Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 *	Routines having to do with the 'struct sk_buff' memory handlers.
   3 *
   4 *	Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   5 *			Florian La Roche <rzsfl@rz.uni-sb.de>
   6 *
   7 *	Fixes:
   8 *		Alan Cox	:	Fixed the worst of the load
   9 *					balancer bugs.
  10 *		Dave Platt	:	Interrupt stacking fix.
  11 *	Richard Kooijman	:	Timestamp fixes.
  12 *		Alan Cox	:	Changed buffer format.
  13 *		Alan Cox	:	destructor hook for AF_UNIX etc.
  14 *		Linus Torvalds	:	Better skb_clone.
  15 *		Alan Cox	:	Added skb_copy.
  16 *		Alan Cox	:	Added all the changed routines Linus
  17 *					only put in the headers
  18 *		Ray VanTassle	:	Fixed --skb->lock in free
  19 *		Alan Cox	:	skb_copy copy arp field
  20 *		Andi Kleen	:	slabified it.
  21 *		Robert Olsson	:	Removed skb_head_pool
  22 *
  23 *	NOTE:
  24 *		The __skb_ routines should be called with interrupts
  25 *	disabled, or you better be *real* sure that the operation is atomic
  26 *	with respect to whatever list is being frobbed (e.g. via lock_sock()
  27 *	or via disabling bottom half handlers, etc).
  28 *
  29 *	This program is free software; you can redistribute it and/or
  30 *	modify it under the terms of the GNU General Public License
  31 *	as published by the Free Software Foundation; either version
  32 *	2 of the License, or (at your option) any later version.
  33 */
  34
  35/*
  36 *	The functions in this file will not compile correctly with gcc 2.4.x
  37 */
  38
  39#include <linux/module.h>
  40#include <linux/types.h>
  41#include <linux/kernel.h>
  42#include <linux/kmemcheck.h>
  43#include <linux/mm.h>
  44#include <linux/interrupt.h>
  45#include <linux/in.h>
  46#include <linux/inet.h>
  47#include <linux/slab.h>
  48#include <linux/netdevice.h>
  49#ifdef CONFIG_NET_CLS_ACT
  50#include <net/pkt_sched.h>
  51#endif
  52#include <linux/string.h>
  53#include <linux/skbuff.h>
  54#include <linux/splice.h>
  55#include <linux/cache.h>
  56#include <linux/rtnetlink.h>
  57#include <linux/init.h>
  58#include <linux/scatterlist.h>
  59#include <linux/errqueue.h>
  60#include <linux/prefetch.h>
  61
  62#include <net/protocol.h>
  63#include <net/dst.h>
  64#include <net/sock.h>
  65#include <net/checksum.h>
  66#include <net/xfrm.h>
  67
  68#include <asm/uaccess.h>
  69#include <asm/system.h>
  70#include <trace/events/skb.h>
  71
  72#include "kmap_skb.h"
  73
  74static struct kmem_cache *skbuff_head_cache __read_mostly;
  75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
  76
  77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
  78				  struct pipe_buffer *buf)
  79{
  80	put_page(buf->page);
  81}
  82
  83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
  84				struct pipe_buffer *buf)
  85{
  86	get_page(buf->page);
  87}
  88
  89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
  90			       struct pipe_buffer *buf)
  91{
  92	return 1;
  93}
  94
  95
  96/* Pipe buffer operations for a socket. */
  97static const struct pipe_buf_operations sock_pipe_buf_ops = {
  98	.can_merge = 0,
  99	.map = generic_pipe_buf_map,
 100	.unmap = generic_pipe_buf_unmap,
 101	.confirm = generic_pipe_buf_confirm,
 102	.release = sock_pipe_buf_release,
 103	.steal = sock_pipe_buf_steal,
 104	.get = sock_pipe_buf_get,
 105};
 106
 107/*
 108 *	Keep out-of-line to prevent kernel bloat.
 109 *	__builtin_return_address is not used because it is not always
 110 *	reliable.
 111 */
 112
 113/**
 114 *	skb_over_panic	- 	private function
 115 *	@skb: buffer
 116 *	@sz: size
 117 *	@here: address
 118 *
 119 *	Out of line support code for skb_put(). Not user callable.
 120 */
 121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
 122{
 123	printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
 124			  "data:%p tail:%#lx end:%#lx dev:%s\n",
 125	       here, skb->len, sz, skb->head, skb->data,
 126	       (unsigned long)skb->tail, (unsigned long)skb->end,
 127	       skb->dev ? skb->dev->name : "<NULL>");
 128	BUG();
 129}
 130
 131/**
 132 *	skb_under_panic	- 	private function
 133 *	@skb: buffer
 134 *	@sz: size
 135 *	@here: address
 136 *
 137 *	Out of line support code for skb_push(). Not user callable.
 138 */
 139
 140static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
 141{
 142	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
 143			  "data:%p tail:%#lx end:%#lx dev:%s\n",
 144	       here, skb->len, sz, skb->head, skb->data,
 145	       (unsigned long)skb->tail, (unsigned long)skb->end,
 146	       skb->dev ? skb->dev->name : "<NULL>");
 147	BUG();
 148}
 149
 150/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 151 *	'private' fields and also do memory statistics to find all the
 152 *	[BEEP] leaks.
 153 *
 154 */
 155
 156/**
 157 *	__alloc_skb	-	allocate a network buffer
 158 *	@size: size to allocate
 159 *	@gfp_mask: allocation mask
 160 *	@fclone: allocate from fclone cache instead of head cache
 161 *		and allocate a cloned (child) skb
 162 *	@node: numa node to allocate memory on
 163 *
 164 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 165 *	tail room of size bytes. The object has a reference count of one.
 166 *	The return is the buffer. On a failure the return is %NULL.
 167 *
 168 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 169 *	%GFP_ATOMIC.
 170 */
 171struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 172			    int fclone, int node)
 173{
 174	struct kmem_cache *cache;
 175	struct skb_shared_info *shinfo;
 176	struct sk_buff *skb;
 177	u8 *data;
 178
 179	cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
 180
 181	/* Get the HEAD */
 182	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
 183	if (!skb)
 184		goto out;
 185	prefetchw(skb);
 186
 187	size = SKB_DATA_ALIGN(size);
 188	data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
 189			gfp_mask, node);
 190	if (!data)
 191		goto nodata;
 192	prefetchw(data + size);
 193
 194	/*
 195	 * Only clear those fields we need to clear, not those that we will
 196	 * actually initialise below. Hence, don't put any more fields after
 197	 * the tail pointer in struct sk_buff!
 198	 */
 199	memset(skb, 0, offsetof(struct sk_buff, tail));
 200	skb->truesize = size + sizeof(struct sk_buff);
 201	atomic_set(&skb->users, 1);
 202	skb->head = data;
 203	skb->data = data;
 204	skb_reset_tail_pointer(skb);
 205	skb->end = skb->tail + size;
 206#ifdef NET_SKBUFF_DATA_USES_OFFSET
 207	skb->mac_header = ~0U;
 208#endif
 209
 210	/* make sure we initialize shinfo sequentially */
 211	shinfo = skb_shinfo(skb);
 212	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 213	atomic_set(&shinfo->dataref, 1);
 214	kmemcheck_annotate_variable(shinfo->destructor_arg);
 215
 216	if (fclone) {
 217		struct sk_buff *child = skb + 1;
 218		atomic_t *fclone_ref = (atomic_t *) (child + 1);
 219
 220		kmemcheck_annotate_bitfield(child, flags1);
 221		kmemcheck_annotate_bitfield(child, flags2);
 222		skb->fclone = SKB_FCLONE_ORIG;
 223		atomic_set(fclone_ref, 1);
 224
 225		child->fclone = SKB_FCLONE_UNAVAILABLE;
 226	}
 227out:
 228	return skb;
 229nodata:
 230	kmem_cache_free(cache, skb);
 231	skb = NULL;
 232	goto out;
 233}
 234EXPORT_SYMBOL(__alloc_skb);
 235
 236/**
 237 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 238 *	@dev: network device to receive on
 239 *	@length: length to allocate
 240 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 241 *
 242 *	Allocate a new &sk_buff and assign it a usage count of one. The
 243 *	buffer has unspecified headroom built in. Users should allocate
 244 *	the headroom they think they need without accounting for the
 245 *	built in space. The built in space is used for optimisations.
 246 *
 247 *	%NULL is returned if there is no free memory.
 248 */
 249struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
 250		unsigned int length, gfp_t gfp_mask)
 251{
 252	struct sk_buff *skb;
 253
 254	skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
 255	if (likely(skb)) {
 256		skb_reserve(skb, NET_SKB_PAD);
 257		skb->dev = dev;
 258	}
 259	return skb;
 260}
 261EXPORT_SYMBOL(__netdev_alloc_skb);
 262
 263void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 264		int size)
 265{
 266	skb_fill_page_desc(skb, i, page, off, size);
 267	skb->len += size;
 268	skb->data_len += size;
 269	skb->truesize += size;
 270}
 271EXPORT_SYMBOL(skb_add_rx_frag);
 272
 273/**
 274 *	dev_alloc_skb - allocate an skbuff for receiving
 275 *	@length: length to allocate
 276 *
 277 *	Allocate a new &sk_buff and assign it a usage count of one. The
 278 *	buffer has unspecified headroom built in. Users should allocate
 279 *	the headroom they think they need without accounting for the
 280 *	built in space. The built in space is used for optimisations.
 281 *
 282 *	%NULL is returned if there is no free memory. Although this function
 283 *	allocates memory it can be called from an interrupt.
 284 */
 285struct sk_buff *dev_alloc_skb(unsigned int length)
 286{
 287	/*
 288	 * There is more code here than it seems:
 289	 * __dev_alloc_skb is an inline
 290	 */
 291	return __dev_alloc_skb(length, GFP_ATOMIC);
 292}
 293EXPORT_SYMBOL(dev_alloc_skb);
 294
 295static void skb_drop_list(struct sk_buff **listp)
 296{
 297	struct sk_buff *list = *listp;
 298
 299	*listp = NULL;
 300
 301	do {
 302		struct sk_buff *this = list;
 303		list = list->next;
 304		kfree_skb(this);
 305	} while (list);
 306}
 307
 308static inline void skb_drop_fraglist(struct sk_buff *skb)
 309{
 310	skb_drop_list(&skb_shinfo(skb)->frag_list);
 311}
 312
 313static void skb_clone_fraglist(struct sk_buff *skb)
 314{
 315	struct sk_buff *list;
 316
 317	skb_walk_frags(skb, list)
 318		skb_get(list);
 319}
 320
 321static void skb_release_data(struct sk_buff *skb)
 322{
 323	if (!skb->cloned ||
 324	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
 325			       &skb_shinfo(skb)->dataref)) {
 326		if (skb_shinfo(skb)->nr_frags) {
 327			int i;
 328			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 329				put_page(skb_shinfo(skb)->frags[i].page);
 330		}
 331
 332		/*
 333		 * If skb buf is from userspace, we need to notify the caller
 334		 * the lower device DMA has done;
 335		 */
 336		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 337			struct ubuf_info *uarg;
 338
 339			uarg = skb_shinfo(skb)->destructor_arg;
 340			if (uarg->callback)
 341				uarg->callback(uarg);
 342		}
 343
 344		if (skb_has_frag_list(skb))
 345			skb_drop_fraglist(skb);
 346
 347		kfree(skb->head);
 348	}
 349}
 350
 351/*
 352 *	Free an skbuff by memory without cleaning the state.
 353 */
 354static void kfree_skbmem(struct sk_buff *skb)
 355{
 356	struct sk_buff *other;
 357	atomic_t *fclone_ref;
 358
 359	switch (skb->fclone) {
 360	case SKB_FCLONE_UNAVAILABLE:
 361		kmem_cache_free(skbuff_head_cache, skb);
 362		break;
 363
 364	case SKB_FCLONE_ORIG:
 365		fclone_ref = (atomic_t *) (skb + 2);
 366		if (atomic_dec_and_test(fclone_ref))
 367			kmem_cache_free(skbuff_fclone_cache, skb);
 368		break;
 369
 370	case SKB_FCLONE_CLONE:
 371		fclone_ref = (atomic_t *) (skb + 1);
 372		other = skb - 1;
 373
 374		/* The clone portion is available for
 375		 * fast-cloning again.
 376		 */
 377		skb->fclone = SKB_FCLONE_UNAVAILABLE;
 378
 379		if (atomic_dec_and_test(fclone_ref))
 380			kmem_cache_free(skbuff_fclone_cache, other);
 381		break;
 382	}
 383}
 384
 385static void skb_release_head_state(struct sk_buff *skb)
 386{
 387	skb_dst_drop(skb);
 388#ifdef CONFIG_XFRM
 389	secpath_put(skb->sp);
 390#endif
 391	if (skb->destructor) {
 392		WARN_ON(in_irq());
 393		skb->destructor(skb);
 394	}
 395#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 396	nf_conntrack_put(skb->nfct);
 397#endif
 398#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
 399	nf_conntrack_put_reasm(skb->nfct_reasm);
 400#endif
 401#ifdef CONFIG_BRIDGE_NETFILTER
 402	nf_bridge_put(skb->nf_bridge);
 403#endif
 404/* XXX: IS this still necessary? - JHS */
 405#ifdef CONFIG_NET_SCHED
 406	skb->tc_index = 0;
 407#ifdef CONFIG_NET_CLS_ACT
 408	skb->tc_verd = 0;
 409#endif
 410#endif
 411}
 412
 413/* Free everything but the sk_buff shell. */
 414static void skb_release_all(struct sk_buff *skb)
 415{
 416	skb_release_head_state(skb);
 417	skb_release_data(skb);
 418}
 419
 420/**
 421 *	__kfree_skb - private function
 422 *	@skb: buffer
 423 *
 424 *	Free an sk_buff. Release anything attached to the buffer.
 425 *	Clean the state. This is an internal helper function. Users should
 426 *	always call kfree_skb
 427 */
 428
 429void __kfree_skb(struct sk_buff *skb)
 430{
 431	skb_release_all(skb);
 432	kfree_skbmem(skb);
 433}
 434EXPORT_SYMBOL(__kfree_skb);
 435
 436/**
 437 *	kfree_skb - free an sk_buff
 438 *	@skb: buffer to free
 439 *
 440 *	Drop a reference to the buffer and free it if the usage count has
 441 *	hit zero.
 442 */
 443void kfree_skb(struct sk_buff *skb)
 444{
 445	if (unlikely(!skb))
 446		return;
 447	if (likely(atomic_read(&skb->users) == 1))
 448		smp_rmb();
 449	else if (likely(!atomic_dec_and_test(&skb->users)))
 450		return;
 451	trace_kfree_skb(skb, __builtin_return_address(0));
 452	__kfree_skb(skb);
 453}
 454EXPORT_SYMBOL(kfree_skb);
 455
 456/**
 457 *	consume_skb - free an skbuff
 458 *	@skb: buffer to free
 459 *
 460 *	Drop a ref to the buffer and free it if the usage count has hit zero
 461 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
 462 *	is being dropped after a failure and notes that
 463 */
 464void consume_skb(struct sk_buff *skb)
 465{
 466	if (unlikely(!skb))
 467		return;
 468	if (likely(atomic_read(&skb->users) == 1))
 469		smp_rmb();
 470	else if (likely(!atomic_dec_and_test(&skb->users)))
 471		return;
 472	trace_consume_skb(skb);
 473	__kfree_skb(skb);
 474}
 475EXPORT_SYMBOL(consume_skb);
 476
 477/**
 478 *	skb_recycle_check - check if skb can be reused for receive
 479 *	@skb: buffer
 480 *	@skb_size: minimum receive buffer size
 481 *
 482 *	Checks that the skb passed in is not shared or cloned, and
 483 *	that it is linear and its head portion at least as large as
 484 *	skb_size so that it can be recycled as a receive buffer.
 485 *	If these conditions are met, this function does any necessary
 486 *	reference count dropping and cleans up the skbuff as if it
 487 *	just came from __alloc_skb().
 488 */
 489bool skb_recycle_check(struct sk_buff *skb, int skb_size)
 490{
 491	struct skb_shared_info *shinfo;
 492
 493	if (irqs_disabled())
 494		return false;
 495
 496	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
 497		return false;
 498
 499	if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
 500		return false;
 501
 502	skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
 503	if (skb_end_pointer(skb) - skb->head < skb_size)
 504		return false;
 505
 506	if (skb_shared(skb) || skb_cloned(skb))
 507		return false;
 508
 509	skb_release_head_state(skb);
 510
 511	shinfo = skb_shinfo(skb);
 512	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 513	atomic_set(&shinfo->dataref, 1);
 514
 515	memset(skb, 0, offsetof(struct sk_buff, tail));
 516	skb->data = skb->head + NET_SKB_PAD;
 517	skb_reset_tail_pointer(skb);
 518
 519	return true;
 520}
 521EXPORT_SYMBOL(skb_recycle_check);
 522
 523static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 524{
 525	new->tstamp		= old->tstamp;
 526	new->dev		= old->dev;
 527	new->transport_header	= old->transport_header;
 528	new->network_header	= old->network_header;
 529	new->mac_header		= old->mac_header;
 530	skb_dst_copy(new, old);
 531	new->rxhash		= old->rxhash;
 532#ifdef CONFIG_XFRM
 533	new->sp			= secpath_get(old->sp);
 534#endif
 535	memcpy(new->cb, old->cb, sizeof(old->cb));
 536	new->csum		= old->csum;
 537	new->local_df		= old->local_df;
 538	new->pkt_type		= old->pkt_type;
 539	new->ip_summed		= old->ip_summed;
 540	skb_copy_queue_mapping(new, old);
 541	new->priority		= old->priority;
 542#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 543	new->ipvs_property	= old->ipvs_property;
 544#endif
 545	new->protocol		= old->protocol;
 546	new->mark		= old->mark;
 547	new->skb_iif		= old->skb_iif;
 548	__nf_copy(new, old);
 549#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 550    defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 551	new->nf_trace		= old->nf_trace;
 552#endif
 553#ifdef CONFIG_NET_SCHED
 554	new->tc_index		= old->tc_index;
 555#ifdef CONFIG_NET_CLS_ACT
 556	new->tc_verd		= old->tc_verd;
 557#endif
 558#endif
 559	new->vlan_tci		= old->vlan_tci;
 560
 561	skb_copy_secmark(new, old);
 562}
 563
 564/*
 565 * You should not add any new code to this function.  Add it to
 566 * __copy_skb_header above instead.
 567 */
 568static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 569{
 570#define C(x) n->x = skb->x
 571
 572	n->next = n->prev = NULL;
 573	n->sk = NULL;
 574	__copy_skb_header(n, skb);
 575
 576	C(len);
 577	C(data_len);
 578	C(mac_len);
 579	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 580	n->cloned = 1;
 581	n->nohdr = 0;
 582	n->destructor = NULL;
 583	C(tail);
 584	C(end);
 585	C(head);
 586	C(data);
 587	C(truesize);
 588	atomic_set(&n->users, 1);
 589
 590	atomic_inc(&(skb_shinfo(skb)->dataref));
 591	skb->cloned = 1;
 592
 593	return n;
 594#undef C
 595}
 596
 597/**
 598 *	skb_morph	-	morph one skb into another
 599 *	@dst: the skb to receive the contents
 600 *	@src: the skb to supply the contents
 601 *
 602 *	This is identical to skb_clone except that the target skb is
 603 *	supplied by the user.
 604 *
 605 *	The target skb is returned upon exit.
 606 */
 607struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 608{
 609	skb_release_all(dst);
 610	return __skb_clone(dst, src);
 611}
 612EXPORT_SYMBOL_GPL(skb_morph);
 613
 614/*	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
 615 *	@skb: the skb to modify
 616 *	@gfp_mask: allocation priority
 617 *
 618 *	This must be called on SKBTX_DEV_ZEROCOPY skb.
 619 *	It will copy all frags into kernel and drop the reference
 620 *	to userspace pages.
 621 *
 622 *	If this function is called from an interrupt gfp_mask() must be
 623 *	%GFP_ATOMIC.
 624 *
 625 *	Returns 0 on success or a negative error code on failure
 626 *	to allocate kernel memory to copy to.
 627 */
 628int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
 629{
 630	int i;
 631	int num_frags = skb_shinfo(skb)->nr_frags;
 632	struct page *page, *head = NULL;
 633	struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
 634
 635	for (i = 0; i < num_frags; i++) {
 636		u8 *vaddr;
 637		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 638
 639		page = alloc_page(GFP_ATOMIC);
 640		if (!page) {
 641			while (head) {
 642				struct page *next = (struct page *)head->private;
 643				put_page(head);
 644				head = next;
 645			}
 646			return -ENOMEM;
 647		}
 648		vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
 649		memcpy(page_address(page),
 650		       vaddr + f->page_offset, f->size);
 651		kunmap_skb_frag(vaddr);
 652		page->private = (unsigned long)head;
 653		head = page;
 654	}
 655
 656	/* skb frags release userspace buffers */
 657	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 658		put_page(skb_shinfo(skb)->frags[i].page);
 659
 660	uarg->callback(uarg);
 661
 662	/* skb frags point to kernel buffers */
 663	for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
 664		skb_shinfo(skb)->frags[i - 1].page_offset = 0;
 665		skb_shinfo(skb)->frags[i - 1].page = head;
 666		head = (struct page *)head->private;
 667	}
 668
 669	skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
 670	return 0;
 671}
 672
 673
 674/**
 675 *	skb_clone	-	duplicate an sk_buff
 676 *	@skb: buffer to clone
 677 *	@gfp_mask: allocation priority
 678 *
 679 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
 680 *	copies share the same packet data but not structure. The new
 681 *	buffer has a reference count of 1. If the allocation fails the
 682 *	function returns %NULL otherwise the new buffer is returned.
 683 *
 684 *	If this function is called from an interrupt gfp_mask() must be
 685 *	%GFP_ATOMIC.
 686 */
 687
 688struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 689{
 690	struct sk_buff *n;
 691
 692	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 693		if (skb_copy_ubufs(skb, gfp_mask))
 694			return NULL;
 695	}
 696
 697	n = skb + 1;
 698	if (skb->fclone == SKB_FCLONE_ORIG &&
 699	    n->fclone == SKB_FCLONE_UNAVAILABLE) {
 700		atomic_t *fclone_ref = (atomic_t *) (n + 1);
 701		n->fclone = SKB_FCLONE_CLONE;
 702		atomic_inc(fclone_ref);
 703	} else {
 704		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
 705		if (!n)
 706			return NULL;
 707
 708		kmemcheck_annotate_bitfield(n, flags1);
 709		kmemcheck_annotate_bitfield(n, flags2);
 710		n->fclone = SKB_FCLONE_UNAVAILABLE;
 711	}
 712
 713	return __skb_clone(n, skb);
 714}
 715EXPORT_SYMBOL(skb_clone);
 716
 717static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 718{
 719#ifndef NET_SKBUFF_DATA_USES_OFFSET
 720	/*
 721	 *	Shift between the two data areas in bytes
 722	 */
 723	unsigned long offset = new->data - old->data;
 724#endif
 725
 726	__copy_skb_header(new, old);
 727
 728#ifndef NET_SKBUFF_DATA_USES_OFFSET
 729	/* {transport,network,mac}_header are relative to skb->head */
 730	new->transport_header += offset;
 731	new->network_header   += offset;
 732	if (skb_mac_header_was_set(new))
 733		new->mac_header	      += offset;
 734#endif
 735	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
 736	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
 737	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
 738}
 739
 740/**
 741 *	skb_copy	-	create private copy of an sk_buff
 742 *	@skb: buffer to copy
 743 *	@gfp_mask: allocation priority
 744 *
 745 *	Make a copy of both an &sk_buff and its data. This is used when the
 746 *	caller wishes to modify the data and needs a private copy of the
 747 *	data to alter. Returns %NULL on failure or the pointer to the buffer
 748 *	on success. The returned buffer has a reference count of 1.
 749 *
 750 *	As by-product this function converts non-linear &sk_buff to linear
 751 *	one, so that &sk_buff becomes completely private and caller is allowed
 752 *	to modify all the data of returned buffer. This means that this
 753 *	function is not recommended for use in circumstances when only
 754 *	header is going to be modified. Use pskb_copy() instead.
 755 */
 756
 757struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
 758{
 759	int headerlen = skb_headroom(skb);
 760	unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
 761	struct sk_buff *n = alloc_skb(size, gfp_mask);
 762
 763	if (!n)
 764		return NULL;
 765
 766	/* Set the data pointer */
 767	skb_reserve(n, headerlen);
 768	/* Set the tail pointer and length */
 769	skb_put(n, skb->len);
 770
 771	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
 772		BUG();
 773
 774	copy_skb_header(n, skb);
 775	return n;
 776}
 777EXPORT_SYMBOL(skb_copy);
 778
 779/**
 780 *	pskb_copy	-	create copy of an sk_buff with private head.
