xref: /openbmc/linux/net/core/skbuff.c (revision bc2cda1e)
11da177e4SLinus Torvalds /*
21da177e4SLinus Torvalds  *	Routines having to do with the 'struct sk_buff' memory handlers.
31da177e4SLinus Torvalds  *
41da177e4SLinus Torvalds  *	Authors:	Alan Cox <iiitac@pyr.swan.ac.uk>
51da177e4SLinus Torvalds  *			Florian La Roche <rzsfl@rz.uni-sb.de>
61da177e4SLinus Torvalds  *
71da177e4SLinus Torvalds  *	Version:	$Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
81da177e4SLinus Torvalds  *
91da177e4SLinus Torvalds  *	Fixes:
101da177e4SLinus Torvalds  *		Alan Cox	:	Fixed the worst of the load
111da177e4SLinus Torvalds  *					balancer bugs.
121da177e4SLinus Torvalds  *		Dave Platt	:	Interrupt stacking fix.
131da177e4SLinus Torvalds  *	Richard Kooijman	:	Timestamp fixes.
141da177e4SLinus Torvalds  *		Alan Cox	:	Changed buffer format.
151da177e4SLinus Torvalds  *		Alan Cox	:	destructor hook for AF_UNIX etc.
161da177e4SLinus Torvalds  *		Linus Torvalds	:	Better skb_clone.
171da177e4SLinus Torvalds  *		Alan Cox	:	Added skb_copy.
181da177e4SLinus Torvalds  *		Alan Cox	:	Added all the changed routines Linus
191da177e4SLinus Torvalds  *					only put in the headers
201da177e4SLinus Torvalds  *		Ray VanTassle	:	Fixed --skb->lock in free
211da177e4SLinus Torvalds  *		Alan Cox	:	skb_copy copy arp field
221da177e4SLinus Torvalds  *		Andi Kleen	:	slabified it.
231da177e4SLinus Torvalds  *		Robert Olsson	:	Removed skb_head_pool
241da177e4SLinus Torvalds  *
251da177e4SLinus Torvalds  *	NOTE:
261da177e4SLinus Torvalds  *		The __skb_ routines should be called with interrupts
271da177e4SLinus Torvalds  *	disabled, or you better be *real* sure that the operation is atomic
281da177e4SLinus Torvalds  *	with respect to whatever list is being frobbed (e.g. via lock_sock()
291da177e4SLinus Torvalds  *	or via disabling bottom half handlers, etc).
301da177e4SLinus Torvalds  *
311da177e4SLinus Torvalds  *	This program is free software; you can redistribute it and/or
321da177e4SLinus Torvalds  *	modify it under the terms of the GNU General Public License
331da177e4SLinus Torvalds  *	as published by the Free Software Foundation; either version
341da177e4SLinus Torvalds  *	2 of the License, or (at your option) any later version.
351da177e4SLinus Torvalds  */
361da177e4SLinus Torvalds 
371da177e4SLinus Torvalds /*
381da177e4SLinus Torvalds  *	The functions in this file will not compile correctly with gcc 2.4.x
391da177e4SLinus Torvalds  */
401da177e4SLinus Torvalds 
411da177e4SLinus Torvalds #include <linux/module.h>
421da177e4SLinus Torvalds #include <linux/types.h>
431da177e4SLinus Torvalds #include <linux/kernel.h>
441da177e4SLinus Torvalds #include <linux/mm.h>
451da177e4SLinus Torvalds #include <linux/interrupt.h>
461da177e4SLinus Torvalds #include <linux/in.h>
471da177e4SLinus Torvalds #include <linux/inet.h>
481da177e4SLinus Torvalds #include <linux/slab.h>
491da177e4SLinus Torvalds #include <linux/netdevice.h>
501da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT
511da177e4SLinus Torvalds #include <net/pkt_sched.h>
521da177e4SLinus Torvalds #endif
531da177e4SLinus Torvalds #include <linux/string.h>
541da177e4SLinus Torvalds #include <linux/skbuff.h>
559c55e01cSJens Axboe #include <linux/splice.h>
561da177e4SLinus Torvalds #include <linux/cache.h>
571da177e4SLinus Torvalds #include <linux/rtnetlink.h>
581da177e4SLinus Torvalds #include <linux/init.h>
59716ea3a7SDavid Howells #include <linux/scatterlist.h>
601da177e4SLinus Torvalds 
611da177e4SLinus Torvalds #include <net/protocol.h>
621da177e4SLinus Torvalds #include <net/dst.h>
631da177e4SLinus Torvalds #include <net/sock.h>
641da177e4SLinus Torvalds #include <net/checksum.h>
651da177e4SLinus Torvalds #include <net/xfrm.h>
661da177e4SLinus Torvalds 
671da177e4SLinus Torvalds #include <asm/uaccess.h>
681da177e4SLinus Torvalds #include <asm/system.h>
691da177e4SLinus Torvalds 
70a1f8e7f7SAl Viro #include "kmap_skb.h"
71a1f8e7f7SAl Viro 
72e18b890bSChristoph Lameter static struct kmem_cache *skbuff_head_cache __read_mostly;
73e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly;
741da177e4SLinus Torvalds 
759c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
769c55e01cSJens Axboe 				  struct pipe_buffer *buf)
779c55e01cSJens Axboe {
789c55e01cSJens Axboe 	struct sk_buff *skb = (struct sk_buff *) buf->private;
799c55e01cSJens Axboe 
809c55e01cSJens Axboe 	kfree_skb(skb);
819c55e01cSJens Axboe }
829c55e01cSJens Axboe 
839c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
849c55e01cSJens Axboe 				struct pipe_buffer *buf)
859c55e01cSJens Axboe {
869c55e01cSJens Axboe 	struct sk_buff *skb = (struct sk_buff *) buf->private;
879c55e01cSJens Axboe 
889c55e01cSJens Axboe 	skb_get(skb);
899c55e01cSJens Axboe }
909c55e01cSJens Axboe 
919c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
929c55e01cSJens Axboe 			       struct pipe_buffer *buf)
939c55e01cSJens Axboe {
949c55e01cSJens Axboe 	return 1;
959c55e01cSJens Axboe }
969c55e01cSJens Axboe 
979c55e01cSJens Axboe 
989c55e01cSJens Axboe /* Pipe buffer operations for a socket. */
999c55e01cSJens Axboe static struct pipe_buf_operations sock_pipe_buf_ops = {
1009c55e01cSJens Axboe 	.can_merge = 0,
1019c55e01cSJens Axboe 	.map = generic_pipe_buf_map,
1029c55e01cSJens Axboe 	.unmap = generic_pipe_buf_unmap,
1039c55e01cSJens Axboe 	.confirm = generic_pipe_buf_confirm,
1049c55e01cSJens Axboe 	.release = sock_pipe_buf_release,
1059c55e01cSJens Axboe 	.steal = sock_pipe_buf_steal,
1069c55e01cSJens Axboe 	.get = sock_pipe_buf_get,
1079c55e01cSJens Axboe };
1089c55e01cSJens Axboe 
1091da177e4SLinus Torvalds /*
1101da177e4SLinus Torvalds  *	Keep out-of-line to prevent kernel bloat.
1111da177e4SLinus Torvalds  *	__builtin_return_address is not used because it is not always
1121da177e4SLinus Torvalds  *	reliable.
1131da177e4SLinus Torvalds  */
1141da177e4SLinus Torvalds 
1151da177e4SLinus Torvalds /**
1161da177e4SLinus Torvalds  *	skb_over_panic	- 	private function
1171da177e4SLinus Torvalds  *	@skb: buffer
1181da177e4SLinus Torvalds  *	@sz: size
1191da177e4SLinus Torvalds  *	@here: address
1201da177e4SLinus Torvalds  *
1211da177e4SLinus Torvalds  *	Out of line support code for skb_put(). Not user callable.
1221da177e4SLinus Torvalds  */
1231da177e4SLinus Torvalds void skb_over_panic(struct sk_buff *skb, int sz, void *here)
1241da177e4SLinus Torvalds {
12526095455SPatrick McHardy 	printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
1264305b541SArnaldo Carvalho de Melo 			  "data:%p tail:%#lx end:%#lx dev:%s\n",
12727a884dcSArnaldo Carvalho de Melo 	       here, skb->len, sz, skb->head, skb->data,
1284305b541SArnaldo Carvalho de Melo 	       (unsigned long)skb->tail, (unsigned long)skb->end,
12926095455SPatrick McHardy 	       skb->dev ? skb->dev->name : "<NULL>");
1301da177e4SLinus Torvalds 	BUG();
1311da177e4SLinus Torvalds }
1321da177e4SLinus Torvalds 
1331da177e4SLinus Torvalds /**
1341da177e4SLinus Torvalds  *	skb_under_panic	- 	private function
1351da177e4SLinus Torvalds  *	@skb: buffer
1361da177e4SLinus Torvalds  *	@sz: size
1371da177e4SLinus Torvalds  *	@here: address
1381da177e4SLinus Torvalds  *
1391da177e4SLinus Torvalds  *	Out of line support code for skb_push(). Not user callable.
1401da177e4SLinus Torvalds  */
1411da177e4SLinus Torvalds 
1421da177e4SLinus Torvalds void skb_under_panic(struct sk_buff *skb, int sz, void *here)
1431da177e4SLinus Torvalds {
14426095455SPatrick McHardy 	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
1454305b541SArnaldo Carvalho de Melo 			  "data:%p tail:%#lx end:%#lx dev:%s\n",
14627a884dcSArnaldo Carvalho de Melo 	       here, skb->len, sz, skb->head, skb->data,
1474305b541SArnaldo Carvalho de Melo 	       (unsigned long)skb->tail, (unsigned long)skb->end,
14826095455SPatrick McHardy 	       skb->dev ? skb->dev->name : "<NULL>");
1491da177e4SLinus Torvalds 	BUG();
1501da177e4SLinus Torvalds }
1511da177e4SLinus Torvalds 
152dc6de336SDavid S. Miller void skb_truesize_bug(struct sk_buff *skb)
153dc6de336SDavid S. Miller {
154dc6de336SDavid S. Miller 	printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
155dc6de336SDavid S. Miller 	       "len=%u, sizeof(sk_buff)=%Zd\n",
156dc6de336SDavid S. Miller 	       skb->truesize, skb->len, sizeof(struct sk_buff));
157dc6de336SDavid S. Miller }
158dc6de336SDavid S. Miller EXPORT_SYMBOL(skb_truesize_bug);
159dc6de336SDavid S. Miller 
1601da177e4SLinus Torvalds /* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
1611da177e4SLinus Torvalds  *	'private' fields and also do memory statistics to find all the
1621da177e4SLinus Torvalds  *	[BEEP] leaks.
1631da177e4SLinus Torvalds  *
1641da177e4SLinus Torvalds  */
1651da177e4SLinus Torvalds 
1661da177e4SLinus Torvalds /**
167d179cd12SDavid S. Miller  *	__alloc_skb	-	allocate a network buffer
1681da177e4SLinus Torvalds  *	@size: size to allocate
1691da177e4SLinus Torvalds  *	@gfp_mask: allocation mask
170c83c2486SRandy Dunlap  *	@fclone: allocate from fclone cache instead of head cache
171c83c2486SRandy Dunlap  *		and allocate a cloned (child) skb
172b30973f8SChristoph Hellwig  *	@node: numa node to allocate memory on
1731da177e4SLinus Torvalds  *
1741da177e4SLinus Torvalds  *	Allocate a new &sk_buff. The returned buffer has no headroom and a
1751da177e4SLinus Torvalds  *	tail room of size bytes. The object has a reference count of one.
1761da177e4SLinus Torvalds  *	The return is the buffer. On a failure the return is %NULL.
1771da177e4SLinus Torvalds  *
1781da177e4SLinus Torvalds  *	Buffers may only be allocated from interrupts using a @gfp_mask of
1791da177e4SLinus Torvalds  *	%GFP_ATOMIC.
1801da177e4SLinus Torvalds  */
181dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
182b30973f8SChristoph Hellwig 			    int fclone, int node)
1831da177e4SLinus Torvalds {
184e18b890bSChristoph Lameter 	struct kmem_cache *cache;
1854947d3efSBenjamin LaHaise 	struct skb_shared_info *shinfo;
1861da177e4SLinus Torvalds 	struct sk_buff *skb;
1871da177e4SLinus Torvalds 	u8 *data;
1881da177e4SLinus Torvalds 
1898798b3fbSHerbert Xu 	cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
1908798b3fbSHerbert Xu 
1911da177e4SLinus Torvalds 	/* Get the HEAD */
192b30973f8SChristoph Hellwig 	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
1931da177e4SLinus Torvalds 	if (!skb)
1941da177e4SLinus Torvalds 		goto out;
1951da177e4SLinus Torvalds 
1961da177e4SLinus Torvalds 	size = SKB_DATA_ALIGN(size);
197b30973f8SChristoph Hellwig 	data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
198b30973f8SChristoph Hellwig 			gfp_mask, node);
1991da177e4SLinus Torvalds 	if (!data)
2001da177e4SLinus Torvalds 		goto nodata;
2011da177e4SLinus Torvalds 
202ca0605a7SArnaldo Carvalho de Melo 	/*
203ca0605a7SArnaldo Carvalho de Melo 	 * See comment in sk_buff definition, just before the 'tail' member
204ca0605a7SArnaldo Carvalho de Melo 	 */
205ca0605a7SArnaldo Carvalho de Melo 	memset(skb, 0, offsetof(struct sk_buff, tail));
2061da177e4SLinus Torvalds 	skb->truesize = size + sizeof(struct sk_buff);
2071da177e4SLinus Torvalds 	atomic_set(&skb->users, 1);
2081da177e4SLinus Torvalds 	skb->head = data;
2091da177e4SLinus Torvalds 	skb->data = data;
21027a884dcSArnaldo Carvalho de Melo 	skb_reset_tail_pointer(skb);
2114305b541SArnaldo Carvalho de Melo 	skb->end = skb->tail + size;
2124947d3efSBenjamin LaHaise 	/* make sure we initialize shinfo sequentially */
2134947d3efSBenjamin LaHaise 	shinfo = skb_shinfo(skb);
2144947d3efSBenjamin LaHaise 	atomic_set(&shinfo->dataref, 1);
2154947d3efSBenjamin LaHaise 	shinfo->nr_frags  = 0;
2167967168cSHerbert Xu 	shinfo->gso_size = 0;
2177967168cSHerbert Xu 	shinfo->gso_segs = 0;
2187967168cSHerbert Xu 	shinfo->gso_type = 0;
2194947d3efSBenjamin LaHaise 	shinfo->ip6_frag_id = 0;
2204947d3efSBenjamin LaHaise 	shinfo->frag_list = NULL;
2214947d3efSBenjamin LaHaise 
222d179cd12SDavid S. Miller 	if (fclone) {
223d179cd12SDavid S. Miller 		struct sk_buff *child = skb + 1;
224d179cd12SDavid S. Miller 		atomic_t *fclone_ref = (atomic_t *) (child + 1);
2251da177e4SLinus Torvalds 
226d179cd12SDavid S. Miller 		skb->fclone = SKB_FCLONE_ORIG;
227d179cd12SDavid S. Miller 		atomic_set(fclone_ref, 1);
228d179cd12SDavid S. Miller 
229d179cd12SDavid S. Miller 		child->fclone = SKB_FCLONE_UNAVAILABLE;
230d179cd12SDavid S. Miller 	}
2311da177e4SLinus Torvalds out:
2321da177e4SLinus Torvalds 	return skb;
2331da177e4SLinus Torvalds nodata:
2348798b3fbSHerbert Xu 	kmem_cache_free(cache, skb);
2351da177e4SLinus Torvalds 	skb = NULL;
2361da177e4SLinus Torvalds 	goto out;
2371da177e4SLinus Torvalds }
2381da177e4SLinus Torvalds 
2391da177e4SLinus Torvalds /**
2408af27456SChristoph Hellwig  *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
2418af27456SChristoph Hellwig  *	@dev: network device to receive on
2428af27456SChristoph Hellwig  *	@length: length to allocate
2438af27456SChristoph Hellwig  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
2448af27456SChristoph Hellwig  *
2458af27456SChristoph Hellwig  *	Allocate a new &sk_buff and assign it a usage count of one. The
2468af27456SChristoph Hellwig  *	buffer has unspecified headroom built in. Users should allocate
2478af27456SChristoph Hellwig  *	the headroom they think they need without accounting for the
2488af27456SChristoph Hellwig  *	built in space. The built in space is used for optimisations.
2498af27456SChristoph Hellwig  *
2508af27456SChristoph Hellwig  *	%NULL is returned if there is no free memory.
2518af27456SChristoph Hellwig  */
2528af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
2538af27456SChristoph Hellwig 		unsigned int length, gfp_t gfp_mask)
2548af27456SChristoph Hellwig {
25543cb76d9SGreg Kroah-Hartman 	int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
2568af27456SChristoph Hellwig 	struct sk_buff *skb;
2578af27456SChristoph Hellwig 
258b30973f8SChristoph Hellwig 	skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
2597b2e497aSChristoph Hellwig 	if (likely(skb)) {
2608af27456SChristoph Hellwig 		skb_reserve(skb, NET_SKB_PAD);
2617b2e497aSChristoph Hellwig 		skb->dev = dev;
2627b2e497aSChristoph Hellwig 	}
2638af27456SChristoph Hellwig 	return skb;
2648af27456SChristoph Hellwig }
2651da177e4SLinus Torvalds 
26627b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp)
2671da177e4SLinus Torvalds {
26827b437c8SHerbert Xu 	struct sk_buff *list = *listp;
2691da177e4SLinus Torvalds 
27027b437c8SHerbert Xu 	*listp = NULL;
2711da177e4SLinus Torvalds 
2721da177e4SLinus Torvalds 	do {
2731da177e4SLinus Torvalds 		struct sk_buff *this = list;
2741da177e4SLinus Torvalds 		list = list->next;
2751da177e4SLinus Torvalds 		kfree_skb(this);
2761da177e4SLinus Torvalds 	} while (list);
2771da177e4SLinus Torvalds }
2781da177e4SLinus Torvalds 
27927b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb)
28027b437c8SHerbert Xu {
28127b437c8SHerbert Xu 	skb_drop_list(&skb_shinfo(skb)->frag_list);
28227b437c8SHerbert Xu }
28327b437c8SHerbert Xu 
2841da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb)
2851da177e4SLinus Torvalds {
2861da177e4SLinus Torvalds 	struct sk_buff *list;
2871da177e4SLinus Torvalds 
2881da177e4SLinus Torvalds 	for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
2891da177e4SLinus Torvalds 		skb_get(list);
2901da177e4SLinus Torvalds }
2911da177e4SLinus Torvalds 
2925bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb)
2931da177e4SLinus Torvalds {
2941da177e4SLinus Torvalds 	if (!skb->cloned ||
2951da177e4SLinus Torvalds 	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
2961da177e4SLinus Torvalds 			       &skb_shinfo(skb)->dataref)) {
2971da177e4SLinus Torvalds 		if (skb_shinfo(skb)->nr_frags) {
2981da177e4SLinus Torvalds 			int i;
2991da177e4SLinus Torvalds 			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3001da177e4SLinus Torvalds 				put_page(skb_shinfo(skb)->frags[i].page);
3011da177e4SLinus Torvalds 		}
3021da177e4SLinus Torvalds 
3031da177e4SLinus Torvalds 		if (skb_shinfo(skb)->frag_list)
3041da177e4SLinus Torvalds 			skb_drop_fraglist(skb);
3051da177e4SLinus Torvalds 
3061da177e4SLinus Torvalds 		kfree(skb->head);
3071da177e4SLinus Torvalds 	}
3081da177e4SLinus Torvalds }
3091da177e4SLinus Torvalds 
3101da177e4SLinus Torvalds /*
3111da177e4SLinus Torvalds  *	Free an skbuff by memory without cleaning the state.
