511 return block_limit - abs_offset;
2514 if (st->frag_data) {
2515 kunmap_atomic(st->frag_data);
2516 st->frag_data = NULL;
2520 st->stepped_offset += skb_frag_size(frag);
2523 if (st->frag_data) {
2524 kunmap_atomic(st->frag_data);
2525 st->frag_data = NULL;
2528 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2529 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2532 } else if (st->cur_skb->next) {
2533 st->cur_skb = st->cur_skb->next;
2540 EXPORT_SYMBOL(skb_seq_read);
2543 * skb_abort_seq_read - Abort a sequential read of skb data
2544 * @st: state variable
2546 * Must be called if skb_seq_read() was not called until it
2549 void skb_abort_seq_read(struct skb_seq_state *st)
2552 kunmap_atomic(st->frag_data);
2554 EXPORT_SYMBOL(skb_abort_seq_read);
2556 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2558 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2559 struct ts_config *conf,
2560 struct ts_state *state)
2562 return skb_seq_read(offset, text, TS_SKB_CB(state));
2565 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2567 skb_abort_seq_read(TS_SKB_CB(state));
2571 * skb_find_text - Find a text pattern in skb data
2572 * @skb: the buffer to look in
2573 * @from: search offset
2575 * @config: textsearch configuration
2576 * @state: uninitialized textsearch state variable
2578 * Finds a pattern in the skb data according to the specified
2579 * textsearch configuration. Use textsearch_next() to retrieve
2580 * subsequent occurrences of the pattern. Returns the offset
2581 * to the first occurrence or UINT_MAX if no match was found.
2583 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2584 unsigned int to, struct ts_config *config,
2585 struct ts_state *state)
2589 config->get_next_block = skb_ts_get_next_block;
2590 config->finish = skb_ts_finish;
2592 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2594 ret = textsearch_find(config, state);
2595 return (ret <= to - from ? ret : UINT_MAX);
2597 EXPORT_SYMBOL(skb_find_text);
2600 * skb_append_datato_frags: - append the user data to a skb
2601 * @sk: sock structure
2602 * @skb: skb structure to be appened with user data.
2603 * @getfrag: call back function to be used for getting the user data
2604 * @from: pointer to user message iov
2605 * @length: length of the iov message
2607 * Description: This procedure append the user data in the fragment part
2608 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2610 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2611 int (*getfrag)(void *from, char *to, int offset,
2612 int len, int odd, struct sk_buff *skb),
2613 void *from, int length)
2616 skb_frag_t *frag = NULL;
2617 struct page *page = NULL;
2623 /* Return error if we don't have space for new frag */
2624 frg_cnt = skb_shinfo(skb)->nr_frags;
2625 if (frg_cnt >= MAX_SKB_FRAGS)
2628 /* allocate a new page for next frag */
2629 page = alloc_pages(sk->sk_allocation, 0);
2631 /* If alloc_page fails just return failure and caller will
2632 * free previous allocated pages by doing kfree_skb()
2637 /* initialize the next frag */
2638 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2639 skb->truesize += PAGE_SIZE;
2640 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2642 /* get the new initialized frag */
2643 frg_cnt = skb_shinfo(skb)->nr_frags;
2644 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2646 /* copy the user data to page */
2647 left = PAGE_SIZE - frag->page_offset;
2648 copy = (length > left)? left : length;
2650 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2651 offset, copy, 0, skb);
2655 /* copy was successful so update the size parameters */
2656 skb_frag_size_add(frag, copy);
2658 skb->data_len += copy;
2662 } while (length > 0);
2666 EXPORT_SYMBOL(skb_append_datato_frags);
2669 * skb_pull_rcsum - pull skb and update receive checksum
2670 * @skb: buffer to update
2671 * @len: length of data pulled
2673 * This function performs an skb_pull on the packet and updates
2674 * the CHECKSUM_COMPLETE checksum. It should be used on
2675 * receive path processing instead of skb_pull unless you know
2676 * that the checksum difference is zero (e.g., a valid IP header)
2677 * or you are setting ip_summed to CHECKSUM_NONE.
2679 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2681 BUG_ON(len > skb->len);
2683 BUG_ON(skb->len < skb->data_len);
2684 skb_postpull_rcsum(skb, skb->data, len);
2685 return skb->data += len;
2687 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2690 * skb_segment - Perform protocol segmentation on skb.