 781 *	@skb: buffer to copy
 782 *	@gfp_mask: allocation priority
 783 *
 784 *	Make a copy of both an &sk_buff and part of its data, located
 785 *	in header. Fragmented data remain shared. This is used when
 786 *	the caller wishes to modify only header of &sk_buff and needs
 787 *	private copy of the header to alter. Returns %NULL on failure
 788 *	or the pointer to the buffer on success.
 789 *	The returned buffer has a reference count of 1.
 790 */
 791
 792struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
 793{
 794	unsigned int size = skb_end_pointer(skb) - skb->head;
 795	struct sk_buff *n = alloc_skb(size, gfp_mask);
 796
 797	if (!n)
 798		goto out;
 799
 800	/* Set the data pointer */
 801	skb_reserve(n, skb_headroom(skb));
 802	/* Set the tail pointer and length */
 803	skb_put(n, skb_headlen(skb));
 804	/* Copy the bytes */
 805	skb_copy_from_linear_data(skb, n->data, n->len);
 806
 807	n->truesize += skb->data_len;
 808	n->data_len  = skb->data_len;
 809	n->len	     = skb->len;
 810
 811	if (skb_shinfo(skb)->nr_frags) {
 812		int i;
 813
 814		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 815			if (skb_copy_ubufs(skb, gfp_mask)) {
 816				kfree_skb(n);
 817				n = NULL;
 818				goto out;
 819			}
 820		}
 821		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 822			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
 823			get_page(skb_shinfo(n)->frags[i].page);
 824		}
 825		skb_shinfo(n)->nr_frags = i;
 826	}
 827
 828	if (skb_has_frag_list(skb)) {
 829		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
 830		skb_clone_fraglist(n);
 831	}
 832
 833	copy_skb_header(n, skb);
 834out:
 835	return n;
 836}
 837EXPORT_SYMBOL(pskb_copy);
 838
 839/**
 840 *	pskb_expand_head - reallocate header of &sk_buff
 841 *	@skb: buffer to reallocate
 842 *	@nhead: room to add at head
 843 *	@ntail: room to add at tail
 844 *	@gfp_mask: allocation priority
 845 *
 846 *	Expands (or creates identical copy, if &nhead and &ntail are zero)
 847 *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
 848 *	reference count of 1. Returns zero in the case of success or error,
 849 *	if expansion failed. In the last case, &sk_buff is not changed.
 850 *
 851 *	All the pointers pointing into skb header may change and must be
 852 *	reloaded after call to this function.
 853 */
 854
 855int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
 856		     gfp_t gfp_mask)
 857{
 858	int i;
 859	u8 *data;
 860	int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
 861	long off;
 862	bool fastpath;
 863
 864	BUG_ON(nhead < 0);
 865
 866	if (skb_shared(skb))
 867		BUG();
 868
 869	size = SKB_DATA_ALIGN(size);
 870
 871	/* Check if we can avoid taking references on fragments if we own
 872	 * the last reference on skb->head. (see skb_release_data())
 873	 */
 874	if (!skb->cloned)
 875		fastpath = true;
 876	else {
 877		int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
 878		fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
 879	}
 880
 881	if (fastpath &&
 882	    size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
 883		memmove(skb->head + size, skb_shinfo(skb),
 884			offsetof(struct skb_shared_info,
 885				 frags[skb_shinfo(skb)->nr_frags]));
 886		memmove(skb->head + nhead, skb->head,
 887			skb_tail_pointer(skb) - skb->head);
 888		off = nhead;
 889		goto adjust_others;
 890	}
 891
 892	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
 893	if (!data)
 894		goto nodata;
 895
 896	/* Copy only real data... and, alas, header. This should be
 897	 * optimized for the cases when header is void.
 898	 */
 899	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
 900
 901	memcpy((struct skb_shared_info *)(data + size),
 902	       skb_shinfo(skb),
 903	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
 904
 905	if (fastpath) {
 906		kfree(skb->head);
 907	} else {
 908		/* copy this zero copy skb frags */
 909		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 910			if (skb_copy_ubufs(skb, gfp_mask))
 911				goto nofrags;
 912		}
 913		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 914			get_page(skb_shinfo(skb)->frags[i].page);
 915
 916		if (skb_has_frag_list(skb))
 917			skb_clone_fraglist(skb);
 918
 919		skb_release_data(skb);
 920	}
 921	off = (data + nhead) - skb->head;
 922
 923	skb->head     = data;
 924adjust_others:
 925	skb->data    += off;
 926#ifdef NET_SKBUFF_DATA_USES_OFFSET
 927	skb->end      = size;
 928	off           = nhead;
 929#else
 930	skb->end      = skb->head + size;
 931#endif
 932	/* {transport,network,mac}_header and tail are relative to skb->head */
 933	skb->tail	      += off;
 934	skb->transport_header += off;
 935	skb->network_header   += off;
 936	if (skb_mac_header_was_set(skb))
 937		skb->mac_header += off;
 938	/* Only adjust this if it actually is csum_start rather than csum */
 939	if (skb->ip_summed == CHECKSUM_PARTIAL)
 940		skb->csum_start += nhead;
 941	skb->cloned   = 0;
 942	skb->hdr_len  = 0;
 943	skb->nohdr    = 0;
 944	atomic_set(&skb_shinfo(skb)->dataref, 1);
 945	return 0;
 946
 947nofrags:
 948	kfree(data);
 949nodata:
 950	return -ENOMEM;
 951}
 952EXPORT_SYMBOL(pskb_expand_head);
 953
 954/* Make private copy of skb with writable head and some headroom */
 955
 956struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
 957{
 958	struct sk_buff *skb2;
 959	int delta = headroom - skb_headroom(skb);
 960
 961	if (delta <= 0)
 962		skb2 = pskb_copy(skb, GFP_ATOMIC);
 963	else {
 964		skb2 = skb_clone(skb, GFP_ATOMIC);
 965		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
 966					     GFP_ATOMIC)) {
 967			kfree_skb(skb2);
 968			skb2 = NULL;
 969		}
 970	}
 971	return skb2;
 972}
 973EXPORT_SYMBOL(skb_realloc_headroom);
 974
 975/**
 976 *	skb_copy_expand	-	copy and expand sk_buff
 977 *	@skb: buffer to copy
 978 *	@newheadroom: new free bytes at head
 979 *	@newtailroom: new free bytes at tail
 980 *	@gfp_mask: allocation priority
 981 *
 982 *	Make a copy of both an &sk_buff and its data and while doing so
 983 *	allocate additional space.
 984 *
 985 *	This is used when the caller wishes to modify the data and needs a
 986 *	private copy of the data to alter as well as more space for new fields.
 987 *	Returns %NULL on failure or the pointer to the buffer
 988 *	on success. The returned buffer has a reference count of 1.
 989 *
 990 *	You must pass %GFP_ATOMIC as the allocation priority if this function
 991 *	is called from an interrupt.
 992 */
 993struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
 994				int newheadroom, int newtailroom,
 995				gfp_t gfp_mask)
 996{
 997	/*
 998	 *	Allocate the copy buffer
 999	 */
1000	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1001				      gfp_mask);
1002	int oldheadroom = skb_headroom(skb);
1003	int head_copy_len, head_copy_off;
1004	int off;
1005
1006	if (!n)
1007		return NULL;
1008
1009	skb_reserve(n, newheadroom);
1010
1011	/* Set the tail pointer and length */
1012	skb_put(n, skb->len);
1013
1014	head_copy_len = oldheadroom;
1015	head_copy_off = 0;
1016	if (newheadroom <= head_copy_len)
1017		head_copy_len = newheadroom;
1018	else
1019		head_copy_off = newheadroom - head_copy_len;
1020
1021	/* Copy the linear header and data. */
1022	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1023			  skb->len + head_copy_len))
1024		BUG();
1025
1026	copy_skb_header(n, skb);
1027
1028	off                  = newheadroom - oldheadroom;
1029	if (n->ip_summed == CHECKSUM_PARTIAL)
1030		n->csum_start += off;
1031#ifdef NET_SKBUFF_DATA_USES_OFFSET
1032	n->transport_header += off;
1033	n->network_header   += off;
1034	if (skb_mac_header_was_set(skb))
1035		n->mac_header += off;
1036#endif
1037
1038	return n;
1039}
1040EXPORT_SYMBOL(skb_copy_expand);
1041
1042/**
1043 *	skb_pad			-	zero pad the tail of an skb
1044 *	@skb: buffer to pad
1045 *	@pad: space to pad
1046 *
1047 *	Ensure that a buffer is followed by a padding area that is zero
1048 *	filled. Used by network drivers which may DMA or transfer data
1049 *	beyond the buffer end onto the wire.
1050 *
1051 *	May return error in out of memory cases. The skb is freed on error.
1052 */
1053
1054int skb_pad(struct sk_buff *skb, int pad)
1055{
1056	int err;
1057	int ntail;
1058
1059	/* If the skbuff is non linear tailroom is always zero.. */
1060	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1061		memset(skb->data+skb->len, 0, pad);
1062		return 0;
1063	}
1064
1065	ntail = skb->data_len + pad - (skb->end - skb->tail);
1066	if (likely(skb_cloned(skb) || ntail > 0)) {
1067		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1068		if (unlikely(err))
1069			goto free_skb;
1070	}
1071
1072	/* FIXME: The use of this function with non-linear skb's really needs
1073	 * to be audited.
1074	 */
1075	err = skb_linearize(skb);
1076	if (unlikely(err))
1077		goto free_skb;
1078
1079	memset(skb->data + skb->len, 0, pad);
1080	return 0;
1081
1082free_skb:
1083	kfree_skb(skb);
1084	return err;
1085}
1086EXPORT_SYMBOL(skb_pad);
1087
1088/**
1089 *	skb_put - add data to a buffer
1090 *	@skb: buffer to use
1091 *	@len: amount of data to add
1092 *
1093 *	This function extends the used data area of the buffer. If this would
1094 *	exceed the total buffer size the kernel will panic. A pointer to the
1095 *	first byte of the extra data is returned.
1096 */
1097unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1098{
1099	unsigned char *tmp = skb_tail_pointer(skb);
1100	SKB_LINEAR_ASSERT(skb);
1101	skb->tail += len;
1102	skb->len  += len;
1103	if (unlikely(skb->tail > skb->end))
1104		skb_over_panic(skb, len, __builtin_return_address(0));
1105	return tmp;
1106}
1107EXPORT_SYMBOL(skb_put);
1108
1109/**
1110 *	skb_push - add data to the start of a buffer
1111 *	@skb: buffer to use
1112 *	@len: amount of data to add
1113 *
1114 *	This function extends the used data area of the buffer at the buffer
1115 *	start. If this would exceed the total buffer headroom the kernel will
1116 *	panic. A pointer to the first byte of the extra data is returned.
1117 */
1118unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1119{
1120	skb->data -= len;
1121	skb->len  += len;
1122	if (unlikely(skb->data<skb->head))
1123		skb_under_panic(skb, len, __builtin_return_address(0));
1124	return skb->data;
1125}
1126EXPORT_SYMBOL(skb_push);
1127
1128/**
1129 *	skb_pull - remove data from the start of a buffer
1130 *	@skb: buffer to use
1131 *	@len: amount of data to remove
1132 *
1133 *	This function removes data from the start of a buffer, returning
1134 *	the memory to the headroom. A pointer to the next data in the buffer
1135 *	is returned. Once the data has been pulled future pushes will overwrite
1136 *	the old data.
1137 */
1138unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1139{
1140	return skb_pull_inline(skb, len);
1141}
1142EXPORT_SYMBOL(skb_pull);
1143
1144/**
1145 *	skb_trim - remove end from a buffer
1146 *	@skb: buffer to alter
1147 *	@len: new length
1148 *
1149 *	Cut the length of a buffer down by removing data from the tail. If
1150 *	the buffer is already under the length specified it is not modified.
1151 *	The skb must be linear.
1152 */
1153void skb_trim(struct sk_buff *skb, unsigned int len)
1154{
1155	if (skb->len > len)
1156		__skb_trim(skb, len);
1157}
1158EXPORT_SYMBOL(skb_trim);
1159
1160/* Trims skb to length len. It can change skb pointers.
1161 */
1162
1163int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1164{
1165	struct sk_buff **fragp;
1166	struct sk_buff *frag;
1167	int offset = skb_headlen(skb);
1168	int nfrags = skb_shinfo(skb)->nr_frags;
1169	int i;
1170	int err;
1171
1172	if (skb_cloned(skb) &&
1173	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1174		return err;
1175
1176	i = 0;
1177	if (offset >= len)
1178		goto drop_pages;
1179
1180	for (; i < nfrags; i++) {
1181		int end = offset + skb_shinfo(skb)->frags[i].size;
1182
1183		if (end < len) {
1184			offset = end;
1185			continue;
1186		}
1187
1188		skb_shinfo(skb)->frags[i++].size = len - offset;
1189
1190drop_pages:
1191		skb_shinfo(skb)->nr_frags = i;
1192
1193		for (; i < nfrags; i++)
1194			put_page(skb_shinfo(skb)->frags[i].page);
1195
1196		if (skb_has_frag_list(skb))
1197			skb_drop_fraglist(skb);
1198		goto done;
1199	}
1200
1201	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1202	     fragp = &frag->next) {
1203		int end = offset + frag->len;
1204
1205		if (skb_shared(frag)) {
1206			struct sk_buff *nfrag;
1207
1208			nfrag = skb_clone(frag, GFP_ATOMIC);
1209			if (unlikely(!nfrag))
1210				return -ENOMEM;
1211
1212			nfrag->next = frag->next;
1213			kfree_skb(frag);
1214			frag = nfrag;
1215			*fragp = frag;
1216		}
1217
1218		if (end < len) {
1219			offset = end;
1220			continue;
1221		}
1222
1223		if (end > len &&
1224		    unlikely((err = pskb_trim(frag, len - offset))))
1225			return err;
1226
1227		if (frag->next)
1228			skb_drop_list(&frag->next);
1229		break;
1230	}
1231
1232done:
1233	if (len > skb_headlen(skb)) {
1234		skb->data_len -= skb->len - len;
1235		skb->len       = len;
1236	} else {
1237		skb->len       = len;
1238		skb->data_len  = 0;
1239		skb_set_tail_pointer(skb, len);
1240	}
1241
1242	return 0;
1243}
1244EXPORT_SYMBOL(___pskb_trim);
1245
1246/**
1247 *	__pskb_pull_tail - advance tail of skb header
1248 *	@skb: buffer to reallocate
1249 *	@delta: number of bytes to advance tail
1250 *
1251 *	The function makes a sense only on a fragmented &sk_buff,
1252 *	it expands header moving its tail forward and copying necessary
1253 *	data from fragmented part.
1254 *
1255 *	&sk_buff MUST have reference count of 1.
1256 *
1257 *	Returns %NULL (and &sk_buff does not change) if pull failed
1258 *	or value of new tail of skb in the case of success.
1259 *
1260 *	All the pointers pointing into skb header may change and must be
1261 *	reloaded after call to this function.
1262 */
1263
1264/* Moves tail of skb head forward, copying data from fragmented part,
1265 * when it is necessary.
1266 * 1. It may fail due to malloc failure.
1267 * 2. It may change skb pointers.
1268 *
1269 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1270 */
1271unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1272{
1273	/* If skb has not enough free space at tail, get new one
1274	 * plus 128 bytes for future expansions. If we have enough
1275	 * room at tail, reallocate without expansion only if skb is cloned.
1276	 */
1277	int i, k, eat = (skb->tail + delta) - skb->end;
1278
1279	if (eat > 0 || skb_cloned(skb)) {
1280		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1281				     GFP_ATOMIC))
1282			return NULL;
1283	}
1284
1285	if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1286		BUG();
1287
1288	/* Optimization: no fragments, no reasons to preestimate
1289	 * size of pulled pages. Superb.
1290	 */
1291	if (!skb_has_frag_list(skb))
1292		goto pull_pages;
1293
1294	/* Estimate size of pulled pages. */
1295	eat = delta;
1296	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1297		if (skb_shinfo(skb)->frags[i].size >= eat)
1298			goto pull_pages;
1299		eat -= skb_shinfo(skb)->frags[i].size;
1300	}
1301
1302	/* If we need update frag list, we are in troubles.
1303	 * Certainly, it possible to add an offset to skb data,
1304	 * but taking into account that pulling is expected to
1305	 * be very rare operation, it is worth to fight against
1306	 * further bloating skb head and crucify ourselves here instead.
1307	 * Pure masohism, indeed. 8)8)
1308	 */
1309	if (eat) {
1310		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311		struct sk_buff *clone = NULL;
1312		struct sk_buff *insp = NULL;
1313
1314		do {
1315			BUG_ON(!list);
1316
1317			if (list->len <= eat) {
1318				/* Eaten as whole. */
1319				eat -= list->len;
1320				list = list->next;
1321				insp = list;
1322			} else {
1323				/* Eaten partially. */
1324
1325				if (skb_shared(list)) {
1326					/* Sucks! We need to fork list. :-( */
1327					clone = skb_clone(list, GFP_ATOMIC);
1328					if (!clone)
1329						return NULL;
1330					insp = list->next;
1331					list = clone;
1332				} else {
1333					/* This may be pulled without
1334					 * problems. */
1335					insp = list;
1336				}
1337				if (!pskb_pull(list, eat)) {
1338					kfree_skb(clone);
1339					return NULL;
1340				}
1341				break;
1342			}
1343		} while (eat);
1344
1345		/* Free pulled out fragments. */
1346		while ((list = skb_shinfo(skb)->frag_list) != insp) {
1347			skb_shinfo(skb)->frag_list = list->next;
1348			kfree_skb(list);
1349		}
1350		/* And insert new clone at head. */
1351		if (clone) {
1352			clone->next = list;
1353			skb_shinfo(skb)->frag_list = clone;
1354		}
1355	}
1356	/* Success! Now we may commit changes to skb data. */
1357
1358pull_pages:
1359	eat = delta;
1360	k = 0;
1361	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362		if (skb_shinfo(skb)->frags[i].size <= eat) {
1363			put_page(skb_shinfo(skb)->frags[i].page);
1364			eat -= skb_shinfo(skb)->frags[i].size;
1365		} else {
1366			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1367			if (eat) {
1368				skb_shinfo(skb)->frags[k].page_offset += eat;
1369				skb_shinfo(skb)->frags[k].size -= eat;
1370				eat = 0;
1371			}
1372			k++;
1373		}
1374	}
1375	skb_shinfo(skb)->nr_frags = k;
1376
1377	skb->tail     += delta;
1378	skb->data_len -= delta;
1379
1380	return skb_tail_pointer(skb);
1381}
1382EXPORT_SYMBOL(__pskb_pull_tail);
1383
1384/**
1385 *	skb_copy_bits - copy bits from skb to kernel buffer
1386 *	@skb: source skb
1387 *	@offset: offset in source
1388 *	@to: destination buffer
1389 *	@len: number of bytes to copy
1390 *
1391 *	Copy the specified number of bytes from the source skb to the
1392 *	destination buffer.
1393 *
1394 *	CAUTION ! :
1395 *		If its prototype is ever changed,
1396 *		check arch/{*}/net/{*}.S files,
1397 *		since it is called from BPF assembly code.
1398 */
1399int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1400{
1401	int start = skb_headlen(skb);
1402	struct sk_buff *frag_iter;
1403	int i, copy;
1404
1405	if (offset > (int)skb->len - len)
1406		goto fault;
1407
1408	/* Copy header. */
1409	if ((copy = start - offset) > 0) {
1410		if (copy > len)
1411			copy = len;
1412		skb_copy_from_linear_data_offset(skb, offset, to, copy);
1413		if ((len -= copy) == 0)
1414			return 0;
1415		offset += copy;
1416		to     += copy;
1417	}
1418
1419	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1420		int end;
1421
1422		WARN_ON(start > offset + len);
1423
1424		end = start + skb_shinfo(skb)->frags[i].size;
1425		if ((copy = end - offset) > 0) {
1426			u8 *vaddr;
1427
1428			if (copy > len)
1429				copy = len;
1430
1431			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1432			memcpy(to,
1433			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
1434			       offset - start, copy);
1435			kunmap_skb_frag(vaddr);
1436
1437			if ((len -= copy) == 0)
1438				return 0;
1439			offset += copy;
1440			to     += copy;
1441		}
1442		start = end;
1443	}
1444
1445	skb_walk_frags(skb, frag_iter) {
1446		int end;
1447
1448		WARN_ON(start > offset + len);
1449
1450		end = start + frag_iter->len;
1451		if ((copy = end - offset) > 0) {
1452			if (copy > len)
1453				copy = len;
1454			if (skb_copy_bits(frag_iter, offset - start, to, copy))
1455				goto fault;
1456			if ((len -= copy) == 0)
1457				return 0;
1458			offset += copy;
1459			to     += copy;
1460		}
1461		start = end;
1462	}
1463
1464	if (!len)
1465		return 0;
1466
1467fault:
1468	return -EFAULT;
1469}
1470EXPORT_SYMBOL(skb_copy_bits);
1471
1472/*
1473 * Callback from splice_to_pipe(), if we need to release some pages
1474 * at the end of the spd in case we error'ed out in filling the pipe.
1475 */
1476static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1477{
1478	put_page(spd->pages[i]);
1479}
1480
1481static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1482					  unsigned int *offset,
1483					  struct sk_buff *skb, struct sock *sk)
1484{
1485	struct page *p = sk->sk_sndmsg_page;
1486	unsigned int off;
1487
1488	if (!p) {
1489new_page:
1490		p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1491		if (!p)
1492			return NULL;
1493
1494		off = sk->sk_sndmsg_off = 0;
1495		/* hold one ref to this page until it's full */
1496	} else {
1497		unsigned int mlen;
1498
1499		off = sk->sk_sndmsg_off;
1500		mlen = PAGE_SIZE - off;
1501		if (mlen < 64 && mlen < *len) {
1502			put_page(p);
1503			goto new_page;
1504		}
1505
1506		*len = min_t(unsigned int, *len, mlen);
1507	}
1508
1509	memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1510	sk->sk_sndmsg_off += *len;
1511	*offset = off;
1512	get_page(p);
1513
1514	return p;
1515}
1516
1517/*
1518 * Fill page/offset/length into spd, if it can hold more pages.
1519 */
1520static inline int spd_fill_page(struct splice_pipe_desc *spd,
1521				struct pipe_inode_info *pipe, struct page *page,
1522				unsigned int *len, unsigned int offset,
1523				struct sk_buff *skb, int linear,
1524				struct sock *sk)
1525{
1526	if (unlikely(spd->nr_pages == pipe->buffers))
1527		return 1;
1528
1529	if (linear) {
1530		page = linear_to_page(page, len, &offset, skb, sk);
1531		if (!page)
1532			return 1;
1533	} else
1534		get_page(page);
1535
1536	spd->pages[spd->nr_pages] = page;
1537	spd->partial[spd->nr_pages].len = *len;
1538	spd->partial[spd->nr_pages].offset = offset;
1539	spd->nr_pages++;
1540
1541	return 0;
1542}
1543
1544static inline void __segment_seek(struct page **page, unsigned int *poff,
1545				  unsigned int *plen, unsigned int off)
1546{
1547	unsigned long n;
1548
1549	*poff += off;
1550	n = *poff / PAGE_SIZE;
1551	if (n)
1552		*page = nth_page(*page, n);
1553
1554	*poff = *poff % PAGE_SIZE;
1555	*plen -= off;
1556}
1557
1558static inline int __splice_segment(struct page *page, unsigned int poff,
1559				   unsigned int plen, unsigned int *off,
1560				   unsigned int *len, struct sk_buff *skb,
1561				   struct splice_pipe_desc *spd, int linear,
1562				   struct sock *sk,
1563				   struct pipe_inode_info *pipe)
1564{
1565	if (!*len)
1566		return 1;
1567
1568	/* skip this segment if already processed */
1569	if (*off >= plen) {
1570		*off -= plen;
1571		return 0;
1572	}
1573
1574	/* ignore any bits we already processed */
1575	if (*off) {
1576		__segment_seek(&page, &poff, &plen, *off);
1577		*off = 0;
1578	}
1579
1580	do {
1581		unsigned int flen = min(*len, plen);
1582
1583		/* the linear region may spread across several pages  */
1584		flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1585
1586		if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1587			return 1;
1588
1589		__segment_seek(&page, &poff, &plen, flen);
1590		*len -= flen;
1591
1592	} while (*len && plen);
1593
1594	return 0;
1595}
1596
1597/*
1598 * Map linear and fragment data from the skb to spd. It reports failure if the
1599 * pipe is full or if we already spliced the requested length.