3121da177e4SLinus Torvalds  */
3132d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb)
3141da177e4SLinus Torvalds {
315d179cd12SDavid S. Miller 	struct sk_buff *other;
316d179cd12SDavid S. Miller 	atomic_t *fclone_ref;
317d179cd12SDavid S. Miller 
318d179cd12SDavid S. Miller 	switch (skb->fclone) {
319d179cd12SDavid S. Miller 	case SKB_FCLONE_UNAVAILABLE:
3201da177e4SLinus Torvalds 		kmem_cache_free(skbuff_head_cache, skb);
321d179cd12SDavid S. Miller 		break;
322d179cd12SDavid S. Miller 
323d179cd12SDavid S. Miller 	case SKB_FCLONE_ORIG:
324d179cd12SDavid S. Miller 		fclone_ref = (atomic_t *) (skb + 2);
325d179cd12SDavid S. Miller 		if (atomic_dec_and_test(fclone_ref))
326d179cd12SDavid S. Miller 			kmem_cache_free(skbuff_fclone_cache, skb);
327d179cd12SDavid S. Miller 		break;
328d179cd12SDavid S. Miller 
329d179cd12SDavid S. Miller 	case SKB_FCLONE_CLONE:
330d179cd12SDavid S. Miller 		fclone_ref = (atomic_t *) (skb + 1);
331d179cd12SDavid S. Miller 		other = skb - 1;
332d179cd12SDavid S. Miller 
333d179cd12SDavid S. Miller 		/* The clone portion is available for
334d179cd12SDavid S. Miller 		 * fast-cloning again.
335d179cd12SDavid S. Miller 		 */
336d179cd12SDavid S. Miller 		skb->fclone = SKB_FCLONE_UNAVAILABLE;
337d179cd12SDavid S. Miller 
338d179cd12SDavid S. Miller 		if (atomic_dec_and_test(fclone_ref))
339d179cd12SDavid S. Miller 			kmem_cache_free(skbuff_fclone_cache, other);
340d179cd12SDavid S. Miller 		break;
3413ff50b79SStephen Hemminger 	}
3421da177e4SLinus Torvalds }
3431da177e4SLinus Torvalds 
3442d4baff8SHerbert Xu /* Free everything but the sk_buff shell. */
3452d4baff8SHerbert Xu static void skb_release_all(struct sk_buff *skb)
3461da177e4SLinus Torvalds {
3471da177e4SLinus Torvalds 	dst_release(skb->dst);
3481da177e4SLinus Torvalds #ifdef CONFIG_XFRM
3491da177e4SLinus Torvalds 	secpath_put(skb->sp);
3501da177e4SLinus Torvalds #endif
3511da177e4SLinus Torvalds 	if (skb->destructor) {
3529c2b3328SStephen Hemminger 		WARN_ON(in_irq());
3531da177e4SLinus Torvalds 		skb->destructor(skb);
3541da177e4SLinus Torvalds 	}
3559fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3565f79e0f9SYasuyuki Kozakai 	nf_conntrack_put(skb->nfct);
3579fb9cbb1SYasuyuki Kozakai 	nf_conntrack_put_reasm(skb->nfct_reasm);
3589fb9cbb1SYasuyuki Kozakai #endif
3591da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER
3601da177e4SLinus Torvalds 	nf_bridge_put(skb->nf_bridge);
3611da177e4SLinus Torvalds #endif
3621da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */
3631da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED
3641da177e4SLinus Torvalds 	skb->tc_index = 0;
3651da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT
3661da177e4SLinus Torvalds 	skb->tc_verd = 0;
3671da177e4SLinus Torvalds #endif
3681da177e4SLinus Torvalds #endif
3692d4baff8SHerbert Xu 	skb_release_data(skb);
3702d4baff8SHerbert Xu }
3711da177e4SLinus Torvalds 
3722d4baff8SHerbert Xu /**
3732d4baff8SHerbert Xu  *	__kfree_skb - private function
3742d4baff8SHerbert Xu  *	@skb: buffer
3752d4baff8SHerbert Xu  *
3762d4baff8SHerbert Xu  *	Free an sk_buff. Release anything attached to the buffer.
3772d4baff8SHerbert Xu  *	Clean the state. This is an internal helper function. Users should
3782d4baff8SHerbert Xu  *	always call kfree_skb
3792d4baff8SHerbert Xu  */
3802d4baff8SHerbert Xu 
3812d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb)
3822d4baff8SHerbert Xu {
3832d4baff8SHerbert Xu 	skb_release_all(skb);
3841da177e4SLinus Torvalds 	kfree_skbmem(skb);
3851da177e4SLinus Torvalds }
3861da177e4SLinus Torvalds 
3871da177e4SLinus Torvalds /**
388231d06aeSJörn Engel  *	kfree_skb - free an sk_buff
389231d06aeSJörn Engel  *	@skb: buffer to free
390231d06aeSJörn Engel  *
391231d06aeSJörn Engel  *	Drop a reference to the buffer and free it if the usage count has
392231d06aeSJörn Engel  *	hit zero.
393231d06aeSJörn Engel  */
394231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb)
395231d06aeSJörn Engel {
396231d06aeSJörn Engel 	if (unlikely(!skb))
397231d06aeSJörn Engel 		return;
398231d06aeSJörn Engel 	if (likely(atomic_read(&skb->users) == 1))
399231d06aeSJörn Engel 		smp_rmb();
400231d06aeSJörn Engel 	else if (likely(!atomic_dec_and_test(&skb->users)))
401231d06aeSJörn Engel 		return;
402231d06aeSJörn Engel 	__kfree_skb(skb);
403231d06aeSJörn Engel }
404231d06aeSJörn Engel 
405dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
406dec18810SHerbert Xu {
407dec18810SHerbert Xu 	new->tstamp		= old->tstamp;
408dec18810SHerbert Xu 	new->dev		= old->dev;
409dec18810SHerbert Xu 	new->transport_header	= old->transport_header;
410dec18810SHerbert Xu 	new->network_header	= old->network_header;
411dec18810SHerbert Xu 	new->mac_header		= old->mac_header;
412dec18810SHerbert Xu 	new->dst		= dst_clone(old->dst);
413dec18810SHerbert Xu #ifdef CONFIG_INET
414dec18810SHerbert Xu 	new->sp			= secpath_get(old->sp);
415dec18810SHerbert Xu #endif
416dec18810SHerbert Xu 	memcpy(new->cb, old->cb, sizeof(old->cb));
417dec18810SHerbert Xu 	new->csum_start		= old->csum_start;
418dec18810SHerbert Xu 	new->csum_offset	= old->csum_offset;
419dec18810SHerbert Xu 	new->local_df		= old->local_df;
420dec18810SHerbert Xu 	new->pkt_type		= old->pkt_type;
421dec18810SHerbert Xu 	new->ip_summed		= old->ip_summed;
422dec18810SHerbert Xu 	skb_copy_queue_mapping(new, old);
423dec18810SHerbert Xu 	new->priority		= old->priority;
424dec18810SHerbert Xu #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
425dec18810SHerbert Xu 	new->ipvs_property	= old->ipvs_property;
426dec18810SHerbert Xu #endif
427dec18810SHerbert Xu 	new->protocol		= old->protocol;
428dec18810SHerbert Xu 	new->mark		= old->mark;
429dec18810SHerbert Xu 	__nf_copy(new, old);
430dec18810SHerbert Xu #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
431dec18810SHerbert Xu     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
432dec18810SHerbert Xu 	new->nf_trace		= old->nf_trace;
433dec18810SHerbert Xu #endif
434dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED
435dec18810SHerbert Xu 	new->tc_index		= old->tc_index;
436dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT
437dec18810SHerbert Xu 	new->tc_verd		= old->tc_verd;
438dec18810SHerbert Xu #endif
439dec18810SHerbert Xu #endif
440dec18810SHerbert Xu 	skb_copy_secmark(new, old);
441dec18810SHerbert Xu }
442dec18810SHerbert Xu 
443e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
4441da177e4SLinus Torvalds {
4451da177e4SLinus Torvalds #define C(x) n->x = skb->x
4461da177e4SLinus Torvalds 
4471da177e4SLinus Torvalds 	n->next = n->prev = NULL;
4481da177e4SLinus Torvalds 	n->sk = NULL;
449dec18810SHerbert Xu 	__copy_skb_header(n, skb);
450dec18810SHerbert Xu 
4511da177e4SLinus Torvalds 	C(len);
4521da177e4SLinus Torvalds 	C(data_len);
4533e6b3b2eSAlexey Dobriyan 	C(mac_len);
454334a8132SPatrick McHardy 	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
45502f1c89dSPaul Moore 	n->cloned = 1;
4561da177e4SLinus Torvalds 	n->nohdr = 0;
4571da177e4SLinus Torvalds 	n->destructor = NULL;
45802f1c89dSPaul Moore 	C(iif);
4591da177e4SLinus Torvalds 	C(tail);
4601da177e4SLinus Torvalds 	C(end);
46102f1c89dSPaul Moore 	C(head);
46202f1c89dSPaul Moore 	C(data);
46302f1c89dSPaul Moore 	C(truesize);
46402f1c89dSPaul Moore 	atomic_set(&n->users, 1);
4651da177e4SLinus Torvalds 
4661da177e4SLinus Torvalds 	atomic_inc(&(skb_shinfo(skb)->dataref));
4671da177e4SLinus Torvalds 	skb->cloned = 1;
4681da177e4SLinus Torvalds 
4691da177e4SLinus Torvalds 	return n;
470e0053ec0SHerbert Xu #undef C
471e0053ec0SHerbert Xu }
472e0053ec0SHerbert Xu 
473e0053ec0SHerbert Xu /**
474e0053ec0SHerbert Xu  *	skb_morph	-	morph one skb into another
475e0053ec0SHerbert Xu  *	@dst: the skb to receive the contents
476e0053ec0SHerbert Xu  *	@src: the skb to supply the contents
477e0053ec0SHerbert Xu  *
478e0053ec0SHerbert Xu  *	This is identical to skb_clone except that the target skb is
479e0053ec0SHerbert Xu  *	supplied by the user.
480e0053ec0SHerbert Xu  *
481e0053ec0SHerbert Xu  *	The target skb is returned upon exit.
482e0053ec0SHerbert Xu  */
483e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
484e0053ec0SHerbert Xu {
4852d4baff8SHerbert Xu 	skb_release_all(dst);
486e0053ec0SHerbert Xu 	return __skb_clone(dst, src);
487e0053ec0SHerbert Xu }
488e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph);
489e0053ec0SHerbert Xu 
490e0053ec0SHerbert Xu /**
491e0053ec0SHerbert Xu  *	skb_clone	-	duplicate an sk_buff
492e0053ec0SHerbert Xu  *	@skb: buffer to clone
493e0053ec0SHerbert Xu  *	@gfp_mask: allocation priority
494e0053ec0SHerbert Xu  *
495e0053ec0SHerbert Xu  *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
496e0053ec0SHerbert Xu  *	copies share the same packet data but not structure. The new
497e0053ec0SHerbert Xu  *	buffer has a reference count of 1. If the allocation fails the
498e0053ec0SHerbert Xu  *	function returns %NULL otherwise the new buffer is returned.
499e0053ec0SHerbert Xu  *
500e0053ec0SHerbert Xu  *	If this function is called from an interrupt gfp_mask() must be
501e0053ec0SHerbert Xu  *	%GFP_ATOMIC.
502e0053ec0SHerbert Xu  */
503e0053ec0SHerbert Xu 
504e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
505e0053ec0SHerbert Xu {
506e0053ec0SHerbert Xu 	struct sk_buff *n;
507e0053ec0SHerbert Xu 
508e0053ec0SHerbert Xu 	n = skb + 1;
509e0053ec0SHerbert Xu 	if (skb->fclone == SKB_FCLONE_ORIG &&
510e0053ec0SHerbert Xu 	    n->fclone == SKB_FCLONE_UNAVAILABLE) {
511e0053ec0SHerbert Xu 		atomic_t *fclone_ref = (atomic_t *) (n + 1);
512e0053ec0SHerbert Xu 		n->fclone = SKB_FCLONE_CLONE;
513e0053ec0SHerbert Xu 		atomic_inc(fclone_ref);
514e0053ec0SHerbert Xu 	} else {
515e0053ec0SHerbert Xu 		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
516e0053ec0SHerbert Xu 		if (!n)
517e0053ec0SHerbert Xu 			return NULL;
518e0053ec0SHerbert Xu 		n->fclone = SKB_FCLONE_UNAVAILABLE;
519e0053ec0SHerbert Xu 	}
520e0053ec0SHerbert Xu 
521e0053ec0SHerbert Xu 	return __skb_clone(n, skb);
5221da177e4SLinus Torvalds }
5231da177e4SLinus Torvalds 
5241da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
5251da177e4SLinus Torvalds {
5262e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET
5271da177e4SLinus Torvalds 	/*
5281da177e4SLinus Torvalds 	 *	Shift between the two data areas in bytes
5291da177e4SLinus Torvalds 	 */
5301da177e4SLinus Torvalds 	unsigned long offset = new->data - old->data;
5312e07fa9cSArnaldo Carvalho de Melo #endif
532dec18810SHerbert Xu 
533dec18810SHerbert Xu 	__copy_skb_header(new, old);
534dec18810SHerbert Xu 
5352e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET
5362e07fa9cSArnaldo Carvalho de Melo 	/* {transport,network,mac}_header are relative to skb->head */
5372e07fa9cSArnaldo Carvalho de Melo 	new->transport_header += offset;
5382e07fa9cSArnaldo Carvalho de Melo 	new->network_header   += offset;
5392e07fa9cSArnaldo Carvalho de Melo 	new->mac_header	      += offset;
5402e07fa9cSArnaldo Carvalho de Melo #endif
5417967168cSHerbert Xu 	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
5427967168cSHerbert Xu 	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
5437967168cSHerbert Xu 	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
5441da177e4SLinus Torvalds }
5451da177e4SLinus Torvalds 
5461da177e4SLinus Torvalds /**
5471da177e4SLinus Torvalds  *	skb_copy	-	create private copy of an sk_buff
5481da177e4SLinus Torvalds  *	@skb: buffer to copy
5491da177e4SLinus Torvalds  *	@gfp_mask: allocation priority
5501da177e4SLinus Torvalds  *
5511da177e4SLinus Torvalds  *	Make a copy of both an &sk_buff and its data. This is used when the
5521da177e4SLinus Torvalds  *	caller wishes to modify the data and needs a private copy of the
5531da177e4SLinus Torvalds  *	data to alter. Returns %NULL on failure or the pointer to the buffer
5541da177e4SLinus Torvalds  *	on success. The returned buffer has a reference count of 1.
5551da177e4SLinus Torvalds  *
5561da177e4SLinus Torvalds  *	As by-product this function converts non-linear &sk_buff to linear
5571da177e4SLinus Torvalds  *	one, so that &sk_buff becomes completely private and caller is allowed
5581da177e4SLinus Torvalds  *	to modify all the data of returned buffer. This means that this
5591da177e4SLinus Torvalds  *	function is not recommended for use in circumstances when only
5601da177e4SLinus Torvalds  *	header is going to be modified. Use pskb_copy() instead.
5611da177e4SLinus Torvalds  */
5621da177e4SLinus Torvalds 
563dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
5641da177e4SLinus Torvalds {
5651da177e4SLinus Torvalds 	int headerlen = skb->data - skb->head;
5661da177e4SLinus Torvalds 	/*
5671da177e4SLinus Torvalds 	 *	Allocate the copy buffer
5681da177e4SLinus Torvalds 	 */
5694305b541SArnaldo Carvalho de Melo 	struct sk_buff *n;
5704305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET
5714305b541SArnaldo Carvalho de Melo 	n = alloc_skb(skb->end + skb->data_len, gfp_mask);
5724305b541SArnaldo Carvalho de Melo #else
5734305b541SArnaldo Carvalho de Melo 	n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
5744305b541SArnaldo Carvalho de Melo #endif
5751da177e4SLinus Torvalds 	if (!n)
5761da177e4SLinus Torvalds 		return NULL;
5771da177e4SLinus Torvalds 
5781da177e4SLinus Torvalds 	/* Set the data pointer */
5791da177e4SLinus Torvalds 	skb_reserve(n, headerlen);
5801da177e4SLinus Torvalds 	/* Set the tail pointer and length */
5811da177e4SLinus Torvalds 	skb_put(n, skb->len);
5821da177e4SLinus Torvalds 
5831da177e4SLinus Torvalds 	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
5841da177e4SLinus Torvalds 		BUG();
5851da177e4SLinus Torvalds 
5861da177e4SLinus Torvalds 	copy_skb_header(n, skb);
5871da177e4SLinus Torvalds 	return n;
5881da177e4SLinus Torvalds }
5891da177e4SLinus Torvalds 
5901da177e4SLinus Torvalds 
5911da177e4SLinus Torvalds /**
5921da177e4SLinus Torvalds  *	pskb_copy	-	create copy of an sk_buff with private head.
5931da177e4SLinus Torvalds  *	@skb: buffer to copy
5941da177e4SLinus Torvalds  *	@gfp_mask: allocation priority
5951da177e4SLinus Torvalds  *
5961da177e4SLinus Torvalds  *	Make a copy of both an &sk_buff and part of its data, located
5971da177e4SLinus Torvalds  *	in header. Fragmented data remain shared. This is used when
5981da177e4SLinus Torvalds  *	the caller wishes to modify only header of &sk_buff and needs
5991da177e4SLinus Torvalds  *	private copy of the header to alter. Returns %NULL on failure
6001da177e4SLinus Torvalds  *	or the pointer to the buffer on success.
6011da177e4SLinus Torvalds  *	The returned buffer has a reference count of 1.
6021da177e4SLinus Torvalds  */
6031da177e4SLinus Torvalds 
604dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
6051da177e4SLinus Torvalds {
6061da177e4SLinus Torvalds 	/*
6071da177e4SLinus Torvalds 	 *	Allocate the copy buffer
6081da177e4SLinus Torvalds 	 */
6094305b541SArnaldo Carvalho de Melo 	struct sk_buff *n;
6104305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET
6114305b541SArnaldo Carvalho de Melo 	n = alloc_skb(skb->end, gfp_mask);
6124305b541SArnaldo Carvalho de Melo #else
6134305b541SArnaldo Carvalho de Melo 	n = alloc_skb(skb->end - skb->head, gfp_mask);
6144305b541SArnaldo Carvalho de Melo #endif
6151da177e4SLinus Torvalds 	if (!n)
6161da177e4SLinus Torvalds 		goto out;
6171da177e4SLinus Torvalds 
6181da177e4SLinus Torvalds 	/* Set the data pointer */
6191da177e4SLinus Torvalds 	skb_reserve(n, skb->data - skb->head);
6201da177e4SLinus Torvalds 	/* Set the tail pointer and length */
6211da177e4SLinus Torvalds 	skb_put(n, skb_headlen(skb));
6221da177e4SLinus Torvalds 	/* Copy the bytes */
623d626f62bSArnaldo Carvalho de Melo 	skb_copy_from_linear_data(skb, n->data, n->len);
6241da177e4SLinus Torvalds 
62525f484a6SHerbert Xu 	n->truesize += skb->data_len;
6261da177e4SLinus Torvalds 	n->data_len  = skb->data_len;
6271da177e4SLinus Torvalds 	n->len	     = skb->len;
6281da177e4SLinus Torvalds 
6291da177e4SLinus Torvalds 	if (skb_shinfo(skb)->nr_frags) {
6301da177e4SLinus Torvalds 		int i;
6311da177e4SLinus Torvalds 
6321da177e4SLinus Torvalds 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
6331da177e4SLinus Torvalds 			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
6341da177e4SLinus Torvalds 			get_page(skb_shinfo(n)->frags[i].page);
6351da177e4SLinus Torvalds 		}
6361da177e4SLinus Torvalds 		skb_shinfo(n)->nr_frags = i;
6371da177e4SLinus Torvalds 	}
6381da177e4SLinus Torvalds 
6391da177e4SLinus Torvalds 	if (skb_shinfo(skb)->frag_list) {
6401da177e4SLinus Torvalds 		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
6411da177e4SLinus Torvalds 		skb_clone_fraglist(n);
6421da177e4SLinus Torvalds 	}
6431da177e4SLinus Torvalds 
6441da177e4SLinus Torvalds 	copy_skb_header(n, skb);
6451da177e4SLinus Torvalds out:
6461da177e4SLinus Torvalds 	return n;
6471da177e4SLinus Torvalds }
6481da177e4SLinus Torvalds 
6491da177e4SLinus Torvalds /**
6501da177e4SLinus Torvalds  *	pskb_expand_head - reallocate header of &sk_buff
6511da177e4SLinus Torvalds  *	@skb: buffer to reallocate
6521da177e4SLinus Torvalds  *	@nhead: room to add at head
6531da177e4SLinus Torvalds  *	@ntail: room to add at tail
6541da177e4SLinus Torvalds  *	@gfp_mask: allocation priority
6551da177e4SLinus Torvalds  *
6561da177e4SLinus Torvalds  *	Expands (or creates identical copy, if &nhead and &ntail are zero)
6571da177e4SLinus Torvalds  *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
6581da177e4SLinus Torvalds  *	reference count of 1. Returns zero in the case of success or error,
6591da177e4SLinus Torvalds  *	if expansion failed. In the last case, &sk_buff is not changed.
