Linux Audio

Check our new training course

Loading...
v4.6
 
   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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  40
  41#include <linux/module.h>
  42#include <linux/types.h>
  43#include <linux/kernel.h>
  44#include <linux/kmemcheck.h>
  45#include <linux/mm.h>
  46#include <linux/interrupt.h>
  47#include <linux/in.h>
  48#include <linux/inet.h>
  49#include <linux/slab.h>
  50#include <linux/tcp.h>
  51#include <linux/udp.h>
 
  52#include <linux/netdevice.h>
  53#ifdef CONFIG_NET_CLS_ACT
  54#include <net/pkt_sched.h>
  55#endif
  56#include <linux/string.h>
  57#include <linux/skbuff.h>
  58#include <linux/splice.h>
  59#include <linux/cache.h>
  60#include <linux/rtnetlink.h>
  61#include <linux/init.h>
  62#include <linux/scatterlist.h>
  63#include <linux/errqueue.h>
  64#include <linux/prefetch.h>
  65#include <linux/if_vlan.h>
 
  66
  67#include <net/protocol.h>
  68#include <net/dst.h>
  69#include <net/sock.h>
  70#include <net/checksum.h>
  71#include <net/ip6_checksum.h>
  72#include <net/xfrm.h>
 
  73
  74#include <asm/uaccess.h>
  75#include <trace/events/skb.h>
  76#include <linux/highmem.h>
  77#include <linux/capability.h>
  78#include <linux/user_namespace.h>
 
 
 
  79
  80struct kmem_cache *skbuff_head_cache __read_mostly;
  81static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 
 
 
  82int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
  83EXPORT_SYMBOL(sysctl_max_skb_frags);
  84
  85/**
  86 *	skb_panic - private function for out-of-line support
  87 *	@skb:	buffer
  88 *	@sz:	size
  89 *	@addr:	address
  90 *	@msg:	skb_over_panic or skb_under_panic
  91 *
  92 *	Out-of-line support for skb_put() and skb_push().
  93 *	Called via the wrapper skb_over_panic() or skb_under_panic().
  94 *	Keep out of line to prevent kernel bloat.
  95 *	__builtin_return_address is not used because it is not always reliable.
  96 */
  97static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
  98		      const char msg[])
  99{
 100	pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 101		 msg, addr, skb->len, sz, skb->head, skb->data,
 102		 (unsigned long)skb->tail, (unsigned long)skb->end,
 103		 skb->dev ? skb->dev->name : "<NULL>");
 104	BUG();
 105}
 106
 107static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 108{
 109	skb_panic(skb, sz, addr, __func__);
 110}
 111
 112static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 113{
 114	skb_panic(skb, sz, addr, __func__);
 115}
 116
 117/*
 118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
 119 * the caller if emergency pfmemalloc reserves are being used. If it is and
 120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
 121 * may be used. Otherwise, the packet data may be discarded until enough
 122 * memory is free
 123 */
 124#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
 125	 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
 126
 127static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
 128			       unsigned long ip, bool *pfmemalloc)
 129{
 130	void *obj;
 131	bool ret_pfmemalloc = false;
 132
 133	/*
 134	 * Try a regular allocation, when that fails and we're not entitled
 135	 * to the reserves, fail.
 136	 */
 137	obj = kmalloc_node_track_caller(size,
 138					flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 139					node);
 140	if (obj || !(gfp_pfmemalloc_allowed(flags)))
 141		goto out;
 142
 143	/* Try again but now we are using pfmemalloc reserves */
 144	ret_pfmemalloc = true;
 145	obj = kmalloc_node_track_caller(size, flags, node);
 146
 147out:
 148	if (pfmemalloc)
 149		*pfmemalloc = ret_pfmemalloc;
 150
 151	return obj;
 152}
 153
 154/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 155 *	'private' fields and also do memory statistics to find all the
 156 *	[BEEP] leaks.
 157 *
 158 */
 159
 160struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
 161{
 162	struct sk_buff *skb;
 163
 164	/* Get the HEAD */
 165	skb = kmem_cache_alloc_node(skbuff_head_cache,
 166				    gfp_mask & ~__GFP_DMA, node);
 167	if (!skb)
 168		goto out;
 169
 170	/*
 171	 * Only clear those fields we need to clear, not those that we will
 172	 * actually initialise below. Hence, don't put any more fields after
 173	 * the tail pointer in struct sk_buff!
 174	 */
 175	memset(skb, 0, offsetof(struct sk_buff, tail));
 176	skb->head = NULL;
 177	skb->truesize = sizeof(struct sk_buff);
 178	atomic_set(&skb->users, 1);
 179
 180	skb->mac_header = (typeof(skb->mac_header))~0U;
 181out:
 182	return skb;
 183}
 184
 185/**
 186 *	__alloc_skb	-	allocate a network buffer
 187 *	@size: size to allocate
 188 *	@gfp_mask: allocation mask
 189 *	@flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
 190 *		instead of head cache and allocate a cloned (child) skb.
 191 *		If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
 192 *		allocations in case the data is required for writeback
 193 *	@node: numa node to allocate memory on
 194 *
 195 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 196 *	tail room of at least size bytes. The object has a reference count
 197 *	of one. The return is the buffer. On a failure the return is %NULL.
 198 *
 199 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 200 *	%GFP_ATOMIC.
 201 */
 202struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 203			    int flags, int node)
 204{
 205	struct kmem_cache *cache;
 206	struct skb_shared_info *shinfo;
 207	struct sk_buff *skb;
 208	u8 *data;
 209	bool pfmemalloc;
 210
 211	cache = (flags & SKB_ALLOC_FCLONE)
 212		? skbuff_fclone_cache : skbuff_head_cache;
 213
 214	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 215		gfp_mask |= __GFP_MEMALLOC;
 216
 217	/* Get the HEAD */
 218	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
 219	if (!skb)
 220		goto out;
 221	prefetchw(skb);
 222
 223	/* We do our best to align skb_shared_info on a separate cache
 224	 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
 225	 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
 226	 * Both skb->head and skb_shared_info are cache line aligned.
 227	 */
 228	size = SKB_DATA_ALIGN(size);
 229	size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 230	data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
 231	if (!data)
 232		goto nodata;
 233	/* kmalloc(size) might give us more room than requested.
 234	 * Put skb_shared_info exactly at the end of allocated zone,
 235	 * to allow max possible filling before reallocation.
 236	 */
 237	size = SKB_WITH_OVERHEAD(ksize(data));
 238	prefetchw(data + size);
 239
 240	/*
 241	 * Only clear those fields we need to clear, not those that we will
 242	 * actually initialise below. Hence, don't put any more fields after
 243	 * the tail pointer in struct sk_buff!
 244	 */
 245	memset(skb, 0, offsetof(struct sk_buff, tail));
 246	/* Account for allocated memory : skb + skb->head */
 247	skb->truesize = SKB_TRUESIZE(size);
 248	skb->pfmemalloc = pfmemalloc;
 249	atomic_set(&skb->users, 1);
 250	skb->head = data;
 251	skb->data = data;
 252	skb_reset_tail_pointer(skb);
 253	skb->end = skb->tail + size;
 254	skb->mac_header = (typeof(skb->mac_header))~0U;
 255	skb->transport_header = (typeof(skb->transport_header))~0U;
 256
 257	/* make sure we initialize shinfo sequentially */
 258	shinfo = skb_shinfo(skb);
 259	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 260	atomic_set(&shinfo->dataref, 1);
 261	kmemcheck_annotate_variable(shinfo->destructor_arg);
 262
 263	if (flags & SKB_ALLOC_FCLONE) {
 264		struct sk_buff_fclones *fclones;
 265
 266		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 267
 268		kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
 269		skb->fclone = SKB_FCLONE_ORIG;
 270		atomic_set(&fclones->fclone_ref, 1);
 271
 272		fclones->skb2.fclone = SKB_FCLONE_CLONE;
 273		fclones->skb2.pfmemalloc = pfmemalloc;
 274	}
 275out:
 276	return skb;
 277nodata:
 278	kmem_cache_free(cache, skb);
 279	skb = NULL;
 280	goto out;
 281}
 282EXPORT_SYMBOL(__alloc_skb);
 283
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 284/**
 285 * __build_skb - build a network buffer
 286 * @data: data buffer provided by caller
 287 * @frag_size: size of data, or 0 if head was kmalloced
 288 *
 289 * Allocate a new &sk_buff. Caller provides space holding head and
 290 * skb_shared_info. @data must have been allocated by kmalloc() only if
 291 * @frag_size is 0, otherwise data should come from the page allocator
 292 *  or vmalloc()
 293 * The return is the new skb buffer.
 294 * On a failure the return is %NULL, and @data is not freed.
 295 * Notes :
 296 *  Before IO, driver allocates only data buffer where NIC put incoming frame
 297 *  Driver should add room at head (NET_SKB_PAD) and
 298 *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
 299 *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
 300 *  before giving packet to stack.
 301 *  RX rings only contains data buffers, not full skbs.
 302 */
 303struct sk_buff *__build_skb(void *data, unsigned int frag_size)
 304{
 305	struct skb_shared_info *shinfo;
 306	struct sk_buff *skb;
 307	unsigned int size = frag_size ? : ksize(data);
 308
 309	skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
 310	if (!skb)
 311		return NULL;
 312
 313	size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 314
 315	memset(skb, 0, offsetof(struct sk_buff, tail));
 316	skb->truesize = SKB_TRUESIZE(size);
 317	atomic_set(&skb->users, 1);
 318	skb->head = data;
 319	skb->data = data;
 320	skb_reset_tail_pointer(skb);
 321	skb->end = skb->tail + size;
 322	skb->mac_header = (typeof(skb->mac_header))~0U;
 323	skb->transport_header = (typeof(skb->transport_header))~0U;
 324
 325	/* make sure we initialize shinfo sequentially */
 326	shinfo = skb_shinfo(skb);
 327	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 328	atomic_set(&shinfo->dataref, 1);
 329	kmemcheck_annotate_variable(shinfo->destructor_arg);
 330
 331	return skb;
 332}
 333
 334/* build_skb() is wrapper over __build_skb(), that specifically
 335 * takes care of skb->head and skb->pfmemalloc
 336 * This means that if @frag_size is not zero, then @data must be backed
 337 * by a page fragment, not kmalloc() or vmalloc()
 338 */
 339struct sk_buff *build_skb(void *data, unsigned int frag_size)
 340{
 341	struct sk_buff *skb = __build_skb(data, frag_size);
 342
 343	if (skb && frag_size) {
 344		skb->head_frag = 1;
 345		if (page_is_pfmemalloc(virt_to_head_page(data)))
 346			skb->pfmemalloc = 1;
 347	}
 348	return skb;
 349}
 350EXPORT_SYMBOL(build_skb);
 351
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 352#define NAPI_SKB_CACHE_SIZE	64
 353
 354struct napi_alloc_cache {
 355	struct page_frag_cache page;
 356	size_t skb_count;
 357	void *skb_cache[NAPI_SKB_CACHE_SIZE];
 358};
 359
 360static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
 361static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
 362
 363static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
 364{
 365	struct page_frag_cache *nc;
 366	unsigned long flags;
 367	void *data;
 368
 369	local_irq_save(flags);
 370	nc = this_cpu_ptr(&netdev_alloc_cache);
 371	data = __alloc_page_frag(nc, fragsz, gfp_mask);
 372	local_irq_restore(flags);
 373	return data;
 
 
 
 374}
 
 375
 376/**
 377 * netdev_alloc_frag - allocate a page fragment
 378 * @fragsz: fragment size
 379 *
 380 * Allocates a frag from a page for receive buffer.
 381 * Uses GFP_ATOMIC allocations.
 382 */
 383void *netdev_alloc_frag(unsigned int fragsz)
 384{
 385	return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
 386}
 387EXPORT_SYMBOL(netdev_alloc_frag);
 388
 389static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
 390{
 391	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 392
 393	return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
 394}
 395
 396void *napi_alloc_frag(unsigned int fragsz)
 397{
 398	return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
 
 
 
 
 
 
 
 399}
 400EXPORT_SYMBOL(napi_alloc_frag);
 401
 402/**
 403 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 404 *	@dev: network device to receive on
 405 *	@len: length to allocate
 406 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 407 *
 408 *	Allocate a new &sk_buff and assign it a usage count of one. The
 409 *	buffer has NET_SKB_PAD headroom built in. Users should allocate
 410 *	the headroom they think they need without accounting for the
 411 *	built in space. The built in space is used for optimisations.
 412 *
 413 *	%NULL is returned if there is no free memory.
 414 */
 415struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
 416				   gfp_t gfp_mask)
 417{
 418	struct page_frag_cache *nc;
 419	unsigned long flags;
 420	struct sk_buff *skb;
 421	bool pfmemalloc;
 422	void *data;
 423
 424	len += NET_SKB_PAD;
 425
 426	if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
 427	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 428		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 429		if (!skb)
 430			goto skb_fail;
 431		goto skb_success;
 432	}
 433
 434	len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 435	len = SKB_DATA_ALIGN(len);
 436
 437	if (sk_memalloc_socks())
 438		gfp_mask |= __GFP_MEMALLOC;
 439
 440	local_irq_save(flags);
 441
 442	nc = this_cpu_ptr(&netdev_alloc_cache);
 443	data = __alloc_page_frag(nc, len, gfp_mask);
 444	pfmemalloc = nc->pfmemalloc;
 445
 446	local_irq_restore(flags);
 
 
 
 
 447
 448	if (unlikely(!data))
 449		return NULL;
 450
 451	skb = __build_skb(data, len);
 452	if (unlikely(!skb)) {
 453		skb_free_frag(data);
 454		return NULL;
 455	}
 456
 457	/* use OR instead of assignment to avoid clearing of bits in mask */
 458	if (pfmemalloc)
 459		skb->pfmemalloc = 1;
 460	skb->head_frag = 1;
 461
 462skb_success:
 463	skb_reserve(skb, NET_SKB_PAD);
 464	skb->dev = dev;
 465
 466skb_fail:
 467	return skb;
 468}
 469EXPORT_SYMBOL(__netdev_alloc_skb);
 470
 471/**
 472 *	__napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
 473 *	@napi: napi instance this buffer was allocated for
 474 *	@len: length to allocate
 475 *	@gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
 476 *
 477 *	Allocate a new sk_buff for use in NAPI receive.  This buffer will
 478 *	attempt to allocate the head from a special reserved region used
 479 *	only for NAPI Rx allocation.  By doing this we can save several
 480 *	CPU cycles by avoiding having to disable and re-enable IRQs.
 481 *
 482 *	%NULL is returned if there is no free memory.
 483 */
 484struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
 485				 gfp_t gfp_mask)
 486{
 487	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 488	struct sk_buff *skb;
 489	void *data;
 490
 491	len += NET_SKB_PAD + NET_IP_ALIGN;
 492
 493	if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
 494	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 495		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 496		if (!skb)
 497			goto skb_fail;
 498		goto skb_success;
 499	}
 500
 501	len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 502	len = SKB_DATA_ALIGN(len);
 503
 504	if (sk_memalloc_socks())
 505		gfp_mask |= __GFP_MEMALLOC;
 506
 507	data = __alloc_page_frag(&nc->page, len, gfp_mask);
 508	if (unlikely(!data))
 509		return NULL;
 510
 511	skb = __build_skb(data, len);
 512	if (unlikely(!skb)) {
 513		skb_free_frag(data);
 514		return NULL;
 515	}
 516
 517	/* use OR instead of assignment to avoid clearing of bits in mask */
 518	if (nc->page.pfmemalloc)
 519		skb->pfmemalloc = 1;
 520	skb->head_frag = 1;
 521
 522skb_success:
 523	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
 524	skb->dev = napi->dev;
 525
 526skb_fail:
 527	return skb;
 528}
 529EXPORT_SYMBOL(__napi_alloc_skb);
 530
 531void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 532		     int size, unsigned int truesize)
 533{
 534	skb_fill_page_desc(skb, i, page, off, size);
 535	skb->len += size;
 536	skb->data_len += size;
 537	skb->truesize += truesize;
 538}
 539EXPORT_SYMBOL(skb_add_rx_frag);
 540
 541void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
 542			  unsigned int truesize)
 543{
 544	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 545
 546	skb_frag_size_add(frag, size);
 547	skb->len += size;
 548	skb->data_len += size;
 549	skb->truesize += truesize;
 550}
 551EXPORT_SYMBOL(skb_coalesce_rx_frag);
 552
 553static void skb_drop_list(struct sk_buff **listp)
 554{
 555	kfree_skb_list(*listp);
 556	*listp = NULL;
 557}
 558
 559static inline void skb_drop_fraglist(struct sk_buff *skb)
 560{
 561	skb_drop_list(&skb_shinfo(skb)->frag_list);
 562}
 563
 564static void skb_clone_fraglist(struct sk_buff *skb)
 565{
 566	struct sk_buff *list;
 567
 568	skb_walk_frags(skb, list)
 569		skb_get(list);
 570}
 571
 572static void skb_free_head(struct sk_buff *skb)
 573{
 574	unsigned char *head = skb->head;
 575
 576	if (skb->head_frag)
 577		skb_free_frag(head);
 578	else
 579		kfree(head);
 580}
 581
 582static void skb_release_data(struct sk_buff *skb)
 583{
 584	struct skb_shared_info *shinfo = skb_shinfo(skb);
 585	int i;
 586
 587	if (skb->cloned &&
 588	    atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
 589			      &shinfo->dataref))
 590		return;
 591
 592	for (i = 0; i < shinfo->nr_frags; i++)
 593		__skb_frag_unref(&shinfo->frags[i]);
 594
 595	/*
 596	 * If skb buf is from userspace, we need to notify the caller
 597	 * the lower device DMA has done;
 598	 */
 599	if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
 600		struct ubuf_info *uarg;
 601
 602		uarg = shinfo->destructor_arg;
 603		if (uarg->callback)
 604			uarg->callback(uarg, true);
 605	}
 606
 607	if (shinfo->frag_list)
 608		kfree_skb_list(shinfo->frag_list);
 609
 
 610	skb_free_head(skb);
 611}
 612
 613/*
 614 *	Free an skbuff by memory without cleaning the state.
 615 */
 616static void kfree_skbmem(struct sk_buff *skb)
 617{
 618	struct sk_buff_fclones *fclones;
 619
 620	switch (skb->fclone) {
 621	case SKB_FCLONE_UNAVAILABLE:
 622		kmem_cache_free(skbuff_head_cache, skb);
 623		return;
 624
 625	case SKB_FCLONE_ORIG:
 626		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 627
 628		/* We usually free the clone (TX completion) before original skb
 629		 * This test would have no chance to be true for the clone,
 630		 * while here, branch prediction will be good.
 631		 */
 632		if (atomic_read(&fclones->fclone_ref) == 1)
 633			goto fastpath;
 634		break;
 635
 636	default: /* SKB_FCLONE_CLONE */
 637		fclones = container_of(skb, struct sk_buff_fclones, skb2);
 638		break;
 639	}
 640	if (!atomic_dec_and_test(&fclones->fclone_ref))
 641		return;
 642fastpath:
 643	kmem_cache_free(skbuff_fclone_cache, fclones);
 644}
 645
 646static void skb_release_head_state(struct sk_buff *skb)
 647{
 648	skb_dst_drop(skb);
 649#ifdef CONFIG_XFRM
 650	secpath_put(skb->sp);
 651#endif
 652	if (skb->destructor) {
 653		WARN_ON(in_irq());
 654		skb->destructor(skb);
 655	}
 656#if IS_ENABLED(CONFIG_NF_CONNTRACK)
 657	nf_conntrack_put(skb->nfct);
 658#endif
 659#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 660	nf_bridge_put(skb->nf_bridge);
 661#endif
 
 662}
 663
 664/* Free everything but the sk_buff shell. */
 665static void skb_release_all(struct sk_buff *skb)
 666{
 667	skb_release_head_state(skb);
 668	if (likely(skb->head))
 669		skb_release_data(skb);
 670}
 671
 672/**
 673 *	__kfree_skb - private function
 674 *	@skb: buffer
 675 *
 676 *	Free an sk_buff. Release anything attached to the buffer.
 677 *	Clean the state. This is an internal helper function. Users should
 678 *	always call kfree_skb
 679 */
 680
 681void __kfree_skb(struct sk_buff *skb)
 682{
 683	skb_release_all(skb);
 684	kfree_skbmem(skb);
 685}
 686EXPORT_SYMBOL(__kfree_skb);
 687
 688/**
 689 *	kfree_skb - free an sk_buff
 690 *	@skb: buffer to free
 691 *
 692 *	Drop a reference to the buffer and free it if the usage count has
 693 *	hit zero.
 694 */
 695void kfree_skb(struct sk_buff *skb)
 696{
 697	if (unlikely(!skb))
 698		return;
 699	if (likely(atomic_read(&skb->users) == 1))
 700		smp_rmb();
 701	else if (likely(!atomic_dec_and_test(&skb->users)))
 702		return;
 
 703	trace_kfree_skb(skb, __builtin_return_address(0));
 704	__kfree_skb(skb);
 705}
 706EXPORT_SYMBOL(kfree_skb);
 707
 708void kfree_skb_list(struct sk_buff *segs)
 709{
 710	while (segs) {
 711		struct sk_buff *next = segs->next;
 712
 713		kfree_skb(segs);
 714		segs = next;
 715	}
 716}
 717EXPORT_SYMBOL(kfree_skb_list);
 718
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 719/**
 720 *	skb_tx_error - report an sk_buff xmit error
 721 *	@skb: buffer that triggered an error
 722 *
 723 *	Report xmit error if a device callback is tracking this skb.
 724 *	skb must be freed afterwards.
 725 */
 726void skb_tx_error(struct sk_buff *skb)
 727{
 728	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 729		struct ubuf_info *uarg;
 730
 731		uarg = skb_shinfo(skb)->destructor_arg;
 732		if (uarg->callback)
 733			uarg->callback(uarg, false);
 734		skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
 735	}
 736}
 737EXPORT_SYMBOL(skb_tx_error);
 738
 739/**
 740 *	consume_skb - free an skbuff
 741 *	@skb: buffer to free
 742 *
 743 *	Drop a ref to the buffer and free it if the usage count has hit zero
 744 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
 745 *	is being dropped after a failure and notes that
 746 */
 747void consume_skb(struct sk_buff *skb)
 748{
 749	if (unlikely(!skb))
 750		return;
 751	if (likely(atomic_read(&skb->users) == 1))
 752		smp_rmb();
 753	else if (likely(!atomic_dec_and_test(&skb->users)))
 754		return;
 
 755	trace_consume_skb(skb);
 756	__kfree_skb(skb);
 757}
 758EXPORT_SYMBOL(consume_skb);
 759
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 760void __kfree_skb_flush(void)
 761{
 762	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 763
 764	/* flush skb_cache if containing objects */
 765	if (nc->skb_count) {
 766		kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
 767				     nc->skb_cache);
 768		nc->skb_count = 0;
 769	}
 770}
 771
 772static inline void _kfree_skb_defer(struct sk_buff *skb)
 773{
 774	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 775
 776	/* drop skb->head and call any destructors for packet */
 777	skb_release_all(skb);
 778
 779	/* record skb to CPU local list */
 780	nc->skb_cache[nc->skb_count++] = skb;
 781
 782#ifdef CONFIG_SLUB
 783	/* SLUB writes into objects when freeing */
 784	prefetchw(skb);
 785#endif
 786
 787	/* flush skb_cache if it is filled */
 788	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
 789		kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
 790				     nc->skb_cache);
 791		nc->skb_count = 0;
 792	}
 793}
 794void __kfree_skb_defer(struct sk_buff *skb)
 795{
 796	_kfree_skb_defer(skb);
 797}
 798
 799void napi_consume_skb(struct sk_buff *skb, int budget)
 800{
 801	if (unlikely(!skb))
 802		return;
 803
 804	/* Zero budget indicate non-NAPI context called us, like netpoll */
 805	if (unlikely(!budget)) {
 806		dev_consume_skb_any(skb);
 807		return;
 808	}
 809
 810	if (likely(atomic_read(&skb->users) == 1))
 811		smp_rmb();
 812	else if (likely(!atomic_dec_and_test(&skb->users)))
 813		return;
 
 814	/* if reaching here SKB is ready to free */
 815	trace_consume_skb(skb);
 816
 817	/* if SKB is a clone, don't handle this case */
 818	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
 819		__kfree_skb(skb);
 820		return;
 821	}
 822
 823	_kfree_skb_defer(skb);
 824}
 825EXPORT_SYMBOL(napi_consume_skb);
 826
 827/* Make sure a field is enclosed inside headers_start/headers_end section */
 828#define CHECK_SKB_FIELD(field) \
 829	BUILD_BUG_ON(offsetof(struct sk_buff, field) <		\
 830		     offsetof(struct sk_buff, headers_start));	\
 831	BUILD_BUG_ON(offsetof(struct sk_buff, field) >		\
 832		     offsetof(struct sk_buff, headers_end));	\
 833
 834static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 835{
 836	new->tstamp		= old->tstamp;
 837	/* We do not copy old->sk */
 838	new->dev		= old->dev;
 839	memcpy(new->cb, old->cb, sizeof(old->cb));
 840	skb_dst_copy(new, old);
 841#ifdef CONFIG_XFRM
 842	new->sp			= secpath_get(old->sp);
 843#endif
 844	__nf_copy(new, old, false);
 845
 846	/* Note : this field could be in headers_start/headers_end section
 847	 * It is not yet because we do not want to have a 16 bit hole
 848	 */
 849	new->queue_mapping = old->queue_mapping;
 850
 851	memcpy(&new->headers_start, &old->headers_start,
 852	       offsetof(struct sk_buff, headers_end) -
 853	       offsetof(struct sk_buff, headers_start));
 854	CHECK_SKB_FIELD(protocol);
 855	CHECK_SKB_FIELD(csum);
 856	CHECK_SKB_FIELD(hash);
 857	CHECK_SKB_FIELD(priority);
 858	CHECK_SKB_FIELD(skb_iif);
 859	CHECK_SKB_FIELD(vlan_proto);
 860	CHECK_SKB_FIELD(vlan_tci);
 861	CHECK_SKB_FIELD(transport_header);
 862	CHECK_SKB_FIELD(network_header);
 863	CHECK_SKB_FIELD(mac_header);
 864	CHECK_SKB_FIELD(inner_protocol);
 865	CHECK_SKB_FIELD(inner_transport_header);
 866	CHECK_SKB_FIELD(inner_network_header);
 867	CHECK_SKB_FIELD(inner_mac_header);
 868	CHECK_SKB_FIELD(mark);
 869#ifdef CONFIG_NETWORK_SECMARK
 870	CHECK_SKB_FIELD(secmark);
 871#endif
 872#ifdef CONFIG_NET_RX_BUSY_POLL
 873	CHECK_SKB_FIELD(napi_id);
 874#endif
 875#ifdef CONFIG_XPS
 876	CHECK_SKB_FIELD(sender_cpu);
 877#endif
 878#ifdef CONFIG_NET_SCHED
 879	CHECK_SKB_FIELD(tc_index);
 880#ifdef CONFIG_NET_CLS_ACT
 881	CHECK_SKB_FIELD(tc_verd);
 882#endif
 883#endif
 884
 885}
 886
 887/*
 888 * You should not add any new code to this function.  Add it to
 889 * __copy_skb_header above instead.
 890 */
 891static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 892{
 893#define C(x) n->x = skb->x
 894
 895	n->next = n->prev = NULL;
 896	n->sk = NULL;
 897	__copy_skb_header(n, skb);
 898
 899	C(len);
 900	C(data_len);
 901	C(mac_len);
 902	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 903	n->cloned = 1;
 904	n->nohdr = 0;
 
 
 905	n->destructor = NULL;
 906	C(tail);
 907	C(end);
 908	C(head);
 909	C(head_frag);
 910	C(data);
 911	C(truesize);
 912	atomic_set(&n->users, 1);
 913
 914	atomic_inc(&(skb_shinfo(skb)->dataref));
 915	skb->cloned = 1;
 916
 917	return n;
 918#undef C
 919}
 920
 921/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 922 *	skb_morph	-	morph one skb into another
 923 *	@dst: the skb to receive the contents
 924 *	@src: the skb to supply the contents
 925 *
 926 *	This is identical to skb_clone except that the target skb is
 927 *	supplied by the user.
 928 *
 929 *	The target skb is returned upon exit.
 930 */
 931struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 932{
 933	skb_release_all(dst);
 934	return __skb_clone(dst, src);
 935}
 936EXPORT_SYMBOL_GPL(skb_morph);
 937
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 938/**
 939 *	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
 940 *	@skb: the skb to modify
 941 *	@gfp_mask: allocation priority
 942 *
 943 *	This must be called on SKBTX_DEV_ZEROCOPY skb.
 944 *	It will copy all frags into kernel and drop the reference
 945 *	to userspace pages.
 946 *
 947 *	If this function is called from an interrupt gfp_mask() must be
 948 *	%GFP_ATOMIC.
 949 *
 950 *	Returns 0 on success or a negative error code on failure
 951 *	to allocate kernel memory to copy to.
 952 */
 953int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
 954{
 955	int i;
 956	int num_frags = skb_shinfo(skb)->nr_frags;
 957	struct page *page, *head = NULL;
 958	struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
 