1600 */
1601static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1602			     unsigned int *offset, unsigned int *len,
1603			     struct splice_pipe_desc *spd, struct sock *sk)
1604{
1605	int seg;
1606
1607	/*
1608	 * map the linear part
1609	 */
1610	if (__splice_segment(virt_to_page(skb->data),
1611			     (unsigned long) skb->data & (PAGE_SIZE - 1),
1612			     skb_headlen(skb),
1613			     offset, len, skb, spd, 1, sk, pipe))
1614		return 1;
1615
1616	/*
1617	 * then map the fragments
1618	 */
1619	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1620		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1621
1622		if (__splice_segment(f->page, f->page_offset, f->size,
1623				     offset, len, skb, spd, 0, sk, pipe))
1624			return 1;
1625	}
1626
1627	return 0;
1628}
1629
1630/*
1631 * Map data from the skb to a pipe. Should handle both the linear part,
1632 * the fragments, and the frag list. It does NOT handle frag lists within
1633 * the frag list, if such a thing exists. We'd probably need to recurse to
1634 * handle that cleanly.
1635 */
1636int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1637		    struct pipe_inode_info *pipe, unsigned int tlen,
1638		    unsigned int flags)
1639{
1640	struct partial_page partial[PIPE_DEF_BUFFERS];
1641	struct page *pages[PIPE_DEF_BUFFERS];
1642	struct splice_pipe_desc spd = {
1643		.pages = pages,
1644		.partial = partial,
1645		.flags = flags,
1646		.ops = &sock_pipe_buf_ops,
1647		.spd_release = sock_spd_release,
1648	};
1649	struct sk_buff *frag_iter;
1650	struct sock *sk = skb->sk;
1651	int ret = 0;
1652
1653	if (splice_grow_spd(pipe, &spd))
1654		return -ENOMEM;
1655
1656	/*
1657	 * __skb_splice_bits() only fails if the output has no room left,
1658	 * so no point in going over the frag_list for the error case.
1659	 */
1660	if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1661		goto done;
1662	else if (!tlen)
1663		goto done;
1664
1665	/*
1666	 * now see if we have a frag_list to map
1667	 */
1668	skb_walk_frags(skb, frag_iter) {
1669		if (!tlen)
1670			break;
1671		if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1672			break;
1673	}
1674
1675done:
1676	if (spd.nr_pages) {
1677		/*
1678		 * Drop the socket lock, otherwise we have reverse
1679		 * locking dependencies between sk_lock and i_mutex
1680		 * here as compared to sendfile(). We enter here
1681		 * with the socket lock held, and splice_to_pipe() will
1682		 * grab the pipe inode lock. For sendfile() emulation,
1683		 * we call into ->sendpage() with the i_mutex lock held
1684		 * and networking will grab the socket lock.
1685		 */
1686		release_sock(sk);
1687		ret = splice_to_pipe(pipe, &spd);
1688		lock_sock(sk);
1689	}
1690
1691	splice_shrink_spd(pipe, &spd);
1692	return ret;
1693}
1694
1695/**
1696 *	skb_store_bits - store bits from kernel buffer to skb
1697 *	@skb: destination buffer
1698 *	@offset: offset in destination
1699 *	@from: source buffer
1700 *	@len: number of bytes to copy
1701 *
1702 *	Copy the specified number of bytes from the source buffer to the
1703 *	destination skb.  This function handles all the messy bits of
1704 *	traversing fragment lists and such.
1705 */
1706
1707int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1708{
1709	int start = skb_headlen(skb);
1710	struct sk_buff *frag_iter;
1711	int i, copy;
1712
1713	if (offset > (int)skb->len - len)
1714		goto fault;
1715
1716	if ((copy = start - offset) > 0) {
1717		if (copy > len)
1718			copy = len;
1719		skb_copy_to_linear_data_offset(skb, offset, from, copy);
1720		if ((len -= copy) == 0)
1721			return 0;
1722		offset += copy;
1723		from += copy;
1724	}
1725
1726	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1728		int end;
1729
1730		WARN_ON(start > offset + len);
1731
1732		end = start + frag->size;
1733		if ((copy = end - offset) > 0) {
1734			u8 *vaddr;
1735
1736			if (copy > len)
1737				copy = len;
1738
1739			vaddr = kmap_skb_frag(frag);
1740			memcpy(vaddr + frag->page_offset + offset - start,
1741			       from, copy);
1742			kunmap_skb_frag(vaddr);
1743
1744			if ((len -= copy) == 0)
1745				return 0;
1746			offset += copy;
1747			from += copy;
1748		}
1749		start = end;
1750	}
1751
1752	skb_walk_frags(skb, frag_iter) {
1753		int end;
1754
1755		WARN_ON(start > offset + len);
1756
1757		end = start + frag_iter->len;
1758		if ((copy = end - offset) > 0) {
1759			if (copy > len)
1760				copy = len;
1761			if (skb_store_bits(frag_iter, offset - start,
1762					   from, copy))
1763				goto fault;
1764			if ((len -= copy) == 0)
1765				return 0;
1766			offset += copy;
1767			from += copy;
1768		}
1769		start = end;
1770	}
1771	if (!len)
1772		return 0;
1773
1774fault:
1775	return -EFAULT;
1776}
1777EXPORT_SYMBOL(skb_store_bits);
1778
1779/* Checksum skb data. */
1780
1781__wsum skb_checksum(const struct sk_buff *skb, int offset,
1782			  int len, __wsum csum)
1783{
1784	int start = skb_headlen(skb);
1785	int i, copy = start - offset;
1786	struct sk_buff *frag_iter;
1787	int pos = 0;
1788
1789	/* Checksum header. */
1790	if (copy > 0) {
1791		if (copy > len)
1792			copy = len;
1793		csum = csum_partial(skb->data + offset, copy, csum);
1794		if ((len -= copy) == 0)
1795			return csum;
1796		offset += copy;
1797		pos	= copy;
1798	}
1799
1800	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1801		int end;
1802
1803		WARN_ON(start > offset + len);
1804
1805		end = start + skb_shinfo(skb)->frags[i].size;
1806		if ((copy = end - offset) > 0) {
1807			__wsum csum2;
1808			u8 *vaddr;
1809			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1810
1811			if (copy > len)
1812				copy = len;
1813			vaddr = kmap_skb_frag(frag);
1814			csum2 = csum_partial(vaddr + frag->page_offset +
1815					     offset - start, copy, 0);
1816			kunmap_skb_frag(vaddr);
1817			csum = csum_block_add(csum, csum2, pos);
1818			if (!(len -= copy))
1819				return csum;
1820			offset += copy;
1821			pos    += copy;
1822		}
1823		start = end;
1824	}
1825
1826	skb_walk_frags(skb, frag_iter) {
1827		int end;
1828
1829		WARN_ON(start > offset + len);
1830
1831		end = start + frag_iter->len;
1832		if ((copy = end - offset) > 0) {
1833			__wsum csum2;
1834			if (copy > len)
1835				copy = len;
1836			csum2 = skb_checksum(frag_iter, offset - start,
1837					     copy, 0);
1838			csum = csum_block_add(csum, csum2, pos);
1839			if ((len -= copy) == 0)
1840				return csum;
1841			offset += copy;
1842			pos    += copy;
1843		}
1844		start = end;
1845	}
1846	BUG_ON(len);
1847
1848	return csum;
1849}
1850EXPORT_SYMBOL(skb_checksum);
1851
1852/* Both of above in one bottle. */
1853
1854__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1855				    u8 *to, int len, __wsum csum)
1856{
1857	int start = skb_headlen(skb);
1858	int i, copy = start - offset;
1859	struct sk_buff *frag_iter;
1860	int pos = 0;
1861
1862	/* Copy header. */
1863	if (copy > 0) {
1864		if (copy > len)
1865			copy = len;
1866		csum = csum_partial_copy_nocheck(skb->data + offset, to,
1867						 copy, csum);
1868		if ((len -= copy) == 0)
1869			return csum;
1870		offset += copy;
1871		to     += copy;
1872		pos	= copy;
1873	}
1874
1875	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876		int end;
1877
1878		WARN_ON(start > offset + len);
1879
1880		end = start + skb_shinfo(skb)->frags[i].size;
1881		if ((copy = end - offset) > 0) {
1882			__wsum csum2;
1883			u8 *vaddr;
1884			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1885
1886			if (copy > len)
1887				copy = len;
1888			vaddr = kmap_skb_frag(frag);
1889			csum2 = csum_partial_copy_nocheck(vaddr +
1890							  frag->page_offset +
1891							  offset - start, to,
1892							  copy, 0);
1893			kunmap_skb_frag(vaddr);
1894			csum = csum_block_add(csum, csum2, pos);
1895			if (!(len -= copy))
1896				return csum;
1897			offset += copy;
1898			to     += copy;
1899			pos    += copy;
1900		}
1901		start = end;
1902	}
1903
1904	skb_walk_frags(skb, frag_iter) {
1905		__wsum csum2;
1906		int end;
1907
1908		WARN_ON(start > offset + len);
1909
1910		end = start + frag_iter->len;
1911		if ((copy = end - offset) > 0) {
1912			if (copy > len)
1913				copy = len;
1914			csum2 = skb_copy_and_csum_bits(frag_iter,
1915						       offset - start,
1916						       to, copy, 0);
1917			csum = csum_block_add(csum, csum2, pos);
1918			if ((len -= copy) == 0)
1919				return csum;
1920			offset += copy;
1921			to     += copy;
1922			pos    += copy;
1923		}
1924		start = end;
1925	}
1926	BUG_ON(len);
1927	return csum;
1928}
1929EXPORT_SYMBOL(skb_copy_and_csum_bits);
1930
1931void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1932{
1933	__wsum csum;
1934	long csstart;
1935
1936	if (skb->ip_summed == CHECKSUM_PARTIAL)
1937		csstart = skb_checksum_start_offset(skb);
1938	else
1939		csstart = skb_headlen(skb);
1940
1941	BUG_ON(csstart > skb_headlen(skb));
1942
1943	skb_copy_from_linear_data(skb, to, csstart);
1944
1945	csum = 0;
1946	if (csstart != skb->len)
1947		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1948					      skb->len - csstart, 0);
1949
1950	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1951		long csstuff = csstart + skb->csum_offset;
1952
1953		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
1954	}
1955}
1956EXPORT_SYMBOL(skb_copy_and_csum_dev);
1957
1958/**
1959 *	skb_dequeue - remove from the head of the queue
1960 *	@list: list to dequeue from
1961 *
1962 *	Remove the head of the list. The list lock is taken so the function
1963 *	may be used safely with other locking list functions. The head item is
1964 *	returned or %NULL if the list is empty.
1965 */
1966
1967struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1968{
1969	unsigned long flags;
1970	struct sk_buff *result;
1971
1972	spin_lock_irqsave(&list->lock, flags);
1973	result = __skb_dequeue(list);
1974	spin_unlock_irqrestore(&list->lock, flags);
1975	return result;
1976}
1977EXPORT_SYMBOL(skb_dequeue);
1978
1979/**
1980 *	skb_dequeue_tail - remove from the tail of the queue
1981 *	@list: list to dequeue from
1982 *
1983 *	Remove the tail of the list. The list lock is taken so the function
1984 *	may be used safely with other locking list functions. The tail item is
1985 *	returned or %NULL if the list is empty.
1986 */
1987struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1988{
1989	unsigned long flags;
1990	struct sk_buff *result;
1991
1992	spin_lock_irqsave(&list->lock, flags);
1993	result = __skb_dequeue_tail(list);
1994	spin_unlock_irqrestore(&list->lock, flags);
1995	return result;
1996}
1997EXPORT_SYMBOL(skb_dequeue_tail);
1998
1999/**
2000 *	skb_queue_purge - empty a list
2001 *	@list: list to empty
2002 *
2003 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 *	the list and one reference dropped. This function takes the list
2005 *	lock and is atomic with respect to other list locking functions.
2006 */
2007void skb_queue_purge(struct sk_buff_head *list)
2008{
2009	struct sk_buff *skb;
2010	while ((skb = skb_dequeue(list)) != NULL)
2011		kfree_skb(skb);
2012}
2013EXPORT_SYMBOL(skb_queue_purge);
2014
2015/**
2016 *	skb_queue_head - queue a buffer at the list head
2017 *	@list: list to use
2018 *	@newsk: buffer to queue
2019 *
2020 *	Queue a buffer at the start of the list. This function takes the
2021 *	list lock and can be used safely with other locking &sk_buff functions
2022 *	safely.
2023 *
2024 *	A buffer cannot be placed on two lists at the same time.
2025 */
2026void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2027{
2028	unsigned long flags;
2029
2030	spin_lock_irqsave(&list->lock, flags);
2031	__skb_queue_head(list, newsk);
2032	spin_unlock_irqrestore(&list->lock, flags);
2033}
2034EXPORT_SYMBOL(skb_queue_head);
2035
2036/**
2037 *	skb_queue_tail - queue a buffer at the list tail
2038 *	@list: list to use
2039 *	@newsk: buffer to queue
2040 *
2041 *	Queue a buffer at the tail of the list. This function takes the
2042 *	list lock and can be used safely with other locking &sk_buff functions
2043 *	safely.
2044 *
2045 *	A buffer cannot be placed on two lists at the same time.
2046 */
2047void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2048{
2049	unsigned long flags;
2050
2051	spin_lock_irqsave(&list->lock, flags);
2052	__skb_queue_tail(list, newsk);
2053	spin_unlock_irqrestore(&list->lock, flags);
2054}
2055EXPORT_SYMBOL(skb_queue_tail);
2056
2057/**
2058 *	skb_unlink	-	remove a buffer from a list
2059 *	@skb: buffer to remove
2060 *	@list: list to use
2061 *
2062 *	Remove a packet from a list. The list locks are taken and this
2063 *	function is atomic with respect to other list locked calls
2064 *
2065 *	You must know what list the SKB is on.
2066 */
2067void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2068{
2069	unsigned long flags;
2070
2071	spin_lock_irqsave(&list->lock, flags);
2072	__skb_unlink(skb, list);
2073	spin_unlock_irqrestore(&list->lock, flags);
2074}
2075EXPORT_SYMBOL(skb_unlink);
2076
2077/**
2078 *	skb_append	-	append a buffer
2079 *	@old: buffer to insert after
2080 *	@newsk: buffer to insert
2081 *	@list: list to use
2082 *
2083 *	Place a packet after a given packet in a list. The list locks are taken
2084 *	and this function is atomic with respect to other list locked calls.
2085 *	A buffer cannot be placed on two lists at the same time.
2086 */
2087void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2088{
2089	unsigned long flags;
2090
2091	spin_lock_irqsave(&list->lock, flags);
2092	__skb_queue_after(list, old, newsk);
2093	spin_unlock_irqrestore(&list->lock, flags);
2094}
2095EXPORT_SYMBOL(skb_append);
2096
2097/**
2098 *	skb_insert	-	insert a buffer
2099 *	@old: buffer to insert before
2100 *	@newsk: buffer to insert
2101 *	@list: list to use
2102 *
2103 *	Place a packet before a given packet in a list. The list locks are
2104 * 	taken and this function is atomic with respect to other list locked
2105 *	calls.
2106 *
2107 *	A buffer cannot be placed on two lists at the same time.
2108 */
2109void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2110{
2111	unsigned long flags;
2112
2113	spin_lock_irqsave(&list->lock, flags);
2114	__skb_insert(newsk, old->prev, old, list);
2115	spin_unlock_irqrestore(&list->lock, flags);
2116}
2117EXPORT_SYMBOL(skb_insert);
2118
2119static inline void skb_split_inside_header(struct sk_buff *skb,
2120					   struct sk_buff* skb1,
2121					   const u32 len, const int pos)
2122{
2123	int i;
2124
2125	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2126					 pos - len);
2127	/* And move data appendix as is. */
2128	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2129		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2130
2131	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2132	skb_shinfo(skb)->nr_frags  = 0;
2133	skb1->data_len		   = skb->data_len;
2134	skb1->len		   += skb1->data_len;
2135	skb->data_len		   = 0;
2136	skb->len		   = len;
2137	skb_set_tail_pointer(skb, len);
2138}
2139
2140static inline void skb_split_no_header(struct sk_buff *skb,
2141				       struct sk_buff* skb1,
2142				       const u32 len, int pos)
2143{
2144	int i, k = 0;
2145	const int nfrags = skb_shinfo(skb)->nr_frags;
2146
2147	skb_shinfo(skb)->nr_frags = 0;
2148	skb1->len		  = skb1->data_len = skb->len - len;
2149	skb->len		  = len;
2150	skb->data_len		  = len - pos;
2151
2152	for (i = 0; i < nfrags; i++) {
2153		int size = skb_shinfo(skb)->frags[i].size;
2154
2155		if (pos + size > len) {
2156			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2157
2158			if (pos < len) {
2159				/* Split frag.
2160				 * We have two variants in this case:
2161				 * 1. Move all the frag to the second
2162				 *    part, if it is possible. F.e.
2163				 *    this approach is mandatory for TUX,
2164				 *    where splitting is expensive.
2165				 * 2. Split is accurately. We make this.
2166				 */
2167				get_page(skb_shinfo(skb)->frags[i].page);
2168				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2169				skb_shinfo(skb1)->frags[0].size -= len - pos;
2170				skb_shinfo(skb)->frags[i].size	= len - pos;
2171				skb_shinfo(skb)->nr_frags++;
2172			}
2173			k++;
2174		} else
2175			skb_shinfo(skb)->nr_frags++;
2176		pos += size;
2177	}
2178	skb_shinfo(skb1)->nr_frags = k;
2179}
2180
2181/**
2182 * skb_split - Split fragmented skb to two parts at length len.
2183 * @skb: the buffer to split
2184 * @skb1: the buffer to receive the second part
2185 * @len: new length for skb
2186 */
2187void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2188{
2189	int pos = skb_headlen(skb);
2190
2191	if (len < pos)	/* Split line is inside header. */
2192		skb_split_inside_header(skb, skb1, len, pos);
2193	else		/* Second chunk has no header, nothing to copy. */
2194		skb_split_no_header(skb, skb1, len, pos);
2195}
2196EXPORT_SYMBOL(skb_split);
2197
2198/* Shifting from/to a cloned skb is a no-go.
2199 *
2200 * Caller cannot keep skb_shinfo related pointers past calling here!
2201 */
2202static int skb_prepare_for_shift(struct sk_buff *skb)
2203{
2204	return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2205}
2206
2207/**
2208 * skb_shift - Shifts paged data partially from skb to another
2209 * @tgt: buffer into which tail data gets added
2210 * @skb: buffer from which the paged data comes from
2211 * @shiftlen: shift up to this many bytes
2212 *
2213 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2214 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2215 * It's up to caller to free skb if everything was shifted.
2216 *
2217 * If @tgt runs out of frags, the whole operation is aborted.
2218 *
2219 * Skb cannot include anything else but paged data while tgt is allowed
2220 * to have non-paged data as well.
2221 *
2222 * TODO: full sized shift could be optimized but that would need
2223 * specialized skb free'er to handle frags without up-to-date nr_frags.
2224 */
2225int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2226{
2227	int from, to, merge, todo;
2228	struct skb_frag_struct *fragfrom, *fragto;
2229
2230	BUG_ON(shiftlen > skb->len);
2231	BUG_ON(skb_headlen(skb));	/* Would corrupt stream */
2232
2233	todo = shiftlen;
2234	from = 0;
2235	to = skb_shinfo(tgt)->nr_frags;
2236	fragfrom = &skb_shinfo(skb)->frags[from];
2237
2238	/* Actual merge is delayed until the point when we know we can
2239	 * commit all, so that we don't have to undo partial changes
2240	 */
2241	if (!to ||
2242	    !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2243		merge = -1;
2244	} else {
2245		merge = to - 1;
2246
2247		todo -= fragfrom->size;
2248		if (todo < 0) {
2249			if (skb_prepare_for_shift(skb) ||
2250			    skb_prepare_for_shift(tgt))
2251				return 0;
2252
2253			/* All previous frag pointers might be stale! */
2254			fragfrom = &skb_shinfo(skb)->frags[from];
2255			fragto = &skb_shinfo(tgt)->frags[merge];
2256
2257			fragto->size += shiftlen;
2258			fragfrom->size -= shiftlen;
2259			fragfrom->page_offset += shiftlen;
2260
2261			goto onlymerged;
2262		}
2263
2264		from++;
2265	}
2266
2267	/* Skip full, not-fitting skb to avoid expensive operations */
2268	if ((shiftlen == skb->len) &&
2269	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2270		return 0;
2271
2272	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2273		return 0;
2274
2275	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2276		if (to == MAX_SKB_FRAGS)
2277			return 0;
2278
2279		fragfrom = &skb_shinfo(skb)->frags[from];
2280		fragto = &skb_shinfo(tgt)->frags[to];
2281
2282		if (todo >= fragfrom->size) {
2283			*fragto = *fragfrom;
2284			todo -= fragfrom->size;
2285			from++;
2286			to++;
2287
2288		} else {
2289			get_page(fragfrom->page);
2290			fragto->page = fragfrom->page;
2291			fragto->page_offset = fragfrom->page_offset;
2292			fragto->size = todo;
2293
2294			fragfrom->page_offset += todo;
2295			fragfrom->size -= todo;
2296			todo = 0;
2297
2298			to++;
2299			break;
2300		}
2301	}
2302
2303	/* Ready to "commit" this state change to tgt */
2304	skb_shinfo(tgt)->nr_frags = to;
2305
2306	if (merge >= 0) {
2307		fragfrom = &skb_shinfo(skb)->frags[0];
2308		fragto = &skb_shinfo(tgt)->frags[merge];
2309
2310		fragto->size += fragfrom->size;
2311		put_page(fragfrom->page);
2312	}
2313
2314	/* Reposition in the original skb */
2315	to = 0;
2316	while (from < skb_shinfo(skb)->nr_frags)
2317		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2318	skb_shinfo(skb)->nr_frags = to;
2319
2320	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2321
2322onlymerged:
2323	/* Most likely the tgt won't ever need its checksum anymore, skb on
2324	 * the other hand might need it if it needs to be resent
2325	 */
2326	tgt->ip_summed = CHECKSUM_PARTIAL;
2327	skb->ip_summed = CHECKSUM_PARTIAL;
2328
2329	/* Yak, is it really working this way? Some helper please? */
2330	skb->len -= shiftlen;
2331	skb->data_len -= shiftlen;
2332	skb->truesize -= shiftlen;
2333	tgt->len += shiftlen;
2334	tgt->data_len += shiftlen;
2335	tgt->truesize += shiftlen;
2336
2337	return shiftlen;
2338}
2339
2340/**
2341 * skb_prepare_seq_read - Prepare a sequential read of skb data
2342 * @skb: the buffer to read
2343 * @from: lower offset of data to be read
2344 * @to: upper offset of data to be read
2345 * @st: state variable
2346 *
2347 * Initializes the specified state variable. Must be called before
2348 * invoking skb_seq_read() for the first time.
2349 */
2350void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2351			  unsigned int to, struct skb_seq_state *st)
2352{
2353	st->lower_offset = from;
2354	st->upper_offset = to;
2355	st->root_skb = st->cur_skb = skb;
2356	st->frag_idx = st->stepped_offset = 0;
2357	st->frag_data = NULL;
2358}
2359EXPORT_SYMBOL(skb_prepare_seq_read);
2360
2361/**
2362 * skb_seq_read - Sequentially read skb data
2363 * @consumed: number of bytes consumed by the caller so far
2364 * @data: destination pointer for data to be returned
2365 * @st: state variable
2366 *
2367 * Reads a block of skb data at &consumed relative to the
2368 * lower offset specified to skb_prepare_seq_read(). Assigns
2369 * the head of the data block to &data and returns the length
2370 * of the block or 0 if the end of the skb data or the upper
2371 * offset has been reached.
2372 *
2373 * The caller is not required to consume all of the data
2374 * returned, i.e. &consumed is typically set to the number
2375 * of bytes already consumed and the next call to
2376 * skb_seq_read() will return the remaining part of the block.
2377 *
2378 * Note 1: The size of each block of data returned can be arbitrary,
2379 *       this limitation is the cost for zerocopy seqeuental
2380 *       reads of potentially non linear data.
2381 *
2382 * Note 2: Fragment lists within fragments are not implemented
2383 *       at the moment, state->root_skb could be replaced with
2384 *       a stack for this purpose.