6601da177e4SLinus Torvalds  *
6611da177e4SLinus Torvalds  *	All the pointers pointing into skb header may change and must be
6621da177e4SLinus Torvalds  *	reloaded after call to this function.
6631da177e4SLinus Torvalds  */
6641da177e4SLinus Torvalds 
66586a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
666dd0fc66fSAl Viro 		     gfp_t gfp_mask)
6671da177e4SLinus Torvalds {
6681da177e4SLinus Torvalds 	int i;
6691da177e4SLinus Torvalds 	u8 *data;
6704305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET
6714305b541SArnaldo Carvalho de Melo 	int size = nhead + skb->end + ntail;
6724305b541SArnaldo Carvalho de Melo #else
6731da177e4SLinus Torvalds 	int size = nhead + (skb->end - skb->head) + ntail;
6744305b541SArnaldo Carvalho de Melo #endif
6751da177e4SLinus Torvalds 	long off;
6761da177e4SLinus Torvalds 
6771da177e4SLinus Torvalds 	if (skb_shared(skb))
6781da177e4SLinus Torvalds 		BUG();
6791da177e4SLinus Torvalds 
6801da177e4SLinus Torvalds 	size = SKB_DATA_ALIGN(size);
6811da177e4SLinus Torvalds 
6821da177e4SLinus Torvalds 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
6831da177e4SLinus Torvalds 	if (!data)
6841da177e4SLinus Torvalds 		goto nodata;
6851da177e4SLinus Torvalds 
6861da177e4SLinus Torvalds 	/* Copy only real data... and, alas, header. This should be
6871da177e4SLinus Torvalds 	 * optimized for the cases when header is void. */
6884305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET
689b6ccc67dSMikael Pettersson 	memcpy(data + nhead, skb->head, skb->tail);
6904305b541SArnaldo Carvalho de Melo #else
691b6ccc67dSMikael Pettersson 	memcpy(data + nhead, skb->head, skb->tail - skb->head);
69227a884dcSArnaldo Carvalho de Melo #endif
6934305b541SArnaldo Carvalho de Melo 	memcpy(data + size, skb_end_pointer(skb),
6944305b541SArnaldo Carvalho de Melo 	       sizeof(struct skb_shared_info));
6951da177e4SLinus Torvalds 
6961da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6971da177e4SLinus Torvalds 		get_page(skb_shinfo(skb)->frags[i].page);
6981da177e4SLinus Torvalds 
6991da177e4SLinus Torvalds 	if (skb_shinfo(skb)->frag_list)
7001da177e4SLinus Torvalds 		skb_clone_fraglist(skb);
7011da177e4SLinus Torvalds 
7021da177e4SLinus Torvalds 	skb_release_data(skb);
7031da177e4SLinus Torvalds 
7041da177e4SLinus Torvalds 	off = (data + nhead) - skb->head;
7051da177e4SLinus Torvalds 
7061da177e4SLinus Torvalds 	skb->head     = data;
7071da177e4SLinus Torvalds 	skb->data    += off;
7084305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET
7094305b541SArnaldo Carvalho de Melo 	skb->end      = size;
71056eb8882SPatrick McHardy 	off           = nhead;
7114305b541SArnaldo Carvalho de Melo #else
7124305b541SArnaldo Carvalho de Melo 	skb->end      = skb->head + size;
71356eb8882SPatrick McHardy #endif
71427a884dcSArnaldo Carvalho de Melo 	/* {transport,network,mac}_header and tail are relative to skb->head */
71527a884dcSArnaldo Carvalho de Melo 	skb->tail	      += off;
716b0e380b1SArnaldo Carvalho de Melo 	skb->transport_header += off;
717b0e380b1SArnaldo Carvalho de Melo 	skb->network_header   += off;
718b0e380b1SArnaldo Carvalho de Melo 	skb->mac_header	      += off;
719172a863fSHerbert Xu 	skb->csum_start       += nhead;
7201da177e4SLinus Torvalds 	skb->cloned   = 0;
721334a8132SPatrick McHardy 	skb->hdr_len  = 0;
7221da177e4SLinus Torvalds 	skb->nohdr    = 0;
7231da177e4SLinus Torvalds 	atomic_set(&skb_shinfo(skb)->dataref, 1);
7241da177e4SLinus Torvalds 	return 0;
7251da177e4SLinus Torvalds 
7261da177e4SLinus Torvalds nodata:
7271da177e4SLinus Torvalds 	return -ENOMEM;
7281da177e4SLinus Torvalds }
7291da177e4SLinus Torvalds 
7301da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */
7311da177e4SLinus Torvalds 
7321da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
7331da177e4SLinus Torvalds {
7341da177e4SLinus Torvalds 	struct sk_buff *skb2;
7351da177e4SLinus Torvalds 	int delta = headroom - skb_headroom(skb);
7361da177e4SLinus Torvalds 
7371da177e4SLinus Torvalds 	if (delta <= 0)
7381da177e4SLinus Torvalds 		skb2 = pskb_copy(skb, GFP_ATOMIC);
7391da177e4SLinus Torvalds 	else {
7401da177e4SLinus Torvalds 		skb2 = skb_clone(skb, GFP_ATOMIC);
7411da177e4SLinus Torvalds 		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
7421da177e4SLinus Torvalds 					     GFP_ATOMIC)) {
7431da177e4SLinus Torvalds 			kfree_skb(skb2);
7441da177e4SLinus Torvalds 			skb2 = NULL;
7451da177e4SLinus Torvalds 		}
7461da177e4SLinus Torvalds 	}
7471da177e4SLinus Torvalds 	return skb2;
7481da177e4SLinus Torvalds }
7491da177e4SLinus Torvalds 
7501da177e4SLinus Torvalds 
7511da177e4SLinus Torvalds /**
7521da177e4SLinus Torvalds  *	skb_copy_expand	-	copy and expand sk_buff
7531da177e4SLinus Torvalds  *	@skb: buffer to copy
7541da177e4SLinus Torvalds  *	@newheadroom: new free bytes at head
7551da177e4SLinus Torvalds  *	@newtailroom: new free bytes at tail
7561da177e4SLinus Torvalds  *	@gfp_mask: allocation priority
7571da177e4SLinus Torvalds  *
7581da177e4SLinus Torvalds  *	Make a copy of both an &sk_buff and its data and while doing so
7591da177e4SLinus Torvalds  *	allocate additional space.
7601da177e4SLinus Torvalds  *
7611da177e4SLinus Torvalds  *	This is used when the caller wishes to modify the data and needs a
7621da177e4SLinus Torvalds  *	private copy of the data to alter as well as more space for new fields.
7631da177e4SLinus Torvalds  *	Returns %NULL on failure or the pointer to the buffer
7641da177e4SLinus Torvalds  *	on success. The returned buffer has a reference count of 1.
7651da177e4SLinus Torvalds  *
7661da177e4SLinus Torvalds  *	You must pass %GFP_ATOMIC as the allocation priority if this function
7671da177e4SLinus Torvalds  *	is called from an interrupt.
7681da177e4SLinus Torvalds  */
7691da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
77086a76cafSVictor Fusco 				int newheadroom, int newtailroom,
771dd0fc66fSAl Viro 				gfp_t gfp_mask)
7721da177e4SLinus Torvalds {
7731da177e4SLinus Torvalds 	/*
7741da177e4SLinus Torvalds 	 *	Allocate the copy buffer
7751da177e4SLinus Torvalds 	 */
7761da177e4SLinus Torvalds 	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
7771da177e4SLinus Torvalds 				      gfp_mask);
778efd1e8d5SPatrick McHardy 	int oldheadroom = skb_headroom(skb);
7791da177e4SLinus Torvalds 	int head_copy_len, head_copy_off;
78052886051SHerbert Xu 	int off;
7811da177e4SLinus Torvalds 
7821da177e4SLinus Torvalds 	if (!n)
7831da177e4SLinus Torvalds 		return NULL;
7841da177e4SLinus Torvalds 
7851da177e4SLinus Torvalds 	skb_reserve(n, newheadroom);
7861da177e4SLinus Torvalds 
7871da177e4SLinus Torvalds 	/* Set the tail pointer and length */
7881da177e4SLinus Torvalds 	skb_put(n, skb->len);
7891da177e4SLinus Torvalds 
790efd1e8d5SPatrick McHardy 	head_copy_len = oldheadroom;
7911da177e4SLinus Torvalds 	head_copy_off = 0;
7921da177e4SLinus Torvalds 	if (newheadroom <= head_copy_len)
7931da177e4SLinus Torvalds 		head_copy_len = newheadroom;
7941da177e4SLinus Torvalds 	else
7951da177e4SLinus Torvalds 		head_copy_off = newheadroom - head_copy_len;
7961da177e4SLinus Torvalds 
7971da177e4SLinus Torvalds 	/* Copy the linear header and data. */
7981da177e4SLinus Torvalds 	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
7991da177e4SLinus Torvalds 			  skb->len + head_copy_len))
8001da177e4SLinus Torvalds 		BUG();
8011da177e4SLinus Torvalds 
8021da177e4SLinus Torvalds 	copy_skb_header(n, skb);
8031da177e4SLinus Torvalds 
804efd1e8d5SPatrick McHardy 	off                  = newheadroom - oldheadroom;
80552886051SHerbert Xu 	n->csum_start       += off;
80652886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET
807efd1e8d5SPatrick McHardy 	n->transport_header += off;
808efd1e8d5SPatrick McHardy 	n->network_header   += off;
809efd1e8d5SPatrick McHardy 	n->mac_header	    += off;
81052886051SHerbert Xu #endif
811efd1e8d5SPatrick McHardy 
8121da177e4SLinus Torvalds 	return n;
8131da177e4SLinus Torvalds }
8141da177e4SLinus Torvalds 
8151da177e4SLinus Torvalds /**
8161da177e4SLinus Torvalds  *	skb_pad			-	zero pad the tail of an skb
8171da177e4SLinus Torvalds  *	@skb: buffer to pad
8181da177e4SLinus Torvalds  *	@pad: space to pad
8191da177e4SLinus Torvalds  *
8201da177e4SLinus Torvalds  *	Ensure that a buffer is followed by a padding area that is zero
8211da177e4SLinus Torvalds  *	filled. Used by network drivers which may DMA or transfer data
8221da177e4SLinus Torvalds  *	beyond the buffer end onto the wire.
8231da177e4SLinus Torvalds  *
8245b057c6bSHerbert Xu  *	May return error in out of memory cases. The skb is freed on error.
8251da177e4SLinus Torvalds  */
8261da177e4SLinus Torvalds 
8275b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad)
8281da177e4SLinus Torvalds {
8295b057c6bSHerbert Xu 	int err;
8305b057c6bSHerbert Xu 	int ntail;
8311da177e4SLinus Torvalds 
8321da177e4SLinus Torvalds 	/* If the skbuff is non linear tailroom is always zero.. */
8335b057c6bSHerbert Xu 	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
8341da177e4SLinus Torvalds 		memset(skb->data+skb->len, 0, pad);
8355b057c6bSHerbert Xu 		return 0;
8361da177e4SLinus Torvalds 	}
8371da177e4SLinus Torvalds 
8384305b541SArnaldo Carvalho de Melo 	ntail = skb->data_len + pad - (skb->end - skb->tail);
8395b057c6bSHerbert Xu 	if (likely(skb_cloned(skb) || ntail > 0)) {
8405b057c6bSHerbert Xu 		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
8415b057c6bSHerbert Xu 		if (unlikely(err))
8425b057c6bSHerbert Xu 			goto free_skb;
8435b057c6bSHerbert Xu 	}
8445b057c6bSHerbert Xu 
8455b057c6bSHerbert Xu 	/* FIXME: The use of this function with non-linear skb's really needs
8465b057c6bSHerbert Xu 	 * to be audited.
8475b057c6bSHerbert Xu 	 */
8485b057c6bSHerbert Xu 	err = skb_linearize(skb);
8495b057c6bSHerbert Xu 	if (unlikely(err))
8505b057c6bSHerbert Xu 		goto free_skb;
8515b057c6bSHerbert Xu 
8525b057c6bSHerbert Xu 	memset(skb->data + skb->len, 0, pad);
8535b057c6bSHerbert Xu 	return 0;
8545b057c6bSHerbert Xu 
8555b057c6bSHerbert Xu free_skb:
8561da177e4SLinus Torvalds 	kfree_skb(skb);
8575b057c6bSHerbert Xu 	return err;
8581da177e4SLinus Torvalds }
8591da177e4SLinus Torvalds 
8603cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers.
8611da177e4SLinus Torvalds  */
8621da177e4SLinus Torvalds 
8633cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len)
8641da177e4SLinus Torvalds {
86527b437c8SHerbert Xu 	struct sk_buff **fragp;
86627b437c8SHerbert Xu 	struct sk_buff *frag;
8671da177e4SLinus Torvalds 	int offset = skb_headlen(skb);
8681da177e4SLinus Torvalds 	int nfrags = skb_shinfo(skb)->nr_frags;
8691da177e4SLinus Torvalds 	int i;
87027b437c8SHerbert Xu 	int err;
87127b437c8SHerbert Xu 
87227b437c8SHerbert Xu 	if (skb_cloned(skb) &&
87327b437c8SHerbert Xu 	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
87427b437c8SHerbert Xu 		return err;
8751da177e4SLinus Torvalds 
876f4d26fb3SHerbert Xu 	i = 0;
877f4d26fb3SHerbert Xu 	if (offset >= len)
878f4d26fb3SHerbert Xu 		goto drop_pages;
879f4d26fb3SHerbert Xu 
880f4d26fb3SHerbert Xu 	for (; i < nfrags; i++) {
8811da177e4SLinus Torvalds 		int end = offset + skb_shinfo(skb)->frags[i].size;
88227b437c8SHerbert Xu 
88327b437c8SHerbert Xu 		if (end < len) {
8841da177e4SLinus Torvalds 			offset = end;
88527b437c8SHerbert Xu 			continue;
8861da177e4SLinus Torvalds 		}
8871da177e4SLinus Torvalds 
88827b437c8SHerbert Xu 		skb_shinfo(skb)->frags[i++].size = len - offset;
88927b437c8SHerbert Xu 
890f4d26fb3SHerbert Xu drop_pages:
89127b437c8SHerbert Xu 		skb_shinfo(skb)->nr_frags = i;
89227b437c8SHerbert Xu 
89327b437c8SHerbert Xu 		for (; i < nfrags; i++)
89427b437c8SHerbert Xu 			put_page(skb_shinfo(skb)->frags[i].page);
89527b437c8SHerbert Xu 
89627b437c8SHerbert Xu 		if (skb_shinfo(skb)->frag_list)
89727b437c8SHerbert Xu 			skb_drop_fraglist(skb);
898f4d26fb3SHerbert Xu 		goto done;
89927b437c8SHerbert Xu 	}
90027b437c8SHerbert Xu 
90127b437c8SHerbert Xu 	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
90227b437c8SHerbert Xu 	     fragp = &frag->next) {
90327b437c8SHerbert Xu 		int end = offset + frag->len;
90427b437c8SHerbert Xu 
90527b437c8SHerbert Xu 		if (skb_shared(frag)) {
90627b437c8SHerbert Xu 			struct sk_buff *nfrag;
90727b437c8SHerbert Xu 
90827b437c8SHerbert Xu 			nfrag = skb_clone(frag, GFP_ATOMIC);
90927b437c8SHerbert Xu 			if (unlikely(!nfrag))
91027b437c8SHerbert Xu 				return -ENOMEM;
91127b437c8SHerbert Xu 
91227b437c8SHerbert Xu 			nfrag->next = frag->next;
913f4d26fb3SHerbert Xu 			kfree_skb(frag);
91427b437c8SHerbert Xu 			frag = nfrag;
91527b437c8SHerbert Xu 			*fragp = frag;
91627b437c8SHerbert Xu 		}
91727b437c8SHerbert Xu 
91827b437c8SHerbert Xu 		if (end < len) {
91927b437c8SHerbert Xu 			offset = end;
92027b437c8SHerbert Xu 			continue;
92127b437c8SHerbert Xu 		}
92227b437c8SHerbert Xu 
92327b437c8SHerbert Xu 		if (end > len &&
92427b437c8SHerbert Xu 		    unlikely((err = pskb_trim(frag, len - offset))))
92527b437c8SHerbert Xu 			return err;
92627b437c8SHerbert Xu 
92727b437c8SHerbert Xu 		if (frag->next)
92827b437c8SHerbert Xu 			skb_drop_list(&frag->next);
92927b437c8SHerbert Xu 		break;
93027b437c8SHerbert Xu 	}
93127b437c8SHerbert Xu 
932f4d26fb3SHerbert Xu done:
93327b437c8SHerbert Xu 	if (len > skb_headlen(skb)) {
9341da177e4SLinus Torvalds 		skb->data_len -= skb->len - len;
9351da177e4SLinus Torvalds 		skb->len       = len;
9361da177e4SLinus Torvalds 	} else {
9371da177e4SLinus Torvalds 		skb->len       = len;
9381da177e4SLinus Torvalds 		skb->data_len  = 0;
93927a884dcSArnaldo Carvalho de Melo 		skb_set_tail_pointer(skb, len);
9401da177e4SLinus Torvalds 	}
9411da177e4SLinus Torvalds 
9421da177e4SLinus Torvalds 	return 0;
9431da177e4SLinus Torvalds }
9441da177e4SLinus Torvalds 
9451da177e4SLinus Torvalds /**
9461da177e4SLinus Torvalds  *	__pskb_pull_tail - advance tail of skb header
9471da177e4SLinus Torvalds  *	@skb: buffer to reallocate
9481da177e4SLinus Torvalds  *	@delta: number of bytes to advance tail
9491da177e4SLinus Torvalds  *
9501da177e4SLinus Torvalds  *	The function makes a sense only on a fragmented &sk_buff,
9511da177e4SLinus Torvalds  *	it expands header moving its tail forward and copying necessary
9521da177e4SLinus Torvalds  *	data from fragmented part.