2691 * @skb: buffer to segment
2692 * @features: features for the output path (see dev->features)
2694 * This function performs segmentation on the given skb. It returns
2695 * a pointer to the first in a list of new skbs for the segments.
2696 * In case of error it returns ERR_PTR(err).
2698 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2700 struct sk_buff *segs = NULL;
2701 struct sk_buff *tail = NULL;
2702 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2703 unsigned int mss = skb_shinfo(skb)->gso_size;
2704 unsigned int doffset = skb->data - skb_mac_header(skb);
2705 unsigned int offset = doffset;
2706 unsigned int headroom;
2708 int sg = !!(features & NETIF_F_SG);
2709 int nfrags = skb_shinfo(skb)->nr_frags;
2714 __skb_push(skb, doffset);
2715 headroom = skb_headroom(skb);
2716 pos = skb_headlen(skb);
2719 struct sk_buff *nskb;
2724 len = skb->len - offset;
2728 hsize = skb_headlen(skb) - offset;
2731 if (hsize > len || !sg)
2734 if (!hsize && i >= nfrags) {
2735 BUG_ON(fskb->len != len);
2738 nskb = skb_clone(fskb, GFP_ATOMIC);
2741 if (unlikely(!nskb))
2744 hsize = skb_end_pointer(nskb) - nskb->head;
2745 if (skb_cow_head(nskb, doffset + headroom)) {
2750 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2752 skb_release_head_state(nskb);
2753 __skb_push(nskb, doffset);
2755 nskb = alloc_skb(hsize + doffset + headroom,
2758 if (unlikely(!nskb))
2761 skb_reserve(nskb, headroom);
2762 __skb_put(nskb, doffset);
2771 __copy_skb_header(nskb, skb);
2772 nskb->mac_len = skb->mac_len;
2774 /* nskb and skb might have different headroom */
2775 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2776 nskb->csum_start += skb_headroom(nskb) - headroom;
2778 skb_reset_mac_header(nskb);
2779 skb_set_network_header(nskb, skb->mac_len);
2780 nskb->transport_header = (nskb->network_header +
2781 skb_network_header_len(skb));
2782 skb_copy_from_linear_data(skb, nskb->data, doffset);
2784 if (fskb != skb_shinfo(skb)->frag_list)
2788 nskb->ip_summed = CHECKSUM_NONE;
2789 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2795 frag = skb_shinfo(nskb)->frags;
2797 skb_copy_from_linear_data_offset(skb, offset,
2798 skb_put(nskb, hsize), hsize);
2800 while (pos < offset + len && i < nfrags) {
2801 *frag = skb_shinfo(skb)->frags[i];
2802 __skb_frag_ref(frag);
2803 size = skb_frag_size(frag);
2806 frag->page_offset += offset - pos;
2807 skb_frag_size_sub(frag, offset - pos);
2810 skb_shinfo(nskb)->nr_frags++;
2812 if (pos + size <= offset + len) {
2816 skb_frag_size_sub(frag, pos + size - (offset + len));
2823 if (pos < offset + len) {
2824 struct sk_buff *fskb2 = fskb;
2826 BUG_ON(pos + fskb->len != offset + len);
2832 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2838 SKB_FRAG_ASSERT(nskb);
2839 skb_shinfo(nskb)->frag_list = fskb2;
2843 nskb->data_len = len - hsize;
2844 nskb->len += nskb->data_len;
2845 nskb->truesize += nskb->data_len;
2846 } while ((offset += len) < skb->len);
2851 while ((skb = segs)) {
2855 return ERR_PTR(err);
2857 EXPORT_SYMBOL_GPL(skb_segment);
2859 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2861 struct sk_buff *p = *head;
2862 struct sk_buff *nskb;
2863 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2864 struct skb_shared_info *pinfo = skb_shinfo(p);
2865 unsigned int headroom;
2866 unsigned int len = skb_gro_len(skb);
2867 unsigned int offset = skb_gro_offset(skb);
2868 unsigned int headlen = skb_headlen(skb);
2870 if (p->len + len >= 65536)
2873 if (pinfo->frag_list)
2875 else if (headlen <= offset) {
2878 int i = skbinfo->nr_frags;
2879 int nr_frags = pinfo->nr_frags + i;
2883 if (nr_frags > MAX_SKB_FRAGS)
2886 pinfo->nr_frags = nr_frags;
2887 skbinfo->nr_frags = 0;
2889 frag = pinfo->frags + nr_frags;
2890 frag2 = skbinfo->frags + i;
2895 frag->page_offset += offset;
2896 skb_frag_size_sub(frag, offset);
2898 skb->truesize -= skb->data_len;
2899 skb->len -= skb->data_len;
2902 NAPI_GRO_CB(skb)->free = 1;
2904 } else if (skb_gro_len(p) != pinfo->gso_size)
2907 headroom = skb_headroom(p);
2908 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2909 if (unlikely(!nskb))
2912 __copy_skb_header(nskb, p);
2913 nskb->mac_len = p->mac_len;
2915 skb_reserve(nskb, headroom);