2385 */
2386unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2387			  struct skb_seq_state *st)
2388{
2389	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2390	skb_frag_t *frag;
2391
2392	if (unlikely(abs_offset >= st->upper_offset))
2393		return 0;
2394
2395next_skb:
2396	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2397
2398	if (abs_offset < block_limit && !st->frag_data) {
2399		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2400		return block_limit - abs_offset;
2401	}
2402
2403	if (st->frag_idx == 0 && !st->frag_data)
2404		st->stepped_offset += skb_headlen(st->cur_skb);
2405
2406	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2407		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2408		block_limit = frag->size + st->stepped_offset;
2409
2410		if (abs_offset < block_limit) {
2411			if (!st->frag_data)
2412				st->frag_data = kmap_skb_frag(frag);
2413
2414			*data = (u8 *) st->frag_data + frag->page_offset +
2415				(abs_offset - st->stepped_offset);
2416
2417			return block_limit - abs_offset;
2418		}
2419
2420		if (st->frag_data) {
2421			kunmap_skb_frag(st->frag_data);
2422			st->frag_data = NULL;
2423		}
2424
2425		st->frag_idx++;
2426		st->stepped_offset += frag->size;
2427	}
2428
2429	if (st->frag_data) {
2430		kunmap_skb_frag(st->frag_data);
2431		st->frag_data = NULL;
2432	}
2433
2434	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2435		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2436		st->frag_idx = 0;
2437		goto next_skb;
2438	} else if (st->cur_skb->next) {
2439		st->cur_skb = st->cur_skb->next;
2440		st->frag_idx = 0;
2441		goto next_skb;
2442	}
2443
2444	return 0;
2445}
2446EXPORT_SYMBOL(skb_seq_read);
2447
2448/**
2449 * skb_abort_seq_read - Abort a sequential read of skb data
2450 * @st: state variable
2451 *
2452 * Must be called if skb_seq_read() was not called until it
2453 * returned 0.
2454 */
2455void skb_abort_seq_read(struct skb_seq_state *st)
2456{
2457	if (st->frag_data)
2458		kunmap_skb_frag(st->frag_data);
2459}
2460EXPORT_SYMBOL(skb_abort_seq_read);
2461
2462#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
2463
2464static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2465					  struct ts_config *conf,
2466					  struct ts_state *state)
2467{
2468	return skb_seq_read(offset, text, TS_SKB_CB(state));
2469}
2470
2471static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2472{
2473	skb_abort_seq_read(TS_SKB_CB(state));
2474}
2475
2476/**
2477 * skb_find_text - Find a text pattern in skb data
2478 * @skb: the buffer to look in
2479 * @from: search offset
2480 * @to: search limit
2481 * @config: textsearch configuration
2482 * @state: uninitialized textsearch state variable
2483 *
2484 * Finds a pattern in the skb data according to the specified
2485 * textsearch configuration. Use textsearch_next() to retrieve
2486 * subsequent occurrences of the pattern. Returns the offset
2487 * to the first occurrence or UINT_MAX if no match was found.
2488 */
2489unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2490			   unsigned int to, struct ts_config *config,
2491			   struct ts_state *state)
2492{
2493	unsigned int ret;
2494
2495	config->get_next_block = skb_ts_get_next_block;
2496	config->finish = skb_ts_finish;
2497
2498	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2499
2500	ret = textsearch_find(config, state);
2501	return (ret <= to - from ? ret : UINT_MAX);
2502}
2503EXPORT_SYMBOL(skb_find_text);
2504
2505/**
2506 * skb_append_datato_frags: - append the user data to a skb
2507 * @sk: sock  structure
2508 * @skb: skb structure to be appened with user data.
2509 * @getfrag: call back function to be used for getting the user data
2510 * @from: pointer to user message iov
2511 * @length: length of the iov message
2512 *
2513 * Description: This procedure append the user data in the fragment part
2514 * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2515 */
2516int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2517			int (*getfrag)(void *from, char *to, int offset,
2518					int len, int odd, struct sk_buff *skb),
2519			void *from, int length)
2520{
2521	int frg_cnt = 0;
2522	skb_frag_t *frag = NULL;
2523	struct page *page = NULL;
2524	int copy, left;
2525	int offset = 0;
2526	int ret;
2527
2528	do {
2529		/* Return error if we don't have space for new frag */
2530		frg_cnt = skb_shinfo(skb)->nr_frags;
2531		if (frg_cnt >= MAX_SKB_FRAGS)
2532			return -EFAULT;
2533
2534		/* allocate a new page for next frag */
2535		page = alloc_pages(sk->sk_allocation, 0);
2536
2537		/* If alloc_page fails just return failure and caller will
2538		 * free previous allocated pages by doing kfree_skb()
2539		 */
2540		if (page == NULL)
2541			return -ENOMEM;
2542
2543		/* initialize the next frag */
2544		skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2545		skb->truesize += PAGE_SIZE;
2546		atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2547
2548		/* get the new initialized frag */
2549		frg_cnt = skb_shinfo(skb)->nr_frags;
2550		frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2551
2552		/* copy the user data to page */
2553		left = PAGE_SIZE - frag->page_offset;
2554		copy = (length > left)? left : length;
2555
2556		ret = getfrag(from, (page_address(frag->page) +
2557			    frag->page_offset + frag->size),
2558			    offset, copy, 0, skb);
2559		if (ret < 0)
2560			return -EFAULT;
2561
2562		/* copy was successful so update the size parameters */
2563		frag->size += copy;
2564		skb->len += copy;
2565		skb->data_len += copy;
2566		offset += copy;
2567		length -= copy;
2568
2569	} while (length > 0);
2570
2571	return 0;
2572}
2573EXPORT_SYMBOL(skb_append_datato_frags);
2574
2575/**
2576 *	skb_pull_rcsum - pull skb and update receive checksum
2577 *	@skb: buffer to update
2578 *	@len: length of data pulled
2579 *
2580 *	This function performs an skb_pull on the packet and updates
2581 *	the CHECKSUM_COMPLETE checksum.  It should be used on
2582 *	receive path processing instead of skb_pull unless you know
2583 *	that the checksum difference is zero (e.g., a valid IP header)
2584 *	or you are setting ip_summed to CHECKSUM_NONE.
2585 */
2586unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2587{
2588	BUG_ON(len > skb->len);
2589	skb->len -= len;
2590	BUG_ON(skb->len < skb->data_len);
2591	skb_postpull_rcsum(skb, skb->data, len);
2592	return skb->data += len;
2593}
2594EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2595
2596/**
2597 *	skb_segment - Perform protocol segmentation on skb.
2598 *	@skb: buffer to segment
2599 *	@features: features for the output path (see dev->features)
2600 *
2601 *	This function performs segmentation on the given skb.  It returns
2602 *	a pointer to the first in a list of new skbs for the segments.
2603 *	In case of error it returns ERR_PTR(err).
2604 */
2605struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2606{
2607	struct sk_buff *segs = NULL;
2608	struct sk_buff *tail = NULL;
2609	struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2610	unsigned int mss = skb_shinfo(skb)->gso_size;
2611	unsigned int doffset = skb->data - skb_mac_header(skb);
2612	unsigned int offset = doffset;
2613	unsigned int headroom;
2614	unsigned int len;
2615	int sg = !!(features & NETIF_F_SG);
2616	int nfrags = skb_shinfo(skb)->nr_frags;
2617	int err = -ENOMEM;
2618	int i = 0;
2619	int pos;
2620
2621	__skb_push(skb, doffset);
2622	headroom = skb_headroom(skb);
2623	pos = skb_headlen(skb);
2624
2625	do {
2626		struct sk_buff *nskb;
2627		skb_frag_t *frag;
2628		int hsize;
2629		int size;
2630
2631		len = skb->len - offset;
2632		if (len > mss)
2633			len = mss;
2634
2635		hsize = skb_headlen(skb) - offset;
2636		if (hsize < 0)
2637			hsize = 0;
2638		if (hsize > len || !sg)
2639			hsize = len;
2640
2641		if (!hsize && i >= nfrags) {
2642			BUG_ON(fskb->len != len);
2643
2644			pos += len;
2645			nskb = skb_clone(fskb, GFP_ATOMIC);
2646			fskb = fskb->next;
2647
2648			if (unlikely(!nskb))
2649				goto err;
2650
2651			hsize = skb_end_pointer(nskb) - nskb->head;
2652			if (skb_cow_head(nskb, doffset + headroom)) {
2653				kfree_skb(nskb);
2654				goto err;
2655			}
2656
2657			nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2658					  hsize;
2659			skb_release_head_state(nskb);
2660			__skb_push(nskb, doffset);
2661		} else {
2662			nskb = alloc_skb(hsize + doffset + headroom,
2663					 GFP_ATOMIC);
2664
2665			if (unlikely(!nskb))
2666				goto err;
2667
2668			skb_reserve(nskb, headroom);
2669			__skb_put(nskb, doffset);
2670		}
2671
2672		if (segs)
2673			tail->next = nskb;
2674		else
2675			segs = nskb;
2676		tail = nskb;
2677
2678		__copy_skb_header(nskb, skb);
2679		nskb->mac_len = skb->mac_len;
2680
2681		/* nskb and skb might have different headroom */
2682		if (nskb->ip_summed == CHECKSUM_PARTIAL)
2683			nskb->csum_start += skb_headroom(nskb) - headroom;
2684
2685		skb_reset_mac_header(nskb);
2686		skb_set_network_header(nskb, skb->mac_len);
2687		nskb->transport_header = (nskb->network_header +
2688					  skb_network_header_len(skb));
2689		skb_copy_from_linear_data(skb, nskb->data, doffset);
2690
2691		if (fskb != skb_shinfo(skb)->frag_list)
2692			continue;
2693
2694		if (!sg) {
2695			nskb->ip_summed = CHECKSUM_NONE;
2696			nskb->csum = skb_copy_and_csum_bits(skb, offset,
2697							    skb_put(nskb, len),
2698							    len, 0);
2699			continue;
2700		}
2701
2702		frag = skb_shinfo(nskb)->frags;
2703
2704		skb_copy_from_linear_data_offset(skb, offset,
2705						 skb_put(nskb, hsize), hsize);
2706
2707		while (pos < offset + len && i < nfrags) {
2708			*frag = skb_shinfo(skb)->frags[i];
2709			get_page(frag->page);
2710			size = frag->size;
2711
2712			if (pos < offset) {
2713				frag->page_offset += offset - pos;
2714				frag->size -= offset - pos;
2715			}
2716
2717			skb_shinfo(nskb)->nr_frags++;
2718
2719			if (pos + size <= offset + len) {
2720				i++;
2721				pos += size;
2722			} else {
2723				frag->size -= pos + size - (offset + len);
2724				goto skip_fraglist;
2725			}
2726
2727			frag++;
2728		}
2729
2730		if (pos < offset + len) {
2731			struct sk_buff *fskb2 = fskb;
2732
2733			BUG_ON(pos + fskb->len != offset + len);
2734
2735			pos += fskb->len;
2736			fskb = fskb->next;
2737
2738			if (fskb2->next) {
2739				fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2740				if (!fskb2)
2741					goto err;
2742			} else
2743				skb_get(fskb2);
2744
2745			SKB_FRAG_ASSERT(nskb);
2746			skb_shinfo(nskb)->frag_list = fskb2;
2747		}
2748
2749skip_fraglist:
2750		nskb->data_len = len - hsize;
2751		nskb->len += nskb->data_len;
2752		nskb->truesize += nskb->data_len;
2753	} while ((offset += len) < skb->len);
2754
2755	return segs;
2756
2757err:
2758	while ((skb = segs)) {
2759		segs = skb->next;
2760		kfree_skb(skb);
2761	}
2762	return ERR_PTR(err);
2763}
2764EXPORT_SYMBOL_GPL(skb_segment);
2765
2766int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2767{
2768	struct sk_buff *p = *head;
2769	struct sk_buff *nskb;
2770	struct skb_shared_info *skbinfo = skb_shinfo(skb);
2771	struct skb_shared_info *pinfo = skb_shinfo(p);
2772	unsigned int headroom;
2773	unsigned int len = skb_gro_len(skb);
2774	unsigned int offset = skb_gro_offset(skb);
2775	unsigned int headlen = skb_headlen(skb);
2776
2777	if (p->len + len >= 65536)
2778		return -E2BIG;
2779
2780	if (pinfo->frag_list)
2781		goto merge;
2782	else if (headlen <= offset) {
2783		skb_frag_t *frag;
2784		skb_frag_t *frag2;
2785		int i = skbinfo->nr_frags;
2786		int nr_frags = pinfo->nr_frags + i;
2787
2788		offset -= headlen;
2789
2790		if (nr_frags > MAX_SKB_FRAGS)
2791			return -E2BIG;
2792
2793		pinfo->nr_frags = nr_frags;
2794		skbinfo->nr_frags = 0;
2795
2796		frag = pinfo->frags + nr_frags;
2797		frag2 = skbinfo->frags + i;
2798		do {
2799			*--frag = *--frag2;
2800		} while (--i);
2801
2802		frag->page_offset += offset;
2803		frag->size -= offset;
2804
2805		skb->truesize -= skb->data_len;
2806		skb->len -= skb->data_len;
2807		skb->data_len = 0;
2808
2809		NAPI_GRO_CB(skb)->free = 1;
2810		goto done;
2811	} else if (skb_gro_len(p) != pinfo->gso_size)
2812		return -E2BIG;
2813
2814	headroom = skb_headroom(p);
2815	nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2816	if (unlikely(!nskb))
2817		return -ENOMEM;
2818
2819	__copy_skb_header(nskb, p);
2820	nskb->mac_len = p->mac_len;
2821
2822	skb_reserve(nskb, headroom);
2823	__skb_put(nskb, skb_gro_offset(p));
2824
2825	skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2826	skb_set_network_header(nskb, skb_network_offset(p));
2827	skb_set_transport_header(nskb, skb_transport_offset(p));
2828
2829	__skb_pull(p, skb_gro_offset(p));
2830	memcpy(skb_mac_header(nskb), skb_mac_header(p),
2831	       p->data - skb_mac_header(p));
2832
2833	*NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2834	skb_shinfo(nskb)->frag_list = p;
2835	skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2836	pinfo->gso_size = 0;
2837	skb_header_release(p);
2838	nskb->prev = p;
2839
2840	nskb->data_len += p->len;
2841	nskb->truesize += p->len;
2842	nskb->len += p->len;
2843
2844	*head = nskb;
2845	nskb->next = p->next;
2846	p->next = NULL;
2847
2848	p = nskb;
2849
2850merge:
2851	if (offset > headlen) {
2852		unsigned int eat = offset - headlen;
2853
2854		skbinfo->frags[0].page_offset += eat;
2855		skbinfo->frags[0].size -= eat;
2856		skb->data_len -= eat;
2857		skb->len -= eat;
2858		offset = headlen;
2859	}
2860
2861	__skb_pull(skb, offset);
2862
2863	p->prev->next = skb;
2864	p->prev = skb;
2865	skb_header_release(skb);
2866
2867done:
2868	NAPI_GRO_CB(p)->count++;
2869	p->data_len += len;
2870	p->truesize += len;
2871	p->len += len;
2872
2873	NAPI_GRO_CB(skb)->same_flow = 1;
2874	return 0;
2875}
2876EXPORT_SYMBOL_GPL(skb_gro_receive);
2877
2878void __init skb_init(void)
2879{
2880	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2881					      sizeof(struct sk_buff),
2882					      0,
2883					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2884					      NULL);
2885	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2886						(2*sizeof(struct sk_buff)) +
2887						sizeof(atomic_t),
2888						0,
2889						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2890						NULL);
2891}
2892
2893/**
2894 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2895 *	@skb: Socket buffer containing the buffers to be mapped
2896 *	@sg: The scatter-gather list to map into
2897 *	@offset: The offset into the buffer's contents to start mapping
2898 *	@len: Length of buffer space to be mapped
2899 *
2900 *	Fill the specified scatter-gather list with mappings/pointers into a
2901 *	region of the buffer space attached to a socket buffer.
2902 */
2903static int
2904__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2905{
2906	int start = skb_headlen(skb);
2907	int i, copy = start - offset;
2908	struct sk_buff *frag_iter;
2909	int elt = 0;
2910
2911	if (copy > 0) {
2912		if (copy > len)
2913			copy = len;
2914		sg_set_buf(sg, skb->data + offset, copy);
2915		elt++;
2916		if ((len -= copy) == 0)
2917			return elt;
2918		offset += copy;
2919	}
2920
2921	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2922		int end;
2923
2924		WARN_ON(start > offset + len);
2925
2926		end = start + skb_shinfo(skb)->frags[i].size;
2927		if ((copy = end - offset) > 0) {
2928			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2929
2930			if (copy > len)
2931				copy = len;
2932			sg_set_page(&sg[elt], frag->page, copy,
2933					frag->page_offset+offset-start);
2934			elt++;
2935			if (!(len -= copy))
2936				return elt;
2937			offset += copy;
2938		}
2939		start = end;
2940	}
2941
2942	skb_walk_frags(skb, frag_iter) {
2943		int end;
2944
2945		WARN_ON(start > offset + len);
2946
2947		end = start + frag_iter->len;
2948		if ((copy = end - offset) > 0) {
2949			if (copy > len)
2950				copy = len;
2951			elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2952					      copy);
2953			if ((len -= copy) == 0)
2954				return elt;
2955			offset += copy;
2956		}
2957		start = end;
2958	}
2959	BUG_ON(len);
2960	return elt;
2961}
2962
2963int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2964{
2965	int nsg = __skb_to_sgvec(skb, sg, offset, len);
2966
2967	sg_mark_end(&sg[nsg - 1]);
2968
2969	return nsg;
2970}
2971EXPORT_SYMBOL_GPL(skb_to_sgvec);
2972
2973/**
2974 *	skb_cow_data - Check that a socket buffer's data buffers are writable
2975 *	@skb: The socket buffer to check.
2976 *	@tailbits: Amount of trailing space to be added
2977 *	@trailer: Returned pointer to the skb where the @tailbits space begins
2978 *
2979 *	Make sure that the data buffers attached to a socket buffer are
2980 *	writable. If they are not, private copies are made of the data buffers
2981 *	and the socket buffer is set to use these instead.
2982 *
2983 *	If @tailbits is given, make sure that there is space to write @tailbits
2984 *	bytes of data beyond current end of socket buffer.  @trailer will be
2985 *	set to point to the skb in which this space begins.
2986 *
2987 *	The number of scatterlist elements required to completely map the
2988 *	COW'd and extended socket buffer will be returned.
2989 */
2990int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2991{
2992	int copyflag;
2993	int elt;
2994	struct sk_buff *skb1, **skb_p;
2995
2996	/* If skb is cloned or its head is paged, reallocate
2997	 * head pulling out all the pages (pages are considered not writable
2998	 * at the moment even if they are anonymous).
2999	 */
3000	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3001	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3002		return -ENOMEM;
3003
3004	/* Easy case. Most of packets will go this way. */
3005	if (!skb_has_frag_list(skb)) {
3006		/* A little of trouble, not enough of space for trailer.
3007		 * This should not happen, when stack is tuned to generate
3008		 * good frames. OK, on miss we reallocate and reserve even more
3009		 * space, 128 bytes is fair. */
3010
3011		if (skb_tailroom(skb) < tailbits &&
3012		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3013			return -ENOMEM;
3014
3015		/* Voila! */
3016		*trailer = skb;
3017		return 1;
3018	}
3019
3020	/* Misery. We are in troubles, going to mincer fragments... */
3021
3022	elt = 1;
3023	skb_p = &skb_shinfo(skb)->frag_list;
3024	copyflag = 0;
3025
3026	while ((skb1 = *skb_p) != NULL) {
3027		int ntail = 0;
3028
3029		/* The fragment is partially pulled by someone,
3030		 * this can happen on input. Copy it and everything
3031		 * after it. */
3032
3033		if (skb_shared(skb1))
3034			copyflag = 1;
3035
3036		/* If the skb is the last, worry about trailer. */
3037
3038		if (skb1->next == NULL && tailbits) {
3039			if (skb_shinfo(skb1)->nr_frags ||
3040			    skb_has_frag_list(skb1) ||
3041			    skb_tailroom(skb1) < tailbits)
3042				ntail = tailbits + 128;
3043		}
3044
3045		if (copyflag ||
3046		    skb_cloned(skb1) ||
3047		    ntail ||
3048		    skb_shinfo(skb1)->nr_frags ||
3049		    skb_has_frag_list(skb1)) {
3050			struct sk_buff *skb2;
3051
3052			/* Fuck, we are miserable poor guys... */
3053			if (ntail == 0)
3054				skb2 = skb_copy(skb1, GFP_ATOMIC);
3055			else
3056				skb2 = skb_copy_expand(skb1,
3057						       skb_headroom(skb1),
3058						       ntail,
3059						       GFP_ATOMIC);
3060			if (unlikely(skb2 == NULL))
3061				return -ENOMEM;
3062
3063			if (skb1->sk)
3064				skb_set_owner_w(skb2, skb1->sk);
3065
3066			/* Looking around. Are we still alive?
3067			 * OK, link new skb, drop old one */
3068
3069			skb2->next = skb1->next;
3070			*skb_p = skb2;
3071			kfree_skb(skb1);
3072			skb1 = skb2;
3073		}
3074		elt++;
3075		*trailer = skb1;
3076		skb_p = &skb1->next;
3077	}
3078
3079	return elt;
3080}
3081EXPORT_SYMBOL_GPL(skb_cow_data);
3082
3083static void sock_rmem_free(struct sk_buff *skb)
3084{
3085	struct sock *sk = skb->sk;
3086
3087	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3088}
3089
3090/*
3091 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3092 */
3093int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3094{
3095	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3096	    (unsigned)sk->sk_rcvbuf)
3097		return -ENOMEM;
3098
3099	skb_orphan(skb);
3100	skb->sk = sk;
3101	skb->destructor = sock_rmem_free;
3102	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3103
3104	/* before exiting rcu section, make sure dst is refcounted */
3105	skb_dst_force(skb);
3106
3107	skb_queue_tail(&sk->sk_error_queue, skb);
3108	if (!sock_flag(sk, SOCK_DEAD))
3109		sk->sk_data_ready(sk, skb->len);
3110	return 0;
3111}
3112EXPORT_SYMBOL(sock_queue_err_skb);
3113
3114void skb_tstamp_tx(struct sk_buff *orig_skb,
3115		struct skb_shared_hwtstamps *hwtstamps)
3116{
3117	struct sock *sk = orig_skb->sk;
3118	struct sock_exterr_skb *serr;
3119	struct sk_buff *skb;
3120	int err;
3121
3122	if (!sk)
3123		return;
3124
3125	skb = skb_clone(orig_skb, GFP_ATOMIC);
3126	if (!skb)
3127		return;
3128
3129	if (hwtstamps) {
3130		*skb_hwtstamps(skb) =
3131			*hwtstamps;
3132	} else {
3133		/*
3134		 * no hardware time stamps available,
3135		 * so keep the shared tx_flags and only
3136		 * store software time stamp
3137		 */
3138		skb->tstamp = ktime_get_real();
3139	}
3140
3141	serr = SKB_EXT_ERR(skb);
3142	memset(serr, 0, sizeof(*serr));
3143	serr->ee.ee_errno = ENOMSG;
3144	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3145
3146	err = sock_queue_err_skb(sk, skb);
3147
3148	if (err)
3149		kfree_skb(skb);
3150}
3151EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3152
3153
3154/**
3155 * skb_partial_csum_set - set up and verify partial csum values for packet
3156 * @skb: the skb to set
3157 * @start: the number of bytes after skb->data to start checksumming.
3158 * @off: the offset from start to place the checksum.
3159 *
3160 * For untrusted partially-checksummed packets, we need to make sure the values
3161 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3162 *
3163 * This function checks and sets those values and skb->ip_summed: if this
3164 * returns false you should drop the packet.
3165 */
3166bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3167{
3168	if (unlikely(start > skb_headlen(skb)) ||
3169	    unlikely((int)start + off > skb_headlen(skb) - 2)) {
3170		if (net_ratelimit())
3171			printk(KERN_WARNING
3172			       "bad partial csum: csum=%u/%u len=%u\n",
3173			       start, off, skb_headlen(skb));
3174		return false;
3175	}
3176	skb->ip_summed = CHECKSUM_PARTIAL;
3177	skb->csum_start = skb_headroom(skb) + start;
3178	skb->csum_offset = off;
3179	return true;
3180}
3181EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3182
3183void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3184{
3185	if (net_ratelimit())
3186		pr_warning("%s: received packets cannot be forwarded"
3187			   " while LRO is enabled\n", skb->dev->name);
3188}
3189EXPORT_SYMBOL(__skb_warn_lro_forwarding);
v3.1
   1/*
   2 *	Routines having to do with the 'struct sk_buff' memory handlers.