 959
 960	for (i = 0; i < num_frags; i++) {
 961		u8 *vaddr;
 962		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 963
 
 
 
 
 
 964		page = alloc_page(gfp_mask);
 965		if (!page) {
 966			while (head) {
 967				struct page *next = (struct page *)page_private(head);
 968				put_page(head);
 969				head = next;
 970			}
 971			return -ENOMEM;
 972		}
 973		vaddr = kmap_atomic(skb_frag_page(f));
 974		memcpy(page_address(page),
 975		       vaddr + f->page_offset, skb_frag_size(f));
 976		kunmap_atomic(vaddr);
 977		set_page_private(page, (unsigned long)head);
 978		head = page;
 979	}
 980
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 981	/* skb frags release userspace buffers */
 982	for (i = 0; i < num_frags; i++)
 983		skb_frag_unref(skb, i);
 984
 985	uarg->callback(uarg, false);
 986
 987	/* skb frags point to kernel buffers */
 988	for (i = num_frags - 1; i >= 0; i--) {
 989		__skb_fill_page_desc(skb, i, head, 0,
 990				     skb_shinfo(skb)->frags[i].size);
 991		head = (struct page *)page_private(head);
 992	}
 
 
 993
 994	skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
 
 995	return 0;
 996}
 997EXPORT_SYMBOL_GPL(skb_copy_ubufs);
 998
 999/**
1000 *	skb_clone	-	duplicate an sk_buff
1001 *	@skb: buffer to clone
1002 *	@gfp_mask: allocation priority
1003 *
1004 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
1005 *	copies share the same packet data but not structure. The new
1006 *	buffer has a reference count of 1. If the allocation fails the
1007 *	function returns %NULL otherwise the new buffer is returned.
1008 *
1009 *	If this function is called from an interrupt gfp_mask() must be
1010 *	%GFP_ATOMIC.
1011 */
1012
1013struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1014{
1015	struct sk_buff_fclones *fclones = container_of(skb,
1016						       struct sk_buff_fclones,
1017						       skb1);
1018	struct sk_buff *n;
1019
1020	if (skb_orphan_frags(skb, gfp_mask))
1021		return NULL;
1022
1023	if (skb->fclone == SKB_FCLONE_ORIG &&
1024	    atomic_read(&fclones->fclone_ref) == 1) {
1025		n = &fclones->skb2;
1026		atomic_set(&fclones->fclone_ref, 2);
1027	} else {
1028		if (skb_pfmemalloc(skb))
1029			gfp_mask |= __GFP_MEMALLOC;
1030
1031		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1032		if (!n)
1033			return NULL;
1034
1035		kmemcheck_annotate_bitfield(n, flags1);
1036		n->fclone = SKB_FCLONE_UNAVAILABLE;
1037	}
1038
1039	return __skb_clone(n, skb);
1040}
1041EXPORT_SYMBOL(skb_clone);
1042
1043static void skb_headers_offset_update(struct sk_buff *skb, int off)
1044{
1045	/* Only adjust this if it actually is csum_start rather than csum */
1046	if (skb->ip_summed == CHECKSUM_PARTIAL)
1047		skb->csum_start += off;
1048	/* {transport,network,mac}_header and tail are relative to skb->head */
1049	skb->transport_header += off;
1050	skb->network_header   += off;
1051	if (skb_mac_header_was_set(skb))
1052		skb->mac_header += off;
1053	skb->inner_transport_header += off;
1054	skb->inner_network_header += off;
1055	skb->inner_mac_header += off;
1056}
 
1057
1058static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1059{
1060	__copy_skb_header(new, old);
1061
1062	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1063	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1064	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1065}
 
1066
1067static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1068{
1069	if (skb_pfmemalloc(skb))
1070		return SKB_ALLOC_RX;
1071	return 0;
1072}
1073
1074/**
1075 *	skb_copy	-	create private copy of an sk_buff
1076 *	@skb: buffer to copy
1077 *	@gfp_mask: allocation priority
1078 *
1079 *	Make a copy of both an &sk_buff and its data. This is used when the
1080 *	caller wishes to modify the data and needs a private copy of the
1081 *	data to alter. Returns %NULL on failure or the pointer to the buffer
1082 *	on success. The returned buffer has a reference count of 1.
1083 *
1084 *	As by-product this function converts non-linear &sk_buff to linear
1085 *	one, so that &sk_buff becomes completely private and caller is allowed
1086 *	to modify all the data of returned buffer. This means that this
1087 *	function is not recommended for use in circumstances when only
1088 *	header is going to be modified. Use pskb_copy() instead.
1089 */
1090
1091struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1092{
1093	int headerlen = skb_headroom(skb);
1094	unsigned int size = skb_end_offset(skb) + skb->data_len;
1095	struct sk_buff *n = __alloc_skb(size, gfp_mask,
1096					skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1097
1098	if (!n)
1099		return NULL;
1100
1101	/* Set the data pointer */
1102	skb_reserve(n, headerlen);
1103	/* Set the tail pointer and length */
1104	skb_put(n, skb->len);
1105
1106	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1107		BUG();
1108
1109	copy_skb_header(n, skb);
1110	return n;
1111}
1112EXPORT_SYMBOL(skb_copy);
1113
1114/**
1115 *	__pskb_copy_fclone	-  create copy of an sk_buff with private head.
1116 *	@skb: buffer to copy
1117 *	@headroom: headroom of new skb
1118 *	@gfp_mask: allocation priority
1119 *	@fclone: if true allocate the copy of the skb from the fclone
1120 *	cache instead of the head cache; it is recommended to set this
1121 *	to true for the cases where the copy will likely be cloned
1122 *
1123 *	Make a copy of both an &sk_buff and part of its data, located
1124 *	in header. Fragmented data remain shared. This is used when
1125 *	the caller wishes to modify only header of &sk_buff and needs
1126 *	private copy of the header to alter. Returns %NULL on failure
1127 *	or the pointer to the buffer on success.
1128 *	The returned buffer has a reference count of 1.
1129 */
1130
1131struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1132				   gfp_t gfp_mask, bool fclone)
1133{
1134	unsigned int size = skb_headlen(skb) + headroom;
1135	int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1136	struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1137
1138	if (!n)
1139		goto out;
1140
1141	/* Set the data pointer */
1142	skb_reserve(n, headroom);
1143	/* Set the tail pointer and length */
1144	skb_put(n, skb_headlen(skb));
1145	/* Copy the bytes */
1146	skb_copy_from_linear_data(skb, n->data, n->len);
1147
1148	n->truesize += skb->data_len;
1149	n->data_len  = skb->data_len;
1150	n->len	     = skb->len;
1151
1152	if (skb_shinfo(skb)->nr_frags) {
1153		int i;
1154
1155		if (skb_orphan_frags(skb, gfp_mask)) {
 
1156			kfree_skb(n);
1157			n = NULL;
1158			goto out;
1159		}
1160		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1162			skb_frag_ref(skb, i);
1163		}
1164		skb_shinfo(n)->nr_frags = i;
1165	}
1166
1167	if (skb_has_frag_list(skb)) {
1168		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1169		skb_clone_fraglist(n);
1170	}
1171
1172	copy_skb_header(n, skb);
1173out:
1174	return n;
1175}
1176EXPORT_SYMBOL(__pskb_copy_fclone);
1177
1178/**
1179 *	pskb_expand_head - reallocate header of &sk_buff
1180 *	@skb: buffer to reallocate
1181 *	@nhead: room to add at head
1182 *	@ntail: room to add at tail
1183 *	@gfp_mask: allocation priority
1184 *
1185 *	Expands (or creates identical copy, if @nhead and @ntail are zero)
1186 *	header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1187 *	reference count of 1. Returns zero in the case of success or error,
1188 *	if expansion failed. In the last case, &sk_buff is not changed.
1189 *
1190 *	All the pointers pointing into skb header may change and must be
1191 *	reloaded after call to this function.
1192 */
1193
1194int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1195		     gfp_t gfp_mask)
1196{
1197	int i;
1198	u8 *data;
1199	int size = nhead + skb_end_offset(skb) + ntail;
1200	long off;
 
1201
1202	BUG_ON(nhead < 0);
1203
1204	if (skb_shared(skb))
1205		BUG();
1206
1207	size = SKB_DATA_ALIGN(size);
1208
1209	if (skb_pfmemalloc(skb))
1210		gfp_mask |= __GFP_MEMALLOC;
1211	data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1212			       gfp_mask, NUMA_NO_NODE, NULL);
1213	if (!data)
1214		goto nodata;
1215	size = SKB_WITH_OVERHEAD(ksize(data));
1216
1217	/* Copy only real data... and, alas, header. This should be
1218	 * optimized for the cases when header is void.
1219	 */
1220	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1221
1222	memcpy((struct skb_shared_info *)(data + size),
1223	       skb_shinfo(skb),
1224	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1225
1226	/*
1227	 * if shinfo is shared we must drop the old head gracefully, but if it
1228	 * is not we can just drop the old head and let the existing refcount
1229	 * be since all we did is relocate the values
1230	 */
1231	if (skb_cloned(skb)) {
1232		/* copy this zero copy skb frags */
1233		if (skb_orphan_frags(skb, gfp_mask))
1234			goto nofrags;
 
 
1235		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1236			skb_frag_ref(skb, i);
1237
1238		if (skb_has_frag_list(skb))
1239			skb_clone_fraglist(skb);
1240
1241		skb_release_data(skb);
1242	} else {
1243		skb_free_head(skb);
1244	}
1245	off = (data + nhead) - skb->head;
1246
1247	skb->head     = data;
1248	skb->head_frag = 0;
1249	skb->data    += off;
1250#ifdef NET_SKBUFF_DATA_USES_OFFSET
1251	skb->end      = size;
1252	off           = nhead;
1253#else
1254	skb->end      = skb->head + size;
1255#endif
1256	skb->tail	      += off;
1257	skb_headers_offset_update(skb, nhead);
1258	skb->cloned   = 0;
1259	skb->hdr_len  = 0;
1260	skb->nohdr    = 0;
1261	atomic_set(&skb_shinfo(skb)->dataref, 1);
 
 
 
 
 
 
 
 
 
 
1262	return 0;
1263
1264nofrags:
1265	kfree(data);
1266nodata:
1267	return -ENOMEM;
1268}
1269EXPORT_SYMBOL(pskb_expand_head);
1270
1271/* Make private copy of skb with writable head and some headroom */
1272
1273struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1274{
1275	struct sk_buff *skb2;
1276	int delta = headroom - skb_headroom(skb);
1277
1278	if (delta <= 0)
1279		skb2 = pskb_copy(skb, GFP_ATOMIC);
1280	else {
1281		skb2 = skb_clone(skb, GFP_ATOMIC);
1282		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1283					     GFP_ATOMIC)) {
1284			kfree_skb(skb2);
1285			skb2 = NULL;
1286		}
1287	}
1288	return skb2;
1289}
1290EXPORT_SYMBOL(skb_realloc_headroom);
1291
1292/**
1293 *	skb_copy_expand	-	copy and expand sk_buff
1294 *	@skb: buffer to copy
1295 *	@newheadroom: new free bytes at head
1296 *	@newtailroom: new free bytes at tail
1297 *	@gfp_mask: allocation priority
1298 *
1299 *	Make a copy of both an &sk_buff and its data and while doing so
1300 *	allocate additional space.
1301 *
1302 *	This is used when the caller wishes to modify the data and needs a
1303 *	private copy of the data to alter as well as more space for new fields.
1304 *	Returns %NULL on failure or the pointer to the buffer
1305 *	on success. The returned buffer has a reference count of 1.
1306 *
1307 *	You must pass %GFP_ATOMIC as the allocation priority if this function
1308 *	is called from an interrupt.
1309 */
1310struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1311				int newheadroom, int newtailroom,
1312				gfp_t gfp_mask)
1313{
1314	/*
1315	 *	Allocate the copy buffer
1316	 */
1317	struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1318					gfp_mask, skb_alloc_rx_flag(skb),
1319					NUMA_NO_NODE);
1320	int oldheadroom = skb_headroom(skb);
1321	int head_copy_len, head_copy_off;
1322
1323	if (!n)
1324		return NULL;
1325
1326	skb_reserve(n, newheadroom);
1327
1328	/* Set the tail pointer and length */
1329	skb_put(n, skb->len);
1330
1331	head_copy_len = oldheadroom;
1332	head_copy_off = 0;
1333	if (newheadroom <= head_copy_len)
1334		head_copy_len = newheadroom;
1335	else
1336		head_copy_off = newheadroom - head_copy_len;
1337
1338	/* Copy the linear header and data. */
1339	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1340			  skb->len + head_copy_len))
1341		BUG();
1342
1343	copy_skb_header(n, skb);
1344
1345	skb_headers_offset_update(n, newheadroom - oldheadroom);
1346
1347	return n;
1348}
1349EXPORT_SYMBOL(skb_copy_expand);
1350
1351/**
1352 *	skb_pad			-	zero pad the tail of an skb
1353 *	@skb: buffer to pad
1354 *	@pad: space to pad
 
1355 *
1356 *	Ensure that a buffer is followed by a padding area that is zero
1357 *	filled. Used by network drivers which may DMA or transfer data
1358 *	beyond the buffer end onto the wire.
1359 *
1360 *	May return error in out of memory cases. The skb is freed on error.
 
1361 */
1362
1363int skb_pad(struct sk_buff *skb, int pad)
1364{
1365	int err;
1366	int ntail;
1367
1368	/* If the skbuff is non linear tailroom is always zero.. */
1369	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1370		memset(skb->data+skb->len, 0, pad);
1371		return 0;
1372	}
1373
1374	ntail = skb->data_len + pad - (skb->end - skb->tail);
1375	if (likely(skb_cloned(skb) || ntail > 0)) {
1376		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1377		if (unlikely(err))
1378			goto free_skb;
1379	}
1380
1381	/* FIXME: The use of this function with non-linear skb's really needs
1382	 * to be audited.
1383	 */
1384	err = skb_linearize(skb);
1385	if (unlikely(err))
1386		goto free_skb;
1387
1388	memset(skb->data + skb->len, 0, pad);
1389	return 0;
1390
1391free_skb:
1392	kfree_skb(skb);
 
1393	return err;
1394}
1395EXPORT_SYMBOL(skb_pad);
1396
1397/**
1398 *	pskb_put - add data to the tail of a potentially fragmented buffer
1399 *	@skb: start of the buffer to use
1400 *	@tail: tail fragment of the buffer to use
1401 *	@len: amount of data to add
1402 *
1403 *	This function extends the used data area of the potentially
1404 *	fragmented buffer. @tail must be the last fragment of @skb -- or
1405 *	@skb itself. If this would exceed the total buffer size the kernel
1406 *	will panic. A pointer to the first byte of the extra data is
1407 *	returned.
1408 */
1409
1410unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1411{
1412	if (tail != skb) {
1413		skb->data_len += len;
1414		skb->len += len;
1415	}
1416	return skb_put(tail, len);
1417}
1418EXPORT_SYMBOL_GPL(pskb_put);
1419
1420/**
1421 *	skb_put - add data to a buffer
1422 *	@skb: buffer to use
1423 *	@len: amount of data to add
1424 *
1425 *	This function extends the used data area of the buffer. If this would
1426 *	exceed the total buffer size the kernel will panic. A pointer to the
1427 *	first byte of the extra data is returned.
1428 */
1429unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1430{
1431	unsigned char *tmp = skb_tail_pointer(skb);
1432	SKB_LINEAR_ASSERT(skb);
1433	skb->tail += len;
1434	skb->len  += len;
1435	if (unlikely(skb->tail > skb->end))
1436		skb_over_panic(skb, len, __builtin_return_address(0));
1437	return tmp;
1438}
1439EXPORT_SYMBOL(skb_put);
1440
1441/**
1442 *	skb_push - add data to the start of a buffer
1443 *	@skb: buffer to use
1444 *	@len: amount of data to add
1445 *
1446 *	This function extends the used data area of the buffer at the buffer
1447 *	start. If this would exceed the total buffer headroom the kernel will
1448 *	panic. A pointer to the first byte of the extra data is returned.
1449 */
1450unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1451{
1452	skb->data -= len;
1453	skb->len  += len;
1454	if (unlikely(skb->data<skb->head))
1455		skb_under_panic(skb, len, __builtin_return_address(0));
1456	return skb->data;
1457}
1458EXPORT_SYMBOL(skb_push);
1459
1460/**
1461 *	skb_pull - remove data from the start of a buffer
1462 *	@skb: buffer to use
1463 *	@len: amount of data to remove
1464 *
1465 *	This function removes data from the start of a buffer, returning
1466 *	the memory to the headroom. A pointer to the next data in the buffer
1467 *	is returned. Once the data has been pulled future pushes will overwrite
1468 *	the old data.
1469 */
1470unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1471{
1472	return skb_pull_inline(skb, len);
1473}
1474EXPORT_SYMBOL(skb_pull);
1475
1476/**
1477 *	skb_trim - remove end from a buffer
1478 *	@skb: buffer to alter
1479 *	@len: new length
1480 *
1481 *	Cut the length of a buffer down by removing data from the tail. If
1482 *	the buffer is already under the length specified it is not modified.
1483 *	The skb must be linear.
1484 */
1485void skb_trim(struct sk_buff *skb, unsigned int len)
1486{
1487	if (skb->len > len)
1488		__skb_trim(skb, len);
1489}
1490EXPORT_SYMBOL(skb_trim);
1491
1492/* Trims skb to length len. It can change skb pointers.
1493 */
1494
1495int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1496{
1497	struct sk_buff **fragp;
1498	struct sk_buff *frag;
1499	int offset = skb_headlen(skb);
1500	int nfrags = skb_shinfo(skb)->nr_frags;
1501	int i;
1502	int err;
1503
1504	if (skb_cloned(skb) &&
1505	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1506		return err;
1507
1508	i = 0;
1509	if (offset >= len)
1510		goto drop_pages;
1511
1512	for (; i < nfrags; i++) {
1513		int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1514
1515		if (end < len) {
1516			offset = end;
1517			continue;
1518		}
1519
1520		skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1521
1522drop_pages:
1523		skb_shinfo(skb)->nr_frags = i;
1524
1525		for (; i < nfrags; i++)
1526			skb_frag_unref(skb, i);
1527
1528		if (skb_has_frag_list(skb))
1529			skb_drop_fraglist(skb);
1530		goto done;
1531	}
1532
1533	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1534	     fragp = &frag->next) {
1535		int end = offset + frag->len;
1536
1537		if (skb_shared(frag)) {
1538			struct sk_buff *nfrag;
1539
1540			nfrag = skb_clone(frag, GFP_ATOMIC);
1541			if (unlikely(!nfrag))
1542				return -ENOMEM;
1543
1544			nfrag->next = frag->next;
1545			consume_skb(frag);
1546			frag = nfrag;
1547			*fragp = frag;
1548		}
1549
1550		if (end < len) {
1551			offset = end;
1552			continue;
1553		}
1554
1555		if (end > len &&
1556		    unlikely((err = pskb_trim(frag, len - offset))))
1557			return err;
1558
1559		if (frag->next)
1560			skb_drop_list(&frag->next);
1561		break;
1562	}
1563
1564done:
1565	if (len > skb_headlen(skb)) {
1566		skb->data_len -= skb->len - len;
1567		skb->len       = len;
1568	} else {
1569		skb->len       = len;
1570		skb->data_len  = 0;
1571		skb_set_tail_pointer(skb, len);
1572	}
1573
 
 
1574	return 0;
1575}
1576EXPORT_SYMBOL(___pskb_trim);
1577
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1578/**
1579 *	__pskb_pull_tail - advance tail of skb header
1580 *	@skb: buffer to reallocate
1581 *	@delta: number of bytes to advance tail
1582 *
1583 *	The function makes a sense only on a fragmented &sk_buff,
1584 *	it expands header moving its tail forward and copying necessary
1585 *	data from fragmented part.
1586 *
1587 *	&sk_buff MUST have reference count of 1.
1588 *
1589 *	Returns %NULL (and &sk_buff does not change) if pull failed
1590 *	or value of new tail of skb in the case of success.
1591 *
1592 *	All the pointers pointing into skb header may change and must be
1593 *	reloaded after call to this function.
1594 */
1595
1596/* Moves tail of skb head forward, copying data from fragmented part,
1597 * when it is necessary.
1598 * 1. It may fail due to malloc failure.
1599 * 2. It may change skb pointers.
1600 *
1601 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1602 */
1603unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1604{
1605	/* If skb has not enough free space at tail, get new one
1606	 * plus 128 bytes for future expansions. If we have enough
1607	 * room at tail, reallocate without expansion only if skb is cloned.
1608	 */
1609	int i, k, eat = (skb->tail + delta) - skb->end;
1610
1611	if (eat > 0 || skb_cloned(skb)) {
1612		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1613				     GFP_ATOMIC))
1614			return NULL;
1615	}
1616
1617	if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1618		BUG();
1619
1620	/* Optimization: no fragments, no reasons to preestimate
1621	 * size of pulled pages. Superb.
1622	 */
1623	if (!skb_has_frag_list(skb))
1624		goto pull_pages;
1625
1626	/* Estimate size of pulled pages. */
1627	eat = delta;
1628	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1629		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1630
1631		if (size >= eat)
1632			goto pull_pages;
1633		eat -= size;
1634	}
1635
1636	/* If we need update frag list, we are in troubles.
1637	 * Certainly, it possible to add an offset to skb data,
1638	 * but taking into account that pulling is expected to
1639	 * be very rare operation, it is worth to fight against
1640	 * further bloating skb head and crucify ourselves here instead.
1641	 * Pure masohism, indeed. 8)8)
1642	 */
1643	if (eat) {
1644		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1645		struct sk_buff *clone = NULL;
1646		struct sk_buff *insp = NULL;
1647
1648		do {
1649			BUG_ON(!list);
1650
1651			if (list->len <= eat) {
1652				/* Eaten as whole. */
1653				eat -= list->len;
1654				list = list->next;
1655				insp = list;
1656			} else {
1657				/* Eaten partially. */
1658
1659				if (skb_shared(list)) {
1660					/* Sucks! We need to fork list. :-( */
1661					clone = skb_clone(list, GFP_ATOMIC);
1662					if (!clone)
1663						return NULL;
1664					insp = list->next;
1665					list = clone;
1666				} else {
1667					/* This may be pulled without
1668					 * problems. */
1669					insp = list;
1670				}
1671				if (!pskb_pull(list, eat)) {
1672					kfree_skb(clone);
1673					return NULL;
1674				}
1675				break;
1676			}
1677		} while (eat);
1678
1679		/* Free pulled out fragments. */
1680		while ((list = skb_shinfo(skb)->frag_list) != insp) {
1681			skb_shinfo(skb)->frag_list = list->next;
1682			kfree_skb(list);
1683		}
1684		/* And insert new clone at head. */
1685		if (clone) {
1686			clone->next = list;
1687			skb_shinfo(skb)->frag_list = clone;
1688		}
1689	}
1690	/* Success! Now we may commit changes to skb data. */
1691
1692pull_pages:
1693	eat = delta;
1694	k = 0;
1695	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1696		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1697
1698		if (size <= eat) {
1699			skb_frag_unref(skb, i);
1700			eat -= size;
1701		} else {
1702			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
 
 
1703			if (eat) {
1704				skb_shinfo(skb)->frags[k].page_offset += eat;
1705				skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
 
 
1706				eat = 0;
1707			}
1708			k++;
1709		}
1710	}
1711	skb_shinfo(skb)->nr_frags = k;
1712
 
1713	skb->tail     += delta;
1714	skb->data_len -= delta;
1715
 
 
 
1716	return skb_tail_pointer(skb);
1717}
1718EXPORT_SYMBOL(__pskb_pull_tail);
1719
1720/**
1721 *	skb_copy_bits - copy bits from skb to kernel buffer
1722 *	@skb: source skb
1723 *	@offset: offset in source
1724 *	@to: destination buffer
1725 *	@len: number of bytes to copy
1726 *
1727 *	Copy the specified number of bytes from the source skb to the
1728 *	destination buffer.
1729 *
1730 *	CAUTION ! :
1731 *		If its prototype is ever changed,
1732 *		check arch/{*}/net/{*}.S files,
1733 *		since it is called from BPF assembly code.
1734 */
1735int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1736{
1737	int start = skb_headlen(skb);
1738	struct sk_buff *frag_iter;
1739	int i, copy;
1740
1741	if (offset > (int)skb->len - len)
1742		goto fault;
1743
1744	/* Copy header. */
1745	if ((copy = start - offset) > 0) {
1746		if (copy > len)
1747			copy = len;
1748		skb_copy_from_linear_data_offset(skb, offset, to, copy);
1749		if ((len -= copy) == 0)
1750			return 0;
1751		offset += copy;
1752		to     += copy;
1753	}
1754
1755	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1756		int end;
1757		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1758
1759		WARN_ON(start > offset + len);
1760
1761		end = start + skb_frag_size(f);
1762		if ((copy = end - offset) > 0) {
 
 
1763			u8 *vaddr;
1764
1765			if (copy > len)
1766				copy = len;
1767
1768			vaddr = kmap_atomic(skb_frag_page(f));
1769			memcpy(to,
1770			       vaddr + f->page_offset + offset - start,
1771			       copy);
1772			kunmap_atomic(vaddr);
 