2916 __skb_put(nskb, skb_gro_offset(p));
2918 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2919 skb_set_network_header(nskb, skb_network_offset(p));
2920 skb_set_transport_header(nskb, skb_transport_offset(p));
2922 __skb_pull(p, skb_gro_offset(p));
2923 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2924 p->data - skb_mac_header(p));
2926 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2927 skb_shinfo(nskb)->frag_list = p;
2928 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2929 pinfo->gso_size = 0;
2930 skb_header_release(p);
2933 nskb->data_len += p->len;
2934 nskb->truesize += p->truesize;
2935 nskb->len += p->len;
2938 nskb->next = p->next;
2944 p->truesize += skb->truesize - len;
2945 if (offset > headlen) {
2946 unsigned int eat = offset - headlen;
2948 skbinfo->frags[0].page_offset += eat;
2949 skb_frag_size_sub(&skbinfo->frags[0], eat);
2950 skb->data_len -= eat;
2955 __skb_pull(skb, offset);
2957 p->prev->next = skb;
2959 skb_header_release(skb);
2962 NAPI_GRO_CB(p)->count++;
2967 NAPI_GRO_CB(skb)->same_flow = 1;
2970 EXPORT_SYMBOL_GPL(skb_gro_receive);
2972 void __init skb_init(void)
2974 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2975 sizeof(struct sk_buff),
2977 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2979 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2980 (2*sizeof(struct sk_buff)) +
2983 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2988 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2989 * @skb: Socket buffer containing the buffers to be mapped
2990 * @sg: The scatter-gather list to map into
2991 * @offset: The offset into the buffer's contents to start mapping
2992 * @len: Length of buffer space to be mapped
2994 * Fill the specified scatter-gather list with mappings/pointers into a
2995 * region of the buffer space attached to a socket buffer.
2998 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3000 int start = skb_headlen(skb);
3001 int i, copy = start - offset;
3002 struct sk_buff *frag_iter;
3008 sg_set_buf(sg, skb->data + offset, copy);
3010 if ((len -= copy) == 0)
3015 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3018 WARN_ON(start > offset + len);
3020 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3021 if ((copy = end - offset) > 0) {
3022 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3026 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3027 frag->page_offset+offset-start);
3036 skb_walk_frags(skb, frag_iter) {
3039 WARN_ON(start > offset + len);
3041 end = start + frag_iter->len;
3042 if ((copy = end - offset) > 0) {
3045 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3047 if ((len -= copy) == 0)
3057 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3059 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3061 sg_mark_end(&sg[nsg - 1]);
3065 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3068 * skb_cow_data - Check that a socket buffer's data buffers are writable
3069 * @skb: The socket buffer to check.
3070 * @tailbits: Amount of trailing space to be added
3071 * @trailer: Returned pointer to the skb where the @tailbits space begins
3073 * Make sure that the data buffers attached to a socket buffer are
3074 * writable. If they are not, private copies are made of the data buffers
3075 * and the socket buffer is set to use these instead.
3077 * If @tailbits is given, make sure that there is space to write @tailbits
3078 * bytes of data beyond current end of socket buffer. @trailer will be
3079 * set to point to the skb in which this space begins.
3081 * The number of scatterlist elements required to completely map the
3082 * COW'd and extended socket buffer will be returned.
3084 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3088 struct sk_buff *skb1, **skb_p;
3090 /* If skb is cloned or its head is paged, reallocate
3091 * head pulling out all the pages (pages are considered not writable
3092 * at the moment even if they are anonymous).
3094 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3095 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3098 /* Easy case. Most of packets will go this way. */
3099 if (!skb_has_frag_list(skb)) {
3100 /* A little of trouble, not enough of space for trailer.