   3 *
   4 *	Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   5 *			Florian La Roche <rzsfl@rz.uni-sb.de>
   6 *
   7 *	Fixes:
   8 *		Alan Cox	:	Fixed the worst of the load
   9 *					balancer bugs.
  10 *		Dave Platt	:	Interrupt stacking fix.
  11 *	Richard Kooijman	:	Timestamp fixes.
  12 *		Alan Cox	:	Changed buffer format.
  13 *		Alan Cox	:	destructor hook for AF_UNIX etc.
  14 *		Linus Torvalds	:	Better skb_clone.
  15 *		Alan Cox	:	Added skb_copy.
  16 *		Alan Cox	:	Added all the changed routines Linus
  17 *					only put in the headers
  18 *		Ray VanTassle	:	Fixed --skb->lock in free
  19 *		Alan Cox	:	skb_copy copy arp field
  20 *		Andi Kleen	:	slabified it.
  21 *		Robert Olsson	:	Removed skb_head_pool
  22 *
  23 *	NOTE:
  24 *		The __skb_ routines should be called with interrupts
  25 *	disabled, or you better be *real* sure that the operation is atomic
  26 *	with respect to whatever list is being frobbed (e.g. via lock_sock()
  27 *	or via disabling bottom half handlers, etc).
  28 *
  29 *	This program is free software; you can redistribute it and/or
  30 *	modify it under the terms of the GNU General Public License
  31 *	as published by the Free Software Foundation; either version
  32 *	2 of the License, or (at your option) any later version.
  33 */
  34
  35/*
  36 *	The functions in this file will not compile correctly with gcc 2.4.x
  37 */
  38
  39#include <linux/module.h>
  40#include <linux/types.h>
  41#include <linux/kernel.h>
  42#include <linux/kmemcheck.h>
  43#include <linux/mm.h>
  44#include <linux/interrupt.h>
  45#include <linux/in.h>
  46#include <linux/inet.h>
  47#include <linux/slab.h>
  48#include <linux/netdevice.h>
  49#ifdef CONFIG_NET_CLS_ACT
  50#include <net/pkt_sched.h>
  51#endif
  52#include <linux/string.h>
  53#include <linux/skbuff.h>
  54#include <linux/splice.h>
  55#include <linux/cache.h>
  56#include <linux/rtnetlink.h>
  57#include <linux/init.h>
  58#include <linux/scatterlist.h>
  59#include <linux/errqueue.h>
  60#include <linux/prefetch.h>
  61
  62#include <net/protocol.h>
  63#include <net/dst.h>
  64#include <net/sock.h>
  65#include <net/checksum.h>
  66#include <net/xfrm.h>
  67
  68#include <asm/uaccess.h>
  69#include <asm/system.h>
  70#include <trace/events/skb.h>
  71
  72#include "kmap_skb.h"
  73
  74static struct kmem_cache *skbuff_head_cache __read_mostly;
  75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
  76
  77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
  78				  struct pipe_buffer *buf)
  79{
  80	put_page(buf->page);
  81}
  82
  83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
  84				struct pipe_buffer *buf)
  85{
  86	get_page(buf->page);
  87}
  88
  89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
  90			       struct pipe_buffer *buf)
  91{
  92	return 1;
  93}
  94
  95
  96/* Pipe buffer operations for a socket. */
  97static const struct pipe_buf_operations sock_pipe_buf_ops = {
  98	.can_merge = 0,
  99	.map = generic_pipe_buf_map,
 100	.unmap = generic_pipe_buf_unmap,
 101	.confirm = generic_pipe_buf_confirm,
 102	.release = sock_pipe_buf_release,
 103	.steal = sock_pipe_buf_steal,
 104	.get = sock_pipe_buf_get,
 105};
 106
 107/*
 108 *	Keep out-of-line to prevent kernel bloat.
 109 *	__builtin_return_address is not used because it is not always
 110 *	reliable.
 111 */
 112
 113/**
 114 *	skb_over_panic	- 	private function
 115 *	@skb: buffer
 116 *	@sz: size
 117 *	@here: address
 118 *
 119 *	Out of line support code for skb_put(). Not user callable.
 120 */
 121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
 122{
 123	printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
 124			  "data:%p tail:%#lx end:%#lx dev:%s\n",
 125	       here, skb->len, sz, skb->head, skb->data,
 126	       (unsigned long)skb->tail, (unsigned long)skb->end,
 127	       skb->dev ? skb->dev->name : "<NULL>");
 128	BUG();
 129}
 130
 131/**
 132 *	skb_under_panic	- 	private function
 133 *	@skb: buffer
 134 *	@sz: size
 135 *	@here: address
 136 *
 137 *	Out of line support code for skb_push(). Not user callable.
 138 */
 139
 140static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
 141{
 142	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
 143			  "data:%p tail:%#lx end:%#lx dev:%s\n",
 144	       here, skb->len, sz, skb->head, skb->data,
 145	       (unsigned long)skb->tail, (unsigned long)skb->end,
 146	       skb->dev ? skb->dev->name : "<NULL>");
 147	BUG();
 148}
 149
 150/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 151 *	'private' fields and also do memory statistics to find all the
 152 *	[BEEP] leaks.
 153 *
 154 */
 155
 156/**
 157 *	__alloc_skb	-	allocate a network buffer
 158 *	@size: size to allocate
 159 *	@gfp_mask: allocation mask
 160 *	@fclone: allocate from fclone cache instead of head cache
 161 *		and allocate a cloned (child) skb
 162 *	@node: numa node to allocate memory on
 163 *
 164 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 165 *	tail room of size bytes. The object has a reference count of one.
 166 *	The return is the buffer. On a failure the return is %NULL.
 167 *
 168 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 169 *	%GFP_ATOMIC.
 170 */
 171struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 172			    int fclone, int node)
 173{
 174	struct kmem_cache *cache;
 175	struct skb_shared_info *shinfo;
 176	struct sk_buff *skb;
 177	u8 *data;
 178
 179	cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
 180
 181	/* Get the HEAD */
 182	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
 183	if (!skb)
 184		goto out;
 185	prefetchw(skb);
 186
 187	size = SKB_DATA_ALIGN(size);
 188	data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
 189			gfp_mask, node);
 190	if (!data)
 191		goto nodata;
 192	prefetchw(data + size);
 193
 194	/*
 195	 * Only clear those fields we need to clear, not those that we will
 196	 * actually initialise below. Hence, don't put any more fields after
 197	 * the tail pointer in struct sk_buff!
 198	 */
 199	memset(skb, 0, offsetof(struct sk_buff, tail));
 200	skb->truesize = size + sizeof(struct sk_buff);
 201	atomic_set(&skb->users, 1);
 202	skb->head = data;
 203	skb->data = data;
 204	skb_reset_tail_pointer(skb);
 205	skb->end = skb->tail + size;
 206#ifdef NET_SKBUFF_DATA_USES_OFFSET
 207	skb->mac_header = ~0U;
 208#endif
 209
 210	/* make sure we initialize shinfo sequentially */
 211	shinfo = skb_shinfo(skb);
 212	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 213	atomic_set(&shinfo->dataref, 1);
 214	kmemcheck_annotate_variable(shinfo->destructor_arg);
 215
 216	if (fclone) {
 217		struct sk_buff *child = skb + 1;
 218		atomic_t *fclone_ref = (atomic_t *) (child + 1);
 219
 220		kmemcheck_annotate_bitfield(child, flags1);
 221		kmemcheck_annotate_bitfield(child, flags2);
 222		skb->fclone = SKB_FCLONE_ORIG;
 223		atomic_set(fclone_ref, 1);
 224
 225		child->fclone = SKB_FCLONE_UNAVAILABLE;
 226	}
 227out:
 228	return skb;
 229nodata:
 230	kmem_cache_free(cache, skb);
 231	skb = NULL;
 232	goto out;
 233}
 234EXPORT_SYMBOL(__alloc_skb);
 235
 236/**
 237 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 238 *	@dev: network device to receive on
 239 *	@length: length to allocate
 240 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 241 *
 242 *	Allocate a new &sk_buff and assign it a usage count of one. The
 243 *	buffer has unspecified headroom built in. Users should allocate
 244 *	the headroom they think they need without accounting for the
 245 *	built in space. The built in space is used for optimisations.
 246 *
 247 *	%NULL is returned if there is no free memory.
 248 */
 249struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
 250		unsigned int length, gfp_t gfp_mask)
 251{
 252	struct sk_buff *skb;
 253
 254	skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
 255	if (likely(skb)) {
 256		skb_reserve(skb, NET_SKB_PAD);
 257		skb->dev = dev;
 258	}
 259	return skb;
 260}
 261EXPORT_SYMBOL(__netdev_alloc_skb);
 262
 263void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 264		int size)
 265{
 266	skb_fill_page_desc(skb, i, page, off, size);
 267	skb->len += size;
 268	skb->data_len += size;
 269	skb->truesize += size;
 270}
 271EXPORT_SYMBOL(skb_add_rx_frag);
 272
 273/**
 274 *	dev_alloc_skb - allocate an skbuff for receiving
 275 *	@length: length to allocate
 276 *
 277 *	Allocate a new &sk_buff and assign it a usage count of one. The
 278 *	buffer has unspecified headroom built in. Users should allocate
 279 *	the headroom they think they need without accounting for the
 280 *	built in space. The built in space is used for optimisations.
 281 *
 282 *	%NULL is returned if there is no free memory. Although this function
 283 *	allocates memory it can be called from an interrupt.
 284 */
 285struct sk_buff *dev_alloc_skb(unsigned int length)
 286{
 287	/*
 288	 * There is more code here than it seems:
 289	 * __dev_alloc_skb is an inline
 290	 */
 291	return __dev_alloc_skb(length, GFP_ATOMIC);
 292}
 293EXPORT_SYMBOL(dev_alloc_skb);
 294
 295static void skb_drop_list(struct sk_buff **listp)
 296{
 297	struct sk_buff *list = *listp;
 298
 299	*listp = NULL;
 300
 301	do {
 302		struct sk_buff *this = list;
 303		list = list->next;
 304		kfree_skb(this);
 305	} while (list);
 306}
 307
 308static inline void skb_drop_fraglist(struct sk_buff *skb)
 309{
 310	skb_drop_list(&skb_shinfo(skb)->frag_list);
 311}
 312
 313static void skb_clone_fraglist(struct sk_buff *skb)
 314{
 315	struct sk_buff *list;
 316
 317	skb_walk_frags(skb, list)
 318		skb_get(list);
 319}
 320
 321static void skb_release_data(struct sk_buff *skb)
 322{
 323	if (!skb->cloned ||
 324	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
 325			       &skb_shinfo(skb)->dataref)) {
 326		if (skb_shinfo(skb)->nr_frags) {
 327			int i;
 328			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 329				put_page(skb_shinfo(skb)->frags[i].page);
 330		}
 331
 332		/*
 333		 * If skb buf is from userspace, we need to notify the caller
 334		 * the lower device DMA has done;
 335		 */
 336		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 337			struct ubuf_info *uarg;
 338
 339			uarg = skb_shinfo(skb)->destructor_arg;
 340			if (uarg->callback)
 341				uarg->callback(uarg);
 342		}
 343
 344		if (skb_has_frag_list(skb))
 345			skb_drop_fraglist(skb);
 346
 347		kfree(skb->head);
 348	}
 349}
 350
 351/*
 352 *	Free an skbuff by memory without cleaning the state.
 353 */
 354static void kfree_skbmem(struct sk_buff *skb)
 355{
 356	struct sk_buff *other;
 357	atomic_t *fclone_ref;
 358
 359	switch (skb->fclone) {
 360	case SKB_FCLONE_UNAVAILABLE:
 361		kmem_cache_free(skbuff_head_cache, skb);
 362		break;
 363
 364	case SKB_FCLONE_ORIG:
 365		fclone_ref = (atomic_t *) (skb + 2);
 366		if (atomic_dec_and_test(fclone_ref))
 367			kmem_cache_free(skbuff_fclone_cache, skb);
 368		break;
 369
 370	case SKB_FCLONE_CLONE:
 371		fclone_ref = (atomic_t *) (skb + 1);
 372		other = skb - 1;
 373
 374		/* The clone portion is available for
 375		 * fast-cloning again.
 376		 */
 377		skb->fclone = SKB_FCLONE_UNAVAILABLE;
 378
 379		if (atomic_dec_and_test(fclone_ref))
 380			kmem_cache_free(skbuff_fclone_cache, other);
 381		break;
 382	}
 383}
 384
 385static void skb_release_head_state(struct sk_buff *skb)
 386{
 387	skb_dst_drop(skb);
 388#ifdef CONFIG_XFRM
 389	secpath_put(skb->sp);
 390#endif
 391	if (skb->destructor) {
 392		WARN_ON(in_irq());
 393		skb->destructor(skb);
 394	}
 395#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 396	nf_conntrack_put(skb->nfct);
 397#endif
 398#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
 399	nf_conntrack_put_reasm(skb->nfct_reasm);
 400#endif
 401#ifdef CONFIG_BRIDGE_NETFILTER
 402	nf_bridge_put(skb->nf_bridge);
 403#endif
 404/* XXX: IS this still necessary? - JHS */
 405#ifdef CONFIG_NET_SCHED
 406	skb->tc_index = 0;
 407#ifdef CONFIG_NET_CLS_ACT
 408	skb->tc_verd = 0;
 409#endif
 410#endif
 411}
 412
 413/* Free everything but the sk_buff shell. */
 414static void skb_release_all(struct sk_buff *skb)
 415{
 416	skb_release_head_state(skb);
 417	skb_release_data(skb);
 418}
 419
 420/**
 421 *	__kfree_skb - private function
 422 *	@skb: buffer
 423 *
 424 *	Free an sk_buff. Release anything attached to the buffer.
 425 *	Clean the state. This is an internal helper function. Users should
 426 *	always call kfree_skb
 427 */
 428
 429void __kfree_skb(struct sk_buff *skb)
 430{
 431	skb_release_all(skb);
 432	kfree_skbmem(skb);
 433}
 434EXPORT_SYMBOL(__kfree_skb);
 435
 436/**
 437 *	kfree_skb - free an sk_buff
 438 *	@skb: buffer to free
 439 *
 440 *	Drop a reference to the buffer and free it if the usage count has
 441 *	hit zero.
 442 */
 443void kfree_skb(struct sk_buff *skb)
 444{
 445	if (unlikely(!skb))
 446		return;
 447	if (likely(atomic_read(&skb->users) == 1))
 448		smp_rmb();
 449	else if (likely(!atomic_dec_and_test(&skb->users)))
 450		return;
 451	trace_kfree_skb(skb, __builtin_return_address(0));
 452	__kfree_skb(skb);
 453}
 454EXPORT_SYMBOL(kfree_skb);
 455
 456/**
 457 *	consume_skb - free an skbuff
 458 *	@skb: buffer to free
 459 *
 460 *	Drop a ref to the buffer and free it if the usage count has hit zero
 461 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
 462 *	is being dropped after a failure and notes that
 463 */
 464void consume_skb(struct sk_buff *skb)
 465{
 466	if (unlikely(!skb))
 467		return;
 468	if (likely(atomic_read(&skb->users) == 1))
 469		smp_rmb();
 470	else if (likely(!atomic_dec_and_test(&skb->users)))
 471		return;
 472	trace_consume_skb(skb);
 473	__kfree_skb(skb);
 474}
 475EXPORT_SYMBOL(consume_skb);
 476
 477/**
 478 *	skb_recycle_check - check if skb can be reused for receive
 479 *	@skb: buffer
 480 *	@skb_size: minimum receive buffer size
 481 *
 482 *	Checks that the skb passed in is not shared or cloned, and
 483 *	that it is linear and its head portion at least as large as
 484 *	skb_size so that it can be recycled as a receive buffer.
 485 *	If these conditions are met, this function does any necessary
 486 *	reference count dropping and cleans up the skbuff as if it
 487 *	just came from __alloc_skb().
 488 */
 489bool skb_recycle_check(struct sk_buff *skb, int skb_size)
 490{
 491	struct skb_shared_info *shinfo;
 492
 493	if (irqs_disabled())
 494		return false;
 495
 496	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
 497		return false;
 498
 499	if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
 500		return false;
 501
 502	skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
 503	if (skb_end_pointer(skb) - skb->head < skb_size)
 504		return false;
 505
 506	if (skb_shared(skb) || skb_cloned(skb))
 507		return false;
 508
 509	skb_release_head_state(skb);
 510
 511	shinfo = skb_shinfo(skb);
 512	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 513	atomic_set(&shinfo->dataref, 1);
 514
 515	memset(skb, 0, offsetof(struct sk_buff, tail));
 516	skb->data = skb->head + NET_SKB_PAD;
 517	skb_reset_tail_pointer(skb);
 518
 519	return true;
 520}
 521EXPORT_SYMBOL(skb_recycle_check);
 522
 523static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 524{
 525	new->tstamp		= old->tstamp;
 526	new->dev		= old->dev;
 527	new->transport_header	= old->transport_header;
 528	new->network_header	= old->network_header;
 529	new->mac_header		= old->mac_header;
 530	skb_dst_copy(new, old);
 531	new->rxhash		= old->rxhash;
 532#ifdef CONFIG_XFRM
 533	new->sp			= secpath_get(old->sp);
 534#endif
 535	memcpy(new->cb, old->cb, sizeof(old->cb));
 536	new->csum		= old->csum;
 537	new->local_df		= old->local_df;
 538	new->pkt_type		= old->pkt_type;
 539	new->ip_summed		= old->ip_summed;
 540	skb_copy_queue_mapping(new, old);
 541	new->priority		= old->priority;
 542#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 543	new->ipvs_property	= old->ipvs_property;
 544#endif
 545	new->protocol		= old->protocol;
 546	new->mark		= old->mark;
 547	new->skb_iif		= old->skb_iif;
 548	__nf_copy(new, old);
 549#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 550    defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 551	new->nf_trace		= old->nf_trace;
 552#endif
 553#ifdef CONFIG_NET_SCHED
 554	new->tc_index		= old->tc_index;
 555#ifdef CONFIG_NET_CLS_ACT
 556	new->tc_verd		= old->tc_verd;
 557#endif
 558#endif
 559	new->vlan_tci		= old->vlan_tci;
 560
 561	skb_copy_secmark(new, old);
 562}
 563
 564/*
 565 * You should not add any new code to this function.  Add it to
 566 * __copy_skb_header above instead.
 567 */
 568static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 569{
 570#define C(x) n->x = skb->x
 571
 572	n->next = n->prev = NULL;
 573	n->sk = NULL;
 574	__copy_skb_header(n, skb);
 575
 576	C(len);
 577	C(data_len);
 578	C(mac_len);
 579	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 580	n->cloned = 1;
 581	n->nohdr = 0;
 582	n->destructor = NULL;
 583	C(tail);
 584	C(end);
 585	C(head);
 586	C(data);
 587	C(truesize);
 588	atomic_set(&n->users, 1);
 589
 590	atomic_inc(&(skb_shinfo(skb)->dataref));
 591	skb->cloned = 1;
 592
 593	return n;
 594#undef C
 595}
 596
 597/**
 598 *	skb_morph	-	morph one skb into another
 599 *	@dst: the skb to receive the contents
 600 *	@src: the skb to supply the contents
 601 *
 602 *	This is identical to skb_clone except that the target skb is
 603 *	supplied by the user.
 604 *
 605 *	The target skb is returned upon exit.
 606 */
 607struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 608{
 609	skb_release_all(dst);
 610	return __skb_clone(dst, src);
 611}
 612EXPORT_SYMBOL_GPL(skb_morph);
 613
 614/*	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
 615 *	@skb: the skb to modify
 616 *	@gfp_mask: allocation priority
 617 *
 618 *	This must be called on SKBTX_DEV_ZEROCOPY skb.
 619 *	It will copy all frags into kernel and drop the reference
 620 *	to userspace pages.
 621 *
 622 *	If this function is called from an interrupt gfp_mask() must be
 623 *	%GFP_ATOMIC.
 624 *
 625 *	Returns 0 on success or a negative error code on failure
 626 *	to allocate kernel memory to copy to.
 627 */
 628int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
 629{
 630	int i;
 631	int num_frags = skb_shinfo(skb)->nr_frags;
 632	struct page *page, *head = NULL;
 633	struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
 634
 635	for (i = 0; i < num_frags; i++) {
 636		u8 *vaddr;
 637		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 638
 639		page = alloc_page(GFP_ATOMIC);
 640		if (!page) {
 641			while (head) {
 642				struct page *next = (struct page *)head->private;
 643				put_page(head);
 644				head = next;
 645			}
 646			return -ENOMEM;
 647		}
 648		vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
 649		memcpy(page_address(page),
 650		       vaddr + f->page_offset, f->size);
 651		kunmap_skb_frag(vaddr);
 652		page->private = (unsigned long)head;
 653		head = page;
 654	}
 655
 656	/* skb frags release userspace buffers */
 657	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 658		put_page(skb_shinfo(skb)->frags[i].page);
 659
 660	uarg->callback(uarg);
 661
 662	/* skb frags point to kernel buffers */
 663	for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
 664		skb_shinfo(skb)->frags[i - 1].page_offset = 0;
 665		skb_shinfo(skb)->frags[i - 1].page = head;
 666		head = (struct page *)head->private;
 667	}
 668
 669	skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
 670	return 0;
 671}
 672
 673
 674/**
 675 *	skb_clone	-	duplicate an sk_buff
 676 *	@skb: buffer to clone
 677 *	@gfp_mask: allocation priority
 678 *
 679 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
 680 *	copies share the same packet data but not structure. The new
 681 *	buffer has a reference count of 1. If the allocation fails the
 682 *	function returns %NULL otherwise the new buffer is returned.
 683 *
 684 *	If this function is called from an interrupt gfp_mask() must be
 685 *	%GFP_ATOMIC.
 686 */
 687
 688struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 689{
 690	struct sk_buff *n;
 691
 692	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 693		if (skb_copy_ubufs(skb, gfp_mask))
 694			return NULL;
 695	}
 696
 697	n = skb + 1;
 698	if (skb->fclone == SKB_FCLONE_ORIG &&
 699	    n->fclone == SKB_FCLONE_UNAVAILABLE) {
 700		atomic_t *fclone_ref = (atomic_t *) (n + 1);
 701		n->fclone = SKB_FCLONE_CLONE;
 702		atomic_inc(fclone_ref);
 703	} else {
 704		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
 705		if (!n)
 706			return NULL;
 707
 708		kmemcheck_annotate_bitfield(n, flags1);
 709		kmemcheck_annotate_bitfield(n, flags2);
 710		n->fclone = SKB_FCLONE_UNAVAILABLE;
 711	}
 712
 713	return __skb_clone(n, skb);
 714}
 715EXPORT_SYMBOL(skb_clone);
 716
 717static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 718{
 719#ifndef NET_SKBUFF_DATA_USES_OFFSET
 720	/*
 721	 *	Shift between the two data areas in bytes
 722	 */
 723	unsigned long offset = new->data - old->data;
 724#endif
 725
 726	__copy_skb_header(new, old);
 727
 728#ifndef NET_SKBUFF_DATA_USES_OFFSET
 729	/* {transport,network,mac}_header are relative to skb->head */
 730	new->transport_header += offset;
 731	new->network_header   += offset;
 732	if (skb_mac_header_was_set(new))
 733		new->mac_header	      += offset;
 734#endif
 735	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
 736	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
 737	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
 738}
 739
 740/**
 741 *	skb_copy	-	create private copy of an sk_buff
 742 *	@skb: buffer to copy
 743 *	@gfp_mask: allocation priority
 744 *
 745 *	Make a copy of both an &sk_buff and its data. This is used when the
 746 *	caller wishes to modify the data and needs a private copy of the
 747 *	data to alter. Returns %NULL on failure or the pointer to the buffer
 748 *	on success. The returned buffer has a reference count of 1.
 749 *
 750 *	As by-product this function converts non-linear &sk_buff to linear
 751 *	one, so that &sk_buff becomes completely private and caller is allowed
 752 *	to modify all the data of returned buffer. This means that this
 753 *	function is not recommended for use in circumstances when only
 754 *	header is going to be modified. Use pskb_copy() instead.
 755 */
 756
 757struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
 758{
 759	int headerlen = skb_headroom(skb);
 760	unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
 761	struct sk_buff *n = alloc_skb(size, gfp_mask);
 762
 763	if (!n)
 764		return NULL;
 765
 766	/* Set the data pointer */
 767	skb_reserve(n, headerlen);
 768	/* Set the tail pointer and length */
 769	skb_put(n, skb->len);
 770
 771	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
 772		BUG();
 773
 774	copy_skb_header(n, skb);
 775	return n;
 776}
 777EXPORT_SYMBOL(skb_copy);
 778
 779/**
 780 *	pskb_copy	-	create copy of an sk_buff with private head.