 
1773
1774			if ((len -= copy) == 0)
1775				return 0;
1776			offset += copy;
1777			to     += copy;
1778		}
1779		start = end;
1780	}
1781
1782	skb_walk_frags(skb, frag_iter) {
1783		int end;
1784
1785		WARN_ON(start > offset + len);
1786
1787		end = start + frag_iter->len;
1788		if ((copy = end - offset) > 0) {
1789			if (copy > len)
1790				copy = len;
1791			if (skb_copy_bits(frag_iter, offset - start, to, copy))
1792				goto fault;
1793			if ((len -= copy) == 0)
1794				return 0;
1795			offset += copy;
1796			to     += copy;
1797		}
1798		start = end;
1799	}
1800
1801	if (!len)
1802		return 0;
1803
1804fault:
1805	return -EFAULT;
1806}
1807EXPORT_SYMBOL(skb_copy_bits);
1808
1809/*
1810 * Callback from splice_to_pipe(), if we need to release some pages
1811 * at the end of the spd in case we error'ed out in filling the pipe.
1812 */
1813static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1814{
1815	put_page(spd->pages[i]);
1816}
1817
1818static struct page *linear_to_page(struct page *page, unsigned int *len,
1819				   unsigned int *offset,
1820				   struct sock *sk)
1821{
1822	struct page_frag *pfrag = sk_page_frag(sk);
1823
1824	if (!sk_page_frag_refill(sk, pfrag))
1825		return NULL;
1826
1827	*len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1828
1829	memcpy(page_address(pfrag->page) + pfrag->offset,
1830	       page_address(page) + *offset, *len);
1831	*offset = pfrag->offset;
1832	pfrag->offset += *len;
1833
1834	return pfrag->page;
1835}
1836
1837static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1838			     struct page *page,
1839			     unsigned int offset)
1840{
1841	return	spd->nr_pages &&
1842		spd->pages[spd->nr_pages - 1] == page &&
1843		(spd->partial[spd->nr_pages - 1].offset +
1844		 spd->partial[spd->nr_pages - 1].len == offset);
1845}
1846
1847/*
1848 * Fill page/offset/length into spd, if it can hold more pages.
1849 */
1850static bool spd_fill_page(struct splice_pipe_desc *spd,
1851			  struct pipe_inode_info *pipe, struct page *page,
1852			  unsigned int *len, unsigned int offset,
1853			  bool linear,
1854			  struct sock *sk)
1855{
1856	if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1857		return true;
1858
1859	if (linear) {
1860		page = linear_to_page(page, len, &offset, sk);
1861		if (!page)
1862			return true;
1863	}
1864	if (spd_can_coalesce(spd, page, offset)) {
1865		spd->partial[spd->nr_pages - 1].len += *len;
1866		return false;
1867	}
1868	get_page(page);
1869	spd->pages[spd->nr_pages] = page;
1870	spd->partial[spd->nr_pages].len = *len;
1871	spd->partial[spd->nr_pages].offset = offset;
1872	spd->nr_pages++;
1873
1874	return false;
1875}
1876
1877static bool __splice_segment(struct page *page, unsigned int poff,
1878			     unsigned int plen, unsigned int *off,
1879			     unsigned int *len,
1880			     struct splice_pipe_desc *spd, bool linear,
1881			     struct sock *sk,
1882			     struct pipe_inode_info *pipe)
1883{
1884	if (!*len)
1885		return true;
1886
1887	/* skip this segment if already processed */
1888	if (*off >= plen) {
1889		*off -= plen;
1890		return false;
1891	}
1892
1893	/* ignore any bits we already processed */
1894	poff += *off;
1895	plen -= *off;
1896	*off = 0;
1897
1898	do {
1899		unsigned int flen = min(*len, plen);
1900
1901		if (spd_fill_page(spd, pipe, page, &flen, poff,
1902				  linear, sk))
1903			return true;
1904		poff += flen;
1905		plen -= flen;
1906		*len -= flen;
1907	} while (*len && plen);
1908
1909	return false;
1910}
1911
1912/*
1913 * Map linear and fragment data from the skb to spd. It reports true if the
1914 * pipe is full or if we already spliced the requested length.
1915 */
1916static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1917			      unsigned int *offset, unsigned int *len,
1918			      struct splice_pipe_desc *spd, struct sock *sk)
1919{
1920	int seg;
1921	struct sk_buff *iter;
1922
1923	/* map the linear part :
1924	 * If skb->head_frag is set, this 'linear' part is backed by a
1925	 * fragment, and if the head is not shared with any clones then
1926	 * we can avoid a copy since we own the head portion of this page.
1927	 */
1928	if (__splice_segment(virt_to_page(skb->data),
1929			     (unsigned long) skb->data & (PAGE_SIZE - 1),
1930			     skb_headlen(skb),
1931			     offset, len, spd,
1932			     skb_head_is_locked(skb),
1933			     sk, pipe))
1934		return true;
1935
1936	/*
1937	 * then map the fragments
1938	 */
1939	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1940		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1941
1942		if (__splice_segment(skb_frag_page(f),
1943				     f->page_offset, skb_frag_size(f),
1944				     offset, len, spd, false, sk, pipe))
1945			return true;
1946	}
1947
1948	skb_walk_frags(skb, iter) {
1949		if (*offset >= iter->len) {
1950			*offset -= iter->len;
1951			continue;
1952		}
1953		/* __skb_splice_bits() only fails if the output has no room
1954		 * left, so no point in going over the frag_list for the error
1955		 * case.
1956		 */
1957		if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1958			return true;
1959	}
1960
1961	return false;
1962}
1963
1964ssize_t skb_socket_splice(struct sock *sk,
1965			  struct pipe_inode_info *pipe,
1966			  struct splice_pipe_desc *spd)
1967{
1968	int ret;
1969
1970	/* Drop the socket lock, otherwise we have reverse
1971	 * locking dependencies between sk_lock and i_mutex
1972	 * here as compared to sendfile(). We enter here
1973	 * with the socket lock held, and splice_to_pipe() will
1974	 * grab the pipe inode lock. For sendfile() emulation,
1975	 * we call into ->sendpage() with the i_mutex lock held
1976	 * and networking will grab the socket lock.
1977	 */
1978	release_sock(sk);
1979	ret = splice_to_pipe(pipe, spd);
1980	lock_sock(sk);
1981
1982	return ret;
1983}
1984
1985/*
1986 * Map data from the skb to a pipe. Should handle both the linear part,
1987 * the fragments, and the frag list.
1988 */
1989int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1990		    struct pipe_inode_info *pipe, unsigned int tlen,
1991		    unsigned int flags,
1992		    ssize_t (*splice_cb)(struct sock *,
1993					 struct pipe_inode_info *,
1994					 struct splice_pipe_desc *))
1995{
1996	struct partial_page partial[MAX_SKB_FRAGS];
1997	struct page *pages[MAX_SKB_FRAGS];
1998	struct splice_pipe_desc spd = {
1999		.pages = pages,
2000		.partial = partial,
2001		.nr_pages_max = MAX_SKB_FRAGS,
2002		.flags = flags,
2003		.ops = &nosteal_pipe_buf_ops,
2004		.spd_release = sock_spd_release,
2005	};
2006	int ret = 0;
2007
2008	__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2009
2010	if (spd.nr_pages)
2011		ret = splice_cb(sk, pipe, &spd);
2012
2013	return ret;
2014}
2015EXPORT_SYMBOL_GPL(skb_splice_bits);
2016
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2017/**
2018 *	skb_store_bits - store bits from kernel buffer to skb
2019 *	@skb: destination buffer
2020 *	@offset: offset in destination
2021 *	@from: source buffer
2022 *	@len: number of bytes to copy
2023 *
2024 *	Copy the specified number of bytes from the source buffer to the
2025 *	destination skb.  This function handles all the messy bits of
2026 *	traversing fragment lists and such.
2027 */
2028
2029int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2030{
2031	int start = skb_headlen(skb);
2032	struct sk_buff *frag_iter;
2033	int i, copy;
2034
2035	if (offset > (int)skb->len - len)
2036		goto fault;
2037
2038	if ((copy = start - offset) > 0) {
2039		if (copy > len)
2040			copy = len;
2041		skb_copy_to_linear_data_offset(skb, offset, from, copy);
2042		if ((len -= copy) == 0)
2043			return 0;
2044		offset += copy;
2045		from += copy;
2046	}
2047
2048	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2049		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2050		int end;
2051
2052		WARN_ON(start > offset + len);
2053
2054		end = start + skb_frag_size(frag);
2055		if ((copy = end - offset) > 0) {
 
 
2056			u8 *vaddr;
2057
2058			if (copy > len)
2059				copy = len;
2060
2061			vaddr = kmap_atomic(skb_frag_page(frag));
2062			memcpy(vaddr + frag->page_offset + offset - start,
2063			       from, copy);
2064			kunmap_atomic(vaddr);
 
 
 
2065
2066			if ((len -= copy) == 0)
2067				return 0;
2068			offset += copy;
2069			from += copy;
2070		}
2071		start = end;
2072	}
2073
2074	skb_walk_frags(skb, frag_iter) {
2075		int end;
2076
2077		WARN_ON(start > offset + len);
2078
2079		end = start + frag_iter->len;
2080		if ((copy = end - offset) > 0) {
2081			if (copy > len)
2082				copy = len;
2083			if (skb_store_bits(frag_iter, offset - start,
2084					   from, copy))
2085				goto fault;
2086			if ((len -= copy) == 0)
2087				return 0;
2088			offset += copy;
2089			from += copy;
2090		}
2091		start = end;
2092	}
2093	if (!len)
2094		return 0;
2095
2096fault:
2097	return -EFAULT;
2098}
2099EXPORT_SYMBOL(skb_store_bits);
2100
2101/* Checksum skb data. */
2102__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2103		      __wsum csum, const struct skb_checksum_ops *ops)
2104{
2105	int start = skb_headlen(skb);
2106	int i, copy = start - offset;
2107	struct sk_buff *frag_iter;
2108	int pos = 0;
2109
2110	/* Checksum header. */
2111	if (copy > 0) {
2112		if (copy > len)
2113			copy = len;
2114		csum = ops->update(skb->data + offset, copy, csum);
 
2115		if ((len -= copy) == 0)
2116			return csum;
2117		offset += copy;
2118		pos	= copy;
2119	}
2120
2121	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2122		int end;
2123		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2124
2125		WARN_ON(start > offset + len);
2126
2127		end = start + skb_frag_size(frag);
2128		if ((copy = end - offset) > 0) {
 
 
2129			__wsum csum2;
2130			u8 *vaddr;
2131
2132			if (copy > len)
2133				copy = len;
2134			vaddr = kmap_atomic(skb_frag_page(frag));
2135			csum2 = ops->update(vaddr + frag->page_offset +
2136					    offset - start, copy, 0);
2137			kunmap_atomic(vaddr);
2138			csum = ops->combine(csum, csum2, pos, copy);
 
 
 
 
 
 
 
 
 
 
2139			if (!(len -= copy))
2140				return csum;
2141			offset += copy;
2142			pos    += copy;
2143		}
2144		start = end;
2145	}
2146
2147	skb_walk_frags(skb, frag_iter) {
2148		int end;
2149
2150		WARN_ON(start > offset + len);
2151
2152		end = start + frag_iter->len;
2153		if ((copy = end - offset) > 0) {
2154			__wsum csum2;
2155			if (copy > len)
2156				copy = len;
2157			csum2 = __skb_checksum(frag_iter, offset - start,
2158					       copy, 0, ops);
2159			csum = ops->combine(csum, csum2, pos, copy);
 
2160			if ((len -= copy) == 0)
2161				return csum;
2162			offset += copy;
2163			pos    += copy;
2164		}
2165		start = end;
2166	}
2167	BUG_ON(len);
2168
2169	return csum;
2170}
2171EXPORT_SYMBOL(__skb_checksum);
2172
2173__wsum skb_checksum(const struct sk_buff *skb, int offset,
2174		    int len, __wsum csum)
2175{
2176	const struct skb_checksum_ops ops = {
2177		.update  = csum_partial_ext,
2178		.combine = csum_block_add_ext,
2179	};
2180
2181	return __skb_checksum(skb, offset, len, csum, &ops);
2182}
2183EXPORT_SYMBOL(skb_checksum);
2184
2185/* Both of above in one bottle. */
2186
2187__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2188				    u8 *to, int len, __wsum csum)
2189{
2190	int start = skb_headlen(skb);
2191	int i, copy = start - offset;
2192	struct sk_buff *frag_iter;
2193	int pos = 0;
2194
2195	/* Copy header. */
2196	if (copy > 0) {
2197		if (copy > len)
2198			copy = len;
2199		csum = csum_partial_copy_nocheck(skb->data + offset, to,
2200						 copy, csum);
2201		if ((len -= copy) == 0)
2202			return csum;
2203		offset += copy;
2204		to     += copy;
2205		pos	= copy;
2206	}
2207
2208	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209		int end;
2210
2211		WARN_ON(start > offset + len);
2212
2213		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2214		if ((copy = end - offset) > 0) {
 
 
 
2215			__wsum csum2;
2216			u8 *vaddr;
2217			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2218
2219			if (copy > len)
2220				copy = len;
2221			vaddr = kmap_atomic(skb_frag_page(frag));
2222			csum2 = csum_partial_copy_nocheck(vaddr +
2223							  frag->page_offset +
2224							  offset - start, to,
2225							  copy, 0);
2226			kunmap_atomic(vaddr);
2227			csum = csum_block_add(csum, csum2, pos);
 
 
 
 
 
 
2228			if (!(len -= copy))
2229				return csum;
2230			offset += copy;
2231			to     += copy;
2232			pos    += copy;
2233		}
2234		start = end;
2235	}
2236
2237	skb_walk_frags(skb, frag_iter) {
2238		__wsum csum2;
2239		int end;
2240
2241		WARN_ON(start > offset + len);
2242
2243		end = start + frag_iter->len;
2244		if ((copy = end - offset) > 0) {
2245			if (copy > len)
2246				copy = len;
2247			csum2 = skb_copy_and_csum_bits(frag_iter,
2248						       offset - start,
2249						       to, copy, 0);
2250			csum = csum_block_add(csum, csum2, pos);
2251			if ((len -= copy) == 0)
2252				return csum;
2253			offset += copy;
2254			to     += copy;
2255			pos    += copy;
2256		}
2257		start = end;
2258	}
2259	BUG_ON(len);
2260	return csum;
2261}
2262EXPORT_SYMBOL(skb_copy_and_csum_bits);
2263
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2264 /**
2265 *	skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2266 *	@from: source buffer
2267 *
2268 *	Calculates the amount of linear headroom needed in the 'to' skb passed
2269 *	into skb_zerocopy().
2270 */
2271unsigned int
2272skb_zerocopy_headlen(const struct sk_buff *from)
2273{
2274	unsigned int hlen = 0;
2275
2276	if (!from->head_frag ||
2277	    skb_headlen(from) < L1_CACHE_BYTES ||
2278	    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2279		hlen = skb_headlen(from);
2280
2281	if (skb_has_frag_list(from))
2282		hlen = from->len;
2283
2284	return hlen;
2285}
2286EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2287
2288/**
2289 *	skb_zerocopy - Zero copy skb to skb
2290 *	@to: destination buffer
2291 *	@from: source buffer
2292 *	@len: number of bytes to copy from source buffer
2293 *	@hlen: size of linear headroom in destination buffer
2294 *
2295 *	Copies up to `len` bytes from `from` to `to` by creating references
2296 *	to the frags in the source buffer.
2297 *
2298 *	The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2299 *	headroom in the `to` buffer.
2300 *
2301 *	Return value:
2302 *	0: everything is OK
2303 *	-ENOMEM: couldn't orphan frags of @from due to lack of memory
2304 *	-EFAULT: skb_copy_bits() found some problem with skb geometry
2305 */
2306int
2307skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2308{
2309	int i, j = 0;
2310	int plen = 0; /* length of skb->head fragment */
2311	int ret;
2312	struct page *page;
2313	unsigned int offset;
2314
2315	BUG_ON(!from->head_frag && !hlen);
2316
2317	/* dont bother with small payloads */
2318	if (len <= skb_tailroom(to))
2319		return skb_copy_bits(from, 0, skb_put(to, len), len);
2320
2321	if (hlen) {
2322		ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2323		if (unlikely(ret))
2324			return ret;
2325		len -= hlen;
2326	} else {
2327		plen = min_t(int, skb_headlen(from), len);
2328		if (plen) {
2329			page = virt_to_head_page(from->head);
2330			offset = from->data - (unsigned char *)page_address(page);
2331			__skb_fill_page_desc(to, 0, page, offset, plen);
2332			get_page(page);
2333			j = 1;
2334			len -= plen;
2335		}
2336	}
2337
2338	to->truesize += len + plen;
2339	to->len += len + plen;
2340	to->data_len += len + plen;
2341
2342	if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2343		skb_tx_error(from);
2344		return -ENOMEM;
2345	}
 
2346
2347	for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
 
 
2348		if (!len)
2349			break;
2350		skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2351		skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2352		len -= skb_shinfo(to)->frags[j].size;
 
 
2353		skb_frag_ref(to, j);
2354		j++;
2355	}
2356	skb_shinfo(to)->nr_frags = j;
2357
2358	return 0;
2359}
2360EXPORT_SYMBOL_GPL(skb_zerocopy);
2361
2362void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2363{
2364	__wsum csum;
2365	long csstart;
2366
2367	if (skb->ip_summed == CHECKSUM_PARTIAL)
2368		csstart = skb_checksum_start_offset(skb);
2369	else
2370		csstart = skb_headlen(skb);
2371
2372	BUG_ON(csstart > skb_headlen(skb));
2373
2374	skb_copy_from_linear_data(skb, to, csstart);
2375
2376	csum = 0;
2377	if (csstart != skb->len)
2378		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2379					      skb->len - csstart, 0);
2380
2381	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2382		long csstuff = csstart + skb->csum_offset;
2383
2384		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
2385	}
2386}
2387EXPORT_SYMBOL(skb_copy_and_csum_dev);
2388
2389/**
2390 *	skb_dequeue - remove from the head of the queue
2391 *	@list: list to dequeue from
2392 *
2393 *	Remove the head of the list. The list lock is taken so the function
2394 *	may be used safely with other locking list functions. The head item is
2395 *	returned or %NULL if the list is empty.
2396 */
2397
2398struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2399{
2400	unsigned long flags;
2401	struct sk_buff *result;
2402
2403	spin_lock_irqsave(&list->lock, flags);
2404	result = __skb_dequeue(list);
2405	spin_unlock_irqrestore(&list->lock, flags);
2406	return result;
2407}
2408EXPORT_SYMBOL(skb_dequeue);
2409
2410/**
2411 *	skb_dequeue_tail - remove from the tail of the queue
2412 *	@list: list to dequeue from
2413 *
2414 *	Remove the tail of the list. The list lock is taken so the function
2415 *	may be used safely with other locking list functions. The tail item is
2416 *	returned or %NULL if the list is empty.
2417 */
2418struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2419{
2420	unsigned long flags;
2421	struct sk_buff *result;
2422
2423	spin_lock_irqsave(&list->lock, flags);
2424	result = __skb_dequeue_tail(list);
2425	spin_unlock_irqrestore(&list->lock, flags);
2426	return result;
2427}
2428EXPORT_SYMBOL(skb_dequeue_tail);
2429
2430/**
2431 *	skb_queue_purge - empty a list
2432 *	@list: list to empty
2433 *
2434 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2435 *	the list and one reference dropped. This function takes the list
2436 *	lock and is atomic with respect to other list locking functions.
2437 */
2438void skb_queue_purge(struct sk_buff_head *list)
2439{
2440	struct sk_buff *skb;
2441	while ((skb = skb_dequeue(list)) != NULL)
2442		kfree_skb(skb);
2443}
2444EXPORT_SYMBOL(skb_queue_purge);
2445
2446/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2447 *	skb_queue_head - queue a buffer at the list head
2448 *	@list: list to use
2449 *	@newsk: buffer to queue
2450 *
2451 *	Queue a buffer at the start of the list. This function takes the
2452 *	list lock and can be used safely with other locking &sk_buff functions
2453 *	safely.
2454 *
2455 *	A buffer cannot be placed on two lists at the same time.
2456 */
2457void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2458{
2459	unsigned long flags;
2460
2461	spin_lock_irqsave(&list->lock, flags);
2462	__skb_queue_head(list, newsk);
2463	spin_unlock_irqrestore(&list->lock, flags);
2464}
2465EXPORT_SYMBOL(skb_queue_head);
2466
2467/**
2468 *	skb_queue_tail - queue a buffer at the list tail
2469 *	@list: list to use
2470 *	@newsk: buffer to queue
2471 *
2472 *	Queue a buffer at the tail of the list. This function takes the
2473 *	list lock and can be used safely with other locking &sk_buff functions
2474 *	safely.
2475 *
2476 *	A buffer cannot be placed on two lists at the same time.
2477 */
2478void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2479{
2480	unsigned long flags;
2481
2482	spin_lock_irqsave(&list->lock, flags);
2483	__skb_queue_tail(list, newsk);
2484	spin_unlock_irqrestore(&list->lock, flags);
2485}
2486EXPORT_SYMBOL(skb_queue_tail);
2487
2488/**
2489 *	skb_unlink	-	remove a buffer from a list
2490 *	@skb: buffer to remove
2491 *	@list: list to use
2492 *
2493 *	Remove a packet from a list. The list locks are taken and this
2494 *	function is atomic with respect to other list locked calls
2495 *
2496 *	You must know what list the SKB is on.
2497 */
2498void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2499{
2500	unsigned long flags;
2501
2502	spin_lock_irqsave(&list->lock, flags);
2503	__skb_unlink(skb, list);
2504	spin_unlock_irqrestore(&list->lock, flags);
2505}
2506EXPORT_SYMBOL(skb_unlink);
2507
2508/**
2509 *	skb_append	-	append a buffer
2510 *	@old: buffer to insert after
2511 *	@newsk: buffer to insert
2512 *	@list: list to use
2513 *
2514 *	Place a packet after a given packet in a list. The list locks are taken
2515 *	and this function is atomic with respect to other list locked calls.
2516 *	A buffer cannot be placed on two lists at the same time.
2517 */
2518void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2519{
2520	unsigned long flags;
2521
2522	spin_lock_irqsave(&list->lock, flags);
2523	__skb_queue_after(list, old, newsk);
2524	spin_unlock_irqrestore(&list->lock, flags);
2525}
2526EXPORT_SYMBOL(skb_append);
2527
2528/**
2529 *	skb_insert	-	insert a buffer
2530 *	@old: buffer to insert before
2531 *	@newsk: buffer to insert
2532 *	@list: list to use
2533 *
2534 *	Place a packet before a given packet in a list. The list locks are
2535 * 	taken and this function is atomic with respect to other list locked
2536 *	calls.
2537 *
2538 *	A buffer cannot be placed on two lists at the same time.
2539 */
2540void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2541{
2542	unsigned long flags;
2543
2544	spin_lock_irqsave(&list->lock, flags);
2545	__skb_insert(newsk, old->prev, old, list);
2546	spin_unlock_irqrestore(&list->lock, flags);
2547}
2548EXPORT_SYMBOL(skb_insert);
2549
2550static inline void skb_split_inside_header(struct sk_buff *skb,
2551					   struct sk_buff* skb1,
2552					   const u32 len, const int pos)
2553{
2554	int i;
2555
2556	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2557					 pos - len);
2558	/* And move data appendix as is. */
2559	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2560		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2561
2562	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2563	skb_shinfo(skb)->nr_frags  = 0;
2564	skb1->data_len		   = skb->data_len;
2565	skb1->len		   += skb1->data_len;
2566	skb->data_len		   = 0;
2567	skb->len		   = len;
2568	skb_set_tail_pointer(skb, len);
2569}
2570
2571static inline void skb_split_no_header(struct sk_buff *skb,
2572				       struct sk_buff* skb1,
2573				       const u32 len, int pos)
2574{
2575	int i, k = 0;
2576	const int nfrags = skb_shinfo(skb)->nr_frags;
2577
2578	skb_shinfo(skb)->nr_frags = 0;
2579	skb1->len		  = skb1->data_len = skb->len - len;
2580	skb->len		  = len;
2581	skb->data_len		  = len - pos;
2582
2583	for (i = 0; i < nfrags; i++) {
2584		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2585
2586		if (pos + size > len) {
2587			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2588
2589			if (pos < len) {
2590				/* Split frag.
2591				 * We have two variants in this case:
2592				 * 1. Move all the frag to the second
2593				 *    part, if it is possible. F.e.
2594				 *    this approach is mandatory for TUX,
2595				 *    where splitting is expensive.
2596				 * 2. Split is accurately. We make this.
2597				 */
2598				skb_frag_ref(skb, i);
2599				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2600				skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2601				skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2602				skb_shinfo(skb)->nr_frags++;
2603			}
2604			k++;
2605		} else
2606			skb_shinfo(skb)->nr_frags++;
2607		pos += size;
2608	}
2609	skb_shinfo(skb1)->nr_frags = k;
2610}
2611
2612/**
2613 * skb_split - Split fragmented skb to two parts at length len.
2614 * @skb: the buffer to split
2615 * @skb1: the buffer to receive the second part
2616 * @len: new length for skb
2617 */
2618void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2619{
2620	int pos = skb_headlen(skb);
2621
2622	skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
 
 
2623	if (len < pos)	/* Split line is inside header. */
2624		skb_split_inside_header(skb, skb1, len, pos);
2625	else		/* Second chunk has no header, nothing to copy. */
2626		skb_split_no_header(skb, skb1, len, pos);
2627}
2628EXPORT_SYMBOL(skb_split);
2629
2630/* Shifting from/to a cloned skb is a no-go.
2631 *
2632 * Caller cannot keep skb_shinfo related pointers past calling here!
2633 */
2634static int skb_prepare_for_shift(struct sk_buff *skb)
2635{
2636	return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2637}
2638
2639/**
2640 * skb_shift - Shifts paged data partially from skb to another
2641 * @tgt: buffer into which tail data gets added
2642 * @skb: buffer from which the paged data comes from
2643 * @shiftlen: shift up to this many bytes
2644 *
2645 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2646 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2647 * It's up to caller to free skb if everything was shifted.
2648 *
2649 * If @tgt runs out of frags, the whole operation is aborted.
2650 *
2651 * Skb cannot include anything else but paged data while tgt is allowed
2652 * to have non-paged data as well.
2653 *
2654 * TODO: full sized shift could be optimized but that would need
2655 * specialized skb free'er to handle frags without up-to-date nr_frags.
2656 */
2657int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2658{
2659	int from, to, merge, todo;
2660	struct skb_frag_struct *fragfrom, *fragto;
2661
2662	BUG_ON(shiftlen > skb->len);
2663	BUG_ON(skb_headlen(skb));	/* Would corrupt stream */
 
 
 