9531da177e4SLinus Torvalds  *
9541da177e4SLinus Torvalds  *	&sk_buff MUST have reference count of 1.
9551da177e4SLinus Torvalds  *
9561da177e4SLinus Torvalds  *	Returns %NULL (and &sk_buff does not change) if pull failed
9571da177e4SLinus Torvalds  *	or value of new tail of skb in the case of success.
9581da177e4SLinus Torvalds  *
9591da177e4SLinus Torvalds  *	All the pointers pointing into skb header may change and must be
9601da177e4SLinus Torvalds  *	reloaded after call to this function.
9611da177e4SLinus Torvalds  */
9621da177e4SLinus Torvalds 
9631da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part,
9641da177e4SLinus Torvalds  * when it is necessary.
9651da177e4SLinus Torvalds  * 1. It may fail due to malloc failure.
9661da177e4SLinus Torvalds  * 2. It may change skb pointers.
9671da177e4SLinus Torvalds  *
9681da177e4SLinus Torvalds  * It is pretty complicated. Luckily, it is called only in exceptional cases.
9691da177e4SLinus Torvalds  */
9701da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
9711da177e4SLinus Torvalds {
9721da177e4SLinus Torvalds 	/* If skb has not enough free space at tail, get new one
9731da177e4SLinus Torvalds 	 * plus 128 bytes for future expansions. If we have enough
9741da177e4SLinus Torvalds 	 * room at tail, reallocate without expansion only if skb is cloned.
9751da177e4SLinus Torvalds 	 */
9764305b541SArnaldo Carvalho de Melo 	int i, k, eat = (skb->tail + delta) - skb->end;
9771da177e4SLinus Torvalds 
9781da177e4SLinus Torvalds 	if (eat > 0 || skb_cloned(skb)) {
9791da177e4SLinus Torvalds 		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
9801da177e4SLinus Torvalds 				     GFP_ATOMIC))
9811da177e4SLinus Torvalds 			return NULL;
9821da177e4SLinus Torvalds 	}
9831da177e4SLinus Torvalds 
98427a884dcSArnaldo Carvalho de Melo 	if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
9851da177e4SLinus Torvalds 		BUG();
9861da177e4SLinus Torvalds 
9871da177e4SLinus Torvalds 	/* Optimization: no fragments, no reasons to preestimate
9881da177e4SLinus Torvalds 	 * size of pulled pages. Superb.
9891da177e4SLinus Torvalds 	 */
9901da177e4SLinus Torvalds 	if (!skb_shinfo(skb)->frag_list)
9911da177e4SLinus Torvalds 		goto pull_pages;
9921da177e4SLinus Torvalds 
9931da177e4SLinus Torvalds 	/* Estimate size of pulled pages. */
9941da177e4SLinus Torvalds 	eat = delta;
9951da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
9961da177e4SLinus Torvalds 		if (skb_shinfo(skb)->frags[i].size >= eat)
9971da177e4SLinus Torvalds 			goto pull_pages;
9981da177e4SLinus Torvalds 		eat -= skb_shinfo(skb)->frags[i].size;
9991da177e4SLinus Torvalds 	}
10001da177e4SLinus Torvalds 
10011da177e4SLinus Torvalds 	/* If we need update frag list, we are in troubles.
10021da177e4SLinus Torvalds 	 * Certainly, it possible to add an offset to skb data,
10031da177e4SLinus Torvalds 	 * but taking into account that pulling is expected to
10041da177e4SLinus Torvalds 	 * be very rare operation, it is worth to fight against
10051da177e4SLinus Torvalds 	 * further bloating skb head and crucify ourselves here instead.
10061da177e4SLinus Torvalds 	 * Pure masohism, indeed. 8)8)
10071da177e4SLinus Torvalds 	 */
10081da177e4SLinus Torvalds 	if (eat) {
10091da177e4SLinus Torvalds 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
10101da177e4SLinus Torvalds 		struct sk_buff *clone = NULL;
10111da177e4SLinus Torvalds 		struct sk_buff *insp = NULL;
10121da177e4SLinus Torvalds 
10131da177e4SLinus Torvalds 		do {
101409a62660SKris Katterjohn 			BUG_ON(!list);
10151da177e4SLinus Torvalds 
10161da177e4SLinus Torvalds 			if (list->len <= eat) {
10171da177e4SLinus Torvalds 				/* Eaten as whole. */
10181da177e4SLinus Torvalds 				eat -= list->len;
10191da177e4SLinus Torvalds 				list = list->next;
10201da177e4SLinus Torvalds 				insp = list;
10211da177e4SLinus Torvalds 			} else {
10221da177e4SLinus Torvalds 				/* Eaten partially. */
10231da177e4SLinus Torvalds 
10241da177e4SLinus Torvalds 				if (skb_shared(list)) {
10251da177e4SLinus Torvalds 					/* Sucks! We need to fork list. :-( */
10261da177e4SLinus Torvalds 					clone = skb_clone(list, GFP_ATOMIC);
10271da177e4SLinus Torvalds 					if (!clone)
10281da177e4SLinus Torvalds 						return NULL;
10291da177e4SLinus Torvalds 					insp = list->next;
10301da177e4SLinus Torvalds 					list = clone;
10311da177e4SLinus Torvalds 				} else {
10321da177e4SLinus Torvalds 					/* This may be pulled without
10331da177e4SLinus Torvalds 					 * problems. */
10341da177e4SLinus Torvalds 					insp = list;
10351da177e4SLinus Torvalds 				}
10361da177e4SLinus Torvalds 				if (!pskb_pull(list, eat)) {
10371da177e4SLinus Torvalds 					if (clone)
10381da177e4SLinus Torvalds 						kfree_skb(clone);
10391da177e4SLinus Torvalds 					return NULL;
10401da177e4SLinus Torvalds 				}
10411da177e4SLinus Torvalds 				break;
10421da177e4SLinus Torvalds 			}
10431da177e4SLinus Torvalds 		} while (eat);
10441da177e4SLinus Torvalds 
10451da177e4SLinus Torvalds 		/* Free pulled out fragments. */
10461da177e4SLinus Torvalds 		while ((list = skb_shinfo(skb)->frag_list) != insp) {
10471da177e4SLinus Torvalds 			skb_shinfo(skb)->frag_list = list->next;
10481da177e4SLinus Torvalds 			kfree_skb(list);
10491da177e4SLinus Torvalds 		}
10501da177e4SLinus Torvalds 		/* And insert new clone at head. */
10511da177e4SLinus Torvalds 		if (clone) {
10521da177e4SLinus Torvalds 			clone->next = list;
10531da177e4SLinus Torvalds 			skb_shinfo(skb)->frag_list = clone;
10541da177e4SLinus Torvalds 		}
10551da177e4SLinus Torvalds 	}
10561da177e4SLinus Torvalds 	/* Success! Now we may commit changes to skb data. */
10571da177e4SLinus Torvalds 
10581da177e4SLinus Torvalds pull_pages:
10591da177e4SLinus Torvalds 	eat = delta;
10601da177e4SLinus Torvalds 	k = 0;
10611da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
10621da177e4SLinus Torvalds 		if (skb_shinfo(skb)->frags[i].size <= eat) {
10631da177e4SLinus Torvalds 			put_page(skb_shinfo(skb)->frags[i].page);
10641da177e4SLinus Torvalds 			eat -= skb_shinfo(skb)->frags[i].size;
10651da177e4SLinus Torvalds 		} else {
10661da177e4SLinus Torvalds 			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
10671da177e4SLinus Torvalds 			if (eat) {
10681da177e4SLinus Torvalds 				skb_shinfo(skb)->frags[k].page_offset += eat;
10691da177e4SLinus Torvalds 				skb_shinfo(skb)->frags[k].size -= eat;
10701da177e4SLinus Torvalds 				eat = 0;
10711da177e4SLinus Torvalds 			}
10721da177e4SLinus Torvalds 			k++;
10731da177e4SLinus Torvalds 		}
10741da177e4SLinus Torvalds 	}
10751da177e4SLinus Torvalds 	skb_shinfo(skb)->nr_frags = k;
10761da177e4SLinus Torvalds 
10771da177e4SLinus Torvalds 	skb->tail     += delta;
10781da177e4SLinus Torvalds 	skb->data_len -= delta;
10791da177e4SLinus Torvalds 
108027a884dcSArnaldo Carvalho de Melo 	return skb_tail_pointer(skb);
10811da177e4SLinus Torvalds }
10821da177e4SLinus Torvalds 
10831da177e4SLinus Torvalds /* Copy some data bits from skb to kernel buffer. */
10841da177e4SLinus Torvalds 
10851da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
10861da177e4SLinus Torvalds {
10871da177e4SLinus Torvalds 	int i, copy;
10881a028e50SDavid S. Miller 	int start = skb_headlen(skb);
10891da177e4SLinus Torvalds 
10901da177e4SLinus Torvalds 	if (offset > (int)skb->len - len)
10911da177e4SLinus Torvalds 		goto fault;
10921da177e4SLinus Torvalds 
10931da177e4SLinus Torvalds 	/* Copy header. */
10941a028e50SDavid S. Miller 	if ((copy = start - offset) > 0) {
10951da177e4SLinus Torvalds 		if (copy > len)
10961da177e4SLinus Torvalds 			copy = len;
1097d626f62bSArnaldo Carvalho de Melo 		skb_copy_from_linear_data_offset(skb, offset, to, copy);
10981da177e4SLinus Torvalds 		if ((len -= copy) == 0)
10991da177e4SLinus Torvalds 			return 0;
11001da177e4SLinus Torvalds 		offset += copy;
11011da177e4SLinus Torvalds 		to     += copy;
11021da177e4SLinus Torvalds 	}
11031da177e4SLinus Torvalds 
11041da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
11051a028e50SDavid S. Miller 		int end;
11061da177e4SLinus Torvalds 
11071a028e50SDavid S. Miller 		BUG_TRAP(start <= offset + len);
11081a028e50SDavid S. Miller 
11091a028e50SDavid S. Miller 		end = start + skb_shinfo(skb)->frags[i].size;
11101da177e4SLinus Torvalds 		if ((copy = end - offset) > 0) {
11111da177e4SLinus Torvalds 			u8 *vaddr;
11121da177e4SLinus Torvalds 
11131da177e4SLinus Torvalds 			if (copy > len)
11141da177e4SLinus Torvalds 				copy = len;
11151da177e4SLinus Torvalds 
11161da177e4SLinus Torvalds 			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
11171da177e4SLinus Torvalds 			memcpy(to,
11181a028e50SDavid S. Miller 			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
11191a028e50SDavid S. Miller 			       offset - start, copy);
11201da177e4SLinus Torvalds 			kunmap_skb_frag(vaddr);
11211da177e4SLinus Torvalds 
11221da177e4SLinus Torvalds 			if ((len -= copy) == 0)
11231da177e4SLinus Torvalds 				return 0;
11241da177e4SLinus Torvalds 			offset += copy;
11251da177e4SLinus Torvalds 			to     += copy;
11261da177e4SLinus Torvalds 		}
11271a028e50SDavid S. Miller 		start = end;
11281da177e4SLinus Torvalds 	}
11291da177e4SLinus Torvalds 
11301da177e4SLinus Torvalds 	if (skb_shinfo(skb)->frag_list) {
11311da177e4SLinus Torvalds 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
11321da177e4SLinus Torvalds 
11331da177e4SLinus Torvalds 		for (; list; list = list->next) {
11341a028e50SDavid S. Miller 			int end;
11351da177e4SLinus Torvalds 
11361a028e50SDavid S. Miller 			BUG_TRAP(start <= offset + len);
11371a028e50SDavid S. Miller 
11381a028e50SDavid S. Miller 			end = start + list->len;
11391da177e4SLinus Torvalds 			if ((copy = end - offset) > 0) {
11401da177e4SLinus Torvalds 				if (copy > len)
11411da177e4SLinus Torvalds 					copy = len;
11421a028e50SDavid S. Miller 				if (skb_copy_bits(list, offset - start,
11431a028e50SDavid S. Miller 						  to, copy))
11441da177e4SLinus Torvalds 					goto fault;
11451da177e4SLinus Torvalds 				if ((len -= copy) == 0)
11461da177e4SLinus Torvalds 					return 0;
11471da177e4SLinus Torvalds 				offset += copy;
11481da177e4SLinus Torvalds 				to     += copy;
11491da177e4SLinus Torvalds 			}
11501a028e50SDavid S. Miller 			start = end;
11511da177e4SLinus Torvalds 		}
11521da177e4SLinus Torvalds 	}
11531da177e4SLinus Torvalds 	if (!len)
11541da177e4SLinus Torvalds 		return 0;
11551da177e4SLinus Torvalds 
11561da177e4SLinus Torvalds fault:
11571da177e4SLinus Torvalds 	return -EFAULT;
11581da177e4SLinus Torvalds }
11591da177e4SLinus Torvalds 
11609c55e01cSJens Axboe /*
11619c55e01cSJens Axboe  * Callback from splice_to_pipe(), if we need to release some pages
11629c55e01cSJens Axboe  * at the end of the spd in case we error'ed out in filling the pipe.
11639c55e01cSJens Axboe  */
11649c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
11659c55e01cSJens Axboe {
11669c55e01cSJens Axboe 	struct sk_buff *skb = (struct sk_buff *) spd->partial[i].private;
11679c55e01cSJens Axboe 
11689c55e01cSJens Axboe 	kfree_skb(skb);
11699c55e01cSJens Axboe }
11709c55e01cSJens Axboe 
11719c55e01cSJens Axboe /*
11729c55e01cSJens Axboe  * Fill page/offset/length into spd, if it can hold more pages.
11739c55e01cSJens Axboe  */
11749c55e01cSJens Axboe static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page,
11759c55e01cSJens Axboe 				unsigned int len, unsigned int offset,
11769c55e01cSJens Axboe 				struct sk_buff *skb)
11779c55e01cSJens Axboe {
11789c55e01cSJens Axboe 	if (unlikely(spd->nr_pages == PIPE_BUFFERS))
11799c55e01cSJens Axboe 		return 1;
11809c55e01cSJens Axboe 
11819c55e01cSJens Axboe 	spd->pages[spd->nr_pages] = page;
11829c55e01cSJens Axboe 	spd->partial[spd->nr_pages].len = len;
11839c55e01cSJens Axboe 	spd->partial[spd->nr_pages].offset = offset;
11849c55e01cSJens Axboe 	spd->partial[spd->nr_pages].private = (unsigned long) skb_get(skb);
11859c55e01cSJens Axboe 	spd->nr_pages++;
11869c55e01cSJens Axboe 	return 0;
11879c55e01cSJens Axboe }
11889c55e01cSJens Axboe 
11899c55e01cSJens Axboe /*
11909c55e01cSJens Axboe  * Map linear and fragment data from the skb to spd. Returns number of
11919c55e01cSJens Axboe  * pages mapped.
11929c55e01cSJens Axboe  */
11939c55e01cSJens Axboe static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset,
11949c55e01cSJens Axboe 			     unsigned int *total_len,
11959c55e01cSJens Axboe 			     struct splice_pipe_desc *spd)
11969c55e01cSJens Axboe {
11979c55e01cSJens Axboe 	unsigned int nr_pages = spd->nr_pages;
11989c55e01cSJens Axboe 	unsigned int poff, plen, len, toff, tlen;
11999c55e01cSJens Axboe 	int headlen, seg;
12009c55e01cSJens Axboe 
12019c55e01cSJens Axboe 	toff = *offset;
12029c55e01cSJens Axboe 	tlen = *total_len;
12039c55e01cSJens Axboe 	if (!tlen)
12049c55e01cSJens Axboe 		goto err;
12059c55e01cSJens Axboe 
12069c55e01cSJens Axboe 	/*
12079c55e01cSJens Axboe 	 * if the offset is greater than the linear part, go directly to
12089c55e01cSJens Axboe 	 * the fragments.
12099c55e01cSJens Axboe 	 */
12109c55e01cSJens Axboe 	headlen = skb_headlen(skb);
12119c55e01cSJens Axboe 	if (toff >= headlen) {
12129c55e01cSJens Axboe 		toff -= headlen;
12139c55e01cSJens Axboe 		goto map_frag;
12149c55e01cSJens Axboe 	}
12159c55e01cSJens Axboe 
12169c55e01cSJens Axboe 	/*
12179c55e01cSJens Axboe 	 * first map the linear region into the pages/partial map, skipping
12189c55e01cSJens Axboe 	 * any potential initial offset.
12199c55e01cSJens Axboe 	 */
12209c55e01cSJens Axboe 	len = 0;
12219c55e01cSJens Axboe 	while (len < headlen) {
12229c55e01cSJens Axboe 		void *p = skb->data + len;
12239c55e01cSJens Axboe 
12249c55e01cSJens Axboe 		poff = (unsigned long) p & (PAGE_SIZE - 1);
12259c55e01cSJens Axboe 		plen = min_t(unsigned int, headlen - len, PAGE_SIZE - poff);
12269c55e01cSJens Axboe 		len += plen;
12279c55e01cSJens Axboe 
12289c55e01cSJens Axboe 		if (toff) {
12299c55e01cSJens Axboe 			if (plen <= toff) {
12309c55e01cSJens Axboe 				toff -= plen;
12319c55e01cSJens Axboe 				continue;
12329c55e01cSJens Axboe 			}
12339c55e01cSJens Axboe 			plen -= toff;
12349c55e01cSJens Axboe 			poff += toff;
12359c55e01cSJens Axboe 			toff = 0;
12369c55e01cSJens Axboe 		}
12379c55e01cSJens Axboe 
12389c55e01cSJens Axboe 		plen = min(plen, tlen);
12399c55e01cSJens Axboe 		if (!plen)
12409c55e01cSJens Axboe 			break;
12419c55e01cSJens Axboe 
12429c55e01cSJens Axboe 		/*
12439c55e01cSJens Axboe 		 * just jump directly to update and return, no point
12449c55e01cSJens Axboe 		 * in going over fragments when the output is full.
12459c55e01cSJens Axboe 		 */
12469c55e01cSJens Axboe 		if (spd_fill_page(spd, virt_to_page(p), plen, poff, skb))
12479c55e01cSJens Axboe 			goto done;
12489c55e01cSJens Axboe 
12499c55e01cSJens Axboe 		tlen -= plen;
12509c55e01cSJens Axboe 	}
12519c55e01cSJens Axboe 
12529c55e01cSJens Axboe 	/*
12539c55e01cSJens Axboe 	 * then map the fragments
12549c55e01cSJens Axboe 	 */
12559c55e01cSJens Axboe map_frag:
12569c55e01cSJens Axboe 	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
12579c55e01cSJens Axboe 		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
12589c55e01cSJens Axboe 
12599c55e01cSJens Axboe 		plen = f->size;
12609c55e01cSJens Axboe 		poff = f->page_offset;
12619c55e01cSJens Axboe 
12629c55e01cSJens Axboe 		if (toff) {
12639c55e01cSJens Axboe 			if (plen <= toff) {
12649c55e01cSJens Axboe 				toff -= plen;
12659c55e01cSJens Axboe 				continue;
12669c55e01cSJens Axboe 			}
12679c55e01cSJens Axboe 			plen -= toff;
12689c55e01cSJens Axboe 			poff += toff;
12699c55e01cSJens Axboe 			toff = 0;
12709c55e01cSJens Axboe 		}
12719c55e01cSJens Axboe 
12729c55e01cSJens Axboe 		plen = min(plen, tlen);
12739c55e01cSJens Axboe 		if (!plen)
12749c55e01cSJens Axboe 			break;
12759c55e01cSJens Axboe 
12769c55e01cSJens Axboe 		if (spd_fill_page(spd, f->page, plen, poff, skb))
12779c55e01cSJens Axboe 			break;
12789c55e01cSJens Axboe 
12799c55e01cSJens Axboe 		tlen -= plen;
12809c55e01cSJens Axboe 	}
12819c55e01cSJens Axboe 
12829c55e01cSJens Axboe done:
12839c55e01cSJens Axboe 	if (spd->nr_pages - nr_pages) {
12849c55e01cSJens Axboe 		*offset = 0;
12859c55e01cSJens Axboe 		*total_len = tlen;
12869c55e01cSJens Axboe 		return 0;
12879c55e01cSJens Axboe 	}
12889c55e01cSJens Axboe err:
12899c55e01cSJens Axboe 	return 1;
12909c55e01cSJens Axboe }
12919c55e01cSJens Axboe 
12929c55e01cSJens Axboe /*
12939c55e01cSJens Axboe  * Map data from the skb to a pipe. Should handle both the linear part,
12949c55e01cSJens Axboe  * the fragments, and the frag list. It does NOT handle frag lists within
12959c55e01cSJens Axboe  * the frag list, if such a thing exists. We'd probably need to recurse to
12969c55e01cSJens Axboe  * handle that cleanly.