 781 *	@skb: buffer to copy
 782 *	@gfp_mask: allocation priority
 783 *
 784 *	Make a copy of both an &sk_buff and part of its data, located
 785 *	in header. Fragmented data remain shared. This is used when
 786 *	the caller wishes to modify only header of &sk_buff and needs
 787 *	private copy of the header to alter. Returns %NULL on failure
 788 *	or the pointer to the buffer on success.
 789 *	The returned buffer has a reference count of 1.
 790 */
 791
 792struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
 793{
 794	unsigned int size = skb_end_pointer(skb) - skb->head;
 795	struct sk_buff *n = alloc_skb(size, gfp_mask);
 796
 797	if (!n)
 798		goto out;
 799
 800	/* Set the data pointer */
 801	skb_reserve(n, skb_headroom(skb));
 802	/* Set the tail pointer and length */
 803	skb_put(n, skb_headlen(skb));
 804	/* Copy the bytes */
 805	skb_copy_from_linear_data(skb, n->data, n->len);
 806
 807	n->truesize += skb->data_len;
 808	n->data_len  = skb->data_len;
 809	n->len	     = skb->len;
 810
 811	if (skb_shinfo(skb)->nr_frags) {
 812		int i;
 813
 814		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 815			if (skb_copy_ubufs(skb, gfp_mask)) {
 816				kfree_skb(n);
 817				n = NULL;
 818				goto out;
 819			}
 820		}
 821		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 822			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
 823			get_page(skb_shinfo(n)->frags[i].page);
 824		}
 825		skb_shinfo(n)->nr_frags = i;
 826	}
 827
 828	if (skb_has_frag_list(skb)) {
 829		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
 830		skb_clone_fraglist(n);
 831	}
 832
 833	copy_skb_header(n, skb);
 834out:
 835	return n;
 836}
 837EXPORT_SYMBOL(pskb_copy);
 838
 839/**
 840 *	pskb_expand_head - reallocate header of &sk_buff
 841 *	@skb: buffer to reallocate
 842 *	@nhead: room to add at head
 843 *	@ntail: room to add at tail
 844 *	@gfp_mask: allocation priority
 845 *
 846 *	Expands (or creates identical copy, if &nhead and &ntail are zero)
 847 *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
 848 *	reference count of 1. Returns zero in the case of success or error,
 849 *	if expansion failed. In the last case, &sk_buff is not changed.
 850 *
 851 *	All the pointers pointing into skb header may change and must be
 852 *	reloaded after call to this function.
 853 */
 854
 855int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
 856		     gfp_t gfp_mask)
 857{
 858	int i;
 859	u8 *data;
 860	int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
 861	long off;
 862	bool fastpath;
 863
 864	BUG_ON(nhead < 0);
 865
 866	if (skb_shared(skb))
 867		BUG();
 868
 869	size = SKB_DATA_ALIGN(size);
 870
 871	/* Check if we can avoid taking references on fragments if we own
 872	 * the last reference on skb->head. (see skb_release_data())
 873	 */
 874	if (!skb->cloned)
 875		fastpath = true;
 876	else {
 877		int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
 878		fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
 879	}
 880
 881	if (fastpath &&
 882	    size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
 883		memmove(skb->head + size, skb_shinfo(skb),
 884			offsetof(struct skb_shared_info,
 885				 frags[skb_shinfo(skb)->nr_frags]));
 886		memmove(skb->head + nhead, skb->head,
 887			skb_tail_pointer(skb) - skb->head);
 888		off = nhead;
 889		goto adjust_others;
 890	}
 891
 892	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
 893	if (!data)
 894		goto nodata;
 895
 896	/* Copy only real data... and, alas, header. This should be
 897	 * optimized for the cases when header is void.
 898	 */
 899	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
 900
 901	memcpy((struct skb_shared_info *)(data + size),
 902	       skb_shinfo(skb),
 903	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
 904
 905	if (fastpath) {
 906		kfree(skb->head);
 907	} else {
 908		/* copy this zero copy skb frags */
 909		if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 910			if (skb_copy_ubufs(skb, gfp_mask))
 911				goto nofrags;
 912		}
 913		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 914			get_page(skb_shinfo(skb)->frags[i].page);
 915
 916		if (skb_has_frag_list(skb))
 917			skb_clone_fraglist(skb);
 918
 919		skb_release_data(skb);
 920	}
 921	off = (data + nhead) - skb->head;
 922
 923	skb->head     = data;
 924adjust_others:
 925	skb->data    += off;
 926#ifdef NET_SKBUFF_DATA_USES_OFFSET
 927	skb->end      = size;
 928	off           = nhead;
 929#else
 930	skb->end      = skb->head + size;
 931#endif
 932	/* {transport,network,mac}_header and tail are relative to skb->head */
 933	skb->tail	      += off;
 934	skb->transport_header += off;
 935	skb->network_header   += off;
 936	if (skb_mac_header_was_set(skb))
 937		skb->mac_header += off;
 938	/* Only adjust this if it actually is csum_start rather than csum */
 939	if (skb->ip_summed == CHECKSUM_PARTIAL)
 940		skb->csum_start += nhead;
 941	skb->cloned   = 0;
 942	skb->hdr_len  = 0;
 943	skb->nohdr    = 0;
 944	atomic_set(&skb_shinfo(skb)->dataref, 1);
 945	return 0;
 946
 947nofrags:
 948	kfree(data);
 949nodata:
 950	return -ENOMEM;
 951}
 952EXPORT_SYMBOL(pskb_expand_head);
 953
 954/* Make private copy of skb with writable head and some headroom */
 955
 956struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
 957{
 958	struct sk_buff *skb2;
 959	int delta = headroom - skb_headroom(skb);
 960
 961	if (delta <= 0)
 962		skb2 = pskb_copy(skb, GFP_ATOMIC);
 963	else {
 964		skb2 = skb_clone(skb, GFP_ATOMIC);
 965		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
 966					     GFP_ATOMIC)) {
 967			kfree_skb(skb2);
 968			skb2 = NULL;
 969		}
 970	}
 971	return skb2;
 972}
 973EXPORT_SYMBOL(skb_realloc_headroom);
 974
 975/**
 976 *	skb_copy_expand	-	copy and expand sk_buff
 977 *	@skb: buffer to copy
 978 *	@newheadroom: new free bytes at head
 979 *	@newtailroom: new free bytes at tail
 980 *	@gfp_mask: allocation priority
 981 *
 982 *	Make a copy of both an &sk_buff and its data and while doing so
 983 *	allocate additional space.
 984 *
 985 *	This is used when the caller wishes to modify the data and needs a
 986 *	private copy of the data to alter as well as more space for new fields.
 987 *	Returns %NULL on failure or the pointer to the buffer
 988 *	on success. The returned buffer has a reference count of 1.
 989 *
 990 *	You must pass %GFP_ATOMIC as the allocation priority if this function
 991 *	is called from an interrupt.
 992 */
 993struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
 994				int newheadroom, int newtailroom,
 995				gfp_t gfp_mask)
 996{
 997	/*
 998	 *	Allocate the copy buffer
 999	 */
1000	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1001				      gfp_mask);
1002	int oldheadroom = skb_headroom(skb);
1003	int head_copy_len, head_copy_off;
1004	int off;
1005
1006	if (!n)
1007		return NULL;
1008
1009	skb_reserve(n, newheadroom);
1010
1011	/* Set the tail pointer and length */
1012	skb_put(n, skb->len);
1013
1014	head_copy_len = oldheadroom;
1015	head_copy_off = 0;
1016	if (newheadroom <= head_copy_len)
1017		head_copy_len = newheadroom;
1018	else
1019		head_copy_off = newheadroom - head_copy_len;
1020
1021	/* Copy the linear header and data. */
1022	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1023			  skb->len + head_copy_len))
1024		BUG();
1025
1026	copy_skb_header(n, skb);
1027
1028	off                  = newheadroom - oldheadroom;
1029	if (n->ip_summed == CHECKSUM_PARTIAL)
1030		n->csum_start += off;
1031#ifdef NET_SKBUFF_DATA_USES_OFFSET
1032	n->transport_header += off;
1033	n->network_header   += off;
1034	if (skb_mac_header_was_set(skb))
1035		n->mac_header += off;
1036#endif
1037
1038	return n;
1039}
1040EXPORT_SYMBOL(skb_copy_expand);
1041
1042/**
1043 *	skb_pad			-	zero pad the tail of an skb
1044 *	@skb: buffer to pad
1045 *	@pad: space to pad
1046 *
1047 *	Ensure that a buffer is followed by a padding area that is zero
1048 *	filled. Used by network drivers which may DMA or transfer data
1049 *	beyond the buffer end onto the wire.
1050 *
1051 *	May return error in out of memory cases. The skb is freed on error.
1052 */
1053
1054int skb_pad(struct sk_buff *skb, int pad)
1055{
1056	int err;
1057	int ntail;
1058
1059	/* If the skbuff is non linear tailroom is always zero.. */
1060	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1061		memset(skb->data+skb->len, 0, pad);
1062		return 0;
1063	}
1064
1065	ntail = skb->data_len + pad - (skb->end - skb->tail);
1066	if (likely(skb_cloned(skb) || ntail > 0)) {
1067		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1068		if (unlikely(err))
1069			goto free_skb;
1070	}
1071
1072	/* FIXME: The use of this function with non-linear skb's really needs
1073	 * to be audited.
1074	 */
1075	err = skb_linearize(skb);
1076	if (unlikely(err))
1077		goto free_skb;
1078
1079	memset(skb->data + skb->len, 0, pad);
1080	return 0;
1081
1082free_skb:
1083	kfree_skb(skb);
1084	return err;
1085}
1086EXPORT_SYMBOL(skb_pad);
1087
1088/**
1089 *	skb_put - add data to a buffer
1090 *	@skb: buffer to use
1091 *	@len: amount of data to add
1092 *
1093 *	This function extends the used data area of the buffer. If this would
1094 *	exceed the total buffer size the kernel will panic. A pointer to the
1095 *	first byte of the extra data is returned.
1096 */
1097unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1098{
1099	unsigned char *tmp = skb_tail_pointer(skb);
1100	SKB_LINEAR_ASSERT(skb);
1101	skb->tail += len;
1102	skb->len  += len;
1103	if (unlikely(skb->tail > skb->end))
1104		skb_over_panic(skb, len, __builtin_return_address(0));
1105	return tmp;
1106}
1107EXPORT_SYMBOL(skb_put);
1108
1109/**
1110 *	skb_push - add data to the start of a buffer
1111 *	@skb: buffer to use
1112 *	@len: amount of data to add
1113 *
1114 *	This function extends the used data area of the buffer at the buffer
1115 *	start. If this would exceed the total buffer headroom the kernel will
1116 *	panic. A pointer to the first byte of the extra data is returned.
1117 */
1118unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1119{
1120	skb->data -= len;
1121	skb->len  += len;
1122	if (unlikely(skb->data<skb->head))
1123		skb_under_panic(skb, len, __builtin_return_address(0));
1124	return skb->data;
1125}
1126EXPORT_SYMBOL(skb_push);
1127
1128/**
1129 *	skb_pull - remove data from the start of a buffer
1130 *	@skb: buffer to use
1131 *	@len: amount of data to remove
1132 *
1133 *	This function removes data from the start of a buffer, returning
1134 *	the memory to the headroom. A pointer to the next data in the buffer
1135 *	is returned. Once the data has been pulled future pushes will overwrite
1136 *	the old data.
1137 */
1138unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1139{
1140	return skb_pull_inline(skb, len);
1141}
1142EXPORT_SYMBOL(skb_pull);
1143
1144/**
1145 *	skb_trim - remove end from a buffer
1146 *	@skb: buffer to alter
1147 *	@len: new length
1148 *
1149 *	Cut the length of a buffer down by removing data from the tail. If
1150 *	the buffer is already under the length specified it is not modified.
1151 *	The skb must be linear.
1152 */
1153void skb_trim(struct sk_buff *skb, unsigned int len)
1154{
1155	if (skb->len > len)
1156		__skb_trim(skb, len);
1157}
1158EXPORT_SYMBOL(skb_trim);
1159
1160/* Trims skb to length len. It can change skb pointers.
1161 */
1162
1163int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1164{
1165	struct sk_buff **fragp;
1166	struct sk_buff *frag;
1167	int offset = skb_headlen(skb);
1168	int nfrags = skb_shinfo(skb)->nr_frags;
1169	int i;
1170	int err;
1171
1172	if (skb_cloned(skb) &&
1173	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1174		return err;
1175
1176	i = 0;
1177	if (offset >= len)
1178		goto drop_pages;
1179
1180	for (; i < nfrags; i++) {
1181		int end = offset + skb_shinfo(skb)->frags[i].size;
1182
1183		if (end < len) {
1184			offset = end;
1185			continue;
1186		}
1187
1188		skb_shinfo(skb)->frags[i++].size = len - offset;
1189
1190drop_pages:
1191		skb_shinfo(skb)->nr_frags = i;
1192
1193		for (; i < nfrags; i++)
1194			put_page(skb_shinfo(skb)->frags[i].page);
1195
1196		if (skb_has_frag_list(skb))
1197			skb_drop_fraglist(skb);
1198		goto done;
1199	}
1200
1201	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1202	     fragp = &frag->next) {
1203		int end = offset + frag->len;
1204
1205		if (skb_shared(frag)) {
1206			struct sk_buff *nfrag;
1207
1208			nfrag = skb_clone(frag, GFP_ATOMIC);
1209			if (unlikely(!nfrag))
1210				return -ENOMEM;
1211
1212			nfrag->next = frag->next;
1213			kfree_skb(frag);
1214			frag = nfrag;
1215			*fragp = frag;
1216		}
1217
1218		if (end < len) {
1219			offset = end;
1220			continue;
1221		}
1222
1223		if (end > len &&
1224		    unlikely((err = pskb_trim(frag, len - offset))))
1225			return err;
1226
1227		if (frag->next)
1228			skb_drop_list(&frag->next);
1229		break;
1230	}
1231
1232done:
1233	if (len > skb_headlen(skb)) {
1234		skb->data_len -= skb->len - len;
1235		skb->len       = len;
1236	} else {
1237		skb->len       = len;
1238		skb->data_len  = 0;
1239		skb_set_tail_pointer(skb, len);
1240	}
1241
1242	return 0;
1243}
1244EXPORT_SYMBOL(___pskb_trim);
1245
1246/**
1247 *	__pskb_pull_tail - advance tail of skb header
1248 *	@skb: buffer to reallocate
1249 *	@delta: number of bytes to advance tail
1250 *
1251 *	The function makes a sense only on a fragmented &sk_buff,
1252 *	it expands header moving its tail forward and copying necessary
1253 *	data from fragmented part.
1254 *
1255 *	&sk_buff MUST have reference count of 1.
1256 *
1257 *	Returns %NULL (and &sk_buff does not change) if pull failed
1258 *	or value of new tail of skb in the case of success.
1259 *
1260 *	All the pointers pointing into skb header may change and must be
1261 *	reloaded after call to this function.
1262 */
1263
1264/* Moves tail of skb head forward, copying data from fragmented part,
1265 * when it is necessary.
1266 * 1. It may fail due to malloc failure.
1267 * 2. It may change skb pointers.
1268 *
1269 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1270 */
1271unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1272{
1273	/* If skb has not enough free space at tail, get new one
1274	 * plus 128 bytes for future expansions. If we have enough
1275	 * room at tail, reallocate without expansion only if skb is cloned.
1276	 */
1277	int i, k, eat = (skb->tail + delta) - skb->end;
1278
1279	if (eat > 0 || skb_cloned(skb)) {
1280		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1281				     GFP_ATOMIC))
1282			return NULL;
1283	}
1284
1285	if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1286		BUG();
1287
1288	/* Optimization: no fragments, no reasons to preestimate
1289	 * size of pulled pages. Superb.
1290	 */
1291	if (!skb_has_frag_list(skb))
1292		goto pull_pages;
1293
1294	/* Estimate size of pulled pages. */
1295	eat = delta;
1296	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1297		if (skb_shinfo(skb)->frags[i].size >= eat)
1298			goto pull_pages;
1299		eat -= skb_shinfo(skb)->frags[i].size;
1300	}
1301
1302	/* If we need update frag list, we are in troubles.
1303	 * Certainly, it possible to add an offset to skb data,
1304	 * but taking into account that pulling is expected to
1305	 * be very rare operation, it is worth to fight against
1306	 * further bloating skb head and crucify ourselves here instead.
1307	 * Pure masohism, indeed. 8)8)
1308	 */
1309	if (eat) {
1310		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311		struct sk_buff *clone = NULL;
1312		struct sk_buff *insp = NULL;
1313
1314		do {
1315			BUG_ON(!list);
1316
1317			if (list->len <= eat) {
1318				/* Eaten as whole. */
1319				eat -= list->len;
1320				list = list->next;
1321				insp = list;
1322			} else {
1323				/* Eaten partially. */
1324
1325				if (skb_shared(list)) {
1326					/* Sucks! We need to fork list. :-( */
1327					clone = skb_clone(list, GFP_ATOMIC);
1328					if (!clone)
1329						return NULL;
1330					insp = list->next;
1331					list = clone;
1332				} else {
1333					/* This may be pulled without
1334					 * problems. */
1335					insp = list;
1336				}
1337				if (!pskb_pull(list, eat)) {
1338					kfree_skb(clone);
1339					return NULL;
1340				}
1341				break;
1342			}
1343		} while (eat);
1344
1345		/* Free pulled out fragments. */
1346		while ((list = skb_shinfo(skb)->frag_list) != insp) {
1347			skb_shinfo(skb)->frag_list = list->next;
1348			kfree_skb(list);
1349		}
1350		/* And insert new clone at head. */
1351		if (clone) {
1352			clone->next = list;
1353			skb_shinfo(skb)->frag_list = clone;
1354		}
1355	}
1356	/* Success! Now we may commit changes to skb data. */
1357
1358pull_pages:
1359	eat = delta;
1360	k = 0;
1361	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362		if (skb_shinfo(skb)->frags[i].size <= eat) {
1363			put_page(skb_shinfo(skb)->frags[i].page);
1364			eat -= skb_shinfo(skb)->frags[i].size;
1365		} else {
1366			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1367			if (eat) {
1368				skb_shinfo(skb)->frags[k].page_offset += eat;
1369				skb_shinfo(skb)->frags[k].size -= eat;
1370				eat = 0;
1371			}
1372			k++;
1373		}
1374	}
1375	skb_shinfo(skb)->nr_frags = k;
1376
1377	skb->tail     += delta;
1378	skb->data_len -= delta;
1379
1380	return skb_tail_pointer(skb);
1381}
1382EXPORT_SYMBOL(__pskb_pull_tail);
1383
1384/**
1385 *	skb_copy_bits - copy bits from skb to kernel buffer
1386 *	@skb: source skb
1387 *	@offset: offset in source
1388 *	@to: destination buffer
1389 *	@len: number of bytes to copy
1390 *
1391 *	Copy the specified number of bytes from the source skb to the
1392 *	destination buffer.
1393 *
1394 *	CAUTION ! :
1395 *		If its prototype is ever changed,
1396 *		check arch/{*}/net/{*}.S files,
1397 *		since it is called from BPF assembly code.
1398 */
1399int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1400{
1401	int start = skb_headlen(skb);
1402	struct sk_buff *frag_iter;
1403	int i, copy;
1404
1405	if (offset > (int)skb->len - len)
1406		goto fault;
1407
1408	/* Copy header. */
1409	if ((copy = start - offset) > 0) {
1410		if (copy > len)
1411			copy = len;
1412		skb_copy_from_linear_data_offset(skb, offset, to, copy);
1413		if ((len -= copy) == 0)
1414			return 0;
1415		offset += copy;
1416		to     += copy;
1417	}
1418
1419	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1420		int end;
1421
1422		WARN_ON(start > offset + len);
1423
1424		end = start + skb_shinfo(skb)->frags[i].size;
1425		if ((copy = end - offset) > 0) {
1426			u8 *vaddr;
1427
1428			if (copy > len)
1429				copy = len;
1430
1431			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1432			memcpy(to,
1433			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
1434			       offset - start, copy);
1435			kunmap_skb_frag(vaddr);
1436
1437			if ((len -= copy) == 0)
1438				return 0;
1439			offset += copy;
1440			to     += copy;
1441		}
1442		start = end;
1443	}
1444
1445	skb_walk_frags(skb, frag_iter) {
1446		int end;
1447
1448		WARN_ON(start > offset + len);
1449
1450		end = start + frag_iter->len;
1451		if ((copy = end - offset) > 0) {
1452			if (copy > len)
1453				copy = len;
1454			if (skb_copy_bits(frag_iter, offset - start, to, copy))
1455				goto fault;
1456			if ((len -= copy) == 0)
1457				return 0;
1458			offset += copy;
1459			to     += copy;
1460		}
1461		start = end;
1462	}
1463
1464	if (!len)
1465		return 0;
1466
1467fault:
1468	return -EFAULT;
1469}
1470EXPORT_SYMBOL(skb_copy_bits);
1471
1472/*
1473 * Callback from splice_to_pipe(), if we need to release some pages
1474 * at the end of the spd in case we error'ed out in filling the pipe.
1475 */
1476static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1477{
1478	put_page(spd->pages[i]);
1479}
1480
1481static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1482					  unsigned int *offset,
1483					  struct sk_buff *skb, struct sock *sk)
1484{
1485	struct page *p = sk->sk_sndmsg_page;
1486	unsigned int off;
1487
1488	if (!p) {
1489new_page:
1490		p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1491		if (!p)
1492			return NULL;
1493
1494		off = sk->sk_sndmsg_off = 0;
1495		/* hold one ref to this page until it's full */
1496	} else {
1497		unsigned int mlen;
1498
1499		off = sk->sk_sndmsg_off;
1500		mlen = PAGE_SIZE - off;
1501		if (mlen < 64 && mlen < *len) {
1502			put_page(p);
1503			goto new_page;
1504		}
1505
1506		*len = min_t(unsigned int, *len, mlen);
1507	}
1508
1509	memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1510	sk->sk_sndmsg_off += *len;
1511	*offset = off;
1512	get_page(p);
1513
1514	return p;
1515}
1516
1517/*
1518 * Fill page/offset/length into spd, if it can hold more pages.
1519 */
1520static inline int spd_fill_page(struct splice_pipe_desc *spd,
1521				struct pipe_inode_info *pipe, struct page *page,
1522				unsigned int *len, unsigned int offset,
1523				struct sk_buff *skb, int linear,
1524				struct sock *sk)
1525{
1526	if (unlikely(spd->nr_pages == pipe->buffers))
1527		return 1;
1528
1529	if (linear) {
1530		page = linear_to_page(page, len, &offset, skb, sk);
1531		if (!page)
1532			return 1;
1533	} else
1534		get_page(page);
1535
1536	spd->pages[spd->nr_pages] = page;
1537	spd->partial[spd->nr_pages].len = *len;
1538	spd->partial[spd->nr_pages].offset = offset;
1539	spd->nr_pages++;
1540
1541	return 0;
1542}
1543
1544static inline void __segment_seek(struct page **page, unsigned int *poff,
1545				  unsigned int *plen, unsigned int off)
1546{
1547	unsigned long n;
1548
1549	*poff += off;
1550	n = *poff / PAGE_SIZE;
1551	if (n)
1552		*page = nth_page(*page, n);
1553
1554	*poff = *poff % PAGE_SIZE;
1555	*plen -= off;
1556}
1557
1558static inline int __splice_segment(struct page *page, unsigned int poff,
1559				   unsigned int plen, unsigned int *off,
1560				   unsigned int *len, struct sk_buff *skb,
1561				   struct splice_pipe_desc *spd, int linear,
1562				   struct sock *sk,
1563				   struct pipe_inode_info *pipe)
1564{
1565	if (!*len)
1566		return 1;
1567
1568	/* skip this segment if already processed */
1569	if (*off >= plen) {
1570		*off -= plen;
1571		return 0;
1572	}
1573
1574	/* ignore any bits we already processed */
1575	if (*off) {
1576		__segment_seek(&page, &poff, &plen, *off);
1577		*off = 0;
1578	}
1579
1580	do {
1581		unsigned int flen = min(*len, plen);
1582
1583		/* the linear region may spread across several pages  */
1584		flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1585
1586		if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1587			return 1;
1588
1589		__segment_seek(&page, &poff, &plen, flen);
1590		*len -= flen;
1591
1592	} while (*len && plen);
1593
1594	return 0;
1595}
1596
1597/*
1598 * Map linear and fragment data from the skb to spd. It reports failure if the
1599 * pipe is full or if we already spliced the requested length.