 
2664
2665	todo = shiftlen;
2666	from = 0;
2667	to = skb_shinfo(tgt)->nr_frags;
2668	fragfrom = &skb_shinfo(skb)->frags[from];
2669
2670	/* Actual merge is delayed until the point when we know we can
2671	 * commit all, so that we don't have to undo partial changes
2672	 */
2673	if (!to ||
2674	    !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675			      fragfrom->page_offset)) {
2676		merge = -1;
2677	} else {
2678		merge = to - 1;
2679
2680		todo -= skb_frag_size(fragfrom);
2681		if (todo < 0) {
2682			if (skb_prepare_for_shift(skb) ||
2683			    skb_prepare_for_shift(tgt))
2684				return 0;
2685
2686			/* All previous frag pointers might be stale! */
2687			fragfrom = &skb_shinfo(skb)->frags[from];
2688			fragto = &skb_shinfo(tgt)->frags[merge];
2689
2690			skb_frag_size_add(fragto, shiftlen);
2691			skb_frag_size_sub(fragfrom, shiftlen);
2692			fragfrom->page_offset += shiftlen;
2693
2694			goto onlymerged;
2695		}
2696
2697		from++;
2698	}
2699
2700	/* Skip full, not-fitting skb to avoid expensive operations */
2701	if ((shiftlen == skb->len) &&
2702	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2703		return 0;
2704
2705	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2706		return 0;
2707
2708	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709		if (to == MAX_SKB_FRAGS)
2710			return 0;
2711
2712		fragfrom = &skb_shinfo(skb)->frags[from];
2713		fragto = &skb_shinfo(tgt)->frags[to];
2714
2715		if (todo >= skb_frag_size(fragfrom)) {
2716			*fragto = *fragfrom;
2717			todo -= skb_frag_size(fragfrom);
2718			from++;
2719			to++;
2720
2721		} else {
2722			__skb_frag_ref(fragfrom);
2723			fragto->page = fragfrom->page;
2724			fragto->page_offset = fragfrom->page_offset;
2725			skb_frag_size_set(fragto, todo);
2726
2727			fragfrom->page_offset += todo;
2728			skb_frag_size_sub(fragfrom, todo);
2729			todo = 0;
2730
2731			to++;
2732			break;
2733		}
2734	}
2735
2736	/* Ready to "commit" this state change to tgt */
2737	skb_shinfo(tgt)->nr_frags = to;
2738
2739	if (merge >= 0) {
2740		fragfrom = &skb_shinfo(skb)->frags[0];
2741		fragto = &skb_shinfo(tgt)->frags[merge];
2742
2743		skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744		__skb_frag_unref(fragfrom);
2745	}
2746
2747	/* Reposition in the original skb */
2748	to = 0;
2749	while (from < skb_shinfo(skb)->nr_frags)
2750		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751	skb_shinfo(skb)->nr_frags = to;
2752
2753	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2754
2755onlymerged:
2756	/* Most likely the tgt won't ever need its checksum anymore, skb on
2757	 * the other hand might need it if it needs to be resent
2758	 */
2759	tgt->ip_summed = CHECKSUM_PARTIAL;
2760	skb->ip_summed = CHECKSUM_PARTIAL;
2761
2762	/* Yak, is it really working this way? Some helper please? */
2763	skb->len -= shiftlen;
2764	skb->data_len -= shiftlen;
2765	skb->truesize -= shiftlen;
2766	tgt->len += shiftlen;
2767	tgt->data_len += shiftlen;
2768	tgt->truesize += shiftlen;
2769
2770	return shiftlen;
2771}
2772
2773/**
2774 * skb_prepare_seq_read - Prepare a sequential read of skb data
2775 * @skb: the buffer to read
2776 * @from: lower offset of data to be read
2777 * @to: upper offset of data to be read
2778 * @st: state variable
2779 *
2780 * Initializes the specified state variable. Must be called before
2781 * invoking skb_seq_read() for the first time.
2782 */
2783void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784			  unsigned int to, struct skb_seq_state *st)
2785{
2786	st->lower_offset = from;
2787	st->upper_offset = to;
2788	st->root_skb = st->cur_skb = skb;
2789	st->frag_idx = st->stepped_offset = 0;
2790	st->frag_data = NULL;
2791}
2792EXPORT_SYMBOL(skb_prepare_seq_read);
2793
2794/**
2795 * skb_seq_read - Sequentially read skb data
2796 * @consumed: number of bytes consumed by the caller so far
2797 * @data: destination pointer for data to be returned
2798 * @st: state variable
2799 *
2800 * Reads a block of skb data at @consumed relative to the
2801 * lower offset specified to skb_prepare_seq_read(). Assigns
2802 * the head of the data block to @data and returns the length
2803 * of the block or 0 if the end of the skb data or the upper
2804 * offset has been reached.
2805 *
2806 * The caller is not required to consume all of the data
2807 * returned, i.e. @consumed is typically set to the number
2808 * of bytes already consumed and the next call to
2809 * skb_seq_read() will return the remaining part of the block.
2810 *
2811 * Note 1: The size of each block of data returned can be arbitrary,
2812 *       this limitation is the cost for zerocopy sequential
2813 *       reads of potentially non linear data.
2814 *
2815 * Note 2: Fragment lists within fragments are not implemented
2816 *       at the moment, state->root_skb could be replaced with
2817 *       a stack for this purpose.
2818 */
2819unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820			  struct skb_seq_state *st)
2821{
2822	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2823	skb_frag_t *frag;
2824
2825	if (unlikely(abs_offset >= st->upper_offset)) {
2826		if (st->frag_data) {
2827			kunmap_atomic(st->frag_data);
2828			st->frag_data = NULL;
2829		}
2830		return 0;
2831	}
2832
2833next_skb:
2834	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2835
2836	if (abs_offset < block_limit && !st->frag_data) {
2837		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838		return block_limit - abs_offset;
2839	}
2840
2841	if (st->frag_idx == 0 && !st->frag_data)
2842		st->stepped_offset += skb_headlen(st->cur_skb);
2843
2844	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846		block_limit = skb_frag_size(frag) + st->stepped_offset;
2847
2848		if (abs_offset < block_limit) {
2849			if (!st->frag_data)
2850				st->frag_data = kmap_atomic(skb_frag_page(frag));
2851
2852			*data = (u8 *) st->frag_data + frag->page_offset +
2853				(abs_offset - st->stepped_offset);
2854
2855			return block_limit - abs_offset;
2856		}
2857
2858		if (st->frag_data) {
2859			kunmap_atomic(st->frag_data);
2860			st->frag_data = NULL;
2861		}
2862
2863		st->frag_idx++;
2864		st->stepped_offset += skb_frag_size(frag);
2865	}
2866
2867	if (st->frag_data) {
2868		kunmap_atomic(st->frag_data);
2869		st->frag_data = NULL;
2870	}
2871
2872	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2874		st->frag_idx = 0;
2875		goto next_skb;
2876	} else if (st->cur_skb->next) {
2877		st->cur_skb = st->cur_skb->next;
2878		st->frag_idx = 0;
2879		goto next_skb;
2880	}
2881
2882	return 0;
2883}
2884EXPORT_SYMBOL(skb_seq_read);
2885
2886/**
2887 * skb_abort_seq_read - Abort a sequential read of skb data
2888 * @st: state variable
2889 *
2890 * Must be called if skb_seq_read() was not called until it
2891 * returned 0.
2892 */
2893void skb_abort_seq_read(struct skb_seq_state *st)
2894{
2895	if (st->frag_data)
2896		kunmap_atomic(st->frag_data);
2897}
2898EXPORT_SYMBOL(skb_abort_seq_read);
2899
2900#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
2901
2902static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903					  struct ts_config *conf,
2904					  struct ts_state *state)
2905{
2906	return skb_seq_read(offset, text, TS_SKB_CB(state));
2907}
2908
2909static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2910{
2911	skb_abort_seq_read(TS_SKB_CB(state));
2912}
2913
2914/**
2915 * skb_find_text - Find a text pattern in skb data
2916 * @skb: the buffer to look in
2917 * @from: search offset
2918 * @to: search limit
2919 * @config: textsearch configuration
2920 *
2921 * Finds a pattern in the skb data according to the specified
2922 * textsearch configuration. Use textsearch_next() to retrieve
2923 * subsequent occurrences of the pattern. Returns the offset
2924 * to the first occurrence or UINT_MAX if no match was found.
2925 */
2926unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927			   unsigned int to, struct ts_config *config)
2928{
2929	struct ts_state state;
2930	unsigned int ret;
2931
2932	config->get_next_block = skb_ts_get_next_block;
2933	config->finish = skb_ts_finish;
2934
2935	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2936
2937	ret = textsearch_find(config, &state);
2938	return (ret <= to - from ? ret : UINT_MAX);
2939}
2940EXPORT_SYMBOL(skb_find_text);
2941
2942/**
2943 * skb_append_datato_frags - append the user data to a skb
2944 * @sk: sock  structure
2945 * @skb: skb structure to be appended with user data.
2946 * @getfrag: call back function to be used for getting the user data
2947 * @from: pointer to user message iov
2948 * @length: length of the iov message
2949 *
2950 * Description: This procedure append the user data in the fragment part
2951 * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2952 */
2953int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954			int (*getfrag)(void *from, char *to, int offset,
2955					int len, int odd, struct sk_buff *skb),
2956			void *from, int length)
2957{
2958	int frg_cnt = skb_shinfo(skb)->nr_frags;
2959	int copy;
2960	int offset = 0;
2961	int ret;
2962	struct page_frag *pfrag = &current->task_frag;
2963
2964	do {
2965		/* Return error if we don't have space for new frag */
2966		if (frg_cnt >= MAX_SKB_FRAGS)
2967			return -EMSGSIZE;
2968
2969		if (!sk_page_frag_refill(sk, pfrag))
2970			return -ENOMEM;
2971
2972		/* copy the user data to page */
2973		copy = min_t(int, length, pfrag->size - pfrag->offset);
2974
2975		ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976			      offset, copy, 0, skb);
2977		if (ret < 0)
2978			return -EFAULT;
2979
2980		/* copy was successful so update the size parameters */
2981		skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2982				   copy);
2983		frg_cnt++;
2984		pfrag->offset += copy;
2985		get_page(pfrag->page);
2986
2987		skb->truesize += copy;
2988		atomic_add(copy, &sk->sk_wmem_alloc);
2989		skb->len += copy;
2990		skb->data_len += copy;
2991		offset += copy;
2992		length -= copy;
2993
2994	} while (length > 0);
2995
2996	return 0;
2997}
2998EXPORT_SYMBOL(skb_append_datato_frags);
2999
3000int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001			 int offset, size_t size)
3002{
3003	int i = skb_shinfo(skb)->nr_frags;
3004
3005	if (skb_can_coalesce(skb, i, page, offset)) {
3006		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007	} else if (i < MAX_SKB_FRAGS) {
3008		get_page(page);
3009		skb_fill_page_desc(skb, i, page, offset, size);
3010	} else {
3011		return -EMSGSIZE;
3012	}
3013
3014	return 0;
3015}
3016EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3017
3018/**
3019 *	skb_push_rcsum - push skb and update receive checksum
3020 *	@skb: buffer to update
3021 *	@len: length of data pulled
3022 *
3023 *	This function performs an skb_push on the packet and updates
3024 *	the CHECKSUM_COMPLETE checksum.  It should be used on
3025 *	receive path processing instead of skb_push unless you know
3026 *	that the checksum difference is zero (e.g., a valid IP header)
3027 *	or you are setting ip_summed to CHECKSUM_NONE.
3028 */
3029static unsigned char *skb_push_rcsum(struct sk_buff *skb, unsigned len)
3030{
3031	skb_push(skb, len);
3032	skb_postpush_rcsum(skb, skb->data, len);
3033	return skb->data;
3034}
3035
3036/**
3037 *	skb_pull_rcsum - pull skb and update receive checksum
3038 *	@skb: buffer to update
3039 *	@len: length of data pulled
3040 *
3041 *	This function performs an skb_pull on the packet and updates
3042 *	the CHECKSUM_COMPLETE checksum.  It should be used on
3043 *	receive path processing instead of skb_pull unless you know
3044 *	that the checksum difference is zero (e.g., a valid IP header)
3045 *	or you are setting ip_summed to CHECKSUM_NONE.
3046 */
3047unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3048{
3049	unsigned char *data = skb->data;
3050
3051	BUG_ON(len > skb->len);
3052	__skb_pull(skb, len);
3053	skb_postpull_rcsum(skb, data, len);
3054	return skb->data;
3055}
3056EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3057
 
 
 
 
 
 
 
 
 
 
 
 
 
3058/**
3059 *	skb_segment - Perform protocol segmentation on skb.
3060 *	@head_skb: buffer to segment
3061 *	@features: features for the output path (see dev->features)
3062 *
3063 *	This function performs segmentation on the given skb.  It returns
3064 *	a pointer to the first in a list of new skbs for the segments.
3065 *	In case of error it returns ERR_PTR(err).
3066 */
3067struct sk_buff *skb_segment(struct sk_buff *head_skb,
3068			    netdev_features_t features)
3069{
3070	struct sk_buff *segs = NULL;
3071	struct sk_buff *tail = NULL;
3072	struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3073	skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3074	unsigned int mss = skb_shinfo(head_skb)->gso_size;
3075	unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3076	struct sk_buff *frag_skb = head_skb;
3077	unsigned int offset = doffset;
3078	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
 
3079	unsigned int headroom;
3080	unsigned int len;
3081	__be16 proto;
3082	bool csum;
3083	int sg = !!(features & NETIF_F_SG);
3084	int nfrags = skb_shinfo(head_skb)->nr_frags;
3085	int err = -ENOMEM;
3086	int i = 0;
3087	int pos;
3088	int dummy;
3089
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3090	__skb_push(head_skb, doffset);
3091	proto = skb_network_protocol(head_skb, &dummy);
3092	if (unlikely(!proto))
3093		return ERR_PTR(-EINVAL);
3094
 
3095	csum = !!can_checksum_protocol(features, proto);
3096
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3097	headroom = skb_headroom(head_skb);
3098	pos = skb_headlen(head_skb);
3099
3100	do {
3101		struct sk_buff *nskb;
3102		skb_frag_t *nskb_frag;
3103		int hsize;
3104		int size;
3105
3106		len = head_skb->len - offset;
3107		if (len > mss)
3108			len = mss;
 
 
 
 
3109
3110		hsize = skb_headlen(head_skb) - offset;
3111		if (hsize < 0)
3112			hsize = 0;
3113		if (hsize > len || !sg)
3114			hsize = len;
3115
3116		if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3117		    (skb_headlen(list_skb) == len || sg)) {
3118			BUG_ON(skb_headlen(list_skb) > len);
3119
3120			i = 0;
3121			nfrags = skb_shinfo(list_skb)->nr_frags;
3122			frag = skb_shinfo(list_skb)->frags;
3123			frag_skb = list_skb;
3124			pos += skb_headlen(list_skb);
3125
3126			while (pos < offset + len) {
3127				BUG_ON(i >= nfrags);
3128
3129				size = skb_frag_size(frag);
3130				if (pos + size > offset + len)
3131					break;
3132
3133				i++;
3134				pos += size;
3135				frag++;
3136			}
3137
3138			nskb = skb_clone(list_skb, GFP_ATOMIC);
3139			list_skb = list_skb->next;
3140
3141			if (unlikely(!nskb))
3142				goto err;
3143
3144			if (unlikely(pskb_trim(nskb, len))) {
3145				kfree_skb(nskb);
3146				goto err;
3147			}
3148
3149			hsize = skb_end_offset(nskb);
3150			if (skb_cow_head(nskb, doffset + headroom)) {
3151				kfree_skb(nskb);
3152				goto err;
3153			}
3154
3155			nskb->truesize += skb_end_offset(nskb) - hsize;
3156			skb_release_head_state(nskb);
3157			__skb_push(nskb, doffset);
3158		} else {
3159			nskb = __alloc_skb(hsize + doffset + headroom,
3160					   GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3161					   NUMA_NO_NODE);
3162
3163			if (unlikely(!nskb))
3164				goto err;
3165
3166			skb_reserve(nskb, headroom);
3167			__skb_put(nskb, doffset);
3168		}
3169
3170		if (segs)
3171			tail->next = nskb;
3172		else
3173			segs = nskb;
3174		tail = nskb;
3175
3176		__copy_skb_header(nskb, head_skb);
3177
3178		skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3179		skb_reset_mac_len(nskb);
3180
3181		skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3182						 nskb->data - tnl_hlen,
3183						 doffset + tnl_hlen);
3184
3185		if (nskb->len == len + doffset)
3186			goto perform_csum_check;
3187
3188		if (!sg) {
3189			if (!nskb->remcsum_offload)
3190				nskb->ip_summed = CHECKSUM_NONE;
3191			SKB_GSO_CB(nskb)->csum =
3192				skb_copy_and_csum_bits(head_skb, offset,
3193						       skb_put(nskb, len),
3194						       len, 0);
3195			SKB_GSO_CB(nskb)->csum_start =
3196				skb_headroom(nskb) + doffset;
3197			continue;
3198		}
3199
3200		nskb_frag = skb_shinfo(nskb)->frags;
3201
3202		skb_copy_from_linear_data_offset(head_skb, offset,
3203						 skb_put(nskb, hsize), hsize);
3204
3205		skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3206			SKBTX_SHARED_FRAG;
 
 
 
 
3207
3208		while (pos < offset + len) {
3209			if (i >= nfrags) {
3210				BUG_ON(skb_headlen(list_skb));
3211
3212				i = 0;
3213				nfrags = skb_shinfo(list_skb)->nr_frags;
3214				frag = skb_shinfo(list_skb)->frags;
3215				frag_skb = list_skb;
 
 
 
 
3216
3217				BUG_ON(!nfrags);
 
 
 
 
 
 
 
3218
3219				list_skb = list_skb->next;
3220			}
3221
3222			if (unlikely(skb_shinfo(nskb)->nr_frags >=
3223				     MAX_SKB_FRAGS)) {
3224				net_warn_ratelimited(
3225					"skb_segment: too many frags: %u %u\n",
3226					pos, mss);
 
3227				goto err;
3228			}
3229
3230			if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3231				goto err;
3232
3233			*nskb_frag = *frag;
3234			__skb_frag_ref(nskb_frag);
3235			size = skb_frag_size(nskb_frag);
3236
3237			if (pos < offset) {
3238				nskb_frag->page_offset += offset - pos;
3239				skb_frag_size_sub(nskb_frag, offset - pos);
3240			}
3241
3242			skb_shinfo(nskb)->nr_frags++;
3243
3244			if (pos + size <= offset + len) {
3245				i++;
3246				frag++;
3247				pos += size;
3248			} else {
3249				skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3250				goto skip_fraglist;
3251			}
3252
3253			nskb_frag++;
3254		}
3255
3256skip_fraglist:
3257		nskb->data_len = len - hsize;
3258		nskb->len += nskb->data_len;
3259		nskb->truesize += nskb->data_len;
3260
3261perform_csum_check:
3262		if (!csum) {
3263			if (skb_has_shared_frag(nskb)) {
3264				err = __skb_linearize(nskb);
3265				if (err)
3266					goto err;
3267			}
3268			if (!nskb->remcsum_offload)
3269				nskb->ip_summed = CHECKSUM_NONE;
3270			SKB_GSO_CB(nskb)->csum =
3271				skb_checksum(nskb, doffset,
3272					     nskb->len - doffset, 0);
3273			SKB_GSO_CB(nskb)->csum_start =
3274				skb_headroom(nskb) + doffset;
3275		}
3276	} while ((offset += len) < head_skb->len);
3277
3278	/* Some callers want to get the end of the list.
3279	 * Put it in segs->prev to avoid walking the list.
3280	 * (see validate_xmit_skb_list() for example)
3281	 */
3282	segs->prev = tail;
3283
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3284	/* Following permits correct backpressure, for protocols
3285	 * using skb_set_owner_w().
3286	 * Idea is to tranfert ownership from head_skb to last segment.
3287	 */
3288	if (head_skb->destructor == sock_wfree) {
3289		swap(tail->truesize, head_skb->truesize);
3290		swap(tail->destructor, head_skb->destructor);
3291		swap(tail->sk, head_skb->sk);
3292	}
3293	return segs;
3294
3295err:
3296	kfree_skb_list(segs);
3297	return ERR_PTR(err);
3298}
3299EXPORT_SYMBOL_GPL(skb_segment);
3300
3301int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3302{
3303	struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3304	unsigned int offset = skb_gro_offset(skb);
3305	unsigned int headlen = skb_headlen(skb);
3306	unsigned int len = skb_gro_len(skb);
3307	struct sk_buff *lp, *p = *head;
3308	unsigned int delta_truesize;
 
3309
3310	if (unlikely(p->len + len >= 65536))
3311		return -E2BIG;
3312
3313	lp = NAPI_GRO_CB(p)->last;
3314	pinfo = skb_shinfo(lp);
3315
3316	if (headlen <= offset) {
3317		skb_frag_t *frag;
3318		skb_frag_t *frag2;
3319		int i = skbinfo->nr_frags;
3320		int nr_frags = pinfo->nr_frags + i;
3321
3322		if (nr_frags > MAX_SKB_FRAGS)
3323			goto merge;
3324
3325		offset -= headlen;
3326		pinfo->nr_frags = nr_frags;
3327		skbinfo->nr_frags = 0;
3328
3329		frag = pinfo->frags + nr_frags;
3330		frag2 = skbinfo->frags + i;
3331		do {
3332			*--frag = *--frag2;
3333		} while (--i);
3334
3335		frag->page_offset += offset;
3336		skb_frag_size_sub(frag, offset);
3337
3338		/* all fragments truesize : remove (head size + sk_buff) */
3339		delta_truesize = skb->truesize -
3340				 SKB_TRUESIZE(skb_end_offset(skb));
3341
3342		skb->truesize -= skb->data_len;
3343		skb->len -= skb->data_len;
3344		skb->data_len = 0;
3345
3346		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3347		goto done;
3348	} else if (skb->head_frag) {
3349		int nr_frags = pinfo->nr_frags;
3350		skb_frag_t *frag = pinfo->frags + nr_frags;
3351		struct page *page = virt_to_head_page(skb->head);
3352		unsigned int first_size = headlen - offset;
3353		unsigned int first_offset;
3354
3355		if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3356			goto merge;
3357
3358		first_offset = skb->data -
3359			       (unsigned char *)page_address(page) +
3360			       offset;
3361
3362		pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3363
3364		frag->page.p	  = page;
3365		frag->page_offset = first_offset;
3366		skb_frag_size_set(frag, first_size);
3367
3368		memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3369		/* We dont need to clear skbinfo->nr_frags here */
3370
3371		delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3372		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3373		goto done;
3374	}
3375
3376merge:
3377	delta_truesize = skb->truesize;
3378	if (offset > headlen) {
3379		unsigned int eat = offset - headlen;
3380
3381		skbinfo->frags[0].page_offset += eat;
3382		skb_frag_size_sub(&skbinfo->frags[0], eat);
3383		skb->data_len -= eat;
3384		skb->len -= eat;
3385		offset = headlen;
3386	}
3387
3388	__skb_pull(skb, offset);
3389
3390	if (NAPI_GRO_CB(p)->last == p)
3391		skb_shinfo(p)->frag_list = skb;
3392	else
3393		NAPI_GRO_CB(p)->last->next = skb;
3394	NAPI_GRO_CB(p)->last = skb;
3395	__skb_header_release(skb);
3396	lp = p;
3397
3398done:
3399	NAPI_GRO_CB(p)->count++;
3400	p->data_len += len;
3401	p->truesize += delta_truesize;
3402	p->len += len;
3403	if (lp != p) {
3404		lp->data_len += len;
3405		lp->truesize += delta_truesize;
3406		lp->len += len;
3407	}
3408	NAPI_GRO_CB(skb)->same_flow = 1;
3409	return 0;
3410}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3411
3412void __init skb_init(void)
3413{
3414	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3415					      sizeof(struct sk_buff),
3416					      0,
3417					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
 
 
3418					      NULL);
3419	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3420						sizeof(struct sk_buff_fclones),
3421						0,
3422						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3423						NULL);
 
3424}
3425
3426/**
3427 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3428 *	@skb: Socket buffer containing the buffers to be mapped
3429 *	@sg: The scatter-gather list to map into
3430 *	@offset: The offset into the buffer's contents to start mapping
3431 *	@len: Length of buffer space to be mapped
3432 *
3433 *	Fill the specified scatter-gather list with mappings/pointers into a
3434 *	region of the buffer space attached to a socket buffer.
3435 */
3436static int
3437__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
 
3438{
3439	int start = skb_headlen(skb);
3440	int i, copy = start - offset;
3441	struct sk_buff *frag_iter;
3442	int elt = 0;
3443
 
 
 
3444	if (copy > 0) {
3445		if (copy > len)
3446			copy = len;
3447		sg_set_buf(sg, skb->data + offset, copy);
3448		elt++;
3449		if ((len -= copy) == 0)
3450			return elt;
3451		offset += copy;
3452	}
3453
3454	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3455		int end;
3456
3457		WARN_ON(start > offset + len);
3458
3459		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3460		if ((copy = end - offset) > 0) {
3461			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 
 
3462
3463			if (copy > len)
3464				copy = len;
3465			sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3466					frag->page_offset+offset-start);
3467			elt++;
3468			if (!(len -= copy))
3469				return elt;
3470			offset += copy;
3471		}
3472		start = end;
3473	}
3474
3475	skb_walk_frags(skb, frag_iter) {
3476		int end;
3477
3478		WARN_ON(start > offset + len);
3479
3480		end = start + frag_iter->len;
3481		if ((copy = end - offset) > 0) {
 
 
 
3482			if (copy > len)
3483				copy = len;
3484			elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3485					      copy);
 
 
 
3486			if ((len -= copy) == 0)
3487				return elt;
3488			offset += copy;
3489		}
3490		start = end;
3491	}
3492	BUG_ON(len);
3493	return elt;
3494}
3495
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3496/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3497 * sglist without mark the sg which contain last skb data as the end.
3498 * So the caller can mannipulate sg list as will when padding new data after
3499 * the first call without calling sg_unmark_end to expend sg list.
3500 *
3501 * Scenario to use skb_to_sgvec_nomark:
3502 * 1. sg_init_table
3503 * 2. skb_to_sgvec_nomark(payload1)
3504 * 3. skb_to_sgvec_nomark(payload2)
3505 *
3506 * This is equivalent to:
3507 * 1. sg_init_table
3508 * 2. skb_to_sgvec(payload1)
3509 * 3. sg_unmark_end
3510 * 4. skb_to_sgvec(payload2)
3511 *
3512 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3513 * is more preferable.
3514 */
3515int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3516			int offset, int len)
3517{
3518	return __skb_to_sgvec(skb, sg, offset, len);
3519}
3520EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3521
3522int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3523{
3524	int nsg = __skb_to_sgvec(skb, sg, offset, len);
3525
3526	sg_mark_end(&sg[nsg - 1]);
3527
3528	return nsg;
3529}
3530EXPORT_SYMBOL_GPL(skb_to_sgvec);
3531
3532/**
3533 *	skb_cow_data - Check that a socket buffer's data buffers are writable
3534 *	@skb: The socket buffer to check.
3535 *	@tailbits: Amount of trailing space to be added
3536 *	@trailer: Returned pointer to the skb where the @tailbits space begins
3537 *
3538 *	Make sure that the data buffers attached to a socket buffer are
3539 *	writable. If they are not, private copies are made of the data buffers
3540 *	and the socket buffer is set to use these instead.
3541 *
3542 *	If @tailbits is given, make sure that there is space to write @tailbits
3543 *	bytes of data beyond current end of socket buffer.  @trailer will be
3544 *	set to point to the skb in which this space begins.
3545 *
3546 *	The number of scatterlist elements required to completely map the
3547 *	COW'd and extended socket buffer will be returned.
3548 */
3549int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3550{
3551	int copyflag;
3552	int elt;
3553	struct sk_buff *skb1, **skb_p;
3554
3555	/* If skb is cloned or its head is paged, reallocate
3556	 * head pulling out all the pages (pages are considered not writable
3557	 * at the moment even if they are anonymous).
3558	 */
3559	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3560	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3561		return -ENOMEM;
3562
3563	/* Easy case. Most of packets will go this way. */
3564	if (!skb_has_frag_list(skb)) {
3565		/* A little of trouble, not enough of space for trailer.
3566		 * This should not happen, when stack is tuned to generate
3567		 * good frames. OK, on miss we reallocate and reserve even more
3568		 * space, 128 bytes is fair. */
3569
3570		if (skb_tailroom(skb) < tailbits &&
3571		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3572			return -ENOMEM;
3573
3574		/* Voila! */
3575		*trailer = skb;
3576		return 1;
3577	}
3578
3579	/* Misery. We are in troubles, going to mincer fragments... */
3580
3581	elt = 1;
3582	skb_p = &skb_shinfo(skb)->frag_list;
3583	copyflag = 0;
3584
3585	while ((skb1 = *skb_p) != NULL) {
3586		int ntail = 0;
3587
3588		/* The fragment is partially pulled by someone,
3589		 * this can happen on input. Copy it and everything
3590		 * after it. */
3591
3592		if (skb_shared(skb1))
3593			copyflag = 1;
3594
3595		/* If the skb is the last, worry about trailer. */
3596
3597		if (skb1->next == NULL && tailbits) {
3598			if (skb_shinfo(skb1)->nr_frags ||
3599			    skb_has_frag_list(skb1) ||
3600			    skb_tailroom(skb1) < tailbits)
3601				ntail = tailbits + 128;
3602		}
3603
3604		if (copyflag ||
3605		    skb_cloned(skb1) ||
3606		    ntail ||
3607		    skb_shinfo(skb1)->nr_frags ||
3608		    skb_has_frag_list(skb1)) {
3609			struct sk_buff *skb2;
3610
3611			/* Fuck, we are miserable poor guys... */
3612			if (ntail == 0)
3613				skb2 = skb_copy(skb1, GFP_ATOMIC);
3614			else
3615				skb2 = skb_copy_expand(skb1,
3616						       skb_headroom(skb1),
3617						       ntail,
3618						       GFP_ATOMIC);
3619			if (unlikely(skb2 == NULL))
3620				return -ENOMEM;
3621
3622			if (skb1->sk)
3623				skb_set_owner_w(skb2, skb1->sk);
3624
3625			/* Looking around. Are we still alive?
3626			 * OK, link new skb, drop old one */
3627
3628			skb2->next = skb1->next;
3629			*skb_p = skb2;
3630			kfree_skb(skb1);
3631			skb1 = skb2;
3632		}
3633		elt++;
3634		*trailer = skb1;
3635		skb_p = &skb1->next;
3636	}
3637
3638	return elt;
3639}
3640EXPORT_SYMBOL_GPL(skb_cow_data);
3641
3642static void sock_rmem_free(struct sk_buff *skb)
3643{
3644	struct sock *sk = skb->sk;
3645
3646	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3647}
3648
 
 
 
 
 
 
 
 
 
3649/*
3650 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3651 */
3652int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3653{
3654	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3655	    (unsigned int)sk->sk_rcvbuf)
3656		return -ENOMEM;
3657
3658	skb_orphan(skb);
3659	skb->sk = sk;
3660	skb->destructor = sock_rmem_free;
3661	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 
3662
3663	/* before exiting rcu section, make sure dst is refcounted */
3664	skb_dst_force(skb);
3665
3666	skb_queue_tail(&sk->sk_error_queue, skb);
3667	if (!sock_flag(sk, SOCK_DEAD))
3668		sk->sk_data_ready(sk);
3669	return 0;
3670}
3671EXPORT_SYMBOL(sock_queue_err_skb);
3672
 
 
 
 
 
 
3673struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3674{
3675	struct sk_buff_head *q = &sk->sk_error_queue;
3676	struct sk_buff *skb, *skb_next;
 
3677	unsigned long flags;
3678	int err = 0;
3679
3680	spin_lock_irqsave(&q->lock, flags);
3681	skb = __skb_dequeue(q);
3682	if (skb && (skb_next = skb_peek(q)))
3683		err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
 
 
 
3684	spin_unlock_irqrestore(&q->lock, flags);
3685
3686	sk->sk_err = err;
3687	if (err)
 
 
3688		sk->sk_error_report(sk);
3689
3690	return skb;
3691}
3692EXPORT_SYMBOL(sock_dequeue_err_skb);
3693
3694/**
3695 * skb_clone_sk - create clone of skb, and take reference to socket
3696 * @skb: the skb to clone
3697 *
3698 * This function creates a clone of a buffer that holds a reference on
3699 * sk_refcnt.  Buffers created via this function are meant to be
3700 * returned using sock_queue_err_skb, or free via kfree_skb.
3701 *
3702 * When passing buffers allocated with this function to sock_queue_err_skb
3703 * it is necessary to wrap the call with sock_hold/sock_put in order to
3704 * prevent the socket from being released prior to being enqueued on
3705 * the sk_error_queue.
3706 */
3707struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3708{
3709	struct sock *sk = skb->sk;
3710	struct sk_buff *clone;
3711
3712	if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3713		return NULL;
3714
3715	clone = skb_clone(skb, GFP_ATOMIC);
3716	if (!clone) {
3717		sock_put(sk);
3718		return NULL;
3719	}
3720
3721	clone->sk = sk;
3722	clone->destructor = sock_efree;
3723
3724	return clone;
3725}
3726EXPORT_SYMBOL(skb_clone_sk);
3727
3728static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3729					struct sock *sk,
3730					int tstype)
 
3731{
3732	struct sock_exterr_skb *serr;
3733	int err;
3734
 
 
3735	serr = SKB_EXT_ERR(skb);
3736	memset(serr, 0, sizeof(*serr));
3737	serr->ee.ee_errno = ENOMSG;
3738	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3739	serr->ee.ee_info = tstype;
 
 
3740	if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3741		serr->ee.ee_data = skb_shinfo(skb)->tskey;
3742		if (sk->sk_protocol == IPPROTO_TCP &&
3743		    sk->sk_type == SOCK_STREAM)
3744			serr->ee.ee_data -= sk->sk_tskey;
3745	}
3746
3747	err = sock_queue_err_skb(sk, skb);
3748
3749	if (err)
3750		kfree_skb(skb);
3751}
3752
3753static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3754{
3755	bool ret;
3756
3757	if (likely(sysctl_tstamp_allow_data || tsonly))
3758		return true;
3759
3760	read_lock_bh(&sk->sk_callback_lock);
3761	ret = sk->sk_socket && sk->sk_socket->file &&
3762	      file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3763	read_unlock_bh(&sk->sk_callback_lock);
3764	return ret;
3765}
3766
3767void skb_complete_tx_timestamp(struct sk_buff *skb,
3768			       struct skb_shared_hwtstamps *hwtstamps)
3769{
3770	struct sock *sk = skb->sk;
3771
3772	if (!skb_may_tx_timestamp(sk, false))
3773		return;
3774
3775	/* take a reference to prevent skb_orphan() from freeing the socket */
3776	sock_hold(sk);
3777
3778	*skb_hwtstamps(skb) = *hwtstamps;
3779	__skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
 