12979c55e01cSJens Axboe  */
12989c55e01cSJens Axboe int skb_splice_bits(struct sk_buff *__skb, unsigned int offset,
12999c55e01cSJens Axboe 		    struct pipe_inode_info *pipe, unsigned int tlen,
13009c55e01cSJens Axboe 		    unsigned int flags)
13019c55e01cSJens Axboe {
13029c55e01cSJens Axboe 	struct partial_page partial[PIPE_BUFFERS];
13039c55e01cSJens Axboe 	struct page *pages[PIPE_BUFFERS];
13049c55e01cSJens Axboe 	struct splice_pipe_desc spd = {
13059c55e01cSJens Axboe 		.pages = pages,
13069c55e01cSJens Axboe 		.partial = partial,
13079c55e01cSJens Axboe 		.flags = flags,
13089c55e01cSJens Axboe 		.ops = &sock_pipe_buf_ops,
13099c55e01cSJens Axboe 		.spd_release = sock_spd_release,
13109c55e01cSJens Axboe 	};
13119c55e01cSJens Axboe 	struct sk_buff *skb;
13129c55e01cSJens Axboe 
13139c55e01cSJens Axboe 	/*
13149c55e01cSJens Axboe 	 * I'd love to avoid the clone here, but tcp_read_sock()
13159c55e01cSJens Axboe 	 * ignores reference counts and unconditonally kills the sk_buff
13169c55e01cSJens Axboe 	 * on return from the actor.
13179c55e01cSJens Axboe 	 */
13189c55e01cSJens Axboe 	skb = skb_clone(__skb, GFP_KERNEL);
13199c55e01cSJens Axboe 	if (unlikely(!skb))
13209c55e01cSJens Axboe 		return -ENOMEM;
13219c55e01cSJens Axboe 
13229c55e01cSJens Axboe 	/*
13239c55e01cSJens Axboe 	 * __skb_splice_bits() only fails if the output has no room left,
13249c55e01cSJens Axboe 	 * so no point in going over the frag_list for the error case.
13259c55e01cSJens Axboe 	 */
13269c55e01cSJens Axboe 	if (__skb_splice_bits(skb, &offset, &tlen, &spd))
13279c55e01cSJens Axboe 		goto done;
13289c55e01cSJens Axboe 	else if (!tlen)
13299c55e01cSJens Axboe 		goto done;
13309c55e01cSJens Axboe 
13319c55e01cSJens Axboe 	/*
13329c55e01cSJens Axboe 	 * now see if we have a frag_list to map
13339c55e01cSJens Axboe 	 */
13349c55e01cSJens Axboe 	if (skb_shinfo(skb)->frag_list) {
13359c55e01cSJens Axboe 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
13369c55e01cSJens Axboe 
13379c55e01cSJens Axboe 		for (; list && tlen; list = list->next) {
13389c55e01cSJens Axboe 			if (__skb_splice_bits(list, &offset, &tlen, &spd))
13399c55e01cSJens Axboe 				break;
13409c55e01cSJens Axboe 		}
13419c55e01cSJens Axboe 	}
13429c55e01cSJens Axboe 
13439c55e01cSJens Axboe done:
13449c55e01cSJens Axboe 	/*
13459c55e01cSJens Axboe 	 * drop our reference to the clone, the pipe consumption will
13469c55e01cSJens Axboe 	 * drop the rest.
13479c55e01cSJens Axboe 	 */
13489c55e01cSJens Axboe 	kfree_skb(skb);
13499c55e01cSJens Axboe 
13509c55e01cSJens Axboe 	if (spd.nr_pages) {
13519c55e01cSJens Axboe 		int ret;
13529c55e01cSJens Axboe 
13539c55e01cSJens Axboe 		/*
13549c55e01cSJens Axboe 		 * Drop the socket lock, otherwise we have reverse
13559c55e01cSJens Axboe 		 * locking dependencies between sk_lock and i_mutex
13569c55e01cSJens Axboe 		 * here as compared to sendfile(). We enter here
13579c55e01cSJens Axboe 		 * with the socket lock held, and splice_to_pipe() will
13589c55e01cSJens Axboe 		 * grab the pipe inode lock. For sendfile() emulation,
13599c55e01cSJens Axboe 		 * we call into ->sendpage() with the i_mutex lock held
13609c55e01cSJens Axboe 		 * and networking will grab the socket lock.
13619c55e01cSJens Axboe 		 */
13629c55e01cSJens Axboe 		release_sock(__skb->sk);
13639c55e01cSJens Axboe 		ret = splice_to_pipe(pipe, &spd);
13649c55e01cSJens Axboe 		lock_sock(__skb->sk);
13659c55e01cSJens Axboe 		return ret;
13669c55e01cSJens Axboe 	}
13679c55e01cSJens Axboe 
13689c55e01cSJens Axboe 	return 0;
13699c55e01cSJens Axboe }
13709c55e01cSJens Axboe 
1371357b40a1SHerbert Xu /**
1372357b40a1SHerbert Xu  *	skb_store_bits - store bits from kernel buffer to skb
1373357b40a1SHerbert Xu  *	@skb: destination buffer
1374357b40a1SHerbert Xu  *	@offset: offset in destination
1375357b40a1SHerbert Xu  *	@from: source buffer
1376357b40a1SHerbert Xu  *	@len: number of bytes to copy
1377357b40a1SHerbert Xu  *
1378357b40a1SHerbert Xu  *	Copy the specified number of bytes from the source buffer to the
1379357b40a1SHerbert Xu  *	destination skb.  This function handles all the messy bits of
1380357b40a1SHerbert Xu  *	traversing fragment lists and such.
1381357b40a1SHerbert Xu  */
1382357b40a1SHerbert Xu 
13830c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1384357b40a1SHerbert Xu {
1385357b40a1SHerbert Xu 	int i, copy;
13861a028e50SDavid S. Miller 	int start = skb_headlen(skb);
1387357b40a1SHerbert Xu 
1388357b40a1SHerbert Xu 	if (offset > (int)skb->len - len)
1389357b40a1SHerbert Xu 		goto fault;
1390357b40a1SHerbert Xu 
13911a028e50SDavid S. Miller 	if ((copy = start - offset) > 0) {
1392357b40a1SHerbert Xu 		if (copy > len)
1393357b40a1SHerbert Xu 			copy = len;
139427d7ff46SArnaldo Carvalho de Melo 		skb_copy_to_linear_data_offset(skb, offset, from, copy);
1395357b40a1SHerbert Xu 		if ((len -= copy) == 0)
1396357b40a1SHerbert Xu 			return 0;
1397357b40a1SHerbert Xu 		offset += copy;
1398357b40a1SHerbert Xu 		from += copy;
1399357b40a1SHerbert Xu 	}
1400357b40a1SHerbert Xu 
1401357b40a1SHerbert Xu 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1402357b40a1SHerbert Xu 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
14031a028e50SDavid S. Miller 		int end;
1404357b40a1SHerbert Xu 
14051a028e50SDavid S. Miller 		BUG_TRAP(start <= offset + len);
14061a028e50SDavid S. Miller 
14071a028e50SDavid S. Miller 		end = start + frag->size;
1408357b40a1SHerbert Xu 		if ((copy = end - offset) > 0) {
1409357b40a1SHerbert Xu 			u8 *vaddr;
1410357b40a1SHerbert Xu 
1411357b40a1SHerbert Xu 			if (copy > len)
1412357b40a1SHerbert Xu 				copy = len;
1413357b40a1SHerbert Xu 
1414357b40a1SHerbert Xu 			vaddr = kmap_skb_frag(frag);
14151a028e50SDavid S. Miller 			memcpy(vaddr + frag->page_offset + offset - start,
14161a028e50SDavid S. Miller 			       from, copy);
1417357b40a1SHerbert Xu 			kunmap_skb_frag(vaddr);
1418357b40a1SHerbert Xu 
1419357b40a1SHerbert Xu 			if ((len -= copy) == 0)
1420357b40a1SHerbert Xu 				return 0;
1421357b40a1SHerbert Xu 			offset += copy;
1422357b40a1SHerbert Xu 			from += copy;
1423357b40a1SHerbert Xu 		}
14241a028e50SDavid S. Miller 		start = end;
1425357b40a1SHerbert Xu 	}
1426357b40a1SHerbert Xu 
1427357b40a1SHerbert Xu 	if (skb_shinfo(skb)->frag_list) {
1428357b40a1SHerbert Xu 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1429357b40a1SHerbert Xu 
1430357b40a1SHerbert Xu 		for (; list; list = list->next) {
14311a028e50SDavid S. Miller 			int end;
1432357b40a1SHerbert Xu 
14331a028e50SDavid S. Miller 			BUG_TRAP(start <= offset + len);
14341a028e50SDavid S. Miller 
14351a028e50SDavid S. Miller 			end = start + list->len;
1436357b40a1SHerbert Xu 			if ((copy = end - offset) > 0) {
1437357b40a1SHerbert Xu 				if (copy > len)
1438357b40a1SHerbert Xu 					copy = len;
14391a028e50SDavid S. Miller 				if (skb_store_bits(list, offset - start,
14401a028e50SDavid S. Miller 						   from, copy))
1441357b40a1SHerbert Xu 					goto fault;
1442357b40a1SHerbert Xu 				if ((len -= copy) == 0)
1443357b40a1SHerbert Xu 					return 0;
1444357b40a1SHerbert Xu 				offset += copy;
1445357b40a1SHerbert Xu 				from += copy;
1446357b40a1SHerbert Xu 			}
14471a028e50SDavid S. Miller 			start = end;
1448357b40a1SHerbert Xu 		}
1449357b40a1SHerbert Xu 	}
1450357b40a1SHerbert Xu 	if (!len)
1451357b40a1SHerbert Xu 		return 0;
1452357b40a1SHerbert Xu 
1453357b40a1SHerbert Xu fault:
1454357b40a1SHerbert Xu 	return -EFAULT;
1455357b40a1SHerbert Xu }
1456357b40a1SHerbert Xu 
1457357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits);
1458357b40a1SHerbert Xu 
14591da177e4SLinus Torvalds /* Checksum skb data. */
14601da177e4SLinus Torvalds 
14612bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset,
14622bbbc868SAl Viro 			  int len, __wsum csum)
14631da177e4SLinus Torvalds {
14641a028e50SDavid S. Miller 	int start = skb_headlen(skb);
14651a028e50SDavid S. Miller 	int i, copy = start - offset;
14661da177e4SLinus Torvalds 	int pos = 0;
14671da177e4SLinus Torvalds 
14681da177e4SLinus Torvalds 	/* Checksum header. */
14691da177e4SLinus Torvalds 	if (copy > 0) {
14701da177e4SLinus Torvalds 		if (copy > len)
14711da177e4SLinus Torvalds 			copy = len;
14721da177e4SLinus Torvalds 		csum = csum_partial(skb->data + offset, copy, csum);
14731da177e4SLinus Torvalds 		if ((len -= copy) == 0)
14741da177e4SLinus Torvalds 			return csum;
14751da177e4SLinus Torvalds 		offset += copy;
14761da177e4SLinus Torvalds 		pos	= copy;
14771da177e4SLinus Torvalds 	}
14781da177e4SLinus Torvalds 
14791da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
14801a028e50SDavid S. Miller 		int end;
14811da177e4SLinus Torvalds 
14821a028e50SDavid S. Miller 		BUG_TRAP(start <= offset + len);
14831a028e50SDavid S. Miller 
14841a028e50SDavid S. Miller 		end = start + skb_shinfo(skb)->frags[i].size;
14851da177e4SLinus Torvalds 		if ((copy = end - offset) > 0) {
148644bb9363SAl Viro 			__wsum csum2;
14871da177e4SLinus Torvalds 			u8 *vaddr;
14881da177e4SLinus Torvalds 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
14891da177e4SLinus Torvalds 
14901da177e4SLinus Torvalds 			if (copy > len)
14911da177e4SLinus Torvalds 				copy = len;
14921da177e4SLinus Torvalds 			vaddr = kmap_skb_frag(frag);
14931a028e50SDavid S. Miller 			csum2 = csum_partial(vaddr + frag->page_offset +
14941a028e50SDavid S. Miller 					     offset - start, copy, 0);
14951da177e4SLinus Torvalds 			kunmap_skb_frag(vaddr);
14961da177e4SLinus Torvalds 			csum = csum_block_add(csum, csum2, pos);
14971da177e4SLinus Torvalds 			if (!(len -= copy))
14981da177e4SLinus Torvalds 				return csum;
14991da177e4SLinus Torvalds 			offset += copy;
15001da177e4SLinus Torvalds 			pos    += copy;
15011da177e4SLinus Torvalds 		}
15021a028e50SDavid S. Miller 		start = end;
15031da177e4SLinus Torvalds 	}
15041da177e4SLinus Torvalds 
15051da177e4SLinus Torvalds 	if (skb_shinfo(skb)->frag_list) {
15061da177e4SLinus Torvalds 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
15071da177e4SLinus Torvalds 
15081da177e4SLinus Torvalds 		for (; list; list = list->next) {
15091a028e50SDavid S. Miller 			int end;
15101da177e4SLinus Torvalds 
15111a028e50SDavid S. Miller 			BUG_TRAP(start <= offset + len);
15121a028e50SDavid S. Miller 
15131a028e50SDavid S. Miller 			end = start + list->len;
15141da177e4SLinus Torvalds 			if ((copy = end - offset) > 0) {
15155f92a738SAl Viro 				__wsum csum2;
15161da177e4SLinus Torvalds 				if (copy > len)
15171da177e4SLinus Torvalds 					copy = len;
15181a028e50SDavid S. Miller 				csum2 = skb_checksum(list, offset - start,
15191a028e50SDavid S. Miller 						     copy, 0);
15201da177e4SLinus Torvalds 				csum = csum_block_add(csum, csum2, pos);
15211da177e4SLinus Torvalds 				if ((len -= copy) == 0)
15221da177e4SLinus Torvalds 					return csum;
15231da177e4SLinus Torvalds 				offset += copy;
15241da177e4SLinus Torvalds 				pos    += copy;
15251da177e4SLinus Torvalds 			}
15261a028e50SDavid S. Miller 			start = end;
15271da177e4SLinus Torvalds 		}
15281da177e4SLinus Torvalds 	}
152909a62660SKris Katterjohn 	BUG_ON(len);
15301da177e4SLinus Torvalds 
15311da177e4SLinus Torvalds 	return csum;
15321da177e4SLinus Torvalds }
15331da177e4SLinus Torvalds 
15341da177e4SLinus Torvalds /* Both of above in one bottle. */
15351da177e4SLinus Torvalds 
153681d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
153781d77662SAl Viro 				    u8 *to, int len, __wsum csum)
15381da177e4SLinus Torvalds {
15391a028e50SDavid S. Miller 	int start = skb_headlen(skb);
15401a028e50SDavid S. Miller 	int i, copy = start - offset;
15411da177e4SLinus Torvalds 	int pos = 0;
15421da177e4SLinus Torvalds 
15431da177e4SLinus Torvalds 	/* Copy header. */
15441da177e4SLinus Torvalds 	if (copy > 0) {
15451da177e4SLinus Torvalds 		if (copy > len)
15461da177e4SLinus Torvalds 			copy = len;
15471da177e4SLinus Torvalds 		csum = csum_partial_copy_nocheck(skb->data + offset, to,
15481da177e4SLinus Torvalds 						 copy, csum);
15491da177e4SLinus Torvalds 		if ((len -= copy) == 0)
15501da177e4SLinus Torvalds 			return csum;
15511da177e4SLinus Torvalds 		offset += copy;
15521da177e4SLinus Torvalds 		to     += copy;
15531da177e4SLinus Torvalds 		pos	= copy;
15541da177e4SLinus Torvalds 	}
15551da177e4SLinus Torvalds 
15561da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
15571a028e50SDavid S. Miller 		int end;
15581da177e4SLinus Torvalds 
15591a028e50SDavid S. Miller 		BUG_TRAP(start <= offset + len);
15601a028e50SDavid S. Miller 
15611a028e50SDavid S. Miller 		end = start + skb_shinfo(skb)->frags[i].size;
15621da177e4SLinus Torvalds 		if ((copy = end - offset) > 0) {
15635084205fSAl Viro 			__wsum csum2;
15641da177e4SLinus Torvalds 			u8 *vaddr;
15651da177e4SLinus Torvalds 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
15661da177e4SLinus Torvalds 
15671da177e4SLinus Torvalds 			if (copy > len)
15681da177e4SLinus Torvalds 				copy = len;
15691da177e4SLinus Torvalds 			vaddr = kmap_skb_frag(frag);
15701da177e4SLinus Torvalds 			csum2 = csum_partial_copy_nocheck(vaddr +
15711a028e50SDavid S. Miller 							  frag->page_offset +
15721a028e50SDavid S. Miller 							  offset - start, to,
15731a028e50SDavid S. Miller 							  copy, 0);
15741da177e4SLinus Torvalds 			kunmap_skb_frag(vaddr);
15751da177e4SLinus Torvalds 			csum = csum_block_add(csum, csum2, pos);
15761da177e4SLinus Torvalds 			if (!(len -= copy))
15771da177e4SLinus Torvalds 				return csum;
15781da177e4SLinus Torvalds 			offset += copy;
15791da177e4SLinus Torvalds 			to     += copy;
15801da177e4SLinus Torvalds 			pos    += copy;
15811da177e4SLinus Torvalds 		}
15821a028e50SDavid S. Miller 		start = end;
15831da177e4SLinus Torvalds 	}
15841da177e4SLinus Torvalds 
15851da177e4SLinus Torvalds 	if (skb_shinfo(skb)->frag_list) {
15861da177e4SLinus Torvalds 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
15871da177e4SLinus Torvalds 
15881da177e4SLinus Torvalds 		for (; list; list = list->next) {
158981d77662SAl Viro 			__wsum csum2;
15901a028e50SDavid S. Miller 			int end;
15911da177e4SLinus Torvalds 
15921a028e50SDavid S. Miller 			BUG_TRAP(start <= offset + len);
15931a028e50SDavid S. Miller 
15941a028e50SDavid S. Miller 			end = start + list->len;
15951da177e4SLinus Torvalds 			if ((copy = end - offset) > 0) {
15961da177e4SLinus Torvalds 				if (copy > len)
15971da177e4SLinus Torvalds 					copy = len;
15981a028e50SDavid S. Miller 				csum2 = skb_copy_and_csum_bits(list,
15991a028e50SDavid S. Miller 							       offset - start,
16001da177e4SLinus Torvalds 							       to, copy, 0);
16011da177e4SLinus Torvalds 				csum = csum_block_add(csum, csum2, pos);
16021da177e4SLinus Torvalds 				if ((len -= copy) == 0)
16031da177e4SLinus Torvalds 					return csum;
16041da177e4SLinus Torvalds 				offset += copy;
16051da177e4SLinus Torvalds 				to     += copy;
16061da177e4SLinus Torvalds 				pos    += copy;
16071da177e4SLinus Torvalds 			}
16081a028e50SDavid S. Miller 			start = end;
16091da177e4SLinus Torvalds 		}
16101da177e4SLinus Torvalds 	}
161109a62660SKris Katterjohn 	BUG_ON(len);
16121da177e4SLinus Torvalds 	return csum;
16131da177e4SLinus Torvalds }
16141da177e4SLinus Torvalds 
16151da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
16161da177e4SLinus Torvalds {
1617d3bc23e7SAl Viro 	__wsum csum;
16181da177e4SLinus Torvalds 	long csstart;
16191da177e4SLinus Torvalds 
162084fa7933SPatrick McHardy 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1621663ead3bSHerbert Xu 		csstart = skb->csum_start - skb_headroom(skb);
16221da177e4SLinus Torvalds 	else
16231da177e4SLinus Torvalds 		csstart = skb_headlen(skb);
16241da177e4SLinus Torvalds 
162509a62660SKris Katterjohn 	BUG_ON(csstart > skb_headlen(skb));
16261da177e4SLinus Torvalds 
1627d626f62bSArnaldo Carvalho de Melo 	skb_copy_from_linear_data(skb, to, csstart);
16281da177e4SLinus Torvalds 
16291da177e4SLinus Torvalds 	csum = 0;
16301da177e4SLinus Torvalds 	if (csstart != skb->len)
16311da177e4SLinus Torvalds 		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
16321da177e4SLinus Torvalds 					      skb->len - csstart, 0);
16331da177e4SLinus Torvalds 
163484fa7933SPatrick McHardy 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1635ff1dcadbSAl Viro 		long csstuff = csstart + skb->csum_offset;
16361da177e4SLinus Torvalds 
1637d3bc23e7SAl Viro 		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
16381da177e4SLinus Torvalds 	}
16391da177e4SLinus Torvalds }
16401da177e4SLinus Torvalds 
16411da177e4SLinus Torvalds /**
16421da177e4SLinus Torvalds  *	skb_dequeue - remove from the head of the queue
16431da177e4SLinus Torvalds  *	@list: list to dequeue from
16441da177e4SLinus Torvalds  *
16451da177e4SLinus Torvalds  *	Remove the head of the list. The list lock is taken so the function
16461da177e4SLinus Torvalds  *	may be used safely with other locking list functions. The head item is
16471da177e4SLinus Torvalds  *	returned or %NULL if the list is empty.