1600 */
1601static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1602			     unsigned int *offset, unsigned int *len,
1603			     struct splice_pipe_desc *spd, struct sock *sk)
1604{
1605	int seg;
1606
1607	/*
1608	 * map the linear part
1609	 */
1610	if (__splice_segment(virt_to_page(skb->data),
1611			     (unsigned long) skb->data & (PAGE_SIZE - 1),
1612			     skb_headlen(skb),
1613			     offset, len, skb, spd, 1, sk, pipe))
1614		return 1;
1615
1616	/*
1617	 * then map the fragments
1618	 */
1619	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1620		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1621
1622		if (__splice_segment(f->page, f->page_offset, f->size,
1623				     offset, len, skb, spd, 0, sk, pipe))
1624			return 1;
1625	}
1626
1627	return 0;
1628}
1629
1630/*
1631 * Map data from the skb to a pipe. Should handle both the linear part,
1632 * the fragments, and the frag list. It does NOT handle frag lists within
1633 * the frag list, if such a thing exists. We'd probably need to recurse to
1634 * handle that cleanly.
1635 */
1636int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1637		    struct pipe_inode_info *pipe, unsigned int tlen,
1638		    unsigned int flags)
1639{
1640	struct partial_page partial[PIPE_DEF_BUFFERS];
1641	struct page *pages[PIPE_DEF_BUFFERS];
1642	struct splice_pipe_desc spd = {
1643		.pages = pages,
1644		.partial = partial,
1645		.flags = flags,
1646		.ops = &sock_pipe_buf_ops,
1647		.spd_release = sock_spd_release,
1648	};
1649	struct sk_buff *frag_iter;
1650	struct sock *sk = skb->sk;
1651	int ret = 0;
1652
1653	if (splice_grow_spd(pipe, &spd))
1654		return -ENOMEM;
1655
1656	/*
1657	 * __skb_splice_bits() only fails if the output has no room left,
1658	 * so no point in going over the frag_list for the error case.
1659	 */
1660	if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1661		goto done;
1662	else if (!tlen)
1663		goto done;
1664
1665	/*
1666	 * now see if we have a frag_list to map
1667	 */
1668	skb_walk_frags(skb, frag_iter) {
1669		if (!tlen)
1670			break;
1671		if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1672			break;
1673	}
1674
1675done:
1676	if (spd.nr_pages) {
1677		/*
1678		 * Drop the socket lock, otherwise we have reverse
1679		 * locking dependencies between sk_lock and i_mutex
1680		 * here as compared to sendfile(). We enter here
1681		 * with the socket lock held, and splice_to_pipe() will
1682		 * grab the pipe inode lock. For sendfile() emulation,
1683		 * we call into ->sendpage() with the i_mutex lock held
1684		 * and networking will grab the socket lock.
1685		 */
1686		release_sock(sk);
1687		ret = splice_to_pipe(pipe, &spd);
1688		lock_sock(sk);
1689	}
1690
1691	splice_shrink_spd(pipe, &spd);
1692	return ret;
1693}
1694
1695/**
1696 *	skb_store_bits - store bits from kernel buffer to skb
1697 *	@skb: destination buffer
1698 *	@offset: offset in destination
1699 *	@from: source buffer
1700 *	@len: number of bytes to copy
1701 *
1702 *	Copy the specified number of bytes from the source buffer to the
1703 *	destination skb.  This function handles all the messy bits of
1704 *	traversing fragment lists and such.
1705 */
1706
1707int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1708{
1709	int start = skb_headlen(skb);
1710	struct sk_buff *frag_iter;
1711	int i, copy;
1712
1713	if (offset > (int)skb->len - len)
1714		goto fault;
1715
1716	if ((copy = start - offset) > 0) {
1717		if (copy > len)
1718			copy = len;
1719		skb_copy_to_linear_data_offset(skb, offset, from, copy);
1720		if ((len -= copy) == 0)
1721			return 0;
1722		offset += copy;
1723		from += copy;
1724	}
1725
1726	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1728		int end;
1729
1730		WARN_ON(start > offset + len);
1731
1732		end = start + frag->size;
1733		if ((copy = end - offset) > 0) {
1734			u8 *vaddr;
1735
1736			if (copy > len)
1737				copy = len;
1738
1739			vaddr = kmap_skb_frag(frag);
1740			memcpy(vaddr + frag->page_offset + offset - start,
1741			       from, copy);
1742			kunmap_skb_frag(vaddr);
1743
1744			if ((len -= copy) == 0)
1745				return 0;
1746			offset += copy;
1747			from += copy;
1748		}
1749		start = end;
1750	}
1751
1752	skb_walk_frags(skb, frag_iter) {
1753		int end;
1754
1755		WARN_ON(start > offset + len);
1756
1757		end = start + frag_iter->len;
1758		if ((copy = end - offset) > 0) {
1759			if (copy > len)
1760				copy = len;
1761			if (skb_store_bits(frag_iter, offset - start,
1762					   from, copy))
1763				goto fault;
1764			if ((len -= copy) == 0)
1765				return 0;
1766			offset += copy;
1767			from += copy;
1768		}
1769		start = end;
1770	}
1771	if (!len)
1772		return 0;
1773
1774fault:
1775	return -EFAULT;
1776}
1777EXPORT_SYMBOL(skb_store_bits);
1778
1779/* Checksum skb data. */
1780
1781__wsum skb_checksum(const struct sk_buff *skb, int offset,
1782			  int len, __wsum csum)
1783{
1784	int start = skb_headlen(skb);
1785	int i, copy = start - offset;
1786	struct sk_buff *frag_iter;
1787	int pos = 0;
1788
1789	/* Checksum header. */
1790	if (copy > 0) {
1791		if (copy > len)
1792			copy = len;
1793		csum = csum_partial(skb->data + offset, copy, csum);
1794		if ((len -= copy) == 0)
1795			return csum;
1796		offset += copy;
1797		pos	= copy;
1798	}
1799
1800	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1801		int end;
1802
1803		WARN_ON(start > offset + len);
1804
1805		end = start + skb_shinfo(skb)->frags[i].size;
1806		if ((copy = end - offset) > 0) {
1807			__wsum csum2;
1808			u8 *vaddr;
1809			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1810
1811			if (copy > len)
1812				copy = len;
1813			vaddr = kmap_skb_frag(frag);
1814			csum2 = csum_partial(vaddr + frag->page_offset +
1815					     offset - start, copy, 0);
1816			kunmap_skb_frag(vaddr);
1817			csum = csum_block_add(csum, csum2, pos);
1818			if (!(len -= copy))
1819				return csum;
1820			offset += copy;
1821			pos    += copy;
1822		}
1823		start = end;
1824	}
1825
1826	skb_walk_frags(skb, frag_iter) {
1827		int end;
1828
1829		WARN_ON(start > offset + len);
1830
1831		end = start + frag_iter->len;
1832		if ((copy = end - offset) > 0) {
1833			__wsum csum2;
1834			if (copy > len)
1835				copy = len;
1836			csum2 = skb_checksum(frag_iter, offset - start,
1837					     copy, 0);
1838			csum = csum_block_add(csum, csum2, pos);
1839			if ((len -= copy) == 0)
1840				return csum;
1841			offset += copy;
1842			pos    += copy;
1843		}
1844		start = end;
1845	}
1846	BUG_ON(len);
1847
1848	return csum;
1849}
1850EXPORT_SYMBOL(skb_checksum);
1851
1852/* Both of above in one bottle. */
1853
1854__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1855				    u8 *to, int len, __wsum csum)
1856{
1857	int start = skb_headlen(skb);
1858	int i, copy = start - offset;
1859	struct sk_buff *frag_iter;
1860	int pos = 0;
1861
1862	/* Copy header. */
1863	if (copy > 0) {
1864		if (copy > len)
1865			copy = len;
1866		csum = csum_partial_copy_nocheck(skb->data + offset, to,
1867						 copy, csum);
1868		if ((len -= copy) == 0)
1869			return csum;
1870		offset += copy;
1871		to     += copy;
1872		pos	= copy;
1873	}
1874
1875	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876		int end;
1877
1878		WARN_ON(start > offset + len);
1879
1880		end = start + skb_shinfo(skb)->frags[i].size;
1881		if ((copy = end - offset) > 0) {
1882			__wsum csum2;
1883			u8 *vaddr;
1884			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1885
1886			if (copy > len)
1887				copy = len;
1888			vaddr = kmap_skb_frag(frag);
1889			csum2 = csum_partial_copy_nocheck(vaddr +
1890							  frag->page_offset +
1891							  offset - start, to,
1892							  copy, 0);
1893			kunmap_skb_frag(vaddr);
1894			csum = csum_block_add(csum, csum2, pos);
1895			if (!(len -= copy))
1896				return csum;
1897			offset += copy;
1898			to     += copy;
1899			pos    += copy;
1900		}
1901		start = end;
1902	}
1903
1904	skb_walk_frags(skb, frag_iter) {
1905		__wsum csum2;
1906		int end;
1907
1908		WARN_ON(start > offset + len);
1909
1910		end = start + frag_iter->len;
1911		if ((copy = end - offset) > 0) {
1912			if (copy > len)
1913				copy = len;
1914			csum2 = skb_copy_and_csum_bits(frag_iter,
1915						       offset - start,
1916						       to, copy, 0);
1917			csum = csum_block_add(csum, csum2, pos);
1918			if ((len -= copy) == 0)
1919				return csum;
1920			offset += copy;
1921			to     += copy;
1922			pos    += copy;
1923		}
1924		start = end;
1925	}
1926	BUG_ON(len);
1927	return csum;
1928}
1929EXPORT_SYMBOL(skb_copy_and_csum_bits);
1930
1931void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1932{
1933	__wsum csum;
1934	long csstart;
1935
1936	if (skb->ip_summed == CHECKSUM_PARTIAL)
1937		csstart = skb_checksum_start_offset(skb);
1938	else
1939		csstart = skb_headlen(skb);
1940
1941	BUG_ON(csstart > skb_headlen(skb));
1942
1943	skb_copy_from_linear_data(skb, to, csstart);
1944
1945	csum = 0;
1946	if (csstart != skb->len)
1947		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1948					      skb->len - csstart, 0);
1949
1950	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1951		long csstuff = csstart + skb->csum_offset;
1952
1953		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
1954	}
1955}
1956EXPORT_SYMBOL(skb_copy_and_csum_dev);
1957
1958/**
1959 *	skb_dequeue - remove from the head of the queue
1960 *	@list: list to dequeue from
1961 *
1962 *	Remove the head of the list. The list lock is taken so the function
1963 *	may be used safely with other locking list functions. The head item is
1964 *	returned or %NULL if the list is empty.
1965 */
1966
1967struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1968{
1969	unsigned long flags;
1970	struct sk_buff *result;
1971
1972	spin_lock_irqsave(&list->lock, flags);
1973	result = __skb_dequeue(list);
1974	spin_unlock_irqrestore(&list->lock, flags);
1975	return result;
1976}
1977EXPORT_SYMBOL(skb_dequeue);
1978
1979/**
1980 *	skb_dequeue_tail - remove from the tail of the queue
1981 *	@list: list to dequeue from
1982 *
1983 *	Remove the tail of the list. The list lock is taken so the function
1984 *	may be used safely with other locking list functions. The tail item is
1985 *	returned or %NULL if the list is empty.
1986 */
1987struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1988{
1989	unsigned long flags;
1990	struct sk_buff *result;
1991
1992	spin_lock_irqsave(&list->lock, flags);
1993	result = __skb_dequeue_tail(list);
1994	spin_unlock_irqrestore(&list->lock, flags);
1995	return result;
1996}
1997EXPORT_SYMBOL(skb_dequeue_tail);
1998
1999/**
2000 *	skb_queue_purge - empty a list
2001 *	@list: list to empty
2002 *
2003 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 *	the list and one reference dropped. This function takes the list
2005 *	lock and is atomic with respect to other list locking functions.
2006 */
2007void skb_queue_purge(struct sk_buff_head *list)
2008{
2009	struct sk_buff *skb;
2010	while ((skb = skb_dequeue(list)) != NULL)
2011		kfree_skb(skb);
2012}
2013EXPORT_SYMBOL(skb_queue_purge);
2014
2015/**
2016 *	skb_queue_head - queue a buffer at the list head
2017 *	@list: list to use
2018 *	@newsk: buffer to queue
2019 *
2020 *	Queue a buffer at the start of the list. This function takes the
2021 *	list lock and can be used safely with other locking &sk_buff functions
2022 *	safely.
2023 *
2024 *	A buffer cannot be placed on two lists at the same time.
2025 */
2026void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2027{
2028	unsigned long flags;
2029
2030	spin_lock_irqsave(&list->lock, flags);
2031	__skb_queue_head(list, newsk);
2032	spin_unlock_irqrestore(&list->lock, flags);
2033}
2034EXPORT_SYMBOL(skb_queue_head);
2035
2036/**
2037 *	skb_queue_tail - queue a buffer at the list tail
2038 *	@list: list to use
2039 *	@newsk: buffer to queue
2040 *
2041 *	Queue a buffer at the tail of the list. This function takes the
2042 *	list lock and can be used safely with other locking &sk_buff functions
2043 *	safely.
2044 *
2045 *	A buffer cannot be placed on two lists at the same time.
2046 */
2047void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2048{
2049	unsigned long flags;
2050
2051	spin_lock_irqsave(&list->lock, flags);
2052	__skb_queue_tail(list, newsk);
2053	spin_unlock_irqrestore(&list->lock, flags);
2054}
2055EXPORT_SYMBOL(skb_queue_tail);
2056
2057/**
2058 *	skb_unlink	-	remove a buffer from a list
2059 *	@skb: buffer to remove
2060 *	@list: list to use
2061 *
2062 *	Remove a packet from a list. The list locks are taken and this
2063 *	function is atomic with respect to other list locked calls
2064 *
2065 *	You must know what list the SKB is on.
2066 */
2067void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2068{
2069	unsigned long flags;
2070
2071	spin_lock_irqsave(&list->lock, flags);
2072	__skb_unlink(skb, list);
2073	spin_unlock_irqrestore(&list->lock, flags);
2074}
2075EXPORT_SYMBOL(skb_unlink);
2076
2077/**
2078 *	skb_append	-	append a buffer
2079 *	@old: buffer to insert after
2080 *	@newsk: buffer to insert
2081 *	@list: list to use
2082 *
2083 *	Place a packet after a given packet in a list. The list locks are taken
2084 *	and this function is atomic with respect to other list locked calls.
2085 *	A buffer cannot be placed on two lists at the same time.
2086 */
2087void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2088{
2089	unsigned long flags;
2090
2091	spin_lock_irqsave(&list->lock, flags);
2092	__skb_queue_after(list, old, newsk);
2093	spin_unlock_irqrestore(&list->lock, flags);
2094}
2095EXPORT_SYMBOL(skb_append);
2096
2097/**
2098 *	skb_insert	-	insert a buffer
2099 *	@old: buffer to insert before
2100 *	@newsk: buffer to insert
2101 *	@list: list to use
2102 *
2103 *	Place a packet before a given packet in a list. The list locks are
2104 * 	taken and this function is atomic with respect to other list locked
2105 *	calls.
2106 *
2107 *	A buffer cannot be placed on two lists at the same time.
2108 */
2109void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2110{
2111	unsigned long flags;
2112
2113	spin_lock_irqsave(&list->lock, flags);
2114	__skb_insert(newsk, old->prev, old, list);
2115	spin_unlock_irqrestore(&list->lock, flags);
2116}
2117EXPORT_SYMBOL(skb_insert);
2118
2119static inline void skb_split_inside_header(struct sk_buff *skb,
2120					   struct sk_buff* skb1,
2121					   const u32 len, const int pos)
2122{
2123	int i;
2124
2125	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2126					 pos - len);
2127	/* And move data appendix as is. */
2128	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2129		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2130
2131	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2132	skb_shinfo(skb)->nr_frags  = 0;
2133	skb1->data_len		   = skb->data_len;
2134	skb1->len		   += skb1->data_len;
2135	skb->data_len		   = 0;
2136	skb->len		   = len;
2137	skb_set_tail_pointer(skb, len);
2138}
2139
2140static inline void skb_split_no_header(struct sk_buff *skb,
2141				       struct sk_buff* skb1,
2142				       const u32 len, int pos)
2143{
2144	int i, k = 0;
2145	const int nfrags = skb_shinfo(skb)->nr_frags;
2146
2147	skb_shinfo(skb)->nr_frags = 0;
2148	skb1->len		  = skb1->data_len = skb->len - len;
2149	skb->len		  = len;
2150	skb->data_len		  = len - pos;
2151
2152	for (i = 0; i < nfrags; i++) {
2153		int size = skb_shinfo(skb)->frags[i].size;
2154
2155		if (pos + size > len) {
2156			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2157
2158			if (pos < len) {
2159				/* Split frag.
2160				 * We have two variants in this case:
2161				 * 1. Move all the frag to the second
2162				 *    part, if it is possible. F.e.
2163				 *    this approach is mandatory for TUX,
2164				 *    where splitting is expensive.
2165				 * 2. Split is accurately. We make this.
2166				 */
2167				get_page(skb_shinfo(skb)->frags[i].page);
2168				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2169				skb_shinfo(skb1)->frags[0].size -= len - pos;
2170				skb_shinfo(skb)->frags[i].size	= len - pos;
2171				skb_shinfo(skb)->nr_frags++;
2172			}
2173			k++;
2174		} else
2175			skb_shinfo(skb)->nr_frags++;
2176		pos += size;
2177	}
2178	skb_shinfo(skb1)->nr_frags = k;
2179}
2180
2181/**
2182 * skb_split - Split fragmented skb to two parts at length len.
2183 * @skb: the buffer to split
2184 * @skb1: the buffer to receive the second part
2185 * @len: new length for skb
2186 */
2187void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2188{
2189	int pos = skb_headlen(skb);
2190
2191	if (len < pos)	/* Split line is inside header. */
2192		skb_split_inside_header(skb, skb1, len, pos);
2193	else		/* Second chunk has no header, nothing to copy. */
2194		skb_split_no_header(skb, skb1, len, pos);
2195}
2196EXPORT_SYMBOL(skb_split);
2197
2198/* Shifting from/to a cloned skb is a no-go.
2199 *
2200 * Caller cannot keep skb_shinfo related pointers past calling here!
2201 */
2202static int skb_prepare_for_shift(struct sk_buff *skb)
2203{
2204	return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2205}
2206
2207/**
2208 * skb_shift - Shifts paged data partially from skb to another
2209 * @tgt: buffer into which tail data gets added
2210 * @skb: buffer from which the paged data comes from
2211 * @shiftlen: shift up to this many bytes
2212 *
2213 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2214 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2215 * It's up to caller to free skb if everything was shifted.
2216 *
2217 * If @tgt runs out of frags, the whole operation is aborted.
2218 *
2219 * Skb cannot include anything else but paged data while tgt is allowed
2220 * to have non-paged data as well.
2221 *
2222 * TODO: full sized shift could be optimized but that would need
2223 * specialized skb free'er to handle frags without up-to-date nr_frags.
2224 */
2225int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2226{
2227	int from, to, merge, todo;
2228	struct skb_frag_struct *fragfrom, *fragto;
2229
2230	BUG_ON(shiftlen > skb->len);
2231	BUG_ON(skb_headlen(skb));	/* Would corrupt stream */
2232
2233	todo = shiftlen;
2234	from = 0;
2235	to = skb_shinfo(tgt)->nr_frags;
2236	fragfrom = &skb_shinfo(skb)->frags[from];
2237
2238	/* Actual merge is delayed until the point when we know we can
2239	 * commit all, so that we don't have to undo partial changes
2240	 */
2241	if (!to ||
2242	    !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2243		merge = -1;
2244	} else {
2245		merge = to - 1;
2246
2247		todo -= fragfrom->size;
2248		if (todo < 0) {
2249			if (skb_prepare_for_shift(skb) ||
2250			    skb_prepare_for_shift(tgt))
2251				return 0;
2252
2253			/* All previous frag pointers might be stale! */
2254			fragfrom = &skb_shinfo(skb)->frags[from];
2255			fragto = &skb_shinfo(tgt)->frags[merge];
2256
2257			fragto->size += shiftlen;
2258			fragfrom->size -= shiftlen;
2259			fragfrom->page_offset += shiftlen;
2260
2261			goto onlymerged;
2262		}
2263
2264		from++;
2265	}
2266
2267	/* Skip full, not-fitting skb to avoid expensive operations */
2268	if ((shiftlen == skb->len) &&
2269	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2270		return 0;
2271
2272	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2273		return 0;
2274
2275	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2276		if (to == MAX_SKB_FRAGS)
2277			return 0;
2278
2279		fragfrom = &skb_shinfo(skb)->frags[from];
2280		fragto = &skb_shinfo(tgt)->frags[to];
2281
2282		if (todo >= fragfrom->size) {
2283			*fragto = *fragfrom;
2284			todo -= fragfrom->size;
2285			from++;
2286			to++;
2287
2288		} else {
2289			get_page(fragfrom->page);
2290			fragto->page = fragfrom->page;
2291			fragto->page_offset = fragfrom->page_offset;
2292			fragto->size = todo;
2293
2294			fragfrom->page_offset += todo;
2295			fragfrom->size -= todo;
2296			todo = 0;
2297
2298			to++;
2299			break;
2300		}
2301	}
2302
2303	/* Ready to "commit" this state change to tgt */
2304	skb_shinfo(tgt)->nr_frags = to;
2305
2306	if (merge >= 0) {
2307		fragfrom = &skb_shinfo(skb)->frags[0];
2308		fragto = &skb_shinfo(tgt)->frags[merge];
2309
2310		fragto->size += fragfrom->size;
2311		put_page(fragfrom->page);
2312	}
2313
2314	/* Reposition in the original skb */
2315	to = 0;
2316	while (from < skb_shinfo(skb)->nr_frags)
2317		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2318	skb_shinfo(skb)->nr_frags = to;
2319
2320	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2321
2322onlymerged:
2323	/* Most likely the tgt won't ever need its checksum anymore, skb on
2324	 * the other hand might need it if it needs to be resent
2325	 */
2326	tgt->ip_summed = CHECKSUM_PARTIAL;
2327	skb->ip_summed = CHECKSUM_PARTIAL;
2328
2329	/* Yak, is it really working this way? Some helper please? */
2330	skb->len -= shiftlen;
2331	skb->data_len -= shiftlen;
2332	skb->truesize -= shiftlen;
2333	tgt->len += shiftlen;
2334	tgt->data_len += shiftlen;
2335	tgt->truesize += shiftlen;
2336
2337	return shiftlen;
2338}
2339
2340/**
2341 * skb_prepare_seq_read - Prepare a sequential read of skb data
2342 * @skb: the buffer to read
2343 * @from: lower offset of data to be read
2344 * @to: upper offset of data to be read
2345 * @st: state variable
2346 *
2347 * Initializes the specified state variable. Must be called before
2348 * invoking skb_seq_read() for the first time.
2349 */
2350void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2351			  unsigned int to, struct skb_seq_state *st)
2352{
2353	st->lower_offset = from;
2354	st->upper_offset = to;
2355	st->root_skb = st->cur_skb = skb;
2356	st->frag_idx = st->stepped_offset = 0;
2357	st->frag_data = NULL;
2358}
2359EXPORT_SYMBOL(skb_prepare_seq_read);
2360
2361/**
2362 * skb_seq_read - Sequentially read skb data
2363 * @consumed: number of bytes consumed by the caller so far
2364 * @data: destination pointer for data to be returned
2365 * @st: state variable
2366 *
2367 * Reads a block of skb data at &consumed relative to the
2368 * lower offset specified to skb_prepare_seq_read(). Assigns
2369 * the head of the data block to &data and returns the length
2370 * of the block or 0 if the end of the skb data or the upper
2371 * offset has been reached.
2372 *
2373 * The caller is not required to consume all of the data
2374 * returned, i.e. &consumed is typically set to the number
2375 * of bytes already consumed and the next call to
2376 * skb_seq_read() will return the remaining part of the block.
2377 *
2378 * Note 1: The size of each block of data returned can be arbitrary,
2379 *       this limitation is the cost for zerocopy seqeuental
2380 *       reads of potentially non linear data.
2381 *
2382 * Note 2: Fragment lists within fragments are not implemented
2383 *       at the moment, state->root_skb could be replaced with
2384 *       a stack for this purpose.