 
 
 
 
 
 
3780
3781	sock_put(sk);
 
3782}
3783EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3784
3785void __skb_tstamp_tx(struct sk_buff *orig_skb,
3786		     struct skb_shared_hwtstamps *hwtstamps,
3787		     struct sock *sk, int tstype)
3788{
3789	struct sk_buff *skb;
3790	bool tsonly;
3791
3792	if (!sk)
3793		return;
3794
 
 
 
 
3795	tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3796	if (!skb_may_tx_timestamp(sk, tsonly))
3797		return;
3798
3799	if (tsonly)
3800		skb = alloc_skb(0, GFP_ATOMIC);
3801	else
 
 
 
 
 
 
 
 
3802		skb = skb_clone(orig_skb, GFP_ATOMIC);
 
3803	if (!skb)
3804		return;
3805
3806	if (tsonly) {
3807		skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
 
3808		skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3809	}
3810
3811	if (hwtstamps)
3812		*skb_hwtstamps(skb) = *hwtstamps;
3813	else
3814		skb->tstamp = ktime_get_real();
3815
3816	__skb_complete_tx_timestamp(skb, sk, tstype);
3817}
3818EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3819
3820void skb_tstamp_tx(struct sk_buff *orig_skb,
3821		   struct skb_shared_hwtstamps *hwtstamps)
3822{
3823	return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3824			       SCM_TSTAMP_SND);
3825}
3826EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3827
3828void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3829{
3830	struct sock *sk = skb->sk;
3831	struct sock_exterr_skb *serr;
3832	int err;
3833
3834	skb->wifi_acked_valid = 1;
3835	skb->wifi_acked = acked;
3836
3837	serr = SKB_EXT_ERR(skb);
3838	memset(serr, 0, sizeof(*serr));
3839	serr->ee.ee_errno = ENOMSG;
3840	serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3841
3842	/* take a reference to prevent skb_orphan() from freeing the socket */
3843	sock_hold(sk);
3844
3845	err = sock_queue_err_skb(sk, skb);
 
 
 
3846	if (err)
3847		kfree_skb(skb);
3848
3849	sock_put(sk);
3850}
3851EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3852
3853/**
3854 * skb_partial_csum_set - set up and verify partial csum values for packet
3855 * @skb: the skb to set
3856 * @start: the number of bytes after skb->data to start checksumming.
3857 * @off: the offset from start to place the checksum.
3858 *
3859 * For untrusted partially-checksummed packets, we need to make sure the values
3860 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3861 *
3862 * This function checks and sets those values and skb->ip_summed: if this
3863 * returns false you should drop the packet.
3864 */
3865bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3866{
3867	if (unlikely(start > skb_headlen(skb)) ||
3868	    unlikely((int)start + off > skb_headlen(skb) - 2)) {
3869		net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3870				     start, off, skb_headlen(skb));
 
 
3871		return false;
3872	}
3873	skb->ip_summed = CHECKSUM_PARTIAL;
3874	skb->csum_start = skb_headroom(skb) + start;
3875	skb->csum_offset = off;
3876	skb_set_transport_header(skb, start);
3877	return true;
3878}
3879EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3880
3881static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3882			       unsigned int max)
3883{
3884	if (skb_headlen(skb) >= len)
3885		return 0;
3886
3887	/* If we need to pullup then pullup to the max, so we
3888	 * won't need to do it again.
3889	 */
3890	if (max > skb->len)
3891		max = skb->len;
3892
3893	if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3894		return -ENOMEM;
3895
3896	if (skb_headlen(skb) < len)
3897		return -EPROTO;
3898
3899	return 0;
3900}
3901
3902#define MAX_TCP_HDR_LEN (15 * 4)
3903
3904static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3905				      typeof(IPPROTO_IP) proto,
3906				      unsigned int off)
3907{
3908	switch (proto) {
3909		int err;
3910
3911	case IPPROTO_TCP:
3912		err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3913					  off + MAX_TCP_HDR_LEN);
3914		if (!err && !skb_partial_csum_set(skb, off,
3915						  offsetof(struct tcphdr,
3916							   check)))
3917			err = -EPROTO;
3918		return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3919
3920	case IPPROTO_UDP:
3921		err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3922					  off + sizeof(struct udphdr));
3923		if (!err && !skb_partial_csum_set(skb, off,
3924						  offsetof(struct udphdr,
3925							   check)))
3926			err = -EPROTO;
3927		return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3928	}
3929
3930	return ERR_PTR(-EPROTO);
3931}
3932
3933/* This value should be large enough to cover a tagged ethernet header plus
3934 * maximally sized IP and TCP or UDP headers.
3935 */
3936#define MAX_IP_HDR_LEN 128
3937
3938static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3939{
3940	unsigned int off;
3941	bool fragment;
3942	__sum16 *csum;
3943	int err;
3944
3945	fragment = false;
3946
3947	err = skb_maybe_pull_tail(skb,
3948				  sizeof(struct iphdr),
3949				  MAX_IP_HDR_LEN);
3950	if (err < 0)
3951		goto out;
3952
3953	if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3954		fragment = true;
3955
3956	off = ip_hdrlen(skb);
3957
3958	err = -EPROTO;
3959
3960	if (fragment)
3961		goto out;
3962
3963	csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3964	if (IS_ERR(csum))
3965		return PTR_ERR(csum);
3966
3967	if (recalculate)
3968		*csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3969					   ip_hdr(skb)->daddr,
3970					   skb->len - off,
3971					   ip_hdr(skb)->protocol, 0);
3972	err = 0;
3973
3974out:
3975	return err;
3976}
3977
3978/* This value should be large enough to cover a tagged ethernet header plus
3979 * an IPv6 header, all options, and a maximal TCP or UDP header.
3980 */
3981#define MAX_IPV6_HDR_LEN 256
3982
3983#define OPT_HDR(type, skb, off) \
3984	(type *)(skb_network_header(skb) + (off))
3985
3986static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3987{
3988	int err;
3989	u8 nexthdr;
3990	unsigned int off;
3991	unsigned int len;
3992	bool fragment;
3993	bool done;
3994	__sum16 *csum;
3995
3996	fragment = false;
3997	done = false;
3998
3999	off = sizeof(struct ipv6hdr);
4000
4001	err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4002	if (err < 0)
4003		goto out;
4004
4005	nexthdr = ipv6_hdr(skb)->nexthdr;
4006
4007	len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4008	while (off <= len && !done) {
4009		switch (nexthdr) {
4010		case IPPROTO_DSTOPTS:
4011		case IPPROTO_HOPOPTS:
4012		case IPPROTO_ROUTING: {
4013			struct ipv6_opt_hdr *hp;
4014
4015			err = skb_maybe_pull_tail(skb,
4016						  off +
4017						  sizeof(struct ipv6_opt_hdr),
4018						  MAX_IPV6_HDR_LEN);
4019			if (err < 0)
4020				goto out;
4021
4022			hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4023			nexthdr = hp->nexthdr;
4024			off += ipv6_optlen(hp);
4025			break;
4026		}
4027		case IPPROTO_AH: {
4028			struct ip_auth_hdr *hp;
4029
4030			err = skb_maybe_pull_tail(skb,
4031						  off +
4032						  sizeof(struct ip_auth_hdr),
4033						  MAX_IPV6_HDR_LEN);
4034			if (err < 0)
4035				goto out;
4036
4037			hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4038			nexthdr = hp->nexthdr;
4039			off += ipv6_authlen(hp);
4040			break;
4041		}
4042		case IPPROTO_FRAGMENT: {
4043			struct frag_hdr *hp;
4044
4045			err = skb_maybe_pull_tail(skb,
4046						  off +
4047						  sizeof(struct frag_hdr),
4048						  MAX_IPV6_HDR_LEN);
4049			if (err < 0)
4050				goto out;
4051
4052			hp = OPT_HDR(struct frag_hdr, skb, off);
4053
4054			if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4055				fragment = true;
4056
4057			nexthdr = hp->nexthdr;
4058			off += sizeof(struct frag_hdr);
4059			break;
4060		}
4061		default:
4062			done = true;
4063			break;
4064		}
4065	}
4066
4067	err = -EPROTO;
4068
4069	if (!done || fragment)
4070		goto out;
4071
4072	csum = skb_checksum_setup_ip(skb, nexthdr, off);
4073	if (IS_ERR(csum))
4074		return PTR_ERR(csum);
4075
4076	if (recalculate)
4077		*csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4078					 &ipv6_hdr(skb)->daddr,
4079					 skb->len - off, nexthdr, 0);
4080	err = 0;
4081
4082out:
4083	return err;
4084}
4085
4086/**
4087 * skb_checksum_setup - set up partial checksum offset
4088 * @skb: the skb to set up
4089 * @recalculate: if true the pseudo-header checksum will be recalculated
4090 */
4091int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4092{
4093	int err;
4094
4095	switch (skb->protocol) {
4096	case htons(ETH_P_IP):
4097		err = skb_checksum_setup_ipv4(skb, recalculate);
4098		break;
4099
4100	case htons(ETH_P_IPV6):
4101		err = skb_checksum_setup_ipv6(skb, recalculate);
4102		break;
4103
4104	default:
4105		err = -EPROTO;
4106		break;
4107	}
4108
4109	return err;
4110}
4111EXPORT_SYMBOL(skb_checksum_setup);
4112
4113/**
4114 * skb_checksum_maybe_trim - maybe trims the given skb
4115 * @skb: the skb to check
4116 * @transport_len: the data length beyond the network header
4117 *
4118 * Checks whether the given skb has data beyond the given transport length.
4119 * If so, returns a cloned skb trimmed to this transport length.
4120 * Otherwise returns the provided skb. Returns NULL in error cases
4121 * (e.g. transport_len exceeds skb length or out-of-memory).
4122 *
4123 * Caller needs to set the skb transport header and free any returned skb if it
4124 * differs from the provided skb.
4125 */
4126static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4127					       unsigned int transport_len)
4128{
4129	struct sk_buff *skb_chk;
4130	unsigned int len = skb_transport_offset(skb) + transport_len;
4131	int ret;
4132
4133	if (skb->len < len)
4134		return NULL;
4135	else if (skb->len == len)
4136		return skb;
4137
4138	skb_chk = skb_clone(skb, GFP_ATOMIC);
4139	if (!skb_chk)
4140		return NULL;
4141
4142	ret = pskb_trim_rcsum(skb_chk, len);
4143	if (ret) {
4144		kfree_skb(skb_chk);
4145		return NULL;
4146	}
4147
4148	return skb_chk;
4149}
4150
4151/**
4152 * skb_checksum_trimmed - validate checksum of an skb
4153 * @skb: the skb to check
4154 * @transport_len: the data length beyond the network header
4155 * @skb_chkf: checksum function to use
4156 *
4157 * Applies the given checksum function skb_chkf to the provided skb.
4158 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4159 *
4160 * If the skb has data beyond the given transport length, then a
4161 * trimmed & cloned skb is checked and returned.
4162 *
4163 * Caller needs to set the skb transport header and free any returned skb if it
4164 * differs from the provided skb.
4165 */
4166struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4167				     unsigned int transport_len,
4168				     __sum16(*skb_chkf)(struct sk_buff *skb))
4169{
4170	struct sk_buff *skb_chk;
4171	unsigned int offset = skb_transport_offset(skb);
4172	__sum16 ret;
4173
4174	skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4175	if (!skb_chk)
4176		goto err;
4177
4178	if (!pskb_may_pull(skb_chk, offset))
4179		goto err;
4180
4181	skb_pull_rcsum(skb_chk, offset);
4182	ret = skb_chkf(skb_chk);
4183	skb_push_rcsum(skb_chk, offset);
4184
4185	if (ret)
4186		goto err;
4187
4188	return skb_chk;
4189
4190err:
4191	if (skb_chk && skb_chk != skb)
4192		kfree_skb(skb_chk);
4193
4194	return NULL;
4195
4196}
4197EXPORT_SYMBOL(skb_checksum_trimmed);
4198
4199void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4200{
4201	net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4202			     skb->dev->name);
4203}
4204EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4205
4206void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4207{
4208	if (head_stolen) {
4209		skb_release_head_state(skb);
4210		kmem_cache_free(skbuff_head_cache, skb);
4211	} else {
4212		__kfree_skb(skb);
4213	}
4214}
4215EXPORT_SYMBOL(kfree_skb_partial);
4216
4217/**
4218 * skb_try_coalesce - try to merge skb to prior one
4219 * @to: prior buffer
4220 * @from: buffer to add
4221 * @fragstolen: pointer to boolean
4222 * @delta_truesize: how much more was allocated than was requested
4223 */
4224bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4225		      bool *fragstolen, int *delta_truesize)
4226{
 
4227	int i, delta, len = from->len;
4228
4229	*fragstolen = false;
4230
4231	if (skb_cloned(to))
4232		return false;
4233
4234	if (len <= skb_tailroom(to)) {
4235		if (len)
4236			BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4237		*delta_truesize = 0;
4238		return true;
4239	}
4240
4241	if (skb_has_frag_list(to) || skb_has_frag_list(from))
 
 
 
 
4242		return false;
4243
4244	if (skb_headlen(from) != 0) {
4245		struct page *page;
4246		unsigned int offset;
4247
4248		if (skb_shinfo(to)->nr_frags +
4249		    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4250			return false;
4251
4252		if (skb_head_is_locked(from))
4253			return false;
4254
4255		delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4256
4257		page = virt_to_head_page(from->head);
4258		offset = from->data - (unsigned char *)page_address(page);
4259
4260		skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4261				   page, offset, skb_headlen(from));
4262		*fragstolen = true;
4263	} else {
4264		if (skb_shinfo(to)->nr_frags +
4265		    skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4266			return false;
4267
4268		delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4269	}
4270
4271	WARN_ON_ONCE(delta < len);
4272
4273	memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4274	       skb_shinfo(from)->frags,
4275	       skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4276	skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4277
4278	if (!skb_cloned(from))
4279		skb_shinfo(from)->nr_frags = 0;
4280
4281	/* if the skb is not cloned this does nothing
4282	 * since we set nr_frags to 0.
4283	 */
4284	for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4285		skb_frag_ref(from, i);
4286
4287	to->truesize += delta;
4288	to->len += len;
4289	to->data_len += len;
4290
4291	*delta_truesize = delta;
4292	return true;
4293}
4294EXPORT_SYMBOL(skb_try_coalesce);
4295
4296/**
4297 * skb_scrub_packet - scrub an skb
4298 *
4299 * @skb: buffer to clean
4300 * @xnet: packet is crossing netns
4301 *
4302 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4303 * into/from a tunnel. Some information have to be cleared during these
4304 * operations.
4305 * skb_scrub_packet can also be used to clean a skb before injecting it in
4306 * another namespace (@xnet == true). We have to clear all information in the
4307 * skb that could impact namespace isolation.
4308 */
4309void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4310{
4311	skb->tstamp.tv64 = 0;
4312	skb->pkt_type = PACKET_HOST;
4313	skb->skb_iif = 0;
4314	skb->ignore_df = 0;
4315	skb_dst_drop(skb);
4316	secpath_reset(skb);
4317	nf_reset(skb);
4318	nf_reset_trace(skb);
4319
 
 
 
 
 
4320	if (!xnet)
4321		return;
4322
4323	skb_orphan(skb);
4324	skb->mark = 0;
 
4325}
4326EXPORT_SYMBOL_GPL(skb_scrub_packet);
4327
4328/**
4329 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4330 *
4331 * @skb: GSO skb
4332 *
4333 * skb_gso_transport_seglen is used to determine the real size of the
4334 * individual segments, including Layer4 headers (TCP/UDP).
4335 *
4336 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4337 */
4338unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4339{
4340	const struct skb_shared_info *shinfo = skb_shinfo(skb);
4341	unsigned int thlen = 0;
4342
4343	if (skb->encapsulation) {
4344		thlen = skb_inner_transport_header(skb) -
4345			skb_transport_header(skb);
4346
4347		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4348			thlen += inner_tcp_hdrlen(skb);
4349	} else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4350		thlen = tcp_hdrlen(skb);
 
 
 
 
4351	}
4352	/* UFO sets gso_size to the size of the fragmentation
4353	 * payload, i.e. the size of the L4 (UDP) header is already
4354	 * accounted for.
4355	 */
4356	return thlen + shinfo->gso_size;
4357}
4358EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4359
4360static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4361{
 
 
 
4362	if (skb_cow(skb, skb_headroom(skb)) < 0) {
4363		kfree_skb(skb);
4364		return NULL;
4365	}
4366
4367	memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4368		2 * ETH_ALEN);
 
 
 
 
 
 
 
 
 
 
4369	skb->mac_header += VLAN_HLEN;
4370	return skb;
4371}
4372
4373struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4374{
4375	struct vlan_hdr *vhdr;
4376	u16 vlan_tci;
4377
4378	if (unlikely(skb_vlan_tag_present(skb))) {
4379		/* vlan_tci is already set-up so leave this for another time */
4380		return skb;
4381	}
4382
4383	skb = skb_share_check(skb, GFP_ATOMIC);
4384	if (unlikely(!skb))
4385		goto err_free;
4386
4387	if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4388		goto err_free;
4389
4390	vhdr = (struct vlan_hdr *)skb->data;
4391	vlan_tci = ntohs(vhdr->h_vlan_TCI);
4392	__vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4393
4394	skb_pull_rcsum(skb, VLAN_HLEN);
4395	vlan_set_encap_proto(skb, vhdr);
4396
4397	skb = skb_reorder_vlan_header(skb);
4398	if (unlikely(!skb))
4399		goto err_free;
4400
4401	skb_reset_network_header(skb);
4402	skb_reset_transport_header(skb);
4403	skb_reset_mac_len(skb);
4404
4405	return skb;
4406
4407err_free:
4408	kfree_skb(skb);
4409	return NULL;
4410}
4411EXPORT_SYMBOL(skb_vlan_untag);
4412
4413int skb_ensure_writable(struct sk_buff *skb, int write_len)
4414{
4415	if (!pskb_may_pull(skb, write_len))
4416		return -ENOMEM;
4417
4418	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4419		return 0;
4420
4421	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4422}
4423EXPORT_SYMBOL(skb_ensure_writable);
4424
4425/* remove VLAN header from packet and update csum accordingly. */
4426static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
 
 
4427{
4428	struct vlan_hdr *vhdr;
4429	unsigned int offset = skb->data - skb_mac_header(skb);
4430	int err;
4431
4432	__skb_push(skb, offset);
 
 
 
 
 
4433	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4434	if (unlikely(err))
4435		goto pull;
4436
4437	skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4438
4439	vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4440	*vlan_tci = ntohs(vhdr->h_vlan_TCI);
4441
4442	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4443	__skb_pull(skb, VLAN_HLEN);
4444
4445	vlan_set_encap_proto(skb, vhdr);
4446	skb->mac_header += VLAN_HLEN;
4447
4448	if (skb_network_offset(skb) < ETH_HLEN)
4449		skb_set_network_header(skb, ETH_HLEN);
4450
4451	skb_reset_mac_len(skb);
4452pull:
4453	__skb_pull(skb, offset);
4454
4455	return err;
4456}
 
4457
 
 
 
4458int skb_vlan_pop(struct sk_buff *skb)
4459{
4460	u16 vlan_tci;
4461	__be16 vlan_proto;
4462	int err;
4463
4464	if (likely(skb_vlan_tag_present(skb))) {
4465		skb->vlan_tci = 0;
4466	} else {
4467		if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4468			      skb->protocol != htons(ETH_P_8021AD)) ||
4469			     skb->len < VLAN_ETH_HLEN))
4470			return 0;
4471
4472		err = __skb_vlan_pop(skb, &vlan_tci);
4473		if (err)
4474			return err;
4475	}
4476	/* move next vlan tag to hw accel tag */
4477	if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4478		    skb->protocol != htons(ETH_P_8021AD)) ||
4479		   skb->len < VLAN_ETH_HLEN))
4480		return 0;
4481
4482	vlan_proto = skb->protocol;
4483	err = __skb_vlan_pop(skb, &vlan_tci);
4484	if (unlikely(err))
4485		return err;
4486
4487	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4488	return 0;
4489}
4490EXPORT_SYMBOL(skb_vlan_pop);
4491
 
 
 
4492int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4493{
4494	if (skb_vlan_tag_present(skb)) {
4495		unsigned int offset = skb->data - skb_mac_header(skb);
4496		int err;
4497
4498		/* __vlan_insert_tag expect skb->data pointing to mac header.
4499		 * So change skb->data before calling it and change back to
4500		 * original position later
4501		 */
4502		__skb_push(skb, offset);
 
4503		err = __vlan_insert_tag(skb, skb->vlan_proto,
4504					skb_vlan_tag_get(skb));
4505		if (err) {
4506			__skb_pull(skb, offset);
4507			return err;
4508		}
4509
4510		skb->protocol = skb->vlan_proto;
4511		skb->mac_len += VLAN_HLEN;
4512
4513		skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4514		__skb_pull(skb, offset);
4515	}
4516	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4517	return 0;
4518}
4519EXPORT_SYMBOL(skb_vlan_push);
4520
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4521/**
4522 * alloc_skb_with_frags - allocate skb with page frags
4523 *
4524 * @header_len: size of linear part
4525 * @data_len: needed length in frags
4526 * @max_page_order: max page order desired.
4527 * @errcode: pointer to error code if any
4528 * @gfp_mask: allocation mask
4529 *
4530 * This can be used to allocate a paged skb, given a maximal order for frags.
4531 */
4532struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4533				     unsigned long data_len,
4534				     int max_page_order,
4535				     int *errcode,
4536				     gfp_t gfp_mask)
4537{
4538	int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4539	unsigned long chunk;
4540	struct sk_buff *skb;
4541	struct page *page;
4542	gfp_t gfp_head;
4543	int i;
4544
4545	*errcode = -EMSGSIZE;
4546	/* Note this test could be relaxed, if we succeed to allocate
4547	 * high order pages...
4548	 */
4549	if (npages > MAX_SKB_FRAGS)
4550		return NULL;
4551
4552	gfp_head = gfp_mask;
4553	if (gfp_head & __GFP_DIRECT_RECLAIM)
4554		gfp_head |= __GFP_REPEAT;
4555
4556	*errcode = -ENOBUFS;
4557	skb = alloc_skb(header_len, gfp_head);
4558	if (!skb)
4559		return NULL;
4560
4561	skb->truesize += npages << PAGE_SHIFT;
4562
4563	for (i = 0; npages > 0; i++) {
4564		int order = max_page_order;
4565
4566		while (order) {
4567			if (npages >= 1 << order) {
4568				page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4569						   __GFP_COMP |
4570						   __GFP_NOWARN |
4571						   __GFP_NORETRY,
4572						   order);
4573				if (page)
4574					goto fill_page;
4575				/* Do not retry other high order allocations */
4576				order = 1;
4577				max_page_order = 0;
4578			}
4579			order--;
4580		}
4581		page = alloc_page(gfp_mask);
4582		if (!page)
4583			goto failure;
4584fill_page:
4585		chunk = min_t(unsigned long, data_len,
4586			      PAGE_SIZE << order);
4587		skb_fill_page_desc(skb, i, page, 0, chunk);
4588		data_len -= chunk;
4589		npages -= 1 << order;
4590	}
4591	return skb;
4592
4593failure:
4594	kfree_skb(skb);
4595	return NULL;
4596}
4597EXPORT_SYMBOL(alloc_skb_with_frags);
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *	Routines having to do with the 'struct sk_buff' memory handlers.
   4 *
   5 *	Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
   6 *			Florian La Roche <rzsfl@rz.uni-sb.de>
   7 *
   8 *	Fixes:
   9 *		Alan Cox	:	Fixed the worst of the load
  10 *					balancer bugs.
  11 *		Dave Platt	:	Interrupt stacking fix.
  12 *	Richard Kooijman	:	Timestamp fixes.
  13 *		Alan Cox	:	Changed buffer format.
  14 *		Alan Cox	:	destructor hook for AF_UNIX etc.
  15 *		Linus Torvalds	:	Better skb_clone.
  16 *		Alan Cox	:	Added skb_copy.
  17 *		Alan Cox	:	Added all the changed routines Linus
  18 *					only put in the headers
  19 *		Ray VanTassle	:	Fixed --skb->lock in free
  20 *		Alan Cox	:	skb_copy copy arp field
  21 *		Andi Kleen	:	slabified it.
  22 *		Robert Olsson	:	Removed skb_head_pool
  23 *
  24 *	NOTE:
  25 *		The __skb_ routines should be called with interrupts
  26 *	disabled, or you better be *real* sure that the operation is atomic
  27 *	with respect to whatever list is being frobbed (e.g. via lock_sock()
  28 *	or via disabling bottom half handlers, etc).
 