16481da177e4SLinus Torvalds  */
16491da177e4SLinus Torvalds 
16501da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list)
16511da177e4SLinus Torvalds {
16521da177e4SLinus Torvalds 	unsigned long flags;
16531da177e4SLinus Torvalds 	struct sk_buff *result;
16541da177e4SLinus Torvalds 
16551da177e4SLinus Torvalds 	spin_lock_irqsave(&list->lock, flags);
16561da177e4SLinus Torvalds 	result = __skb_dequeue(list);
16571da177e4SLinus Torvalds 	spin_unlock_irqrestore(&list->lock, flags);
16581da177e4SLinus Torvalds 	return result;
16591da177e4SLinus Torvalds }
16601da177e4SLinus Torvalds 
16611da177e4SLinus Torvalds /**
16621da177e4SLinus Torvalds  *	skb_dequeue_tail - remove from the tail of the queue
16631da177e4SLinus Torvalds  *	@list: list to dequeue from
16641da177e4SLinus Torvalds  *
16651da177e4SLinus Torvalds  *	Remove the tail of the list. The list lock is taken so the function
16661da177e4SLinus Torvalds  *	may be used safely with other locking list functions. The tail item is
16671da177e4SLinus Torvalds  *	returned or %NULL if the list is empty.
16681da177e4SLinus Torvalds  */
16691da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
16701da177e4SLinus Torvalds {
16711da177e4SLinus Torvalds 	unsigned long flags;
16721da177e4SLinus Torvalds 	struct sk_buff *result;
16731da177e4SLinus Torvalds 
16741da177e4SLinus Torvalds 	spin_lock_irqsave(&list->lock, flags);
16751da177e4SLinus Torvalds 	result = __skb_dequeue_tail(list);
16761da177e4SLinus Torvalds 	spin_unlock_irqrestore(&list->lock, flags);
16771da177e4SLinus Torvalds 	return result;
16781da177e4SLinus Torvalds }
16791da177e4SLinus Torvalds 
16801da177e4SLinus Torvalds /**
16811da177e4SLinus Torvalds  *	skb_queue_purge - empty a list
16821da177e4SLinus Torvalds  *	@list: list to empty
16831da177e4SLinus Torvalds  *
16841da177e4SLinus Torvalds  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
16851da177e4SLinus Torvalds  *	the list and one reference dropped. This function takes the list
16861da177e4SLinus Torvalds  *	lock and is atomic with respect to other list locking functions.
16871da177e4SLinus Torvalds  */
16881da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list)
16891da177e4SLinus Torvalds {
16901da177e4SLinus Torvalds 	struct sk_buff *skb;
16911da177e4SLinus Torvalds 	while ((skb = skb_dequeue(list)) != NULL)
16921da177e4SLinus Torvalds 		kfree_skb(skb);
16931da177e4SLinus Torvalds }
16941da177e4SLinus Torvalds 
16951da177e4SLinus Torvalds /**
16961da177e4SLinus Torvalds  *	skb_queue_head - queue a buffer at the list head
16971da177e4SLinus Torvalds  *	@list: list to use
16981da177e4SLinus Torvalds  *	@newsk: buffer to queue
16991da177e4SLinus Torvalds  *
17001da177e4SLinus Torvalds  *	Queue a buffer at the start of the list. This function takes the
17011da177e4SLinus Torvalds  *	list lock and can be used safely with other locking &sk_buff functions
17021da177e4SLinus Torvalds  *	safely.
17031da177e4SLinus Torvalds  *
17041da177e4SLinus Torvalds  *	A buffer cannot be placed on two lists at the same time.
17051da177e4SLinus Torvalds  */
17061da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
17071da177e4SLinus Torvalds {
17081da177e4SLinus Torvalds 	unsigned long flags;
17091da177e4SLinus Torvalds 
17101da177e4SLinus Torvalds 	spin_lock_irqsave(&list->lock, flags);
17111da177e4SLinus Torvalds 	__skb_queue_head(list, newsk);
17121da177e4SLinus Torvalds 	spin_unlock_irqrestore(&list->lock, flags);
17131da177e4SLinus Torvalds }
17141da177e4SLinus Torvalds 
17151da177e4SLinus Torvalds /**
17161da177e4SLinus Torvalds  *	skb_queue_tail - queue a buffer at the list tail
17171da177e4SLinus Torvalds  *	@list: list to use
17181da177e4SLinus Torvalds  *	@newsk: buffer to queue
17191da177e4SLinus Torvalds  *
17201da177e4SLinus Torvalds  *	Queue a buffer at the tail of the list. This function takes the
17211da177e4SLinus Torvalds  *	list lock and can be used safely with other locking &sk_buff functions
17221da177e4SLinus Torvalds  *	safely.
17231da177e4SLinus Torvalds  *
17241da177e4SLinus Torvalds  *	A buffer cannot be placed on two lists at the same time.
17251da177e4SLinus Torvalds  */
17261da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
17271da177e4SLinus Torvalds {
17281da177e4SLinus Torvalds 	unsigned long flags;
17291da177e4SLinus Torvalds 
17301da177e4SLinus Torvalds 	spin_lock_irqsave(&list->lock, flags);
17311da177e4SLinus Torvalds 	__skb_queue_tail(list, newsk);
17321da177e4SLinus Torvalds 	spin_unlock_irqrestore(&list->lock, flags);
17331da177e4SLinus Torvalds }
17348728b834SDavid S. Miller 
17351da177e4SLinus Torvalds /**
17361da177e4SLinus Torvalds  *	skb_unlink	-	remove a buffer from a list
17371da177e4SLinus Torvalds  *	@skb: buffer to remove
17388728b834SDavid S. Miller  *	@list: list to use
17391da177e4SLinus Torvalds  *
17408728b834SDavid S. Miller  *	Remove a packet from a list. The list locks are taken and this
17418728b834SDavid S. Miller  *	function is atomic with respect to other list locked calls
17421da177e4SLinus Torvalds  *
17438728b834SDavid S. Miller  *	You must know what list the SKB is on.
17441da177e4SLinus Torvalds  */
17458728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
17461da177e4SLinus Torvalds {
17471da177e4SLinus Torvalds 	unsigned long flags;
17481da177e4SLinus Torvalds 
17491da177e4SLinus Torvalds 	spin_lock_irqsave(&list->lock, flags);
17508728b834SDavid S. Miller 	__skb_unlink(skb, list);
17511da177e4SLinus Torvalds 	spin_unlock_irqrestore(&list->lock, flags);
17521da177e4SLinus Torvalds }
17531da177e4SLinus Torvalds 
17541da177e4SLinus Torvalds /**
17551da177e4SLinus Torvalds  *	skb_append	-	append a buffer
17561da177e4SLinus Torvalds  *	@old: buffer to insert after
17571da177e4SLinus Torvalds  *	@newsk: buffer to insert
17588728b834SDavid S. Miller  *	@list: list to use
17591da177e4SLinus Torvalds  *
17601da177e4SLinus Torvalds  *	Place a packet after a given packet in a list. The list locks are taken
17611da177e4SLinus Torvalds  *	and this function is atomic with respect to other list locked calls.
17621da177e4SLinus Torvalds  *	A buffer cannot be placed on two lists at the same time.
17631da177e4SLinus Torvalds  */
17648728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
17651da177e4SLinus Torvalds {
17661da177e4SLinus Torvalds 	unsigned long flags;
17671da177e4SLinus Torvalds 
17688728b834SDavid S. Miller 	spin_lock_irqsave(&list->lock, flags);
17698728b834SDavid S. Miller 	__skb_append(old, newsk, list);
17708728b834SDavid S. Miller 	spin_unlock_irqrestore(&list->lock, flags);
17711da177e4SLinus Torvalds }
17721da177e4SLinus Torvalds 
17731da177e4SLinus Torvalds 
17741da177e4SLinus Torvalds /**
17751da177e4SLinus Torvalds  *	skb_insert	-	insert a buffer
17761da177e4SLinus Torvalds  *	@old: buffer to insert before
17771da177e4SLinus Torvalds  *	@newsk: buffer to insert
17788728b834SDavid S. Miller  *	@list: list to use
17791da177e4SLinus Torvalds  *
17808728b834SDavid S. Miller  *	Place a packet before a given packet in a list. The list locks are
17818728b834SDavid S. Miller  * 	taken and this function is atomic with respect to other list locked
17828728b834SDavid S. Miller  *	calls.
17838728b834SDavid S. Miller  *
17841da177e4SLinus Torvalds  *	A buffer cannot be placed on two lists at the same time.
17851da177e4SLinus Torvalds  */
17868728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
17871da177e4SLinus Torvalds {
17881da177e4SLinus Torvalds 	unsigned long flags;
17891da177e4SLinus Torvalds 
17908728b834SDavid S. Miller 	spin_lock_irqsave(&list->lock, flags);
17918728b834SDavid S. Miller 	__skb_insert(newsk, old->prev, old, list);
17928728b834SDavid S. Miller 	spin_unlock_irqrestore(&list->lock, flags);
17931da177e4SLinus Torvalds }
17941da177e4SLinus Torvalds 
17951da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb,
17961da177e4SLinus Torvalds 					   struct sk_buff* skb1,
17971da177e4SLinus Torvalds 					   const u32 len, const int pos)
17981da177e4SLinus Torvalds {
17991da177e4SLinus Torvalds 	int i;
18001da177e4SLinus Torvalds 
1801d626f62bSArnaldo Carvalho de Melo 	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1802d626f62bSArnaldo Carvalho de Melo 					 pos - len);
18031da177e4SLinus Torvalds 	/* And move data appendix as is. */
18041da177e4SLinus Torvalds 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
18051da177e4SLinus Torvalds 		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
18061da177e4SLinus Torvalds 
18071da177e4SLinus Torvalds 	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
18081da177e4SLinus Torvalds 	skb_shinfo(skb)->nr_frags  = 0;
18091da177e4SLinus Torvalds 	skb1->data_len		   = skb->data_len;
18101da177e4SLinus Torvalds 	skb1->len		   += skb1->data_len;
18111da177e4SLinus Torvalds 	skb->data_len		   = 0;
18121da177e4SLinus Torvalds 	skb->len		   = len;
181327a884dcSArnaldo Carvalho de Melo 	skb_set_tail_pointer(skb, len);
18141da177e4SLinus Torvalds }
18151da177e4SLinus Torvalds 
18161da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb,
18171da177e4SLinus Torvalds 				       struct sk_buff* skb1,
18181da177e4SLinus Torvalds 				       const u32 len, int pos)
18191da177e4SLinus Torvalds {
18201da177e4SLinus Torvalds 	int i, k = 0;
18211da177e4SLinus Torvalds 	const int nfrags = skb_shinfo(skb)->nr_frags;
18221da177e4SLinus Torvalds 
18231da177e4SLinus Torvalds 	skb_shinfo(skb)->nr_frags = 0;
18241da177e4SLinus Torvalds 	skb1->len		  = skb1->data_len = skb->len - len;
18251da177e4SLinus Torvalds 	skb->len		  = len;
18261da177e4SLinus Torvalds 	skb->data_len		  = len - pos;
18271da177e4SLinus Torvalds 
18281da177e4SLinus Torvalds 	for (i = 0; i < nfrags; i++) {
18291da177e4SLinus Torvalds 		int size = skb_shinfo(skb)->frags[i].size;
18301da177e4SLinus Torvalds 
18311da177e4SLinus Torvalds 		if (pos + size > len) {
18321da177e4SLinus Torvalds 			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
18331da177e4SLinus Torvalds 
18341da177e4SLinus Torvalds 			if (pos < len) {
18351da177e4SLinus Torvalds 				/* Split frag.
18361da177e4SLinus Torvalds 				 * We have two variants in this case:
18371da177e4SLinus Torvalds 				 * 1. Move all the frag to the second
18381da177e4SLinus Torvalds 				 *    part, if it is possible. F.e.
18391da177e4SLinus Torvalds 				 *    this approach is mandatory for TUX,
18401da177e4SLinus Torvalds 				 *    where splitting is expensive.
18411da177e4SLinus Torvalds 				 * 2. Split is accurately. We make this.
18421da177e4SLinus Torvalds 				 */
18431da177e4SLinus Torvalds 				get_page(skb_shinfo(skb)->frags[i].page);
18441da177e4SLinus Torvalds 				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
18451da177e4SLinus Torvalds 				skb_shinfo(skb1)->frags[0].size -= len - pos;
18461da177e4SLinus Torvalds 				skb_shinfo(skb)->frags[i].size	= len - pos;
18471da177e4SLinus Torvalds 				skb_shinfo(skb)->nr_frags++;
18481da177e4SLinus Torvalds 			}
18491da177e4SLinus Torvalds 			k++;
18501da177e4SLinus Torvalds 		} else
18511da177e4SLinus Torvalds 			skb_shinfo(skb)->nr_frags++;
18521da177e4SLinus Torvalds 		pos += size;
18531da177e4SLinus Torvalds 	}
18541da177e4SLinus Torvalds 	skb_shinfo(skb1)->nr_frags = k;
18551da177e4SLinus Torvalds }
18561da177e4SLinus Torvalds 
18571da177e4SLinus Torvalds /**
18581da177e4SLinus Torvalds  * skb_split - Split fragmented skb to two parts at length len.
18591da177e4SLinus Torvalds  * @skb: the buffer to split
18601da177e4SLinus Torvalds  * @skb1: the buffer to receive the second part
18611da177e4SLinus Torvalds  * @len: new length for skb
18621da177e4SLinus Torvalds  */
18631da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
18641da177e4SLinus Torvalds {
18651da177e4SLinus Torvalds 	int pos = skb_headlen(skb);
18661da177e4SLinus Torvalds 
18671da177e4SLinus Torvalds 	if (len < pos)	/* Split line is inside header. */
18681da177e4SLinus Torvalds 		skb_split_inside_header(skb, skb1, len, pos);
18691da177e4SLinus Torvalds 	else		/* Second chunk has no header, nothing to copy. */
18701da177e4SLinus Torvalds 		skb_split_no_header(skb, skb1, len, pos);
18711da177e4SLinus Torvalds }
18721da177e4SLinus Torvalds 
1873677e90edSThomas Graf /**
1874677e90edSThomas Graf  * skb_prepare_seq_read - Prepare a sequential read of skb data
1875677e90edSThomas Graf  * @skb: the buffer to read
1876677e90edSThomas Graf  * @from: lower offset of data to be read
1877677e90edSThomas Graf  * @to: upper offset of data to be read
1878677e90edSThomas Graf  * @st: state variable
1879677e90edSThomas Graf  *
1880677e90edSThomas Graf  * Initializes the specified state variable. Must be called before
1881677e90edSThomas Graf  * invoking skb_seq_read() for the first time.
1882677e90edSThomas Graf  */
1883677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1884677e90edSThomas Graf 			  unsigned int to, struct skb_seq_state *st)
1885677e90edSThomas Graf {
1886677e90edSThomas Graf 	st->lower_offset = from;
1887677e90edSThomas Graf 	st->upper_offset = to;
1888677e90edSThomas Graf 	st->root_skb = st->cur_skb = skb;
1889677e90edSThomas Graf 	st->frag_idx = st->stepped_offset = 0;
1890677e90edSThomas Graf 	st->frag_data = NULL;
1891677e90edSThomas Graf }
1892677e90edSThomas Graf 
1893677e90edSThomas Graf /**
1894677e90edSThomas Graf  * skb_seq_read - Sequentially read skb data
1895677e90edSThomas Graf  * @consumed: number of bytes consumed by the caller so far
1896677e90edSThomas Graf  * @data: destination pointer for data to be returned
1897677e90edSThomas Graf  * @st: state variable
1898677e90edSThomas Graf  *
1899677e90edSThomas Graf  * Reads a block of skb data at &consumed relative to the
1900677e90edSThomas Graf  * lower offset specified to skb_prepare_seq_read(). Assigns
1901677e90edSThomas Graf  * the head of the data block to &data and returns the length
1902677e90edSThomas Graf  * of the block or 0 if the end of the skb data or the upper
1903677e90edSThomas Graf  * offset has been reached.
1904677e90edSThomas Graf  *
1905677e90edSThomas Graf  * The caller is not required to consume all of the data
1906677e90edSThomas Graf  * returned, i.e. &consumed is typically set to the number
1907677e90edSThomas Graf  * of bytes already consumed and the next call to
1908677e90edSThomas Graf  * skb_seq_read() will return the remaining part of the block.
1909677e90edSThomas Graf  *
1910bc2cda1eSRandy Dunlap  * Note 1: The size of each block of data returned can be arbitary,
1911677e90edSThomas Graf  *       this limitation is the cost for zerocopy seqeuental
1912677e90edSThomas Graf  *       reads of potentially non linear data.
1913677e90edSThomas Graf  *
1914bc2cda1eSRandy Dunlap  * Note 2: Fragment lists within fragments are not implemented
1915677e90edSThomas Graf  *       at the moment, state->root_skb could be replaced with
1916677e90edSThomas Graf  *       a stack for this purpose.