3101 * This should not happen, when stack is tuned to generate
3102 * good frames. OK, on miss we reallocate and reserve even more
3103 * space, 128 bytes is fair. */
3105 if (skb_tailroom(skb) < tailbits &&
3106 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3114 /* Misery. We are in troubles, going to mincer fragments... */
3117 skb_p = &skb_shinfo(skb)->frag_list;
3120 while ((skb1 = *skb_p) != NULL) {
3123 /* The fragment is partially pulled by someone,
3124 * this can happen on input. Copy it and everything
3127 if (skb_shared(skb1))
3130 /* If the skb is the last, worry about trailer. */
3132 if (skb1->next == NULL && tailbits) {
3133 if (skb_shinfo(skb1)->nr_frags ||
3134 skb_has_frag_list(skb1) ||
3135 skb_tailroom(skb1) < tailbits)
3136 ntail = tailbits + 128;
3142 skb_shinfo(skb1)->nr_frags ||
3143 skb_has_frag_list(skb1)) {
3144 struct sk_buff *skb2;
3146 /* Fuck, we are miserable poor guys... */
3148 skb2 = skb_copy(skb1, GFP_ATOMIC);
3150 skb2 = skb_copy_expand(skb1,
3154 if (unlikely(skb2 == NULL))
3158 skb_set_owner_w(skb2, skb1->sk);
3160 /* Looking around. Are we still alive?
3161 * OK, link new skb, drop old one */
3163 skb2->next = skb1->next;
3170 skb_p = &skb1->next;
3175 EXPORT_SYMBOL_GPL(skb_cow_data);
3177 static void sock_rmem_free(struct sk_buff *skb)
3179 struct sock *sk = skb->sk;
3181 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3185 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3187 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3191 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3192 (unsigned int)sk->sk_rcvbuf)
3197 skb->destructor = sock_rmem_free;
3198 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3200 /* before exiting rcu section, make sure dst is refcounted */
3203 skb_queue_tail(&sk->sk_error_queue, skb);
3204 if (!sock_flag(sk, SOCK_DEAD))
3205 sk->sk_data_ready(sk, len);
3208 EXPORT_SYMBOL(sock_queue_err_skb);
3210 void skb_tstamp_tx(struct sk_buff *orig_skb,
3211 struct skb_shared_hwtstamps *hwtstamps)
3213 struct sock *sk = orig_skb->sk;
3214 struct sock_exterr_skb *serr;
3215 struct sk_buff *skb;
3221 skb = skb_clone(orig_skb, GFP_ATOMIC);
3226 *skb_hwtstamps(skb) =
3230 * no hardware time stamps available,
3231 * so keep the shared tx_flags and only
3232 * store software time stamp
3234 skb->tstamp = ktime_get_real();
3237 serr = SKB_EXT_ERR(skb);
3238 memset(serr, 0, sizeof(*serr));
3239 serr->ee.ee_errno = ENOMSG;
3240 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3242 err = sock_queue_err_skb(sk, skb);
3247 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3249 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3251 struct sock *sk = skb->sk;
3252 struct sock_exterr_skb *serr;
3255 skb->wifi_acked_valid = 1;
3256 skb->wifi_acked = acked;
3258 serr = SKB_EXT_ERR(skb);
3259 memset(serr, 0, sizeof(*serr));
3260 serr->ee.ee_errno = ENOMSG;
3261 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3263 err = sock_queue_err_skb(sk, skb);
3267 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3271 * skb_partial_csum_set - set up and verify partial csum values for packet
3272 * @skb: the skb to set
3273 * @start: the number of bytes after skb->data to start checksumming.
3274 * @off: the offset from start to place the checksum.
3276 * For untrusted partially-checksummed packets, we need to make sure the values
3277 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3279 * This function checks and sets those values and skb->ip_summed: if this
3280 * returns false you should drop the packet.
3282 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3284 if (unlikely(start > skb_headlen(skb)) ||
3285 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3286 if (net_ratelimit())
3288 "bad partial csum: csum=%u/%u len=%u\n",
3289 start, off, skb_headlen(skb));
3292 skb->ip_summed = CHECKSUM_PARTIAL;
3293 skb->csum_start = skb_headroom(skb) + start;
3294 skb->csum_offset = off;
3297 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3299 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3301 if (net_ratelimit())
3302 pr_warning("%s: received packets cannot be forwarded"
3303 " while LRO is enabled\n", skb->dev->name);
3305 EXPORT_SYMBOL(__skb_warn_lro_forwarding);