 
 
 
 
  29 */
  30
  31/*
  32 *	The functions in this file will not compile correctly with gcc 2.4.x
  33 */
  34
  35#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  36
  37#include <linux/module.h>
  38#include <linux/types.h>
  39#include <linux/kernel.h>
 
  40#include <linux/mm.h>
  41#include <linux/interrupt.h>
  42#include <linux/in.h>
  43#include <linux/inet.h>
  44#include <linux/slab.h>
  45#include <linux/tcp.h>
  46#include <linux/udp.h>
  47#include <linux/sctp.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#include <linux/if_vlan.h>
  62#include <linux/mpls.h>
  63
  64#include <net/protocol.h>
  65#include <net/dst.h>
  66#include <net/sock.h>
  67#include <net/checksum.h>
  68#include <net/ip6_checksum.h>
  69#include <net/xfrm.h>
  70#include <net/mpls.h>
  71
  72#include <linux/uaccess.h>
  73#include <trace/events/skb.h>
  74#include <linux/highmem.h>
  75#include <linux/capability.h>
  76#include <linux/user_namespace.h>
  77#include <linux/indirect_call_wrapper.h>
  78
  79#include "datagram.h"
  80
  81struct kmem_cache *skbuff_head_cache __ro_after_init;
  82static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
  83#ifdef CONFIG_SKB_EXTENSIONS
  84static struct kmem_cache *skbuff_ext_cache __ro_after_init;
  85#endif
  86int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
  87EXPORT_SYMBOL(sysctl_max_skb_frags);
  88
  89/**
  90 *	skb_panic - private function for out-of-line support
  91 *	@skb:	buffer
  92 *	@sz:	size
  93 *	@addr:	address
  94 *	@msg:	skb_over_panic or skb_under_panic
  95 *
  96 *	Out-of-line support for skb_put() and skb_push().
  97 *	Called via the wrapper skb_over_panic() or skb_under_panic().
  98 *	Keep out of line to prevent kernel bloat.
  99 *	__builtin_return_address is not used because it is not always reliable.
 100 */
 101static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 102		      const char msg[])
 103{
 104	pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 105		 msg, addr, skb->len, sz, skb->head, skb->data,
 106		 (unsigned long)skb->tail, (unsigned long)skb->end,
 107		 skb->dev ? skb->dev->name : "<NULL>");
 108	BUG();
 109}
 110
 111static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 112{
 113	skb_panic(skb, sz, addr, __func__);
 114}
 115
 116static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 117{
 118	skb_panic(skb, sz, addr, __func__);
 119}
 120
 121/*
 122 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
 123 * the caller if emergency pfmemalloc reserves are being used. If it is and
 124 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
 125 * may be used. Otherwise, the packet data may be discarded until enough
 126 * memory is free
 127 */
 128#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
 129	 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
 130
 131static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
 132			       unsigned long ip, bool *pfmemalloc)
 133{
 134	void *obj;
 135	bool ret_pfmemalloc = false;
 136
 137	/*
 138	 * Try a regular allocation, when that fails and we're not entitled
 139	 * to the reserves, fail.
 140	 */
 141	obj = kmalloc_node_track_caller(size,
 142					flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 143					node);
 144	if (obj || !(gfp_pfmemalloc_allowed(flags)))
 145		goto out;
 146
 147	/* Try again but now we are using pfmemalloc reserves */
 148	ret_pfmemalloc = true;
 149	obj = kmalloc_node_track_caller(size, flags, node);
 150
 151out:
 152	if (pfmemalloc)
 153		*pfmemalloc = ret_pfmemalloc;
 154
 155	return obj;
 156}
 157
 158/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 159 *	'private' fields and also do memory statistics to find all the
 160 *	[BEEP] leaks.
 161 *
 162 */
 163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 164/**
 165 *	__alloc_skb	-	allocate a network buffer
 166 *	@size: size to allocate
 167 *	@gfp_mask: allocation mask
 168 *	@flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
 169 *		instead of head cache and allocate a cloned (child) skb.
 170 *		If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
 171 *		allocations in case the data is required for writeback
 172 *	@node: numa node to allocate memory on
 173 *
 174 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 175 *	tail room of at least size bytes. The object has a reference count
 176 *	of one. The return is the buffer. On a failure the return is %NULL.
 177 *
 178 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 179 *	%GFP_ATOMIC.
 180 */
 181struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 182			    int flags, int node)
 183{
 184	struct kmem_cache *cache;
 185	struct skb_shared_info *shinfo;
 186	struct sk_buff *skb;
 187	u8 *data;
 188	bool pfmemalloc;
 189
 190	cache = (flags & SKB_ALLOC_FCLONE)
 191		? skbuff_fclone_cache : skbuff_head_cache;
 192
 193	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 194		gfp_mask |= __GFP_MEMALLOC;
 195
 196	/* Get the HEAD */
 197	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
 198	if (!skb)
 199		goto out;
 200	prefetchw(skb);
 201
 202	/* We do our best to align skb_shared_info on a separate cache
 203	 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
 204	 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
 205	 * Both skb->head and skb_shared_info are cache line aligned.
 206	 */
 207	size = SKB_DATA_ALIGN(size);
 208	size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 209	data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
 210	if (!data)
 211		goto nodata;
 212	/* kmalloc(size) might give us more room than requested.
 213	 * Put skb_shared_info exactly at the end of allocated zone,
 214	 * to allow max possible filling before reallocation.
 215	 */
 216	size = SKB_WITH_OVERHEAD(ksize(data));
 217	prefetchw(data + size);
 218
 219	/*
 220	 * Only clear those fields we need to clear, not those that we will
 221	 * actually initialise below. Hence, don't put any more fields after
 222	 * the tail pointer in struct sk_buff!
 223	 */
 224	memset(skb, 0, offsetof(struct sk_buff, tail));
 225	/* Account for allocated memory : skb + skb->head */
 226	skb->truesize = SKB_TRUESIZE(size);
 227	skb->pfmemalloc = pfmemalloc;
 228	refcount_set(&skb->users, 1);
 229	skb->head = data;
 230	skb->data = data;
 231	skb_reset_tail_pointer(skb);
 232	skb->end = skb->tail + size;
 233	skb->mac_header = (typeof(skb->mac_header))~0U;
 234	skb->transport_header = (typeof(skb->transport_header))~0U;
 235
 236	/* make sure we initialize shinfo sequentially */
 237	shinfo = skb_shinfo(skb);
 238	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 239	atomic_set(&shinfo->dataref, 1);
 
 240
 241	if (flags & SKB_ALLOC_FCLONE) {
 242		struct sk_buff_fclones *fclones;
 243
 244		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 245
 
 246		skb->fclone = SKB_FCLONE_ORIG;
 247		refcount_set(&fclones->fclone_ref, 1);
 248
 249		fclones->skb2.fclone = SKB_FCLONE_CLONE;
 
 250	}
 251out:
 252	return skb;
 253nodata:
 254	kmem_cache_free(cache, skb);
 255	skb = NULL;
 256	goto out;
 257}
 258EXPORT_SYMBOL(__alloc_skb);
 259
 260/* Caller must provide SKB that is memset cleared */
 261static struct sk_buff *__build_skb_around(struct sk_buff *skb,
 262					  void *data, unsigned int frag_size)
 263{
 264	struct skb_shared_info *shinfo;
 265	unsigned int size = frag_size ? : ksize(data);
 266
 267	size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 268
 269	/* Assumes caller memset cleared SKB */
 270	skb->truesize = SKB_TRUESIZE(size);
 271	refcount_set(&skb->users, 1);
 272	skb->head = data;
 273	skb->data = data;
 274	skb_reset_tail_pointer(skb);
 275	skb->end = skb->tail + size;
 276	skb->mac_header = (typeof(skb->mac_header))~0U;
 277	skb->transport_header = (typeof(skb->transport_header))~0U;
 278
 279	/* make sure we initialize shinfo sequentially */
 280	shinfo = skb_shinfo(skb);
 281	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 282	atomic_set(&shinfo->dataref, 1);
 283
 284	return skb;
 285}
 286
 287/**
 288 * __build_skb - build a network buffer
 289 * @data: data buffer provided by caller
 290 * @frag_size: size of data, or 0 if head was kmalloced
 291 *
 292 * Allocate a new &sk_buff. Caller provides space holding head and
 293 * skb_shared_info. @data must have been allocated by kmalloc() only if
 294 * @frag_size is 0, otherwise data should come from the page allocator
 295 *  or vmalloc()
 296 * The return is the new skb buffer.
 297 * On a failure the return is %NULL, and @data is not freed.
 298 * Notes :
 299 *  Before IO, driver allocates only data buffer where NIC put incoming frame
 300 *  Driver should add room at head (NET_SKB_PAD) and
 301 *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
 302 *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
 303 *  before giving packet to stack.
 304 *  RX rings only contains data buffers, not full skbs.
 305 */
 306struct sk_buff *__build_skb(void *data, unsigned int frag_size)
 307{
 
 308	struct sk_buff *skb;
 
 309
 310	skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
 311	if (unlikely(!skb))
 312		return NULL;
 313
 
 
 314	memset(skb, 0, offsetof(struct sk_buff, tail));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 315
 316	return __build_skb_around(skb, data, frag_size);
 317}
 318
 319/* build_skb() is wrapper over __build_skb(), that specifically
 320 * takes care of skb->head and skb->pfmemalloc
 321 * This means that if @frag_size is not zero, then @data must be backed
 322 * by a page fragment, not kmalloc() or vmalloc()
 323 */
 324struct sk_buff *build_skb(void *data, unsigned int frag_size)
 325{
 326	struct sk_buff *skb = __build_skb(data, frag_size);
 327
 328	if (skb && frag_size) {
 329		skb->head_frag = 1;
 330		if (page_is_pfmemalloc(virt_to_head_page(data)))
 331			skb->pfmemalloc = 1;
 332	}
 333	return skb;
 334}
 335EXPORT_SYMBOL(build_skb);
 336
 337/**
 338 * build_skb_around - build a network buffer around provided skb
 339 * @skb: sk_buff provide by caller, must be memset cleared
 340 * @data: data buffer provided by caller
 341 * @frag_size: size of data, or 0 if head was kmalloced
 342 */
 343struct sk_buff *build_skb_around(struct sk_buff *skb,
 344				 void *data, unsigned int frag_size)
 345{
 346	if (unlikely(!skb))
 347		return NULL;
 348
 349	skb = __build_skb_around(skb, data, frag_size);
 350
 351	if (skb && frag_size) {
 352		skb->head_frag = 1;
 353		if (page_is_pfmemalloc(virt_to_head_page(data)))
 354			skb->pfmemalloc = 1;
 355	}
 356	return skb;
 357}
 358EXPORT_SYMBOL(build_skb_around);
 359
 360#define NAPI_SKB_CACHE_SIZE	64
 361
 362struct napi_alloc_cache {
 363	struct page_frag_cache page;
 364	unsigned int skb_count;
 365	void *skb_cache[NAPI_SKB_CACHE_SIZE];
 366};
 367
 368static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
 369static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
 370
 371static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
 372{
 373	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 
 
 374
 375	return page_frag_alloc(&nc->page, fragsz, gfp_mask);
 376}
 377
 378void *napi_alloc_frag(unsigned int fragsz)
 379{
 380	fragsz = SKB_DATA_ALIGN(fragsz);
 381
 382	return __napi_alloc_frag(fragsz, GFP_ATOMIC);
 383}
 384EXPORT_SYMBOL(napi_alloc_frag);
 385
 386/**
 387 * netdev_alloc_frag - allocate a page fragment
 388 * @fragsz: fragment size
 389 *
 390 * Allocates a frag from a page for receive buffer.
 391 * Uses GFP_ATOMIC allocations.
 392 */
 393void *netdev_alloc_frag(unsigned int fragsz)
 394{
 395	struct page_frag_cache *nc;
 396	void *data;
 
 
 
 
 
 
 
 
 397
 398	fragsz = SKB_DATA_ALIGN(fragsz);
 399	if (in_irq() || irqs_disabled()) {
 400		nc = this_cpu_ptr(&netdev_alloc_cache);
 401		data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
 402	} else {
 403		local_bh_disable();
 404		data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
 405		local_bh_enable();
 406	}
 407	return data;
 408}
 409EXPORT_SYMBOL(netdev_alloc_frag);
 410
 411/**
 412 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 413 *	@dev: network device to receive on
 414 *	@len: length to allocate
 415 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 416 *
 417 *	Allocate a new &sk_buff and assign it a usage count of one. The
 418 *	buffer has NET_SKB_PAD headroom built in. Users should allocate
 419 *	the headroom they think they need without accounting for the
 420 *	built in space. The built in space is used for optimisations.
 421 *
 422 *	%NULL is returned if there is no free memory.
 423 */
 424struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
 425				   gfp_t gfp_mask)
 426{
 427	struct page_frag_cache *nc;
 
 428	struct sk_buff *skb;
 429	bool pfmemalloc;
 430	void *data;
 431
 432	len += NET_SKB_PAD;
 433
 434	if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
 435	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 436		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 437		if (!skb)
 438			goto skb_fail;
 439		goto skb_success;
 440	}
 441
 442	len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 443	len = SKB_DATA_ALIGN(len);
 444
 445	if (sk_memalloc_socks())
 446		gfp_mask |= __GFP_MEMALLOC;
 447
 448	if (in_irq() || irqs_disabled()) {
 449		nc = this_cpu_ptr(&netdev_alloc_cache);
 450		data = page_frag_alloc(nc, len, gfp_mask);
 451		pfmemalloc = nc->pfmemalloc;
 452	} else {
 453		local_bh_disable();
 454		nc = this_cpu_ptr(&napi_alloc_cache.page);
 455		data = page_frag_alloc(nc, len, gfp_mask);
 456		pfmemalloc = nc->pfmemalloc;
 457		local_bh_enable();
 458	}
 459
 460	if (unlikely(!data))
 461		return NULL;
 462
 463	skb = __build_skb(data, len);
 464	if (unlikely(!skb)) {
 465		skb_free_frag(data);
 466		return NULL;
 467	}
 468
 469	/* use OR instead of assignment to avoid clearing of bits in mask */
 470	if (pfmemalloc)
 471		skb->pfmemalloc = 1;
 472	skb->head_frag = 1;
 473
 474skb_success:
 475	skb_reserve(skb, NET_SKB_PAD);
 476	skb->dev = dev;
 477
 478skb_fail:
 479	return skb;
 480}
 481EXPORT_SYMBOL(__netdev_alloc_skb);
 482
 483/**
 484 *	__napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
 485 *	@napi: napi instance this buffer was allocated for
 486 *	@len: length to allocate
 487 *	@gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
 488 *
 489 *	Allocate a new sk_buff for use in NAPI receive.  This buffer will
 490 *	attempt to allocate the head from a special reserved region used
 491 *	only for NAPI Rx allocation.  By doing this we can save several
 492 *	CPU cycles by avoiding having to disable and re-enable IRQs.
 493 *
 494 *	%NULL is returned if there is no free memory.
 495 */
 496struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
 497				 gfp_t gfp_mask)
 498{
 499	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 500	struct sk_buff *skb;
 501	void *data;
 502
 503	len += NET_SKB_PAD + NET_IP_ALIGN;
 504
 505	if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
 506	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 507		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 508		if (!skb)
 509			goto skb_fail;
 510		goto skb_success;
 511	}
 512
 513	len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 514	len = SKB_DATA_ALIGN(len);
 515
 516	if (sk_memalloc_socks())
 517		gfp_mask |= __GFP_MEMALLOC;
 518
 519	data = page_frag_alloc(&nc->page, len, gfp_mask);
 520	if (unlikely(!data))
 521		return NULL;
 522
 523	skb = __build_skb(data, len);
 524	if (unlikely(!skb)) {
 525		skb_free_frag(data);
 526		return NULL;
 527	}
 528
 529	/* use OR instead of assignment to avoid clearing of bits in mask */
 530	if (nc->page.pfmemalloc)
 531		skb->pfmemalloc = 1;
 532	skb->head_frag = 1;
 533
 534skb_success:
 535	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
 536	skb->dev = napi->dev;
 537
 538skb_fail:
 539	return skb;
 540}
 541EXPORT_SYMBOL(__napi_alloc_skb);
 542
 543void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 544		     int size, unsigned int truesize)
 545{
 546	skb_fill_page_desc(skb, i, page, off, size);
 547	skb->len += size;
 548	skb->data_len += size;
 549	skb->truesize += truesize;
 550}
 551EXPORT_SYMBOL(skb_add_rx_frag);
 552
 553void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
 554			  unsigned int truesize)
 555{
 556	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 557
 558	skb_frag_size_add(frag, size);
 559	skb->len += size;
 560	skb->data_len += size;
 561	skb->truesize += truesize;
 562}
 563EXPORT_SYMBOL(skb_coalesce_rx_frag);
 564
 565static void skb_drop_list(struct sk_buff **listp)
 566{
 567	kfree_skb_list(*listp);
 568	*listp = NULL;
 569}
 570
 571static inline void skb_drop_fraglist(struct sk_buff *skb)
 572{
 573	skb_drop_list(&skb_shinfo(skb)->frag_list);
 574}
 575
 576static void skb_clone_fraglist(struct sk_buff *skb)
 577{
 578	struct sk_buff *list;
 579
 580	skb_walk_frags(skb, list)
 581		skb_get(list);
 582}
 583
 584static void skb_free_head(struct sk_buff *skb)
 585{
 586	unsigned char *head = skb->head;
 587
 588	if (skb->head_frag)
 589		skb_free_frag(head);
 590	else
 591		kfree(head);
 592}
 593
 594static void skb_release_data(struct sk_buff *skb)
 595{
 596	struct skb_shared_info *shinfo = skb_shinfo(skb);
 597	int i;
 598
 599	if (skb->cloned &&
 600	    atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
 601			      &shinfo->dataref))
 602		return;
 603
 604	for (i = 0; i < shinfo->nr_frags; i++)
 605		__skb_frag_unref(&shinfo->frags[i]);
 606
 
 
 
 
 
 
 
 
 
 
 
 
 607	if (shinfo->frag_list)
 608		kfree_skb_list(shinfo->frag_list);
 609
 610	skb_zcopy_clear(skb, true);
 611	skb_free_head(skb);
 612}
 613
 614/*
 615 *	Free an skbuff by memory without cleaning the state.
 616 */
 617static void kfree_skbmem(struct sk_buff *skb)
 618{
 619	struct sk_buff_fclones *fclones;
 620
 621	switch (skb->fclone) {
 622	case SKB_FCLONE_UNAVAILABLE:
 623		kmem_cache_free(skbuff_head_cache, skb);
 624		return;
 625
 626	case SKB_FCLONE_ORIG:
 627		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 628
 629		/* We usually free the clone (TX completion) before original skb
 630		 * This test would have no chance to be true for the clone,
 631		 * while here, branch prediction will be good.
 632		 */
 633		if (refcount_read(&fclones->fclone_ref) == 1)
 634			goto fastpath;
 635		break;
 636
 637	default: /* SKB_FCLONE_CLONE */
 638		fclones = container_of(skb, struct sk_buff_fclones, skb2);
 639		break;
 640	}
 641	if (!refcount_dec_and_test(&fclones->fclone_ref))
 642		return;
 643fastpath:
 644	kmem_cache_free(skbuff_fclone_cache, fclones);
 645}
 646
 647void skb_release_head_state(struct sk_buff *skb)
 648{
 649	skb_dst_drop(skb);
 
 
 
 650	if (skb->destructor) {
 651		WARN_ON(in_irq());
 652		skb->destructor(skb);
 653	}
 654#if IS_ENABLED(CONFIG_NF_CONNTRACK)
 655	nf_conntrack_put(skb_nfct(skb));
 
 
 
 656#endif
 657	skb_ext_put(skb);
 658}
 659
 660/* Free everything but the sk_buff shell. */
 661static void skb_release_all(struct sk_buff *skb)
 662{
 663	skb_release_head_state(skb);
 664	if (likely(skb->head))
 665		skb_release_data(skb);
 666}
 667
 668/**
 669 *	__kfree_skb - private function
 670 *	@skb: buffer
 671 *
 672 *	Free an sk_buff. Release anything attached to the buffer.
 673 *	Clean the state. This is an internal helper function. Users should
 674 *	always call kfree_skb
 675 */
 676
 677void __kfree_skb(struct sk_buff *skb)
 678{
 679	skb_release_all(skb);
 680	kfree_skbmem(skb);
 681}
 682EXPORT_SYMBOL(__kfree_skb);
 683
 684/**
 685 *	kfree_skb - free an sk_buff
 686 *	@skb: buffer to free
 687 *
 688 *	Drop a reference to the buffer and free it if the usage count has
 689 *	hit zero.
 690 */
 691void kfree_skb(struct sk_buff *skb)
 692{
 693	if (!skb_unref(skb))
 
 
 
 
 694		return;
 695
 696	trace_kfree_skb(skb, __builtin_return_address(0));
 697	__kfree_skb(skb);
 698}
 699EXPORT_SYMBOL(kfree_skb);
 700
 701void kfree_skb_list(struct sk_buff *segs)
 702{
 703	while (segs) {
 704		struct sk_buff *next = segs->next;
 705
 706		kfree_skb(segs);
 707		segs = next;
 708	}
 709}
 710EXPORT_SYMBOL(kfree_skb_list);
 711
 712/* Dump skb information and contents.
 713 *
 714 * Must only be called from net_ratelimit()-ed paths.
 715 *
 716 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
 717 */
 718void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
 719{
 720	static atomic_t can_dump_full = ATOMIC_INIT(5);
 721	struct skb_shared_info *sh = skb_shinfo(skb);
 722	struct net_device *dev = skb->dev;
 723	struct sock *sk = skb->sk;
 724	struct sk_buff *list_skb;
 725	bool has_mac, has_trans;
 726	int headroom, tailroom;
 727	int i, len, seg_len;
 728
 729	if (full_pkt)
 730		full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
 731
 732	if (full_pkt)
 733		len = skb->len;
 734	else
 735		len = min_t(int, skb->len, MAX_HEADER + 128);
 736
 737	headroom = skb_headroom(skb);
 738	tailroom = skb_tailroom(skb);
 739
 740	has_mac = skb_mac_header_was_set(skb);
 741	has_trans = skb_transport_header_was_set(skb);
 742
 743	printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
 744	       "mac=(%d,%d) net=(%d,%d) trans=%d\n"
 745	       "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
 746	       "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
 747	       "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
 748	       level, skb->len, headroom, skb_headlen(skb), tailroom,
 749	       has_mac ? skb->mac_header : -1,
 750	       has_mac ? skb_mac_header_len(skb) : -1,
 751	       skb->network_header,
 752	       has_trans ? skb_network_header_len(skb) : -1,
 753	       has_trans ? skb->transport_header : -1,
 754	       sh->tx_flags, sh->nr_frags,
 755	       sh->gso_size, sh->gso_type, sh->gso_segs,
 756	       skb->csum, skb->ip_summed, skb->csum_complete_sw,
 757	       skb->csum_valid, skb->csum_level,
 758	       skb->hash, skb->sw_hash, skb->l4_hash,
 759	       ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
 760
 761	if (dev)
 762		printk("%sdev name=%s feat=0x%pNF\n",
 763		       level, dev->name, &dev->features);
 764	if (sk)
 765		printk("%ssk family=%hu type=%u proto=%u\n",
 766		       level, sk->sk_family, sk->sk_type, sk->sk_protocol);
 767
 768	if (full_pkt && headroom)
 769		print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
 770			       16, 1, skb->head, headroom, false);
 771
 772	seg_len = min_t(int, skb_headlen(skb), len);
 773	if (seg_len)
 774		print_hex_dump(level, "skb linear:   ", DUMP_PREFIX_OFFSET,
 775			       16, 1, skb->data, seg_len, false);
 776	len -= seg_len;
 777
 778	if (full_pkt && tailroom)
 779		print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
 780			       16, 1, skb_tail_pointer(skb), tailroom, false);
 781
 782	for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
 783		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 784		u32 p_off, p_len, copied;
 785		struct page *p;
 786		u8 *vaddr;
 787
 788		skb_frag_foreach_page(frag, skb_frag_off(frag),
 789				      skb_frag_size(frag), p, p_off, p_len,
 790				      copied) {
 791			seg_len = min_t(int, p_len, len);
 792			vaddr = kmap_atomic(p);
 793			print_hex_dump(level, "skb frag:     ",
 794				       DUMP_PREFIX_OFFSET,
 795				       16, 1, vaddr + p_off, seg_len, false);
 796			kunmap_atomic(vaddr);
 797			len -= seg_len;
 798			if (!len)
 799				break;
 800		}
 801	}
 802
 803	if (full_pkt && skb_has_frag_list(skb)) {
 804		printk("skb fraglist:\n");
 805		skb_walk_frags(skb, list_skb)
 806			skb_dump(level, list_skb, true);
 807	}
 808}
 809EXPORT_SYMBOL(skb_dump);
 810
 811/**
 812 *	skb_tx_error - report an sk_buff xmit error
 813 *	@skb: buffer that triggered an error
 814 *
 815 *	Report xmit error if a device callback is tracking this skb.
 816 *	skb must be freed afterwards.
 817 */
 818void skb_tx_error(struct sk_buff *skb)
 819{
 820	skb_zcopy_clear(skb, true);
 
 
 
 
 
 
 
 821}
 822EXPORT_SYMBOL(skb_tx_error);
 823
 824/**
 825 *	consume_skb - free an skbuff
 826 *	@skb: buffer to free
 827 *
 828 *	Drop a ref to the buffer and free it if the usage count has hit zero
 829 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
 830 *	is being dropped after a failure and notes that
 831 */
 832void consume_skb(struct sk_buff *skb)
 833{
 834	if (!skb_unref(skb))
 
 
 
 
 835		return;
 836
 837	trace_consume_skb(skb);
 838	__kfree_skb(skb);
 839}
 840EXPORT_SYMBOL(consume_skb);
 841
 842/**
 843 *	consume_stateless_skb - free an skbuff, assuming it is stateless
 844 *	@skb: buffer to free
 845 *
 846 *	Alike consume_skb(), but this variant assumes that this is the last
 847 *	skb reference and all the head states have been already dropped
 848 */
 849void __consume_stateless_skb(struct sk_buff *skb)
 850{
 851	trace_consume_skb(skb);
 852	skb_release_data(skb);
 853	kfree_skbmem(skb);
 854}
 855
 856void __kfree_skb_flush(void)
 857{
 858	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 859
 860	/* flush skb_cache if containing objects */
 861	if (nc->skb_count) {
 862		kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
 863				     nc->skb_cache);
 864		nc->skb_count = 0;
 865	}
 866}
 867
 868static inline void _kfree_skb_defer(struct sk_buff *skb)
 869{
 870	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 871
 872	/* drop skb->head and call any destructors for packet */
 873	skb_release_all(skb);
 874
 875	/* record skb to CPU local list */
 876	nc->skb_cache[nc->skb_count++] = skb;
 877
 878#ifdef CONFIG_SLUB
 879	/* SLUB writes into objects when freeing */
 880	prefetchw(skb);
 881#endif
 882
 883	/* flush skb_cache if it is filled */
 884	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
 885		kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
 886				     nc->skb_cache);
 887		nc->skb_count = 0;
 888	}
 889}
 890void __kfree_skb_defer(struct sk_buff *skb)
 891{
 892	_kfree_skb_defer(skb);
 893}
 894
 895void napi_consume_skb(struct sk_buff *skb, int budget)
 896{
 897	if (unlikely(!skb))
 898		return;
 899
 900	/* Zero budget indicate non-NAPI context called us, like netpoll */
 901	if (unlikely(!budget)) {
 902		dev_consume_skb_any(skb);
 903		return;
 904	}
 905
 906	if (!skb_unref(skb))
 
 
 907		return;
 908
 909	/* if reaching here SKB is ready to free */
 910	trace_consume_skb(skb);
 911
 912	/* if SKB is a clone, don't handle this case */
 913	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
 914		__kfree_skb(skb);
 915		return;
 916	}
 917
 918	_kfree_skb_defer(skb);
 919}
 920EXPORT_SYMBOL(napi_consume_skb);
 921
 922/* Make sure a field is enclosed inside headers_start/headers_end section */
 923#define CHECK_SKB_FIELD(field) \
 924	BUILD_BUG_ON(offsetof(struct sk_buff, field) <		\
 925		     offsetof(struct sk_buff, headers_start));	\
 926	BUILD_BUG_ON(offsetof(struct sk_buff, field) >		\
 927		     offsetof(struct sk_buff, headers_end));	\
 928
 929static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 930{
 931	new->tstamp		= old->tstamp;
 932	/* We do not copy old->sk */
 933	new->dev		= old->dev;
 934	memcpy(new->cb, old->cb, sizeof(old->cb));
 935	skb_dst_copy(new, old);
 936	__skb_ext_copy(new, old);
 
 
 937	__nf_copy(new, old, false);
 938
 939	/* Note : this field could be in headers_start/headers_end section
 940	 * It is not yet because we do not want to have a 16 bit hole
 941	 */
 942	new->queue_mapping = old->queue_mapping;
 943
 944	memcpy(&new->headers_start, &old->headers_start,
 945	       offsetof(struct sk_buff, headers_end) -
 946	       offsetof(struct sk_buff, headers_start));
 947	CHECK_SKB_FIELD(protocol);
 948	CHECK_SKB_FIELD(csum);
 949	CHECK_SKB_FIELD(hash);
 950	CHECK_SKB_FIELD(priority);
 951	CHECK_SKB_FIELD(skb_iif);
 952	CHECK_SKB_FIELD(vlan_proto);
 953	CHECK_SKB_FIELD(vlan_tci);
 954	CHECK_SKB_FIELD(transport_header);
 955	CHECK_SKB_FIELD(network_header);
 956	CHECK_SKB_FIELD(mac_header);
 957	CHECK_SKB_FIELD(inner_protocol);
 958	CHECK_SKB_FIELD(inner_transport_header);
 959	CHECK_SKB_FIELD(inner_network_header);
 960	CHECK_SKB_FIELD(inner_mac_header);
 961	CHECK_SKB_FIELD(mark);
 962#ifdef CONFIG_NETWORK_SECMARK
 963	CHECK_SKB_FIELD(secmark);
 964#endif
 965#ifdef CONFIG_NET_RX_BUSY_POLL
 966	CHECK_SKB_FIELD(napi_id);
 967#endif
 968#ifdef CONFIG_XPS
 969	CHECK_SKB_FIELD(sender_cpu);
 970#endif
 971#ifdef CONFIG_NET_SCHED
 972	CHECK_SKB_FIELD(tc_index);
 
 
 