1917677e90edSThomas Graf  */
1918677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1919677e90edSThomas Graf 			  struct skb_seq_state *st)
1920677e90edSThomas Graf {
1921677e90edSThomas Graf 	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1922677e90edSThomas Graf 	skb_frag_t *frag;
1923677e90edSThomas Graf 
1924677e90edSThomas Graf 	if (unlikely(abs_offset >= st->upper_offset))
1925677e90edSThomas Graf 		return 0;
1926677e90edSThomas Graf 
1927677e90edSThomas Graf next_skb:
1928677e90edSThomas Graf 	block_limit = skb_headlen(st->cur_skb);
1929677e90edSThomas Graf 
1930677e90edSThomas Graf 	if (abs_offset < block_limit) {
1931677e90edSThomas Graf 		*data = st->cur_skb->data + abs_offset;
1932677e90edSThomas Graf 		return block_limit - abs_offset;
1933677e90edSThomas Graf 	}
1934677e90edSThomas Graf 
1935677e90edSThomas Graf 	if (st->frag_idx == 0 && !st->frag_data)
1936677e90edSThomas Graf 		st->stepped_offset += skb_headlen(st->cur_skb);
1937677e90edSThomas Graf 
1938677e90edSThomas Graf 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1939677e90edSThomas Graf 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1940677e90edSThomas Graf 		block_limit = frag->size + st->stepped_offset;
1941677e90edSThomas Graf 
1942677e90edSThomas Graf 		if (abs_offset < block_limit) {
1943677e90edSThomas Graf 			if (!st->frag_data)
1944677e90edSThomas Graf 				st->frag_data = kmap_skb_frag(frag);
1945677e90edSThomas Graf 
1946677e90edSThomas Graf 			*data = (u8 *) st->frag_data + frag->page_offset +
1947677e90edSThomas Graf 				(abs_offset - st->stepped_offset);
1948677e90edSThomas Graf 
1949677e90edSThomas Graf 			return block_limit - abs_offset;
1950677e90edSThomas Graf 		}
1951677e90edSThomas Graf 
1952677e90edSThomas Graf 		if (st->frag_data) {
1953677e90edSThomas Graf 			kunmap_skb_frag(st->frag_data);
1954677e90edSThomas Graf 			st->frag_data = NULL;
1955677e90edSThomas Graf 		}
1956677e90edSThomas Graf 
1957677e90edSThomas Graf 		st->frag_idx++;
1958677e90edSThomas Graf 		st->stepped_offset += frag->size;
1959677e90edSThomas Graf 	}
1960677e90edSThomas Graf 
19615b5a60daSOlaf Kirch 	if (st->frag_data) {
19625b5a60daSOlaf Kirch 		kunmap_skb_frag(st->frag_data);
19635b5a60daSOlaf Kirch 		st->frag_data = NULL;
19645b5a60daSOlaf Kirch 	}
19655b5a60daSOlaf Kirch 
1966677e90edSThomas Graf 	if (st->cur_skb->next) {
1967677e90edSThomas Graf 		st->cur_skb = st->cur_skb->next;
1968677e90edSThomas Graf 		st->frag_idx = 0;
1969677e90edSThomas Graf 		goto next_skb;
1970677e90edSThomas Graf 	} else if (st->root_skb == st->cur_skb &&
1971677e90edSThomas Graf 		   skb_shinfo(st->root_skb)->frag_list) {
1972677e90edSThomas Graf 		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1973677e90edSThomas Graf 		goto next_skb;
1974677e90edSThomas Graf 	}
1975677e90edSThomas Graf 
1976677e90edSThomas Graf 	return 0;
1977677e90edSThomas Graf }
1978677e90edSThomas Graf 
1979677e90edSThomas Graf /**
1980677e90edSThomas Graf  * skb_abort_seq_read - Abort a sequential read of skb data
1981677e90edSThomas Graf  * @st: state variable
1982677e90edSThomas Graf  *
1983677e90edSThomas Graf  * Must be called if skb_seq_read() was not called until it
1984677e90edSThomas Graf  * returned 0.
1985677e90edSThomas Graf  */
1986677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st)
1987677e90edSThomas Graf {
1988677e90edSThomas Graf 	if (st->frag_data)
1989677e90edSThomas Graf 		kunmap_skb_frag(st->frag_data);
1990677e90edSThomas Graf }
1991677e90edSThomas Graf 
19923fc7e8a6SThomas Graf #define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
19933fc7e8a6SThomas Graf 
19943fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
19953fc7e8a6SThomas Graf 					  struct ts_config *conf,
19963fc7e8a6SThomas Graf 					  struct ts_state *state)
19973fc7e8a6SThomas Graf {
19983fc7e8a6SThomas Graf 	return skb_seq_read(offset, text, TS_SKB_CB(state));
19993fc7e8a6SThomas Graf }
20003fc7e8a6SThomas Graf 
20013fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
20023fc7e8a6SThomas Graf {
20033fc7e8a6SThomas Graf 	skb_abort_seq_read(TS_SKB_CB(state));
20043fc7e8a6SThomas Graf }
20053fc7e8a6SThomas Graf 
20063fc7e8a6SThomas Graf /**
20073fc7e8a6SThomas Graf  * skb_find_text - Find a text pattern in skb data
20083fc7e8a6SThomas Graf  * @skb: the buffer to look in
20093fc7e8a6SThomas Graf  * @from: search offset
20103fc7e8a6SThomas Graf  * @to: search limit
20113fc7e8a6SThomas Graf  * @config: textsearch configuration
20123fc7e8a6SThomas Graf  * @state: uninitialized textsearch state variable
20133fc7e8a6SThomas Graf  *
20143fc7e8a6SThomas Graf  * Finds a pattern in the skb data according to the specified
20153fc7e8a6SThomas Graf  * textsearch configuration. Use textsearch_next() to retrieve
20163fc7e8a6SThomas Graf  * subsequent occurrences of the pattern. Returns the offset
20173fc7e8a6SThomas Graf  * to the first occurrence or UINT_MAX if no match was found.
20183fc7e8a6SThomas Graf  */
20193fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
20203fc7e8a6SThomas Graf 			   unsigned int to, struct ts_config *config,
20213fc7e8a6SThomas Graf 			   struct ts_state *state)
20223fc7e8a6SThomas Graf {
2023f72b948dSPhil Oester 	unsigned int ret;
2024f72b948dSPhil Oester 
20253fc7e8a6SThomas Graf 	config->get_next_block = skb_ts_get_next_block;
20263fc7e8a6SThomas Graf 	config->finish = skb_ts_finish;
20273fc7e8a6SThomas Graf 
20283fc7e8a6SThomas Graf 	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
20293fc7e8a6SThomas Graf 
2030f72b948dSPhil Oester 	ret = textsearch_find(config, state);
2031f72b948dSPhil Oester 	return (ret <= to - from ? ret : UINT_MAX);
20323fc7e8a6SThomas Graf }
20333fc7e8a6SThomas Graf 
2034e89e9cf5SAnanda Raju /**
2035e89e9cf5SAnanda Raju  * skb_append_datato_frags: - append the user data to a skb
2036e89e9cf5SAnanda Raju  * @sk: sock  structure
2037e89e9cf5SAnanda Raju  * @skb: skb structure to be appened with user data.
2038e89e9cf5SAnanda Raju  * @getfrag: call back function to be used for getting the user data
2039e89e9cf5SAnanda Raju  * @from: pointer to user message iov
2040e89e9cf5SAnanda Raju  * @length: length of the iov message
2041e89e9cf5SAnanda Raju  *
2042e89e9cf5SAnanda Raju  * Description: This procedure append the user data in the fragment part
2043e89e9cf5SAnanda Raju  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2044e89e9cf5SAnanda Raju  */
2045e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2046dab9630fSMartin Waitz 			int (*getfrag)(void *from, char *to, int offset,
2047e89e9cf5SAnanda Raju 					int len, int odd, struct sk_buff *skb),
2048e89e9cf5SAnanda Raju 			void *from, int length)
2049e89e9cf5SAnanda Raju {
2050e89e9cf5SAnanda Raju 	int frg_cnt = 0;
2051e89e9cf5SAnanda Raju 	skb_frag_t *frag = NULL;
2052e89e9cf5SAnanda Raju 	struct page *page = NULL;
2053e89e9cf5SAnanda Raju 	int copy, left;
2054e89e9cf5SAnanda Raju 	int offset = 0;
2055e89e9cf5SAnanda Raju 	int ret;
2056e89e9cf5SAnanda Raju 
2057e89e9cf5SAnanda Raju 	do {
2058e89e9cf5SAnanda Raju 		/* Return error if we don't have space for new frag */
2059e89e9cf5SAnanda Raju 		frg_cnt = skb_shinfo(skb)->nr_frags;
2060e89e9cf5SAnanda Raju 		if (frg_cnt >= MAX_SKB_FRAGS)
2061e89e9cf5SAnanda Raju 			return -EFAULT;
2062e89e9cf5SAnanda Raju 
2063e89e9cf5SAnanda Raju 		/* allocate a new page for next frag */
2064e89e9cf5SAnanda Raju 		page = alloc_pages(sk->sk_allocation, 0);
2065e89e9cf5SAnanda Raju 
2066e89e9cf5SAnanda Raju 		/* If alloc_page fails just return failure and caller will
2067e89e9cf5SAnanda Raju 		 * free previous allocated pages by doing kfree_skb()
2068e89e9cf5SAnanda Raju 		 */
2069e89e9cf5SAnanda Raju 		if (page == NULL)
2070e89e9cf5SAnanda Raju 			return -ENOMEM;
2071e89e9cf5SAnanda Raju 
2072e89e9cf5SAnanda Raju 		/* initialize the next frag */
2073e89e9cf5SAnanda Raju 		sk->sk_sndmsg_page = page;
2074e89e9cf5SAnanda Raju 		sk->sk_sndmsg_off = 0;
2075e89e9cf5SAnanda Raju 		skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2076e89e9cf5SAnanda Raju 		skb->truesize += PAGE_SIZE;
2077e89e9cf5SAnanda Raju 		atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2078e89e9cf5SAnanda Raju 
2079e89e9cf5SAnanda Raju 		/* get the new initialized frag */
2080e89e9cf5SAnanda Raju 		frg_cnt = skb_shinfo(skb)->nr_frags;
2081e89e9cf5SAnanda Raju 		frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2082e89e9cf5SAnanda Raju 
2083e89e9cf5SAnanda Raju 		/* copy the user data to page */
2084e89e9cf5SAnanda Raju 		left = PAGE_SIZE - frag->page_offset;
2085e89e9cf5SAnanda Raju 		copy = (length > left)? left : length;
2086e89e9cf5SAnanda Raju 
2087e89e9cf5SAnanda Raju 		ret = getfrag(from, (page_address(frag->page) +
2088e89e9cf5SAnanda Raju 			    frag->page_offset + frag->size),
2089e89e9cf5SAnanda Raju 			    offset, copy, 0, skb);
2090e89e9cf5SAnanda Raju 		if (ret < 0)
2091e89e9cf5SAnanda Raju 			return -EFAULT;
2092e89e9cf5SAnanda Raju 
2093e89e9cf5SAnanda Raju 		/* copy was successful so update the size parameters */
2094e89e9cf5SAnanda Raju 		sk->sk_sndmsg_off += copy;
2095e89e9cf5SAnanda Raju 		frag->size += copy;
2096e89e9cf5SAnanda Raju 		skb->len += copy;
2097e89e9cf5SAnanda Raju 		skb->data_len += copy;
2098e89e9cf5SAnanda Raju 		offset += copy;
2099e89e9cf5SAnanda Raju 		length -= copy;
2100e89e9cf5SAnanda Raju 
2101e89e9cf5SAnanda Raju 	} while (length > 0);
2102e89e9cf5SAnanda Raju 
2103e89e9cf5SAnanda Raju 	return 0;
2104e89e9cf5SAnanda Raju }
2105e89e9cf5SAnanda Raju 
2106cbb042f9SHerbert Xu /**
2107cbb042f9SHerbert Xu  *	skb_pull_rcsum - pull skb and update receive checksum
2108cbb042f9SHerbert Xu  *	@skb: buffer to update
2109cbb042f9SHerbert Xu  *	@start: start of data before pull
2110cbb042f9SHerbert Xu  *	@len: length of data pulled
2111cbb042f9SHerbert Xu  *
2112cbb042f9SHerbert Xu  *	This function performs an skb_pull on the packet and updates
211384fa7933SPatrick McHardy  *	update the CHECKSUM_COMPLETE checksum.  It should be used on
211484fa7933SPatrick McHardy  *	receive path processing instead of skb_pull unless you know
211584fa7933SPatrick McHardy  *	that the checksum difference is zero (e.g., a valid IP header)
211684fa7933SPatrick McHardy  *	or you are setting ip_summed to CHECKSUM_NONE.
2117cbb042f9SHerbert Xu  */
2118cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2119cbb042f9SHerbert Xu {
2120cbb042f9SHerbert Xu 	BUG_ON(len > skb->len);
2121cbb042f9SHerbert Xu 	skb->len -= len;
2122cbb042f9SHerbert Xu 	BUG_ON(skb->len < skb->data_len);
2123cbb042f9SHerbert Xu 	skb_postpull_rcsum(skb, skb->data, len);
2124cbb042f9SHerbert Xu 	return skb->data += len;
2125cbb042f9SHerbert Xu }
2126cbb042f9SHerbert Xu 
2127f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2128f94691acSArnaldo Carvalho de Melo 
2129f4c50d99SHerbert Xu /**
2130f4c50d99SHerbert Xu  *	skb_segment - Perform protocol segmentation on skb.
2131f4c50d99SHerbert Xu  *	@skb: buffer to segment
2132576a30ebSHerbert Xu  *	@features: features for the output path (see dev->features)
2133f4c50d99SHerbert Xu  *
2134f4c50d99SHerbert Xu  *	This function performs segmentation on the given skb.  It returns
2135f4c50d99SHerbert Xu  *	the segment at the given position.  It returns NULL if there are
2136f4c50d99SHerbert Xu  *	no more segments to generate, or when an error is encountered.
2137f4c50d99SHerbert Xu  */
2138576a30ebSHerbert Xu struct sk_buff *skb_segment(struct sk_buff *skb, int features)
2139f4c50d99SHerbert Xu {
2140f4c50d99SHerbert Xu 	struct sk_buff *segs = NULL;
2141f4c50d99SHerbert Xu 	struct sk_buff *tail = NULL;
2142f4c50d99SHerbert Xu 	unsigned int mss = skb_shinfo(skb)->gso_size;
214398e399f8SArnaldo Carvalho de Melo 	unsigned int doffset = skb->data - skb_mac_header(skb);
2144f4c50d99SHerbert Xu 	unsigned int offset = doffset;
2145f4c50d99SHerbert Xu 	unsigned int headroom;
2146f4c50d99SHerbert Xu 	unsigned int len;
2147576a30ebSHerbert Xu 	int sg = features & NETIF_F_SG;
2148f4c50d99SHerbert Xu 	int nfrags = skb_shinfo(skb)->nr_frags;
2149f4c50d99SHerbert Xu 	int err = -ENOMEM;
2150f4c50d99SHerbert Xu 	int i = 0;
2151f4c50d99SHerbert Xu 	int pos;
2152f4c50d99SHerbert Xu 
2153f4c50d99SHerbert Xu 	__skb_push(skb, doffset);
2154f4c50d99SHerbert Xu 	headroom = skb_headroom(skb);
2155f4c50d99SHerbert Xu 	pos = skb_headlen(skb);
2156f4c50d99SHerbert Xu 
2157f4c50d99SHerbert Xu 	do {
2158f4c50d99SHerbert Xu 		struct sk_buff *nskb;
2159f4c50d99SHerbert Xu 		skb_frag_t *frag;
2160c8884eddSHerbert Xu 		int hsize;
2161f4c50d99SHerbert Xu 		int k;
2162f4c50d99SHerbert Xu 		int size;
2163f4c50d99SHerbert Xu 
2164f4c50d99SHerbert Xu 		len = skb->len - offset;
2165f4c50d99SHerbert Xu 		if (len > mss)
2166f4c50d99SHerbert Xu 			len = mss;
2167f4c50d99SHerbert Xu 
2168f4c50d99SHerbert Xu 		hsize = skb_headlen(skb) - offset;
2169f4c50d99SHerbert Xu 		if (hsize < 0)
2170f4c50d99SHerbert Xu 			hsize = 0;
2171c8884eddSHerbert Xu 		if (hsize > len || !sg)
2172c8884eddSHerbert Xu 			hsize = len;
2173f4c50d99SHerbert Xu 
2174c8884eddSHerbert Xu 		nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
2175f4c50d99SHerbert Xu 		if (unlikely(!nskb))
2176f4c50d99SHerbert Xu 			goto err;
2177f4c50d99SHerbert Xu 
2178f4c50d99SHerbert Xu 		if (segs)
2179f4c50d99SHerbert Xu 			tail->next = nskb;
2180f4c50d99SHerbert Xu 		else
2181f4c50d99SHerbert Xu 			segs = nskb;
2182f4c50d99SHerbert Xu 		tail = nskb;
2183f4c50d99SHerbert Xu 
2184f4c50d99SHerbert Xu 		nskb->dev = skb->dev;
2185f25f4e44SPeter P Waskiewicz Jr 		skb_copy_queue_mapping(nskb, skb);
2186f4c50d99SHerbert Xu 		nskb->priority = skb->priority;
2187f4c50d99SHerbert Xu 		nskb->protocol = skb->protocol;
2188f4c50d99SHerbert Xu 		nskb->dst = dst_clone(skb->dst);
2189f4c50d99SHerbert Xu 		memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
2190f4c50d99SHerbert Xu 		nskb->pkt_type = skb->pkt_type;
2191f4c50d99SHerbert Xu 		nskb->mac_len = skb->mac_len;
2192f4c50d99SHerbert Xu 
2193f4c50d99SHerbert Xu 		skb_reserve(nskb, headroom);
2194459a98edSArnaldo Carvalho de Melo 		skb_reset_mac_header(nskb);
2195ddc7b8e3SArnaldo Carvalho de Melo 		skb_set_network_header(nskb, skb->mac_len);
2196b0e380b1SArnaldo Carvalho de Melo 		nskb->transport_header = (nskb->network_header +
2197b0e380b1SArnaldo Carvalho de Melo 					  skb_network_header_len(skb));
2198d626f62bSArnaldo Carvalho de Melo 		skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
2199d626f62bSArnaldo Carvalho de Melo 					  doffset);
2200f4c50d99SHerbert Xu 		if (!sg) {
2201f4c50d99SHerbert Xu 			nskb->csum = skb_copy_and_csum_bits(skb, offset,
2202f4c50d99SHerbert Xu 							    skb_put(nskb, len),
2203f4c50d99SHerbert Xu 							    len, 0);
2204f4c50d99SHerbert Xu 			continue;
2205f4c50d99SHerbert Xu 		}
2206f4c50d99SHerbert Xu 
2207f4c50d99SHerbert Xu 		frag = skb_shinfo(nskb)->frags;
2208f4c50d99SHerbert Xu 		k = 0;
2209f4c50d99SHerbert Xu 
221084fa7933SPatrick McHardy 		nskb->ip_summed = CHECKSUM_PARTIAL;
2211f4c50d99SHerbert Xu 		nskb->csum = skb->csum;
2212d626f62bSArnaldo Carvalho de Melo 		skb_copy_from_linear_data_offset(skb, offset,
2213d626f62bSArnaldo Carvalho de Melo 						 skb_put(nskb, hsize), hsize);
2214f4c50d99SHerbert Xu 
2215f4c50d99SHerbert Xu 		while (pos < offset + len) {
2216f4c50d99SHerbert Xu 			BUG_ON(i >= nfrags);
2217f4c50d99SHerbert Xu 
2218f4c50d99SHerbert Xu 			*frag = skb_shinfo(skb)->frags[i];
2219f4c50d99SHerbert Xu 			get_page(frag->page);
2220f4c50d99SHerbert Xu 			size = frag->size;
2221f4c50d99SHerbert Xu 
2222f4c50d99SHerbert Xu 			if (pos < offset) {
2223f4c50d99SHerbert Xu 				frag->page_offset += offset - pos;
2224f4c50d99SHerbert Xu 				frag->size -= offset - pos;
2225f4c50d99SHerbert Xu 			}
2226f4c50d99SHerbert Xu 
2227f4c50d99SHerbert Xu 			k++;
2228f4c50d99SHerbert Xu 
2229f4c50d99SHerbert Xu 			if (pos + size <= offset + len) {
2230f4c50d99SHerbert Xu 				i++;
2231f4c50d99SHerbert Xu 				pos += size;
2232f4c50d99SHerbert Xu 			} else {
2233f4c50d99SHerbert Xu 				frag->size -= pos + size - (offset + len);
2234f4c50d99SHerbert Xu 				break;
2235f4c50d99SHerbert Xu 			}
2236f4c50d99SHerbert Xu 
2237f4c50d99SHerbert Xu 			frag++;
2238f4c50d99SHerbert Xu 		}
2239f4c50d99SHerbert Xu 
2240f4c50d99SHerbert Xu 		skb_shinfo(nskb)->nr_frags = k;
2241f4c50d99SHerbert Xu 		nskb->data_len = len - hsize;
2242f4c50d99SHerbert Xu 		nskb->len += nskb->data_len;
2243f4c50d99SHerbert Xu 		nskb->truesize += nskb->data_len;
2244f4c50d99SHerbert Xu 	} while ((offset += len) < skb->len);
2245f4c50d99SHerbert Xu 
2246f4c50d99SHerbert Xu 	return segs;
2247f4c50d99SHerbert Xu 
2248f4c50d99SHerbert Xu err:
2249f4c50d99SHerbert Xu 	while ((skb = segs)) {
2250f4c50d99SHerbert Xu 		segs = skb->next;
2251b08d5840SPatrick McHardy 		kfree_skb(skb);
2252f4c50d99SHerbert Xu 	}
2253f4c50d99SHerbert Xu 	return ERR_PTR(err);
2254f4c50d99SHerbert Xu }
2255f4c50d99SHerbert Xu 
2256f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment);
2257f4c50d99SHerbert Xu 
22581da177e4SLinus Torvalds void __init skb_init(void)
22591da177e4SLinus Torvalds {
22601da177e4SLinus Torvalds 	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
22611da177e4SLinus Torvalds 					      sizeof(struct sk_buff),
22621da177e4SLinus Torvalds 					      0,
2263e5d679f3SAlexey Dobriyan 					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
226420c2df83SPaul Mundt 					      NULL);
2265d179cd12SDavid S. Miller 	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2266d179cd12SDavid S. Miller 						(2*sizeof(struct sk_buff)) +
2267d179cd12SDavid S. Miller 						sizeof(atomic_t),
2268d179cd12SDavid S. Miller 						0,
2269e5d679f3SAlexey Dobriyan 						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
227020c2df83SPaul Mundt 						NULL);
22711da177e4SLinus Torvalds }
22721da177e4SLinus Torvalds 
2273716ea3a7SDavid Howells /**
2274716ea3a7SDavid Howells  *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2275716ea3a7SDavid Howells  *	@skb: Socket buffer containing the buffers to be mapped
2276716ea3a7SDavid Howells  *	@sg: The scatter-gather list to map into
2277716ea3a7SDavid Howells  *	@offset: The offset into the buffer's contents to start mapping
2278716ea3a7SDavid Howells  *	@len: Length of buffer space to be mapped
2279716ea3a7SDavid Howells  *
2280716ea3a7SDavid Howells  *	Fill the specified scatter-gather list with mappings/pointers into a
2281716ea3a7SDavid Howells  *	region of the buffer space attached to a socket buffer.