2385 */
2386unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2387			  struct skb_seq_state *st)
2388{
2389	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2390	skb_frag_t *frag;
2391
2392	if (unlikely(abs_offset >= st->upper_offset))
2393		return 0;
2394
2395next_skb:
2396	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2397
2398	if (abs_offset < block_limit && !st->frag_data) {
2399		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2400		return block_limit - abs_offset;
2401	}
2402
2403	if (st->frag_idx == 0 && !st->frag_data)
2404		st->stepped_offset += skb_headlen(st->cur_skb);
2405
2406	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2407		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2408		block_limit = frag->size + st->stepped_offset;
2409
2410		if (abs_offset < block_limit) {
2411			if (!st->frag_data)
2412				st->frag_data = kmap_skb_frag(frag);
2413
2414			*data = (u8 *) st->frag_data + frag->page_offset +
2415				(abs_offset - st->stepped_offset);
2416
2417			return block_limit - abs_offset;
2418		}
2419
2420		if (st->frag_data) {
2421			kunmap_skb_frag(st->frag_data);
2422			st->frag_data = NULL;
2423		}
2424
2425		st->frag_idx++;
2426		st->stepped_offset += frag->size;
2427	}
2428
2429	if (st->frag_data) {
2430		kunmap_skb_frag(st->frag_data);
2431		st->frag_data = NULL;
2432	}
2433
2434	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2435		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2436		st->frag_idx = 0;
2437		goto next_skb;
2438	} else if (st->cur_skb->next) {
2439		st->cur_skb = st->cur_skb->next;
2440		st->frag_idx = 0;
2441		goto next_skb;
2442	}
2443
2444	return 0;
2445}
2446EXPORT_SYMBOL(skb_seq_read);
2447
2448/**
2449 * skb_abort_seq_read - Abort a sequential read of skb data
2450 * @st: state variable
2451 *
2452 * Must be called if skb_seq_read() was not called until it
2453 * returned 0.
2454 */
2455void skb_abort_seq_read(struct skb_seq_state *st)
2456{
2457	if (st->frag_data)
2458		kunmap_skb_frag(st->frag_data);
2459}
2460EXPORT_SYMBOL(skb_abort_seq_read);
2461
2462#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
2463
2464static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2465					  struct ts_config *conf,
2466					  struct ts_state *state)
2467{
2468	return skb_seq_read(offset, text, TS_SKB_CB(state));
2469}
2470
2471static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2472{
2473	skb_abort_seq_read(TS_SKB_CB(state));
2474}
2475
2476/**
2477 * skb_find_text - Find a text pattern in skb data
2478 * @skb: the buffer to look in
2479 * @from: search offset
2480 * @to: search limit
2481 * @config: textsearch configuration
2482 * @state: uninitialized textsearch state variable
2483 *
2484 * Finds a pattern in the skb data according to the specified
2485 * textsearch configuration. Use textsearch_next() to retrieve
2486 * subsequent occurrences of the pattern. Returns the offset
2487 * to the first occurrence or UINT_MAX if no match was found.
2488 */
2489unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2490			   unsigned int to, struct ts_config *config,
2491			   struct ts_state *state)
2492{
2493	unsigned int ret;
2494
2495	config->get_next_block = skb_ts_get_next_block;
2496	config->finish = skb_ts_finish;
2497
2498	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2499
2500	ret = textsearch_find(config, state);
2501	return (ret <= to - from ? ret : UINT_MAX);
2502}
2503EXPORT_SYMBOL(skb_find_text);
2504
2505/**
2506 * skb_append_datato_frags: - append the user data to a skb
2507 * @sk: sock  structure
2508 * @skb: skb structure to be appened with user data.
2509 * @getfrag: call back function to be used for getting the user data
2510 * @from: pointer to user message iov
2511 * @length: length of the iov message
2512 *
2513 * Description: This procedure append the user data in the fragment part
2514 * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2515 */
2516int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2517			int (*getfrag)(void *from, char *to, int offset,
2518					int len, int odd, struct sk_buff *skb),
2519			void *from, int length)
2520{
2521	int frg_cnt = 0;
2522	skb_frag_t *frag = NULL;
2523	struct page *page = NULL;
2524	int copy, left;
2525	int offset = 0;
2526	int ret;
2527
2528	do {
2529		/* Return error if we don't have space for new frag */
2530		frg_cnt = skb_shinfo(skb)->nr_frags;
2531		if (frg_cnt >= MAX_SKB_FRAGS)
2532			return -EFAULT;
2533
2534		/* allocate a new page for next frag */
2535		page = alloc_pages(sk->sk_allocation, 0);
2536
2537		/* If alloc_page fails just return failure and caller will
2538		 * free previous allocated pages by doing kfree_skb()
2539		 */
2540		if (page == NULL)
2541			return -ENOMEM;
2542
2543		/* initialize the next frag */
2544		skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2545		skb->truesize += PAGE_SIZE;
2546		atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2547
2548		/* get the new initialized frag */
2549		frg_cnt = skb_shinfo(skb)->nr_frags;
2550		frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2551
2552		/* copy the user data to page */
2553		left = PAGE_SIZE - frag->page_offset;
2554		copy = (length > left)? left : length;
2555
2556		ret = getfrag(from, (page_address(frag->page) +
2557			    frag->page_offset + frag->size),
2558			    offset, copy, 0, skb);
2559		if (ret < 0)
2560			return -EFAULT;
2561
2562		/* copy was successful so update the size parameters */
2563		frag->size += copy;
2564		skb->len += copy;
2565		skb->data_len += copy;
2566		offset += copy;
2567		length -= copy;
2568
2569	} while (length > 0);
2570
2571	return 0;
2572}
2573EXPORT_SYMBOL(skb_append_datato_frags);
2574
2575/**
2576 *	skb_pull_rcsum - pull skb and update receive checksum
2577 *	@skb: buffer to update
2578 *	@len: length of data pulled
2579 *
2580 *	This function performs an skb_pull on the packet and updates
2581 *	the CHECKSUM_COMPLETE checksum.  It should be used on
2582 *	receive path processing instead of skb_pull unless you know
2583 *	that the checksum difference is zero (e.g., a valid IP header)
2584 *	or you are setting ip_summed to CHECKSUM_NONE.
2585 */
2586unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2587{
2588	BUG_ON(len > skb->len);
2589	skb->len -= len;
2590	BUG_ON(skb->len < skb->data_len);
2591	skb_postpull_rcsum(skb, skb->data, len);
2592	return skb->data += len;
2593}
2594EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2595
2596/**
2597 *	skb_segment - Perform protocol segmentation on skb.
2598 *	@skb: buffer to segment
2599 *	@features: features for the output path (see dev->features)
2600 *
2601 *	This function performs segmentation on the given skb.  It returns
2602 *	a pointer to the first in a list of new skbs for the segments.
2603 *	In case of error it returns ERR_PTR(err).
2604 */
2605struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2606{
2607	struct sk_buff *segs = NULL;
2608	struct sk_buff *tail = NULL;
2609	struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2610	unsigned int mss = skb_shinfo(skb)->gso_size;
2611	unsigned int doffset = skb->data - skb_mac_header(skb);
2612	unsigned int offset = doffset;
2613	unsigned int headroom;
2614	unsigned int len;
2615	int sg = !!(features & NETIF_F_SG);
2616	int nfrags = skb_shinfo(skb)->nr_frags;
2617	int err = -ENOMEM;
2618	int i = 0;
2619	int pos;
2620
2621	__skb_push(skb, doffset);
2622	headroom = skb_headroom(skb);
2623	pos = skb_headlen(skb);
2624
2625	do {
2626		struct sk_buff *nskb;
2627		skb_frag_t *frag;
2628		int hsize;
2629		int size;
2630
2631		len = skb->len - offset;
2632		if (len > mss)
2633			len = mss;
2634
2635		hsize = skb_headlen(skb) - offset;
2636		if (hsize < 0)
2637			hsize = 0;
2638		if (hsize > len || !sg)
2639			hsize = len;
2640
2641		if (!hsize && i >= nfrags) {
2642			BUG_ON(fskb->len != len);
2643
2644			pos += len;
2645			nskb = skb_clone(fskb, GFP_ATOMIC);
2646			fskb = fskb->next;
2647
2648			if (unlikely(!nskb))
2649				goto err;
2650
2651			hsize = skb_end_pointer(nskb) - nskb->head;
2652			if (skb_cow_head(nskb, doffset + headroom)) {
2653				kfree_skb(nskb);
2654				goto err;
2655			}
2656
2657			nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2658					  hsize;
2659			skb_release_head_state(nskb);
2660			__skb_push(nskb, doffset);
2661		} else {
2662			nskb = alloc_skb(hsize + doffset + headroom,
2663					 GFP_ATOMIC);
2664
2665			if (unlikely(!nskb))
2666				goto err;
2667
2668			skb_reserve(nskb, headroom);
2669			__skb_put(nskb, doffset);
2670		}
2671
2672		if (segs)
2673			tail->next = nskb;
2674		else
2675			segs = nskb;
2676		tail = nskb;
2677
2678		__copy_skb_header(nskb, skb);
2679		nskb->mac_len = skb->mac_len;
2680
2681		/* nskb and skb might have different headroom */
2682		if (nskb->ip_summed == CHECKSUM_PARTIAL)
2683			nskb->csum_start += skb_headroom(nskb) - headroom;
2684
2685		skb_reset_mac_header(nskb);
2686		skb_set_network_header(nskb, skb->mac_len);
2687		nskb->transport_header = (nskb->network_header +
2688					  skb_network_header_len(skb));
2689		skb_copy_from_linear_data(skb, nskb->data, doffset);
2690
2691		if (fskb != skb_shinfo(skb)->frag_list)
2692			continue;
2693
2694		if (!sg) {
2695			nskb->ip_summed = CHECKSUM_NONE;
2696			nskb->csum = skb_copy_and_csum_bits(skb, offset,
2697							    skb_put(nskb, len),
2698							    len, 0);
2699			continue;
2700		}
2701
2702		frag = skb_shinfo(nskb)->frags;
2703
2704		skb_copy_from_linear_data_offset(skb, offset,
2705						 skb_put(nskb, hsize), hsize);
2706
2707		while (pos < offset + len && i < nfrags) {
2708			*frag = skb_shinfo(skb)->frags[i];
2709			get_page(frag->page);
2710			size = frag->size;
2711
2712			if (pos < offset) {
2713				frag->page_offset += offset - pos;
2714				frag->size -= offset - pos;
2715			}
2716
2717			skb_shinfo(nskb)->nr_frags++;
2718
2719			if (pos + size <= offset + len) {
2720				i++;
2721				pos += size;
2722			} else {
2723				frag->size -= pos + size - (offset + len);
2724				goto skip_fraglist;
2725			}
2726
2727			frag++;
2728		}
2729
2730		if (pos < offset + len) {
2731			struct sk_buff *fskb2 = fskb;
2732
2733			BUG_ON(pos + fskb->len != offset + len);
2734
2735			pos += fskb->len;
2736			fskb = fskb->next;
2737
2738			if (fskb2->next) {
2739				fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2740				if (!fskb2)
2741					goto err;
2742			} else
2743				skb_get(fskb2);
2744
2745			SKB_FRAG_ASSERT(nskb);
2746			skb_shinfo(nskb)->frag_list = fskb2;
2747		}
2748
2749skip_fraglist:
2750		nskb->data_len = len - hsize;
2751		nskb->len += nskb->data_len;
2752		nskb->truesize += nskb->data_len;
2753	} while ((offset += len) < skb->len);
2754
2755	return segs;
2756
2757err:
2758	while ((skb = segs)) {
2759		segs = skb->next;
2760		kfree_skb(skb);
2761	}
2762	return ERR_PTR(err);
2763}
2764EXPORT_SYMBOL_GPL(skb_segment);
2765
2766int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2767{
2768	struct sk_buff *p = *head;
2769	struct sk_buff *nskb;
2770	struct skb_shared_info *skbinfo = skb_shinfo(skb);
2771	struct skb_shared_info *pinfo = skb_shinfo(p);
2772	unsigned int headroom;
2773	unsigned int len = skb_gro_len(skb);
2774	unsigned int offset = skb_gro_offset(skb);
2775	unsigned int headlen = skb_headlen(skb);
2776
2777	if (p->len + len >= 65536)
2778		return -E2BIG;
2779
2780	if (pinfo->frag_list)
2781		goto merge;
2782	else if (headlen <= offset) {
2783		skb_frag_t *frag;
2784		skb_frag_t *frag2;
2785		int i = skbinfo->nr_frags;
2786		int nr_frags = pinfo->nr_frags + i;
2787
2788		offset -= headlen;
2789
2790		if (nr_frags > MAX_SKB_FRAGS)
2791			return -E2BIG;
2792
2793		pinfo->nr_frags = nr_frags;
2794		skbinfo->nr_frags = 0;
2795
2796		frag = pinfo->frags + nr_frags;
2797		frag2 = skbinfo->frags + i;
2798		do {
2799			*--frag = *--frag2;
2800		} while (--i);
2801
2802		frag->page_offset += offset;
2803		frag->size -= offset;
2804
2805		skb->truesize -= skb->data_len;
2806		skb->len -= skb->data_len;
2807		skb->data_len = 0;
2808
2809		NAPI_GRO_CB(skb)->free = 1;
2810		goto done;
2811	} else if (skb_gro_len(p) != pinfo->gso_size)
2812		return -E2BIG;
2813
2814	headroom = skb_headroom(p);
2815	nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2816	if (unlikely(!nskb))
2817		return -ENOMEM;
2818
2819	__copy_skb_header(nskb, p);
2820	nskb->mac_len = p->mac_len;
2821
2822	skb_reserve(nskb, headroom);
2823	__skb_put(nskb, skb_gro_offset(p));
2824
2825	skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2826	skb_set_network_header(nskb, skb_network_offset(p));
2827	skb_set_transport_header(nskb, skb_transport_offset(p));
2828
2829	__skb_pull(p, skb_gro_offset(p));
2830	memcpy(skb_mac_header(nskb), skb_mac_header(p),
2831	       p->data - skb_mac_header(p));
2832
2833	*NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2834	skb_shinfo(nskb)->frag_list = p;
2835	skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2836	pinfo->gso_size = 0;
2837	skb_header_release(p);
2838	nskb->prev = p;
2839
2840	nskb->data_len += p->len;
2841	nskb->truesize += p->len;
2842	nskb->len += p->len;
2843
2844	*head = nskb;
2845	nskb->next = p->next;
2846	p->next = NULL;
2847
2848	p = nskb;
2849
2850merge:
2851	if (offset > headlen) {
2852		unsigned int eat = offset - headlen;
2853
2854		skbinfo->frags[0].page_offset += eat;
2855		skbinfo->frags[0].size -= eat;
2856		skb->data_len -= eat;
2857		skb->len -= eat;
2858		offset = headlen;
2859	}
2860
2861	__skb_pull(skb, offset);
2862
2863	p->prev->next = skb;
2864	p->prev = skb;
2865	skb_header_release(skb);
2866
2867done:
2868	NAPI_GRO_CB(p)->count++;
2869	p->data_len += len;
2870	p->truesize += len;
2871	p->len += len;
2872
2873	NAPI_GRO_CB(skb)->same_flow = 1;
2874	return 0;
2875}
2876EXPORT_SYMBOL_GPL(skb_gro_receive);
2877
2878void __init skb_init(void)
2879{
2880	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2881					      sizeof(struct sk_buff),
2882					      0,
2883					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2884					      NULL);
2885	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2886						(2*sizeof(struct sk_buff)) +
2887						sizeof(atomic_t),
2888						0,
2889						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2890						NULL);
2891}
2892
2893/**
2894 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2895 *	@skb: Socket buffer containing the buffers to be mapped
2896 *	@sg: The scatter-gather list to map into
2897 *	@offset: The offset into the buffer's contents to start mapping
2898 *	@len: Length of buffer space to be mapped
2899 *
2900 *	Fill the specified scatter-gather list with mappings/pointers into a
2901 *	region of the buffer space attached to a socket buffer.
2902 */
2903static int
2904__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2905{
2906	int start = skb_headlen(skb);
2907	int i, copy = start - offset;
2908	struct sk_buff *frag_iter;
2909	int elt = 0;
2910
2911	if (copy > 0) {
2912		if (copy > len)
2913			copy = len;
2914		sg_set_buf(sg, skb->data + offset, copy);
2915		elt++;
2916		if ((len -= copy) == 0)
2917			return elt;
2918		offset += copy;
2919	}
2920
2921	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2922		int end;
2923
2924		WARN_ON(start > offset + len);
2925
2926		end = start + skb_shinfo(skb)->frags[i].size;
2927		if ((copy = end - offset) > 0) {
2928			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2929
2930			if (copy > len)
2931				copy = len;
2932			sg_set_page(&sg[elt], frag->page, copy,
2933					frag->page_offset+offset-start);
2934			elt++;
2935			if (!(len -= copy))
2936				return elt;
2937			offset += copy;
2938		}
2939		start = end;
2940	}
2941
2942	skb_walk_frags(skb, frag_iter) {
2943		int end;
2944
2945		WARN_ON(start > offset + len);
2946
2947		end = start + frag_iter->len;
2948		if ((copy = end - offset) > 0) {
2949			if (copy > len)
2950				copy = len;
2951			elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2952					      copy);
2953			if ((len -= copy) == 0)
2954				return elt;
2955			offset += copy;
2956		}
2957		start = end;
2958	}
2959	BUG_ON(len);
2960	return elt;
2961}
2962
2963int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2964{
2965	int nsg = __skb_to_sgvec(skb, sg, offset, len);
2966
2967	sg_mark_end(&sg[nsg - 1]);
2968
2969	return nsg;
2970}
2971EXPORT_SYMBOL_GPL(skb_to_sgvec);
2972
2973/**
2974 *	skb_cow_data - Check that a socket buffer's data buffers are writable
2975 *	@skb: The socket buffer to check.
2976 *	@tailbits: Amount of trailing space to be added
2977 *	@trailer: Returned pointer to the skb where the @tailbits space begins
2978 *
2979 *	Make sure that the data buffers attached to a socket buffer are
2980 *	writable. If they are not, private copies are made of the data buffers
2981 *	and the socket buffer is set to use these instead.
2982 *
2983 *	If @tailbits is given, make sure that there is space to write @tailbits
2984 *	bytes of data beyond current end of socket buffer.  @trailer will be
2985 *	set to point to the skb in which this space begins.
2986 *
2987 *	The number of scatterlist elements required to completely map the
2988 *	COW'd and extended socket buffer will be returned.
2989 */
2990int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2991{
2992	int copyflag;
2993	int elt;
2994	struct sk_buff *skb1, **skb_p;
2995
2996	/* If skb is cloned or its head is paged, reallocate
2997	 * head pulling out all the pages (pages are considered not writable
2998	 * at the moment even if they are anonymous).
2999	 */
3000	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3001	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3002		return -ENOMEM;
3003
3004	/* Easy case. Most of packets will go this way. */
3005	if (!skb_has_frag_list(skb)) {
3006		/* A little of trouble, not enough of space for trailer.
3007		 * This should not happen, when stack is tuned to generate
3008		 * good frames. OK, on miss we reallocate and reserve even more
3009		 * space, 128 bytes is fair. */
3010
3011		if (skb_tailroom(skb) < tailbits &&
3012		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3013			return -ENOMEM;
3014
3015		/* Voila! */
3016		*trailer = skb;
3017		return 1;
3018	}
3019
3020	/* Misery. We are in troubles, going to mincer fragments... */
3021
3022	elt = 1;
3023	skb_p = &skb_shinfo(skb)->frag_list;
3024	copyflag = 0;
3025
3026	while ((skb1 = *skb_p) != NULL) {
3027		int ntail = 0;
3028
3029		/* The fragment is partially pulled by someone,
3030		 * this can happen on input. Copy it and everything
3031		 * after it. */
3032
3033		if (skb_shared(skb1))
3034			copyflag = 1;
3035
3036		/* If the skb is the last, worry about trailer. */
3037
3038		if (skb1->next == NULL && tailbits) {
3039			if (skb_shinfo(skb1)->nr_frags ||
3040			    skb_has_frag_list(skb1) ||
3041			    skb_tailroom(skb1) < tailbits)
3042				ntail = tailbits + 128;
3043		}
3044
3045		if (copyflag ||
3046		    skb_cloned(skb1) ||
3047		    ntail ||
3048		    skb_shinfo(skb1)->nr_frags ||
3049		    skb_has_frag_list(skb1)) {
3050			struct sk_buff *skb2;
3051
3052			/* Fuck, we are miserable poor guys... */
3053			if (ntail == 0)
3054				skb2 = skb_copy(skb1, GFP_ATOMIC);
3055			else
3056				skb2 = skb_copy_expand(skb1,
3057						       skb_headroom(skb1),
3058						       ntail,
3059						       GFP_ATOMIC);
3060			if (unlikely(skb2 == NULL))
3061				return -ENOMEM;
3062
3063			if (skb1->sk)
3064				skb_set_owner_w(skb2, skb1->sk);
3065
3066			/* Looking around. Are we still alive?
3067			 * OK, link new skb, drop old one */
3068
3069			skb2->next = skb1->next;
3070			*skb_p = skb2;
3071			kfree_skb(skb1);
3072			skb1 = skb2;
3073		}
3074		elt++;
3075		*trailer = skb1;
3076		skb_p = &skb1->next;
3077	}
3078
3079	return elt;
3080}
3081EXPORT_SYMBOL_GPL(skb_cow_data);
3082
3083static void sock_rmem_free(struct sk_buff *skb)
3084{
3085	struct sock *sk = skb->sk;
3086
3087	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3088}
3089
3090/*
3091 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3092 */
3093int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3094{
3095	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3096	    (unsigned)sk->sk_rcvbuf)
3097		return -ENOMEM;
3098
3099	skb_orphan(skb);
3100	skb->sk = sk;
3101	skb->destructor = sock_rmem_free;
3102	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3103
3104	/* before exiting rcu section, make sure dst is refcounted */
3105	skb_dst_force(skb);
3106
3107	skb_queue_tail(&sk->sk_error_queue, skb);
3108	if (!sock_flag(sk, SOCK_DEAD))
3109		sk->sk_data_ready(sk, skb->len);
3110	return 0;
3111}
3112EXPORT_SYMBOL(sock_queue_err_skb);
3113
3114void skb_tstamp_tx(struct sk_buff *orig_skb,
3115		struct skb_shared_hwtstamps *hwtstamps)
3116{
3117	struct sock *sk = orig_skb->sk;
3118	struct sock_exterr_skb *serr;
3119	struct sk_buff *skb;
3120	int err;
3121
3122	if (!sk)
3123		return;
3124
3125	skb = skb_clone(orig_skb, GFP_ATOMIC);
3126	if (!skb)
3127		return;
3128
3129	if (hwtstamps) {
3130		*skb_hwtstamps(skb) =
3131			*hwtstamps;
3132	} else {
3133		/*
3134		 * no hardware time stamps available,
3135		 * so keep the shared tx_flags and only
3136		 * store software time stamp
3137		 */
3138		skb->tstamp = ktime_get_real();
3139	}
3140
3141	serr = SKB_EXT_ERR(skb);
3142	memset(serr, 0, sizeof(*serr));
3143	serr->ee.ee_errno = ENOMSG;
3144	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3145
3146	err = sock_queue_err_skb(sk, skb);
3147
3148	if (err)
3149		kfree_skb(skb);
3150}
3151EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3152
3153
3154/**
3155 * skb_partial_csum_set - set up and verify partial csum values for packet
3156 * @skb: the skb to set
3157 * @start: the number of bytes after skb->data to start checksumming.
3158 * @off: the offset from start to place the checksum.
3159 *
3160 * For untrusted partially-checksummed packets, we need to make sure the values
3161 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3162 *
3163 * This function checks and sets those values and skb->ip_summed: if this
3164 * returns false you should drop the packet.
3165 */
3166bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3167{
3168	if (unlikely(start > skb_headlen(skb)) ||
3169	    unlikely((int)start + off > skb_headlen(skb) - 2)) {
3170		if (net_ratelimit())
3171			printk(KERN_WARNING
3172			       "bad partial csum: csum=%u/%u len=%u\n",
3173			       start, off, skb_headlen(skb));
3174		return false;
3175	}
3176	skb->ip_summed = CHECKSUM_PARTIAL;
3177	skb->csum_start = skb_headroom(skb) + start;
3178	skb->csum_offset = off;
3179	return true;
3180}
3181EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3182
3183void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3184{
3185	if (net_ratelimit())
3186		pr_warning("%s: received packets cannot be forwarded"
3187			   " while LRO is enabled\n", skb->dev->name);
3188}
3189EXPORT_SYMBOL(__skb_warn_lro_forwarding);