 973#endif
 974
 975}
 976
 977/*
 978 * You should not add any new code to this function.  Add it to
 979 * __copy_skb_header above instead.
 980 */
 981static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 982{
 983#define C(x) n->x = skb->x
 984
 985	n->next = n->prev = NULL;
 986	n->sk = NULL;
 987	__copy_skb_header(n, skb);
 988
 989	C(len);
 990	C(data_len);
 991	C(mac_len);
 992	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 993	n->cloned = 1;
 994	n->nohdr = 0;
 995	n->peeked = 0;
 996	C(pfmemalloc);
 997	n->destructor = NULL;
 998	C(tail);
 999	C(end);
1000	C(head);
1001	C(head_frag);
1002	C(data);
1003	C(truesize);
1004	refcount_set(&n->users, 1);
1005
1006	atomic_inc(&(skb_shinfo(skb)->dataref));
1007	skb->cloned = 1;
1008
1009	return n;
1010#undef C
1011}
1012
1013/**
1014 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1015 * @first: first sk_buff of the msg
1016 */
1017struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1018{
1019	struct sk_buff *n;
1020
1021	n = alloc_skb(0, GFP_ATOMIC);
1022	if (!n)
1023		return NULL;
1024
1025	n->len = first->len;
1026	n->data_len = first->len;
1027	n->truesize = first->truesize;
1028
1029	skb_shinfo(n)->frag_list = first;
1030
1031	__copy_skb_header(n, first);
1032	n->destructor = NULL;
1033
1034	return n;
1035}
1036EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1037
1038/**
1039 *	skb_morph	-	morph one skb into another
1040 *	@dst: the skb to receive the contents
1041 *	@src: the skb to supply the contents
1042 *
1043 *	This is identical to skb_clone except that the target skb is
1044 *	supplied by the user.
1045 *
1046 *	The target skb is returned upon exit.
1047 */
1048struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1049{
1050	skb_release_all(dst);
1051	return __skb_clone(dst, src);
1052}
1053EXPORT_SYMBOL_GPL(skb_morph);
1054
1055int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1056{
1057	unsigned long max_pg, num_pg, new_pg, old_pg;
1058	struct user_struct *user;
1059
1060	if (capable(CAP_IPC_LOCK) || !size)
1061		return 0;
1062
1063	num_pg = (size >> PAGE_SHIFT) + 2;	/* worst case */
1064	max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1065	user = mmp->user ? : current_user();
1066
1067	do {
1068		old_pg = atomic_long_read(&user->locked_vm);
1069		new_pg = old_pg + num_pg;
1070		if (new_pg > max_pg)
1071			return -ENOBUFS;
1072	} while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1073		 old_pg);
1074
1075	if (!mmp->user) {
1076		mmp->user = get_uid(user);
1077		mmp->num_pg = num_pg;
1078	} else {
1079		mmp->num_pg += num_pg;
1080	}
1081
1082	return 0;
1083}
1084EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1085
1086void mm_unaccount_pinned_pages(struct mmpin *mmp)
1087{
1088	if (mmp->user) {
1089		atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1090		free_uid(mmp->user);
1091	}
1092}
1093EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1094
1095struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1096{
1097	struct ubuf_info *uarg;
1098	struct sk_buff *skb;
1099
1100	WARN_ON_ONCE(!in_task());
1101
1102	skb = sock_omalloc(sk, 0, GFP_KERNEL);
1103	if (!skb)
1104		return NULL;
1105
1106	BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1107	uarg = (void *)skb->cb;
1108	uarg->mmp.user = NULL;
1109
1110	if (mm_account_pinned_pages(&uarg->mmp, size)) {
1111		kfree_skb(skb);
1112		return NULL;
1113	}
1114
1115	uarg->callback = sock_zerocopy_callback;
1116	uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1117	uarg->len = 1;
1118	uarg->bytelen = size;
1119	uarg->zerocopy = 1;
1120	refcount_set(&uarg->refcnt, 1);
1121	sock_hold(sk);
1122
1123	return uarg;
1124}
1125EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1126
1127static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1128{
1129	return container_of((void *)uarg, struct sk_buff, cb);
1130}
1131
1132struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1133					struct ubuf_info *uarg)
1134{
1135	if (uarg) {
1136		const u32 byte_limit = 1 << 19;		/* limit to a few TSO */
1137		u32 bytelen, next;
1138
1139		/* realloc only when socket is locked (TCP, UDP cork),
1140		 * so uarg->len and sk_zckey access is serialized
1141		 */
1142		if (!sock_owned_by_user(sk)) {
1143			WARN_ON_ONCE(1);
1144			return NULL;
1145		}
1146
1147		bytelen = uarg->bytelen + size;
1148		if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1149			/* TCP can create new skb to attach new uarg */
1150			if (sk->sk_type == SOCK_STREAM)
1151				goto new_alloc;
1152			return NULL;
1153		}
1154
1155		next = (u32)atomic_read(&sk->sk_zckey);
1156		if ((u32)(uarg->id + uarg->len) == next) {
1157			if (mm_account_pinned_pages(&uarg->mmp, size))
1158				return NULL;
1159			uarg->len++;
1160			uarg->bytelen = bytelen;
1161			atomic_set(&sk->sk_zckey, ++next);
1162
1163			/* no extra ref when appending to datagram (MSG_MORE) */
1164			if (sk->sk_type == SOCK_STREAM)
1165				sock_zerocopy_get(uarg);
1166
1167			return uarg;
1168		}
1169	}
1170
1171new_alloc:
1172	return sock_zerocopy_alloc(sk, size);
1173}
1174EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1175
1176static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1177{
1178	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1179	u32 old_lo, old_hi;
1180	u64 sum_len;
1181
1182	old_lo = serr->ee.ee_info;
1183	old_hi = serr->ee.ee_data;
1184	sum_len = old_hi - old_lo + 1ULL + len;
1185
1186	if (sum_len >= (1ULL << 32))
1187		return false;
1188
1189	if (lo != old_hi + 1)
1190		return false;
1191
1192	serr->ee.ee_data += len;
1193	return true;
1194}
1195
1196void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1197{
1198	struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1199	struct sock_exterr_skb *serr;
1200	struct sock *sk = skb->sk;
1201	struct sk_buff_head *q;
1202	unsigned long flags;
1203	u32 lo, hi;
1204	u16 len;
1205
1206	mm_unaccount_pinned_pages(&uarg->mmp);
1207
1208	/* if !len, there was only 1 call, and it was aborted
1209	 * so do not queue a completion notification
1210	 */
1211	if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1212		goto release;
1213
1214	len = uarg->len;
1215	lo = uarg->id;
1216	hi = uarg->id + len - 1;
1217
1218	serr = SKB_EXT_ERR(skb);
1219	memset(serr, 0, sizeof(*serr));
1220	serr->ee.ee_errno = 0;
1221	serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1222	serr->ee.ee_data = hi;
1223	serr->ee.ee_info = lo;
1224	if (!success)
1225		serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1226
1227	q = &sk->sk_error_queue;
1228	spin_lock_irqsave(&q->lock, flags);
1229	tail = skb_peek_tail(q);
1230	if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1231	    !skb_zerocopy_notify_extend(tail, lo, len)) {
1232		__skb_queue_tail(q, skb);
1233		skb = NULL;
1234	}
1235	spin_unlock_irqrestore(&q->lock, flags);
1236
1237	sk->sk_error_report(sk);
1238
1239release:
1240	consume_skb(skb);
1241	sock_put(sk);
1242}
1243EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1244
1245void sock_zerocopy_put(struct ubuf_info *uarg)
1246{
1247	if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1248		if (uarg->callback)
1249			uarg->callback(uarg, uarg->zerocopy);
1250		else
1251			consume_skb(skb_from_uarg(uarg));
1252	}
1253}
1254EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1255
1256void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1257{
1258	if (uarg) {
1259		struct sock *sk = skb_from_uarg(uarg)->sk;
1260
1261		atomic_dec(&sk->sk_zckey);
1262		uarg->len--;
1263
1264		if (have_uref)
1265			sock_zerocopy_put(uarg);
1266	}
1267}
1268EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1269
1270int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1271{
1272	return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1273}
1274EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1275
1276int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1277			     struct msghdr *msg, int len,
1278			     struct ubuf_info *uarg)
1279{
1280	struct ubuf_info *orig_uarg = skb_zcopy(skb);
1281	struct iov_iter orig_iter = msg->msg_iter;
1282	int err, orig_len = skb->len;
1283
1284	/* An skb can only point to one uarg. This edge case happens when
1285	 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1286	 */
1287	if (orig_uarg && uarg != orig_uarg)
1288		return -EEXIST;
1289
1290	err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1291	if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1292		struct sock *save_sk = skb->sk;
1293
1294		/* Streams do not free skb on error. Reset to prev state. */
1295		msg->msg_iter = orig_iter;
1296		skb->sk = sk;
1297		___pskb_trim(skb, orig_len);
1298		skb->sk = save_sk;
1299		return err;
1300	}
1301
1302	skb_zcopy_set(skb, uarg, NULL);
1303	return skb->len - orig_len;
1304}
1305EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1306
1307static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1308			      gfp_t gfp_mask)
1309{
1310	if (skb_zcopy(orig)) {
1311		if (skb_zcopy(nskb)) {
1312			/* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1313			if (!gfp_mask) {
1314				WARN_ON_ONCE(1);
1315				return -ENOMEM;
1316			}
1317			if (skb_uarg(nskb) == skb_uarg(orig))
1318				return 0;
1319			if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1320				return -EIO;
1321		}
1322		skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1323	}
1324	return 0;
1325}
1326
1327/**
1328 *	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
1329 *	@skb: the skb to modify
1330 *	@gfp_mask: allocation priority
1331 *
1332 *	This must be called on SKBTX_DEV_ZEROCOPY skb.
1333 *	It will copy all frags into kernel and drop the reference
1334 *	to userspace pages.
1335 *
1336 *	If this function is called from an interrupt gfp_mask() must be
1337 *	%GFP_ATOMIC.
1338 *
1339 *	Returns 0 on success or a negative error code on failure
1340 *	to allocate kernel memory to copy to.
1341 */
1342int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1343{
 
1344	int num_frags = skb_shinfo(skb)->nr_frags;
1345	struct page *page, *head = NULL;
1346	int i, new_frags;
1347	u32 d_off;
1348
1349	if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1350		return -EINVAL;
 
1351
1352	if (!num_frags)
1353		goto release;
1354
1355	new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1356	for (i = 0; i < new_frags; i++) {
1357		page = alloc_page(gfp_mask);
1358		if (!page) {
1359			while (head) {
1360				struct page *next = (struct page *)page_private(head);
1361				put_page(head);
1362				head = next;
1363			}
1364			return -ENOMEM;
1365		}
 
 
 
 
1366		set_page_private(page, (unsigned long)head);
1367		head = page;
1368	}
1369
1370	page = head;
1371	d_off = 0;
1372	for (i = 0; i < num_frags; i++) {
1373		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1374		u32 p_off, p_len, copied;
1375		struct page *p;
1376		u8 *vaddr;
1377
1378		skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1379				      p, p_off, p_len, copied) {
1380			u32 copy, done = 0;
1381			vaddr = kmap_atomic(p);
1382
1383			while (done < p_len) {
1384				if (d_off == PAGE_SIZE) {
1385					d_off = 0;
1386					page = (struct page *)page_private(page);
1387				}
1388				copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1389				memcpy(page_address(page) + d_off,
1390				       vaddr + p_off + done, copy);
1391				done += copy;
1392				d_off += copy;
1393			}
1394			kunmap_atomic(vaddr);
1395		}
1396	}
1397
1398	/* skb frags release userspace buffers */
1399	for (i = 0; i < num_frags; i++)
1400		skb_frag_unref(skb, i);
1401
 
 
1402	/* skb frags point to kernel buffers */
1403	for (i = 0; i < new_frags - 1; i++) {
1404		__skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
 
1405		head = (struct page *)page_private(head);
1406	}
1407	__skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1408	skb_shinfo(skb)->nr_frags = new_frags;
1409
1410release:
1411	skb_zcopy_clear(skb, false);
1412	return 0;
1413}
1414EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1415
1416/**
1417 *	skb_clone	-	duplicate an sk_buff
1418 *	@skb: buffer to clone
1419 *	@gfp_mask: allocation priority
1420 *
1421 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
1422 *	copies share the same packet data but not structure. The new
1423 *	buffer has a reference count of 1. If the allocation fails the
1424 *	function returns %NULL otherwise the new buffer is returned.
1425 *
1426 *	If this function is called from an interrupt gfp_mask() must be
1427 *	%GFP_ATOMIC.
1428 */
1429
1430struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1431{
1432	struct sk_buff_fclones *fclones = container_of(skb,
1433						       struct sk_buff_fclones,
1434						       skb1);
1435	struct sk_buff *n;
1436
1437	if (skb_orphan_frags(skb, gfp_mask))
1438		return NULL;
1439
1440	if (skb->fclone == SKB_FCLONE_ORIG &&
1441	    refcount_read(&fclones->fclone_ref) == 1) {
1442		n = &fclones->skb2;
1443		refcount_set(&fclones->fclone_ref, 2);
1444	} else {
1445		if (skb_pfmemalloc(skb))
1446			gfp_mask |= __GFP_MEMALLOC;
1447
1448		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1449		if (!n)
1450			return NULL;
1451
 
1452		n->fclone = SKB_FCLONE_UNAVAILABLE;
1453	}
1454
1455	return __skb_clone(n, skb);
1456}
1457EXPORT_SYMBOL(skb_clone);
1458
1459void skb_headers_offset_update(struct sk_buff *skb, int off)
1460{
1461	/* Only adjust this if it actually is csum_start rather than csum */
1462	if (skb->ip_summed == CHECKSUM_PARTIAL)
1463		skb->csum_start += off;
1464	/* {transport,network,mac}_header and tail are relative to skb->head */
1465	skb->transport_header += off;
1466	skb->network_header   += off;
1467	if (skb_mac_header_was_set(skb))
1468		skb->mac_header += off;
1469	skb->inner_transport_header += off;
1470	skb->inner_network_header += off;
1471	skb->inner_mac_header += off;
1472}
1473EXPORT_SYMBOL(skb_headers_offset_update);
1474
1475void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1476{
1477	__copy_skb_header(new, old);
1478
1479	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1480	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1481	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1482}
1483EXPORT_SYMBOL(skb_copy_header);
1484
1485static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1486{
1487	if (skb_pfmemalloc(skb))
1488		return SKB_ALLOC_RX;
1489	return 0;
1490}
1491
1492/**
1493 *	skb_copy	-	create private copy of an sk_buff
1494 *	@skb: buffer to copy
1495 *	@gfp_mask: allocation priority
1496 *
1497 *	Make a copy of both an &sk_buff and its data. This is used when the
1498 *	caller wishes to modify the data and needs a private copy of the
1499 *	data to alter. Returns %NULL on failure or the pointer to the buffer
1500 *	on success. The returned buffer has a reference count of 1.
1501 *
1502 *	As by-product this function converts non-linear &sk_buff to linear
1503 *	one, so that &sk_buff becomes completely private and caller is allowed
1504 *	to modify all the data of returned buffer. This means that this
1505 *	function is not recommended for use in circumstances when only
1506 *	header is going to be modified. Use pskb_copy() instead.
1507 */
1508
1509struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1510{
1511	int headerlen = skb_headroom(skb);
1512	unsigned int size = skb_end_offset(skb) + skb->data_len;
1513	struct sk_buff *n = __alloc_skb(size, gfp_mask,
1514					skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1515
1516	if (!n)
1517		return NULL;
1518
1519	/* Set the data pointer */
1520	skb_reserve(n, headerlen);
1521	/* Set the tail pointer and length */
1522	skb_put(n, skb->len);
1523
1524	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
 
1525
1526	skb_copy_header(n, skb);
1527	return n;
1528}
1529EXPORT_SYMBOL(skb_copy);
1530
1531/**
1532 *	__pskb_copy_fclone	-  create copy of an sk_buff with private head.
1533 *	@skb: buffer to copy
1534 *	@headroom: headroom of new skb
1535 *	@gfp_mask: allocation priority
1536 *	@fclone: if true allocate the copy of the skb from the fclone
1537 *	cache instead of the head cache; it is recommended to set this
1538 *	to true for the cases where the copy will likely be cloned
1539 *
1540 *	Make a copy of both an &sk_buff and part of its data, located
1541 *	in header. Fragmented data remain shared. This is used when
1542 *	the caller wishes to modify only header of &sk_buff and needs
1543 *	private copy of the header to alter. Returns %NULL on failure
1544 *	or the pointer to the buffer on success.
1545 *	The returned buffer has a reference count of 1.
1546 */
1547
1548struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1549				   gfp_t gfp_mask, bool fclone)
1550{
1551	unsigned int size = skb_headlen(skb) + headroom;
1552	int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1553	struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1554
1555	if (!n)
1556		goto out;
1557
1558	/* Set the data pointer */
1559	skb_reserve(n, headroom);
1560	/* Set the tail pointer and length */
1561	skb_put(n, skb_headlen(skb));
1562	/* Copy the bytes */
1563	skb_copy_from_linear_data(skb, n->data, n->len);
1564
1565	n->truesize += skb->data_len;
1566	n->data_len  = skb->data_len;
1567	n->len	     = skb->len;
1568
1569	if (skb_shinfo(skb)->nr_frags) {
1570		int i;
1571
1572		if (skb_orphan_frags(skb, gfp_mask) ||
1573		    skb_zerocopy_clone(n, skb, gfp_mask)) {
1574			kfree_skb(n);
1575			n = NULL;
1576			goto out;
1577		}
1578		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1579			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1580			skb_frag_ref(skb, i);
1581		}
1582		skb_shinfo(n)->nr_frags = i;
1583	}
1584
1585	if (skb_has_frag_list(skb)) {
1586		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1587		skb_clone_fraglist(n);
1588	}
1589
1590	skb_copy_header(n, skb);
1591out:
1592	return n;
1593}
1594EXPORT_SYMBOL(__pskb_copy_fclone);
1595
1596/**
1597 *	pskb_expand_head - reallocate header of &sk_buff
1598 *	@skb: buffer to reallocate
1599 *	@nhead: room to add at head
1600 *	@ntail: room to add at tail
1601 *	@gfp_mask: allocation priority
1602 *
1603 *	Expands (or creates identical copy, if @nhead and @ntail are zero)
1604 *	header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1605 *	reference count of 1. Returns zero in the case of success or error,
1606 *	if expansion failed. In the last case, &sk_buff is not changed.
1607 *
1608 *	All the pointers pointing into skb header may change and must be
1609 *	reloaded after call to this function.
1610 */
1611
1612int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1613		     gfp_t gfp_mask)
1614{
1615	int i, osize = skb_end_offset(skb);
1616	int size = osize + nhead + ntail;
 
1617	long off;
1618	u8 *data;
1619
1620	BUG_ON(nhead < 0);
1621
1622	BUG_ON(skb_shared(skb));
 
1623
1624	size = SKB_DATA_ALIGN(size);
1625
1626	if (skb_pfmemalloc(skb))
1627		gfp_mask |= __GFP_MEMALLOC;
1628	data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1629			       gfp_mask, NUMA_NO_NODE, NULL);
1630	if (!data)
1631		goto nodata;
1632	size = SKB_WITH_OVERHEAD(ksize(data));
1633
1634	/* Copy only real data... and, alas, header. This should be
1635	 * optimized for the cases when header is void.
1636	 */
1637	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1638
1639	memcpy((struct skb_shared_info *)(data + size),
1640	       skb_shinfo(skb),
1641	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1642
1643	/*
1644	 * if shinfo is shared we must drop the old head gracefully, but if it
1645	 * is not we can just drop the old head and let the existing refcount
1646	 * be since all we did is relocate the values
1647	 */
1648	if (skb_cloned(skb)) {
 
1649		if (skb_orphan_frags(skb, gfp_mask))
1650			goto nofrags;
1651		if (skb_zcopy(skb))
1652			refcount_inc(&skb_uarg(skb)->refcnt);
1653		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1654			skb_frag_ref(skb, i);
1655
1656		if (skb_has_frag_list(skb))
1657			skb_clone_fraglist(skb);
1658
1659		skb_release_data(skb);
1660	} else {
1661		skb_free_head(skb);
1662	}
1663	off = (data + nhead) - skb->head;
1664
1665	skb->head     = data;
1666	skb->head_frag = 0;
1667	skb->data    += off;
1668#ifdef NET_SKBUFF_DATA_USES_OFFSET
1669	skb->end      = size;
1670	off           = nhead;
1671#else
1672	skb->end      = skb->head + size;
1673#endif
1674	skb->tail	      += off;
1675	skb_headers_offset_update(skb, nhead);
1676	skb->cloned   = 0;
1677	skb->hdr_len  = 0;
1678	skb->nohdr    = 0;
1679	atomic_set(&skb_shinfo(skb)->dataref, 1);
1680
1681	skb_metadata_clear(skb);
1682
1683	/* It is not generally safe to change skb->truesize.
1684	 * For the moment, we really care of rx path, or
1685	 * when skb is orphaned (not attached to a socket).
1686	 */
1687	if (!skb->sk || skb->destructor == sock_edemux)
1688		skb->truesize += size - osize;
1689
1690	return 0;
1691
1692nofrags:
1693	kfree(data);
1694nodata:
1695	return -ENOMEM;
1696}
1697EXPORT_SYMBOL(pskb_expand_head);
1698
1699/* Make private copy of skb with writable head and some headroom */
1700
1701struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1702{
1703	struct sk_buff *skb2;
1704	int delta = headroom - skb_headroom(skb);
1705
1706	if (delta <= 0)
1707		skb2 = pskb_copy(skb, GFP_ATOMIC);
1708	else {
1709		skb2 = skb_clone(skb, GFP_ATOMIC);
1710		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1711					     GFP_ATOMIC)) {
1712			kfree_skb(skb2);
1713			skb2 = NULL;
1714		}
1715	}
1716	return skb2;
1717}
1718EXPORT_SYMBOL(skb_realloc_headroom);
1719
1720/**
1721 *	skb_copy_expand	-	copy and expand sk_buff
1722 *	@skb: buffer to copy
1723 *	@newheadroom: new free bytes at head
1724 *	@newtailroom: new free bytes at tail
1725 *	@gfp_mask: allocation priority
1726 *
1727 *	Make a copy of both an &sk_buff and its data and while doing so
1728 *	allocate additional space.
1729 *
1730 *	This is used when the caller wishes to modify the data and needs a
1731 *	private copy of the data to alter as well as more space for new fields.
1732 *	Returns %NULL on failure or the pointer to the buffer
1733 *	on success. The returned buffer has a reference count of 1.
1734 *
1735 *	You must pass %GFP_ATOMIC as the allocation priority if this function
1736 *	is called from an interrupt.
1737 */
1738struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1739				int newheadroom, int newtailroom,
1740				gfp_t gfp_mask)
1741{
1742	/*
1743	 *	Allocate the copy buffer
1744	 */
1745	struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1746					gfp_mask, skb_alloc_rx_flag(skb),
1747					NUMA_NO_NODE);
1748	int oldheadroom = skb_headroom(skb);
1749	int head_copy_len, head_copy_off;
1750
1751	if (!n)
1752		return NULL;
1753
1754	skb_reserve(n, newheadroom);
1755
1756	/* Set the tail pointer and length */
1757	skb_put(n, skb->len);
1758
1759	head_copy_len = oldheadroom;
1760	head_copy_off = 0;
1761	if (newheadroom <= head_copy_len)
1762		head_copy_len = newheadroom;
1763	else
1764		head_copy_off = newheadroom - head_copy_len;
1765
1766	/* Copy the linear header and data. */
1767	BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1768			     skb->len + head_copy_len));
 
1769
1770	skb_copy_header(n, skb);
1771
1772	skb_headers_offset_update(n, newheadroom - oldheadroom);
1773
1774	return n;
1775}
1776EXPORT_SYMBOL(skb_copy_expand);
1777
1778/**
1779 *	__skb_pad		-	zero pad the tail of an skb
1780 *	@skb: buffer to pad
1781 *	@pad: space to pad
1782 *	@free_on_error: free buffer on error
1783 *
1784 *	Ensure that a buffer is followed by a padding area that is zero
1785 *	filled. Used by network drivers which may DMA or transfer data
1786 *	beyond the buffer end onto the wire.
1787 *
1788 *	May return error in out of memory cases. The skb is freed on error
1789 *	if @free_on_error is true.
1790 */
1791
1792int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1793{
1794	int err;
1795	int ntail;
1796
1797	/* If the skbuff is non linear tailroom is always zero.. */
1798	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1799		memset(skb->data+skb->len, 0, pad);
1800		return 0;
1801	}
1802
1803	ntail = skb->data_len + pad - (skb->end - skb->tail);
1804	if (likely(skb_cloned(skb) || ntail > 0)) {
1805		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1806		if (unlikely(err))
1807			goto free_skb;
1808	}
1809
1810	/* FIXME: The use of this function with non-linear skb's really needs
1811	 * to be audited.
1812	 */
1813	err = skb_linearize(skb);
1814	if (unlikely(err))
1815		goto free_skb;
1816
1817	memset(skb->data + skb->len, 0, pad);
1818	return 0;
1819
1820free_skb:
1821	if (free_on_error)
1822		kfree_skb(skb);
1823	return err;
1824}
1825EXPORT_SYMBOL(__skb_pad);
1826
1827/**
1828 *	pskb_put - add data to the tail of a potentially fragmented buffer
1829 *	@skb: start of the buffer to use
1830 *	@tail: tail fragment of the buffer to use
1831 *	@len: amount of data to add
1832 *
1833 *	This function extends the used data area of the potentially
1834 *	fragmented buffer. @tail must be the last fragment of @skb -- or
1835 *	@skb itself. If this would exceed the total buffer size the kernel
1836 *	will panic. A pointer to the first byte of the extra data is
1837 *	returned.
1838 */
1839
1840void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1841{
1842	if (tail != skb) {
1843		skb->data_len += len;
1844		skb->len += len;
1845	}
1846	return skb_put(tail, len);
1847}
1848EXPORT_SYMBOL_GPL(pskb_put);
1849
1850/**
1851 *	skb_put - add data to a buffer
1852 *	@skb: buffer to use
1853 *	@len: amount of data to add
1854 *
1855 *	This function extends the used data area of the buffer. If this would
1856 *	exceed the total buffer size the kernel will panic. A pointer to the
1857 *	first byte of the extra data is returned.
1858 */
1859void *skb_put(struct sk_buff *skb, unsigned int len)
1860{
1861	void *tmp = skb_tail_pointer(skb);
1862	SKB_LINEAR_ASSERT(skb);
1863	skb->tail += len;
1864	skb->len  += len;
1865	if (unlikely(skb->tail > skb->end))
1866		skb_over_panic(skb, len, __builtin_return_address(0));
1867	return tmp;
1868}
1869EXPORT_SYMBOL(skb_put);
1870
1871/**
1872 *	skb_push - add data to the start of a buffer
1873 *	@skb: buffer to use
1874 *	@len: amount of data to add
1875 *
1876 *	This function extends the used data area of the buffer at the buffer
1877 *	start. If this would exceed the total buffer headroom the kernel will
1878 *	panic. A pointer to the first byte of the extra data is returned.
1879 */
1880void *skb_push(struct sk_buff *skb, unsigned int len)
1881{
1882	skb->data -= len;
1883	skb->len  += len;
1884	if (unlikely(skb->data < skb->head))
1885		skb_under_panic(skb, len, __builtin_return_address(0));
1886	return skb->data;
1887}
1888EXPORT_SYMBOL(skb_push);
1889
1890/**
1891 *	skb_pull - remove data from the start of a buffer
1892 *	@skb: buffer to use
1893 *	@len: amount of data to remove
1894 *
1895 *	This function removes data from the start of a buffer, returning
1896 *	the memory to the headroom. A pointer to the next data in the buffer
1897 *	is returned. Once the data has been pulled future pushes will overwrite
1898 *	the old data.
1899 */
1900void *skb_pull(struct sk_buff *skb, unsigned int len)
1901{
1902	return skb_pull_inline(skb, len);
1903}
1904EXPORT_SYMBOL(skb_pull);
1905
1906/**
1907 *	skb_trim - remove end from a buffer
1908 *	@skb: buffer to alter
1909 *	@len: new length
1910 *
1911 *	Cut the length of a buffer down by removing data from the tail. If
1912 *	the buffer is already under the length specified it is not modified.
1913 *	The skb must be linear.
1914 */
1915void skb_trim(struct sk_buff *skb, unsigned int len)
1916{
1917	if (skb->len > len)
1918		__skb_trim(skb, len);
1919}
1920EXPORT_SYMBOL(skb_trim);
1921
1922/* Trims skb to length len. It can change skb pointers.
1923 */
1924
1925int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1926{
1927	struct sk_buff **fragp;
1928	struct sk_buff *frag;
1929	int offset = skb_headlen(skb);
1930	int nfrags = skb_shinfo(skb)->nr_frags;
1931	int i;
1932	int err;
1933
1934	if (skb_cloned(skb) &&
1935	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1936		return err;
1937
1938	i = 0;
1939	if (offset >= len)
1940		goto drop_pages;
1941
1942	for (; i < nfrags; i++) {
1943		int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1944
1945		if (end < len) {
1946			offset = end;
1947			continue;
1948		}
1949
1950		skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1951
1952drop_pages:
1953		skb_shinfo(skb)->nr_frags = i;
1954
1955		for (; i < nfrags; i++)
1956			skb_frag_unref(skb, i);
1957
1958		if (skb_has_frag_list(skb))
1959			skb_drop_fraglist(skb);
1960		goto done;
1961	}
1962
1963	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1964	     fragp = &frag->next) {
1965		int end = offset + frag->len;
1966
1967		if (skb_shared(frag)) {
1968			struct sk_buff *nfrag;
1969
1970			nfrag = skb_clone(frag, GFP_ATOMIC);
1971			if (unlikely(!nfrag))
1972				return -ENOMEM;
1973
1974			nfrag->next = frag->next;
1975			consume_skb(frag);
1976			frag = nfrag;
1977			*fragp = frag;
1978		}
1979
1980		if (end < len) {
1981			offset = end;
1982			continue;
1983		}
1984
1985		if (end > len &&
1986		    unlikely((err = pskb_trim(frag, len - offset))))
1987			return err;
1988
1989		if (frag->next)
1990			skb_drop_list(&frag->next);
1991		break;
1992	}
1993
1994done:
1995	if (len > skb_headlen(skb)) {
1996		skb->data_len -= skb->len - len;
1997		skb->len       = len;
1998	} else {
1999		skb->len       = len;
2000		skb->data_len  = 0;
2001		skb_set_tail_pointer(skb, len);
2002	}
2003
2004	if (!skb->sk || skb->destructor == sock_edemux)
2005		skb_condense(skb);
2006	return 0;
2007}
2008EXPORT_SYMBOL(___pskb_trim);
2009
2010/* Note : use pskb_trim_rcsum() instead of calling this directly
2011 */
2012int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2013{
2014	if (skb->ip_summed == CHECKSUM_COMPLETE) {
2015		int delta = skb->len - len;
2016
2017		skb->csum = csum_block_sub(skb->csum,
2018					   skb_checksum(skb, len, delta, 0),
2019					   len);
2020	}
2021	return __pskb_trim(skb, len);
2022}
2023EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2024
2025/**
2026 *	__pskb_pull_tail - advance tail of skb header
2027 *	@skb: buffer to reallocate
2028 *	@delta: number of bytes to advance tail
2029 *
2030 *	The function makes a sense only on a fragmented &sk_buff,
2031 *	it expands header moving its tail forward and copying necessary
2032 *	data from fragmented part.
2033 *
2034 *	&sk_buff MUST have reference count of 1.
2035 *
2036 *	Returns %NULL (and &sk_buff does not change) if pull failed
2037 *	or value of new tail of skb in the case of success.
2038 *
2039 *	All the pointers pointing into skb header may change and must be
2040 *	reloaded after call to this function.
2041 */
2042
2043/* Moves tail of skb head forward, copying data from fragmented part,
2044 * when it is necessary.
2045 * 1. It may fail due to malloc failure.
2046 * 2. It may change skb pointers.
2047 *
2048 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2049 */
2050void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2051{
2052	/* If skb has not enough free space at tail, get new one
2053	 * plus 128 bytes for future expansions. If we have enough
2054	 * room at tail, reallocate without expansion only if skb is cloned.
2055	 */
2056	int i, k, eat = (skb->tail + delta) - skb->end;
2057
2058	if (eat > 0 || skb_cloned(skb)) {
2059		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2060				     GFP_ATOMIC))
2061			return NULL;
2062	}
2063
2064	BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2065			     skb_tail_pointer(skb), delta));
2066
2067	/* Optimization: no fragments, no reasons to preestimate
2068	 * size of pulled pages. Superb.
2069	 */
2070	if (!skb_has_frag_list(skb))
2071		goto pull_pages;
2072
2073	/* Estimate size of pulled pages. */
2074	eat = delta;
2075	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2076		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2077
2078		if (size >= eat)
2079			goto pull_pages;
2080		eat -= size;
2081	}
2082
2083	/* If we need update frag list, we are in troubles.
2084	 * Certainly, it is possible to add an offset to skb data,
2085	 * but taking into account that pulling is expected to
2086	 * be very rare operation, it is worth to fight against
2087	 * further bloating skb head and crucify ourselves here instead.
2088	 * Pure masohism, indeed. 8)8)
2089	 */
2090	if (eat) {
2091		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2092		struct sk_buff *clone = NULL;
2093		struct sk_buff *insp = NULL;
2094
2095		do {
 