2282716ea3a7SDavid Howells  */
228351c739d1SDavid S. Miller static int
228451c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2285716ea3a7SDavid Howells {
22861a028e50SDavid S. Miller 	int start = skb_headlen(skb);
22871a028e50SDavid S. Miller 	int i, copy = start - offset;
2288716ea3a7SDavid Howells 	int elt = 0;
2289716ea3a7SDavid Howells 
2290716ea3a7SDavid Howells 	if (copy > 0) {
2291716ea3a7SDavid Howells 		if (copy > len)
2292716ea3a7SDavid Howells 			copy = len;
2293642f1490SJens Axboe 		sg_set_buf(sg, skb->data + offset, copy);
2294716ea3a7SDavid Howells 		elt++;
2295716ea3a7SDavid Howells 		if ((len -= copy) == 0)
2296716ea3a7SDavid Howells 			return elt;
2297716ea3a7SDavid Howells 		offset += copy;
2298716ea3a7SDavid Howells 	}
2299716ea3a7SDavid Howells 
2300716ea3a7SDavid Howells 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
23011a028e50SDavid S. Miller 		int end;
2302716ea3a7SDavid Howells 
23031a028e50SDavid S. Miller 		BUG_TRAP(start <= offset + len);
23041a028e50SDavid S. Miller 
23051a028e50SDavid S. Miller 		end = start + skb_shinfo(skb)->frags[i].size;
2306716ea3a7SDavid Howells 		if ((copy = end - offset) > 0) {
2307716ea3a7SDavid Howells 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2308716ea3a7SDavid Howells 
2309716ea3a7SDavid Howells 			if (copy > len)
2310716ea3a7SDavid Howells 				copy = len;
2311642f1490SJens Axboe 			sg_set_page(&sg[elt], frag->page, copy,
2312642f1490SJens Axboe 					frag->page_offset+offset-start);
2313716ea3a7SDavid Howells 			elt++;
2314716ea3a7SDavid Howells 			if (!(len -= copy))
2315716ea3a7SDavid Howells 				return elt;
2316716ea3a7SDavid Howells 			offset += copy;
2317716ea3a7SDavid Howells 		}
23181a028e50SDavid S. Miller 		start = end;
2319716ea3a7SDavid Howells 	}
2320716ea3a7SDavid Howells 
2321716ea3a7SDavid Howells 	if (skb_shinfo(skb)->frag_list) {
2322716ea3a7SDavid Howells 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2323716ea3a7SDavid Howells 
2324716ea3a7SDavid Howells 		for (; list; list = list->next) {
23251a028e50SDavid S. Miller 			int end;
2326716ea3a7SDavid Howells 
23271a028e50SDavid S. Miller 			BUG_TRAP(start <= offset + len);
23281a028e50SDavid S. Miller 
23291a028e50SDavid S. Miller 			end = start + list->len;
2330716ea3a7SDavid Howells 			if ((copy = end - offset) > 0) {
2331716ea3a7SDavid Howells 				if (copy > len)
2332716ea3a7SDavid Howells 					copy = len;
233351c739d1SDavid S. Miller 				elt += __skb_to_sgvec(list, sg+elt, offset - start,
233451c739d1SDavid S. Miller 						      copy);
2335716ea3a7SDavid Howells 				if ((len -= copy) == 0)
2336716ea3a7SDavid Howells 					return elt;
2337716ea3a7SDavid Howells 				offset += copy;
2338716ea3a7SDavid Howells 			}
23391a028e50SDavid S. Miller 			start = end;
2340716ea3a7SDavid Howells 		}
2341716ea3a7SDavid Howells 	}
2342716ea3a7SDavid Howells 	BUG_ON(len);
2343716ea3a7SDavid Howells 	return elt;
2344716ea3a7SDavid Howells }
2345716ea3a7SDavid Howells 
234651c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
234751c739d1SDavid S. Miller {
234851c739d1SDavid S. Miller 	int nsg = __skb_to_sgvec(skb, sg, offset, len);
234951c739d1SDavid S. Miller 
2350c46f2334SJens Axboe 	sg_mark_end(&sg[nsg - 1]);
235151c739d1SDavid S. Miller 
235251c739d1SDavid S. Miller 	return nsg;
235351c739d1SDavid S. Miller }
235451c739d1SDavid S. Miller 
2355716ea3a7SDavid Howells /**
2356716ea3a7SDavid Howells  *	skb_cow_data - Check that a socket buffer's data buffers are writable
2357716ea3a7SDavid Howells  *	@skb: The socket buffer to check.
2358716ea3a7SDavid Howells  *	@tailbits: Amount of trailing space to be added
2359716ea3a7SDavid Howells  *	@trailer: Returned pointer to the skb where the @tailbits space begins
2360716ea3a7SDavid Howells  *
2361716ea3a7SDavid Howells  *	Make sure that the data buffers attached to a socket buffer are
2362716ea3a7SDavid Howells  *	writable. If they are not, private copies are made of the data buffers
2363716ea3a7SDavid Howells  *	and the socket buffer is set to use these instead.
2364716ea3a7SDavid Howells  *
2365716ea3a7SDavid Howells  *	If @tailbits is given, make sure that there is space to write @tailbits
2366716ea3a7SDavid Howells  *	bytes of data beyond current end of socket buffer.  @trailer will be
2367716ea3a7SDavid Howells  *	set to point to the skb in which this space begins.
2368716ea3a7SDavid Howells  *
2369716ea3a7SDavid Howells  *	The number of scatterlist elements required to completely map the
2370716ea3a7SDavid Howells  *	COW'd and extended socket buffer will be returned.
2371716ea3a7SDavid Howells  */
2372716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2373716ea3a7SDavid Howells {
2374716ea3a7SDavid Howells 	int copyflag;
2375716ea3a7SDavid Howells 	int elt;
2376716ea3a7SDavid Howells 	struct sk_buff *skb1, **skb_p;
2377716ea3a7SDavid Howells 
2378716ea3a7SDavid Howells 	/* If skb is cloned or its head is paged, reallocate
2379716ea3a7SDavid Howells 	 * head pulling out all the pages (pages are considered not writable
2380716ea3a7SDavid Howells 	 * at the moment even if they are anonymous).
2381716ea3a7SDavid Howells 	 */
2382716ea3a7SDavid Howells 	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2383716ea3a7SDavid Howells 	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2384716ea3a7SDavid Howells 		return -ENOMEM;
2385716ea3a7SDavid Howells 
2386716ea3a7SDavid Howells 	/* Easy case. Most of packets will go this way. */
2387716ea3a7SDavid Howells 	if (!skb_shinfo(skb)->frag_list) {
2388716ea3a7SDavid Howells 		/* A little of trouble, not enough of space for trailer.
2389716ea3a7SDavid Howells 		 * This should not happen, when stack is tuned to generate
2390716ea3a7SDavid Howells 		 * good frames. OK, on miss we reallocate and reserve even more
2391716ea3a7SDavid Howells 		 * space, 128 bytes is fair. */
2392716ea3a7SDavid Howells 
2393716ea3a7SDavid Howells 		if (skb_tailroom(skb) < tailbits &&
2394716ea3a7SDavid Howells 		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2395716ea3a7SDavid Howells 			return -ENOMEM;
2396716ea3a7SDavid Howells 
2397716ea3a7SDavid Howells 		/* Voila! */
2398716ea3a7SDavid Howells 		*trailer = skb;
2399716ea3a7SDavid Howells 		return 1;
2400716ea3a7SDavid Howells 	}
2401716ea3a7SDavid Howells 
2402716ea3a7SDavid Howells 	/* Misery. We are in troubles, going to mincer fragments... */
2403716ea3a7SDavid Howells 
2404716ea3a7SDavid Howells 	elt = 1;
2405716ea3a7SDavid Howells 	skb_p = &skb_shinfo(skb)->frag_list;
2406716ea3a7SDavid Howells 	copyflag = 0;
2407716ea3a7SDavid Howells 
2408716ea3a7SDavid Howells 	while ((skb1 = *skb_p) != NULL) {
2409716ea3a7SDavid Howells 		int ntail = 0;
2410716ea3a7SDavid Howells 
2411716ea3a7SDavid Howells 		/* The fragment is partially pulled by someone,
2412716ea3a7SDavid Howells 		 * this can happen on input. Copy it and everything
2413716ea3a7SDavid Howells 		 * after it. */
2414716ea3a7SDavid Howells 
2415716ea3a7SDavid Howells 		if (skb_shared(skb1))
2416716ea3a7SDavid Howells 			copyflag = 1;
2417716ea3a7SDavid Howells 
2418716ea3a7SDavid Howells 		/* If the skb is the last, worry about trailer. */
2419716ea3a7SDavid Howells 
2420716ea3a7SDavid Howells 		if (skb1->next == NULL && tailbits) {
2421716ea3a7SDavid Howells 			if (skb_shinfo(skb1)->nr_frags ||
2422716ea3a7SDavid Howells 			    skb_shinfo(skb1)->frag_list ||
2423716ea3a7SDavid Howells 			    skb_tailroom(skb1) < tailbits)
2424716ea3a7SDavid Howells 				ntail = tailbits + 128;
2425716ea3a7SDavid Howells 		}
2426716ea3a7SDavid Howells 
2427716ea3a7SDavid Howells 		if (copyflag ||
2428716ea3a7SDavid Howells 		    skb_cloned(skb1) ||
2429716ea3a7SDavid Howells 		    ntail ||
2430716ea3a7SDavid Howells 		    skb_shinfo(skb1)->nr_frags ||
2431716ea3a7SDavid Howells 		    skb_shinfo(skb1)->frag_list) {
2432716ea3a7SDavid Howells 			struct sk_buff *skb2;
2433716ea3a7SDavid Howells 
2434716ea3a7SDavid Howells 			/* Fuck, we are miserable poor guys... */
2435716ea3a7SDavid Howells 			if (ntail == 0)
2436716ea3a7SDavid Howells 				skb2 = skb_copy(skb1, GFP_ATOMIC);
2437716ea3a7SDavid Howells 			else
2438716ea3a7SDavid Howells 				skb2 = skb_copy_expand(skb1,
2439716ea3a7SDavid Howells 						       skb_headroom(skb1),
2440716ea3a7SDavid Howells 						       ntail,
2441716ea3a7SDavid Howells 						       GFP_ATOMIC);
2442716ea3a7SDavid Howells 			if (unlikely(skb2 == NULL))
2443716ea3a7SDavid Howells 				return -ENOMEM;
2444716ea3a7SDavid Howells 
2445716ea3a7SDavid Howells 			if (skb1->sk)
2446716ea3a7SDavid Howells 				skb_set_owner_w(skb2, skb1->sk);
2447716ea3a7SDavid Howells 
2448716ea3a7SDavid Howells 			/* Looking around. Are we still alive?
2449716ea3a7SDavid Howells 			 * OK, link new skb, drop old one */
2450716ea3a7SDavid Howells 
2451716ea3a7SDavid Howells 			skb2->next = skb1->next;
2452716ea3a7SDavid Howells 			*skb_p = skb2;
2453716ea3a7SDavid Howells 			kfree_skb(skb1);
2454716ea3a7SDavid Howells 			skb1 = skb2;
2455716ea3a7SDavid Howells 		}
2456716ea3a7SDavid Howells 		elt++;
2457716ea3a7SDavid Howells 		*trailer = skb1;
2458716ea3a7SDavid Howells 		skb_p = &skb1->next;
2459716ea3a7SDavid Howells 	}
2460716ea3a7SDavid Howells 
2461716ea3a7SDavid Howells 	return elt;
2462716ea3a7SDavid Howells }
2463716ea3a7SDavid Howells 
2464f35d9d8aSRusty Russell /**
2465f35d9d8aSRusty Russell  * skb_partial_csum_set - set up and verify partial csum values for packet
2466f35d9d8aSRusty Russell  * @skb: the skb to set
2467f35d9d8aSRusty Russell  * @start: the number of bytes after skb->data to start checksumming.
2468f35d9d8aSRusty Russell  * @off: the offset from start to place the checksum.
2469f35d9d8aSRusty Russell  *
2470f35d9d8aSRusty Russell  * For untrusted partially-checksummed packets, we need to make sure the values
2471f35d9d8aSRusty Russell  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
2472f35d9d8aSRusty Russell  *
2473f35d9d8aSRusty Russell  * This function checks and sets those values and skb->ip_summed: if this
2474f35d9d8aSRusty Russell  * returns false you should drop the packet.
2475f35d9d8aSRusty Russell  */
2476f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
2477f35d9d8aSRusty Russell {
2478f35d9d8aSRusty Russell 	if (unlikely(start > skb->len - 2) ||
2479f35d9d8aSRusty Russell 	    unlikely((int)start + off > skb->len - 2)) {
2480f35d9d8aSRusty Russell 		if (net_ratelimit())
2481f35d9d8aSRusty Russell 			printk(KERN_WARNING
2482f35d9d8aSRusty Russell 			       "bad partial csum: csum=%u/%u len=%u\n",
2483f35d9d8aSRusty Russell 			       start, off, skb->len);
2484f35d9d8aSRusty Russell 		return false;
2485f35d9d8aSRusty Russell 	}
2486f35d9d8aSRusty Russell 	skb->ip_summed = CHECKSUM_PARTIAL;
2487f35d9d8aSRusty Russell 	skb->csum_start = skb_headroom(skb) + start;
2488f35d9d8aSRusty Russell 	skb->csum_offset = off;
2489f35d9d8aSRusty Russell 	return true;
2490f35d9d8aSRusty Russell }
2491f35d9d8aSRusty Russell 
24921da177e4SLinus Torvalds EXPORT_SYMBOL(___pskb_trim);
24931da177e4SLinus Torvalds EXPORT_SYMBOL(__kfree_skb);
2494231d06aeSJörn Engel EXPORT_SYMBOL(kfree_skb);
24951da177e4SLinus Torvalds EXPORT_SYMBOL(__pskb_pull_tail);
2496d179cd12SDavid S. Miller EXPORT_SYMBOL(__alloc_skb);
24978af27456SChristoph Hellwig EXPORT_SYMBOL(__netdev_alloc_skb);
24981da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_copy);
24991da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_expand_head);
25001da177e4SLinus Torvalds EXPORT_SYMBOL(skb_checksum);
25011da177e4SLinus Torvalds EXPORT_SYMBOL(skb_clone);
25021da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy);
25031da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_bits);
25041da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_dev);
25051da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_bits);
25061da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_expand);
25071da177e4SLinus Torvalds EXPORT_SYMBOL(skb_over_panic);
25081da177e4SLinus Torvalds EXPORT_SYMBOL(skb_pad);
25091da177e4SLinus Torvalds EXPORT_SYMBOL(skb_realloc_headroom);
25101da177e4SLinus Torvalds EXPORT_SYMBOL(skb_under_panic);
25111da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue);
25121da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue_tail);
25131da177e4SLinus Torvalds EXPORT_SYMBOL(skb_insert);
25141da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_purge);
25151da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_head);
25161da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_tail);
25171da177e4SLinus Torvalds EXPORT_SYMBOL(skb_unlink);
25181da177e4SLinus Torvalds EXPORT_SYMBOL(skb_append);
25191da177e4SLinus Torvalds EXPORT_SYMBOL(skb_split);
2520677e90edSThomas Graf EXPORT_SYMBOL(skb_prepare_seq_read);
2521677e90edSThomas Graf EXPORT_SYMBOL(skb_seq_read);
2522677e90edSThomas Graf EXPORT_SYMBOL(skb_abort_seq_read);
25233fc7e8a6SThomas Graf EXPORT_SYMBOL(skb_find_text);
2524e89e9cf5SAnanda Raju EXPORT_SYMBOL(skb_append_datato_frags);
2525716ea3a7SDavid Howells 
2526716ea3a7SDavid Howells EXPORT_SYMBOL_GPL(skb_to_sgvec);
2527716ea3a7SDavid Howells EXPORT_SYMBOL_GPL(skb_cow_data);
2528f35d9d8aSRusty Russell EXPORT_SYMBOL_GPL(skb_partial_csum_set);
2529