 
2096			if (list->len <= eat) {
2097				/* Eaten as whole. */
2098				eat -= list->len;
2099				list = list->next;
2100				insp = list;
2101			} else {
2102				/* Eaten partially. */
2103
2104				if (skb_shared(list)) {
2105					/* Sucks! We need to fork list. :-( */
2106					clone = skb_clone(list, GFP_ATOMIC);
2107					if (!clone)
2108						return NULL;
2109					insp = list->next;
2110					list = clone;
2111				} else {
2112					/* This may be pulled without
2113					 * problems. */
2114					insp = list;
2115				}
2116				if (!pskb_pull(list, eat)) {
2117					kfree_skb(clone);
2118					return NULL;
2119				}
2120				break;
2121			}
2122		} while (eat);
2123
2124		/* Free pulled out fragments. */
2125		while ((list = skb_shinfo(skb)->frag_list) != insp) {
2126			skb_shinfo(skb)->frag_list = list->next;
2127			kfree_skb(list);
2128		}
2129		/* And insert new clone at head. */
2130		if (clone) {
2131			clone->next = list;
2132			skb_shinfo(skb)->frag_list = clone;
2133		}
2134	}
2135	/* Success! Now we may commit changes to skb data. */
2136
2137pull_pages:
2138	eat = delta;
2139	k = 0;
2140	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2141		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2142
2143		if (size <= eat) {
2144			skb_frag_unref(skb, i);
2145			eat -= size;
2146		} else {
2147			skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2148
2149			*frag = skb_shinfo(skb)->frags[i];
2150			if (eat) {
2151				skb_frag_off_add(frag, eat);
2152				skb_frag_size_sub(frag, eat);
2153				if (!i)
2154					goto end;
2155				eat = 0;
2156			}
2157			k++;
2158		}
2159	}
2160	skb_shinfo(skb)->nr_frags = k;
2161
2162end:
2163	skb->tail     += delta;
2164	skb->data_len -= delta;
2165
2166	if (!skb->data_len)
2167		skb_zcopy_clear(skb, false);
2168
2169	return skb_tail_pointer(skb);
2170}
2171EXPORT_SYMBOL(__pskb_pull_tail);
2172
2173/**
2174 *	skb_copy_bits - copy bits from skb to kernel buffer
2175 *	@skb: source skb
2176 *	@offset: offset in source
2177 *	@to: destination buffer
2178 *	@len: number of bytes to copy
2179 *
2180 *	Copy the specified number of bytes from the source skb to the
2181 *	destination buffer.
2182 *
2183 *	CAUTION ! :
2184 *		If its prototype is ever changed,
2185 *		check arch/{*}/net/{*}.S files,
2186 *		since it is called from BPF assembly code.
2187 */
2188int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2189{
2190	int start = skb_headlen(skb);
2191	struct sk_buff *frag_iter;
2192	int i, copy;
2193
2194	if (offset > (int)skb->len - len)
2195		goto fault;
2196
2197	/* Copy header. */
2198	if ((copy = start - offset) > 0) {
2199		if (copy > len)
2200			copy = len;
2201		skb_copy_from_linear_data_offset(skb, offset, to, copy);
2202		if ((len -= copy) == 0)
2203			return 0;
2204		offset += copy;
2205		to     += copy;
2206	}
2207
2208	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209		int end;
2210		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2211
2212		WARN_ON(start > offset + len);
2213
2214		end = start + skb_frag_size(f);
2215		if ((copy = end - offset) > 0) {
2216			u32 p_off, p_len, copied;
2217			struct page *p;
2218			u8 *vaddr;
2219
2220			if (copy > len)
2221				copy = len;
2222
2223			skb_frag_foreach_page(f,
2224					      skb_frag_off(f) + offset - start,
2225					      copy, p, p_off, p_len, copied) {
2226				vaddr = kmap_atomic(p);
2227				memcpy(to + copied, vaddr + p_off, p_len);
2228				kunmap_atomic(vaddr);
2229			}
2230
2231			if ((len -= copy) == 0)
2232				return 0;
2233			offset += copy;
2234			to     += copy;
2235		}
2236		start = end;
2237	}
2238
2239	skb_walk_frags(skb, frag_iter) {
2240		int end;
2241
2242		WARN_ON(start > offset + len);
2243
2244		end = start + frag_iter->len;
2245		if ((copy = end - offset) > 0) {
2246			if (copy > len)
2247				copy = len;
2248			if (skb_copy_bits(frag_iter, offset - start, to, copy))
2249				goto fault;
2250			if ((len -= copy) == 0)
2251				return 0;
2252			offset += copy;
2253			to     += copy;
2254		}
2255		start = end;
2256	}
2257
2258	if (!len)
2259		return 0;
2260
2261fault:
2262	return -EFAULT;
2263}
2264EXPORT_SYMBOL(skb_copy_bits);
2265
2266/*
2267 * Callback from splice_to_pipe(), if we need to release some pages
2268 * at the end of the spd in case we error'ed out in filling the pipe.
2269 */
2270static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2271{
2272	put_page(spd->pages[i]);
2273}
2274
2275static struct page *linear_to_page(struct page *page, unsigned int *len,
2276				   unsigned int *offset,
2277				   struct sock *sk)
2278{
2279	struct page_frag *pfrag = sk_page_frag(sk);
2280
2281	if (!sk_page_frag_refill(sk, pfrag))
2282		return NULL;
2283
2284	*len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2285
2286	memcpy(page_address(pfrag->page) + pfrag->offset,
2287	       page_address(page) + *offset, *len);
2288	*offset = pfrag->offset;
2289	pfrag->offset += *len;
2290
2291	return pfrag->page;
2292}
2293
2294static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2295			     struct page *page,
2296			     unsigned int offset)
2297{
2298	return	spd->nr_pages &&
2299		spd->pages[spd->nr_pages - 1] == page &&
2300		(spd->partial[spd->nr_pages - 1].offset +
2301		 spd->partial[spd->nr_pages - 1].len == offset);
2302}
2303
2304/*
2305 * Fill page/offset/length into spd, if it can hold more pages.
2306 */
2307static bool spd_fill_page(struct splice_pipe_desc *spd,
2308			  struct pipe_inode_info *pipe, struct page *page,
2309			  unsigned int *len, unsigned int offset,
2310			  bool linear,
2311			  struct sock *sk)
2312{
2313	if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2314		return true;
2315
2316	if (linear) {
2317		page = linear_to_page(page, len, &offset, sk);
2318		if (!page)
2319			return true;
2320	}
2321	if (spd_can_coalesce(spd, page, offset)) {
2322		spd->partial[spd->nr_pages - 1].len += *len;
2323		return false;
2324	}
2325	get_page(page);
2326	spd->pages[spd->nr_pages] = page;
2327	spd->partial[spd->nr_pages].len = *len;
2328	spd->partial[spd->nr_pages].offset = offset;
2329	spd->nr_pages++;
2330
2331	return false;
2332}
2333
2334static bool __splice_segment(struct page *page, unsigned int poff,
2335			     unsigned int plen, unsigned int *off,
2336			     unsigned int *len,
2337			     struct splice_pipe_desc *spd, bool linear,
2338			     struct sock *sk,
2339			     struct pipe_inode_info *pipe)
2340{
2341	if (!*len)
2342		return true;
2343
2344	/* skip this segment if already processed */
2345	if (*off >= plen) {
2346		*off -= plen;
2347		return false;
2348	}
2349
2350	/* ignore any bits we already processed */
2351	poff += *off;
2352	plen -= *off;
2353	*off = 0;
2354
2355	do {
2356		unsigned int flen = min(*len, plen);
2357
2358		if (spd_fill_page(spd, pipe, page, &flen, poff,
2359				  linear, sk))
2360			return true;
2361		poff += flen;
2362		plen -= flen;
2363		*len -= flen;
2364	} while (*len && plen);
2365
2366	return false;
2367}
2368
2369/*
2370 * Map linear and fragment data from the skb to spd. It reports true if the
2371 * pipe is full or if we already spliced the requested length.
2372 */
2373static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2374			      unsigned int *offset, unsigned int *len,
2375			      struct splice_pipe_desc *spd, struct sock *sk)
2376{
2377	int seg;
2378	struct sk_buff *iter;
2379
2380	/* map the linear part :
2381	 * If skb->head_frag is set, this 'linear' part is backed by a
2382	 * fragment, and if the head is not shared with any clones then
2383	 * we can avoid a copy since we own the head portion of this page.
2384	 */
2385	if (__splice_segment(virt_to_page(skb->data),
2386			     (unsigned long) skb->data & (PAGE_SIZE - 1),
2387			     skb_headlen(skb),
2388			     offset, len, spd,
2389			     skb_head_is_locked(skb),
2390			     sk, pipe))
2391		return true;
2392
2393	/*
2394	 * then map the fragments
2395	 */
2396	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2397		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2398
2399		if (__splice_segment(skb_frag_page(f),
2400				     skb_frag_off(f), skb_frag_size(f),
2401				     offset, len, spd, false, sk, pipe))
2402			return true;
2403	}
2404
2405	skb_walk_frags(skb, iter) {
2406		if (*offset >= iter->len) {
2407			*offset -= iter->len;
2408			continue;
2409		}
2410		/* __skb_splice_bits() only fails if the output has no room
2411		 * left, so no point in going over the frag_list for the error
2412		 * case.
2413		 */
2414		if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2415			return true;
2416	}
2417
2418	return false;
2419}
2420
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2421/*
2422 * Map data from the skb to a pipe. Should handle both the linear part,
2423 * the fragments, and the frag list.
2424 */
2425int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2426		    struct pipe_inode_info *pipe, unsigned int tlen,
2427		    unsigned int flags)
 
 
 
2428{
2429	struct partial_page partial[MAX_SKB_FRAGS];
2430	struct page *pages[MAX_SKB_FRAGS];
2431	struct splice_pipe_desc spd = {
2432		.pages = pages,
2433		.partial = partial,
2434		.nr_pages_max = MAX_SKB_FRAGS,
 
2435		.ops = &nosteal_pipe_buf_ops,
2436		.spd_release = sock_spd_release,
2437	};
2438	int ret = 0;
2439
2440	__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2441
2442	if (spd.nr_pages)
2443		ret = splice_to_pipe(pipe, &spd);
2444
2445	return ret;
2446}
2447EXPORT_SYMBOL_GPL(skb_splice_bits);
2448
2449/* Send skb data on a socket. Socket must be locked. */
2450int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2451			 int len)
2452{
2453	unsigned int orig_len = len;
2454	struct sk_buff *head = skb;
2455	unsigned short fragidx;
2456	int slen, ret;
2457
2458do_frag_list:
2459
2460	/* Deal with head data */
2461	while (offset < skb_headlen(skb) && len) {
2462		struct kvec kv;
2463		struct msghdr msg;
2464
2465		slen = min_t(int, len, skb_headlen(skb) - offset);
2466		kv.iov_base = skb->data + offset;
2467		kv.iov_len = slen;
2468		memset(&msg, 0, sizeof(msg));
2469		msg.msg_flags = MSG_DONTWAIT;
2470
2471		ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2472		if (ret <= 0)
2473			goto error;
2474
2475		offset += ret;
2476		len -= ret;
2477	}
2478
2479	/* All the data was skb head? */
2480	if (!len)
2481		goto out;
2482
2483	/* Make offset relative to start of frags */
2484	offset -= skb_headlen(skb);
2485
2486	/* Find where we are in frag list */
2487	for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2488		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
2489
2490		if (offset < skb_frag_size(frag))
2491			break;
2492
2493		offset -= skb_frag_size(frag);
2494	}
2495
2496	for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2497		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
2498
2499		slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2500
2501		while (slen) {
2502			ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2503						     skb_frag_off(frag) + offset,
2504						     slen, MSG_DONTWAIT);
2505			if (ret <= 0)
2506				goto error;
2507
2508			len -= ret;
2509			offset += ret;
2510			slen -= ret;
2511		}
2512
2513		offset = 0;
2514	}
2515
2516	if (len) {
2517		/* Process any frag lists */
2518
2519		if (skb == head) {
2520			if (skb_has_frag_list(skb)) {
2521				skb = skb_shinfo(skb)->frag_list;
2522				goto do_frag_list;
2523			}
2524		} else if (skb->next) {
2525			skb = skb->next;
2526			goto do_frag_list;
2527		}
2528	}
2529
2530out:
2531	return orig_len - len;
2532
2533error:
2534	return orig_len == len ? ret : orig_len - len;
2535}
2536EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2537
2538/**
2539 *	skb_store_bits - store bits from kernel buffer to skb
2540 *	@skb: destination buffer
2541 *	@offset: offset in destination
2542 *	@from: source buffer
2543 *	@len: number of bytes to copy
2544 *
2545 *	Copy the specified number of bytes from the source buffer to the
2546 *	destination skb.  This function handles all the messy bits of
2547 *	traversing fragment lists and such.
2548 */
2549
2550int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2551{
2552	int start = skb_headlen(skb);
2553	struct sk_buff *frag_iter;
2554	int i, copy;
2555
2556	if (offset > (int)skb->len - len)
2557		goto fault;
2558
2559	if ((copy = start - offset) > 0) {
2560		if (copy > len)
2561			copy = len;
2562		skb_copy_to_linear_data_offset(skb, offset, from, copy);
2563		if ((len -= copy) == 0)
2564			return 0;
2565		offset += copy;
2566		from += copy;
2567	}
2568
2569	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2570		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2571		int end;
2572
2573		WARN_ON(start > offset + len);
2574
2575		end = start + skb_frag_size(frag);
2576		if ((copy = end - offset) > 0) {
2577			u32 p_off, p_len, copied;
2578			struct page *p;
2579			u8 *vaddr;
2580
2581			if (copy > len)
2582				copy = len;
2583
2584			skb_frag_foreach_page(frag,
2585					      skb_frag_off(frag) + offset - start,
2586					      copy, p, p_off, p_len, copied) {
2587				vaddr = kmap_atomic(p);
2588				memcpy(vaddr + p_off, from + copied, p_len);
2589				kunmap_atomic(vaddr);
2590			}
2591
2592			if ((len -= copy) == 0)
2593				return 0;
2594			offset += copy;
2595			from += copy;
2596		}
2597		start = end;
2598	}
2599
2600	skb_walk_frags(skb, frag_iter) {
2601		int end;
2602
2603		WARN_ON(start > offset + len);
2604
2605		end = start + frag_iter->len;
2606		if ((copy = end - offset) > 0) {
2607			if (copy > len)
2608				copy = len;
2609			if (skb_store_bits(frag_iter, offset - start,
2610					   from, copy))
2611				goto fault;
2612			if ((len -= copy) == 0)
2613				return 0;
2614			offset += copy;
2615			from += copy;
2616		}
2617		start = end;
2618	}
2619	if (!len)
2620		return 0;
2621
2622fault:
2623	return -EFAULT;
2624}
2625EXPORT_SYMBOL(skb_store_bits);
2626
2627/* Checksum skb data. */
2628__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2629		      __wsum csum, const struct skb_checksum_ops *ops)
2630{
2631	int start = skb_headlen(skb);
2632	int i, copy = start - offset;
2633	struct sk_buff *frag_iter;
2634	int pos = 0;
2635
2636	/* Checksum header. */
2637	if (copy > 0) {
2638		if (copy > len)
2639			copy = len;
2640		csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2641				       skb->data + offset, copy, csum);
2642		if ((len -= copy) == 0)
2643			return csum;
2644		offset += copy;
2645		pos	= copy;
2646	}
2647
2648	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2649		int end;
2650		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2651
2652		WARN_ON(start > offset + len);
2653
2654		end = start + skb_frag_size(frag);
2655		if ((copy = end - offset) > 0) {
2656			u32 p_off, p_len, copied;
2657			struct page *p;
2658			__wsum csum2;
2659			u8 *vaddr;
2660
2661			if (copy > len)
2662				copy = len;
2663
2664			skb_frag_foreach_page(frag,
2665					      skb_frag_off(frag) + offset - start,
2666					      copy, p, p_off, p_len, copied) {
2667				vaddr = kmap_atomic(p);
2668				csum2 = INDIRECT_CALL_1(ops->update,
2669							csum_partial_ext,
2670							vaddr + p_off, p_len, 0);
2671				kunmap_atomic(vaddr);
2672				csum = INDIRECT_CALL_1(ops->combine,
2673						       csum_block_add_ext, csum,
2674						       csum2, pos, p_len);
2675				pos += p_len;
2676			}
2677
2678			if (!(len -= copy))
2679				return csum;
2680			offset += copy;
 
2681		}
2682		start = end;
2683	}
2684
2685	skb_walk_frags(skb, frag_iter) {
2686		int end;
2687
2688		WARN_ON(start > offset + len);
2689
2690		end = start + frag_iter->len;
2691		if ((copy = end - offset) > 0) {
2692			__wsum csum2;
2693			if (copy > len)
2694				copy = len;
2695			csum2 = __skb_checksum(frag_iter, offset - start,
2696					       copy, 0, ops);
2697			csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2698					       csum, csum2, pos, copy);
2699			if ((len -= copy) == 0)
2700				return csum;
2701			offset += copy;
2702			pos    += copy;
2703		}
2704		start = end;
2705	}
2706	BUG_ON(len);
2707
2708	return csum;
2709}
2710EXPORT_SYMBOL(__skb_checksum);
2711
2712__wsum skb_checksum(const struct sk_buff *skb, int offset,
2713		    int len, __wsum csum)
2714{
2715	const struct skb_checksum_ops ops = {
2716		.update  = csum_partial_ext,
2717		.combine = csum_block_add_ext,
2718	};
2719
2720	return __skb_checksum(skb, offset, len, csum, &ops);
2721}
2722EXPORT_SYMBOL(skb_checksum);
2723
2724/* Both of above in one bottle. */
2725
2726__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2727				    u8 *to, int len, __wsum csum)
2728{
2729	int start = skb_headlen(skb);
2730	int i, copy = start - offset;
2731	struct sk_buff *frag_iter;
2732	int pos = 0;
2733
2734	/* Copy header. */
2735	if (copy > 0) {
2736		if (copy > len)
2737			copy = len;
2738		csum = csum_partial_copy_nocheck(skb->data + offset, to,
2739						 copy, csum);
2740		if ((len -= copy) == 0)
2741			return csum;
2742		offset += copy;
2743		to     += copy;
2744		pos	= copy;
2745	}
2746
2747	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2748		int end;
2749
2750		WARN_ON(start > offset + len);
2751
2752		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2753		if ((copy = end - offset) > 0) {
2754			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2755			u32 p_off, p_len, copied;
2756			struct page *p;
2757			__wsum csum2;
2758			u8 *vaddr;
 
2759
2760			if (copy > len)
2761				copy = len;
2762
2763			skb_frag_foreach_page(frag,
2764					      skb_frag_off(frag) + offset - start,
2765					      copy, p, p_off, p_len, copied) {
2766				vaddr = kmap_atomic(p);
2767				csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2768								  to + copied,
2769								  p_len, 0);
2770				kunmap_atomic(vaddr);
2771				csum = csum_block_add(csum, csum2, pos);
2772				pos += p_len;
2773			}
2774
2775			if (!(len -= copy))
2776				return csum;
2777			offset += copy;
2778			to     += copy;
 
2779		}
2780		start = end;
2781	}
2782
2783	skb_walk_frags(skb, frag_iter) {
2784		__wsum csum2;
2785		int end;
2786
2787		WARN_ON(start > offset + len);
2788
2789		end = start + frag_iter->len;
2790		if ((copy = end - offset) > 0) {
2791			if (copy > len)
2792				copy = len;
2793			csum2 = skb_copy_and_csum_bits(frag_iter,
2794						       offset - start,
2795						       to, copy, 0);
2796			csum = csum_block_add(csum, csum2, pos);
2797			if ((len -= copy) == 0)
2798				return csum;
2799			offset += copy;
2800			to     += copy;
2801			pos    += copy;
2802		}
2803		start = end;
2804	}
2805	BUG_ON(len);
2806	return csum;
2807}
2808EXPORT_SYMBOL(skb_copy_and_csum_bits);
2809
2810__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2811{
2812	__sum16 sum;
2813
2814	sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2815	/* See comments in __skb_checksum_complete(). */
2816	if (likely(!sum)) {
2817		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2818		    !skb->csum_complete_sw)
2819			netdev_rx_csum_fault(skb->dev, skb);
2820	}
2821	if (!skb_shared(skb))
2822		skb->csum_valid = !sum;
2823	return sum;
2824}
2825EXPORT_SYMBOL(__skb_checksum_complete_head);
2826
2827/* This function assumes skb->csum already holds pseudo header's checksum,
2828 * which has been changed from the hardware checksum, for example, by
2829 * __skb_checksum_validate_complete(). And, the original skb->csum must
2830 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2831 *
2832 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2833 * zero. The new checksum is stored back into skb->csum unless the skb is
2834 * shared.
2835 */
2836__sum16 __skb_checksum_complete(struct sk_buff *skb)
2837{
2838	__wsum csum;
2839	__sum16 sum;
2840
2841	csum = skb_checksum(skb, 0, skb->len, 0);
2842
2843	sum = csum_fold(csum_add(skb->csum, csum));
2844	/* This check is inverted, because we already knew the hardware
2845	 * checksum is invalid before calling this function. So, if the
2846	 * re-computed checksum is valid instead, then we have a mismatch
2847	 * between the original skb->csum and skb_checksum(). This means either
2848	 * the original hardware checksum is incorrect or we screw up skb->csum
2849	 * when moving skb->data around.
2850	 */
2851	if (likely(!sum)) {
2852		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2853		    !skb->csum_complete_sw)
2854			netdev_rx_csum_fault(skb->dev, skb);
2855	}
2856
2857	if (!skb_shared(skb)) {
2858		/* Save full packet checksum */
2859		skb->csum = csum;
2860		skb->ip_summed = CHECKSUM_COMPLETE;
2861		skb->csum_complete_sw = 1;
2862		skb->csum_valid = !sum;
2863	}
2864
2865	return sum;
2866}
2867EXPORT_SYMBOL(__skb_checksum_complete);
2868
2869static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2870{
2871	net_warn_ratelimited(
2872		"%s: attempt to compute crc32c without libcrc32c.ko\n",
2873		__func__);
2874	return 0;
2875}
2876
2877static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2878				       int offset, int len)
2879{
2880	net_warn_ratelimited(
2881		"%s: attempt to compute crc32c without libcrc32c.ko\n",
2882		__func__);
2883	return 0;
2884}
2885
2886static const struct skb_checksum_ops default_crc32c_ops = {
2887	.update  = warn_crc32c_csum_update,
2888	.combine = warn_crc32c_csum_combine,
2889};
2890
2891const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2892	&default_crc32c_ops;
2893EXPORT_SYMBOL(crc32c_csum_stub);
2894
2895 /**
2896 *	skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2897 *	@from: source buffer
2898 *
2899 *	Calculates the amount of linear headroom needed in the 'to' skb passed
2900 *	into skb_zerocopy().
2901 */
2902unsigned int
2903skb_zerocopy_headlen(const struct sk_buff *from)
2904{
2905	unsigned int hlen = 0;
2906
2907	if (!from->head_frag ||
2908	    skb_headlen(from) < L1_CACHE_BYTES ||
2909	    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2910		hlen = skb_headlen(from);
2911
2912	if (skb_has_frag_list(from))
2913		hlen = from->len;
2914
2915	return hlen;
2916}
2917EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2918
2919/**
2920 *	skb_zerocopy - Zero copy skb to skb
2921 *	@to: destination buffer
2922 *	@from: source buffer
2923 *	@len: number of bytes to copy from source buffer
2924 *	@hlen: size of linear headroom in destination buffer
2925 *
2926 *	Copies up to `len` bytes from `from` to `to` by creating references
2927 *	to the frags in the source buffer.
2928 *
2929 *	The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2930 *	headroom in the `to` buffer.
2931 *
2932 *	Return value:
2933 *	0: everything is OK
2934 *	-ENOMEM: couldn't orphan frags of @from due to lack of memory
2935 *	-EFAULT: skb_copy_bits() found some problem with skb geometry
2936 */
2937int
2938skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2939{
2940	int i, j = 0;
2941	int plen = 0; /* length of skb->head fragment */
2942	int ret;
2943	struct page *page;
2944	unsigned int offset;
2945
2946	BUG_ON(!from->head_frag && !hlen);
2947
2948	/* dont bother with small payloads */
2949	if (len <= skb_tailroom(to))
2950		return skb_copy_bits(from, 0, skb_put(to, len), len);
2951
2952	if (hlen) {
2953		ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2954		if (unlikely(ret))
2955			return ret;
2956		len -= hlen;
2957	} else {
2958		plen = min_t(int, skb_headlen(from), len);
2959		if (plen) {
2960			page = virt_to_head_page(from->head);
2961			offset = from->data - (unsigned char *)page_address(page);
2962			__skb_fill_page_desc(to, 0, page, offset, plen);
2963			get_page(page);
2964			j = 1;
2965			len -= plen;
2966		}
2967	}
2968
2969	to->truesize += len + plen;
2970	to->len += len + plen;
2971	to->data_len += len + plen;
2972
2973	if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2974		skb_tx_error(from);
2975		return -ENOMEM;
2976	}
2977	skb_zerocopy_clone(to, from, GFP_ATOMIC);
2978
2979	for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2980		int size;
2981
2982		if (!len)
2983			break;
2984		skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2985		size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2986					len);
2987		skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2988		len -= size;
2989		skb_frag_ref(to, j);
2990		j++;
2991	}
2992	skb_shinfo(to)->nr_frags = j;
2993
2994	return 0;
2995}
2996EXPORT_SYMBOL_GPL(skb_zerocopy);
2997
2998void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2999{
3000	__wsum csum;
3001	long csstart;
3002
3003	if (skb->ip_summed == CHECKSUM_PARTIAL)
3004		csstart = skb_checksum_start_offset(skb);
3005	else
3006		csstart = skb_headlen(skb);
3007
3008	BUG_ON(csstart > skb_headlen(skb));
3009
3010	skb_copy_from_linear_data(skb, to, csstart);
3011
3012	csum = 0;
3013	if (csstart != skb->len)
3014		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3015					      skb->len - csstart, 0);
3016
3017	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3018		long csstuff = csstart + skb->csum_offset;
3019
3020		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
3021	}
3022}
3023EXPORT_SYMBOL(skb_copy_and_csum_dev);
3024
3025/**
3026 *	skb_dequeue - remove from the head of the queue
3027 *	@list: list to dequeue from
3028 *
3029 *	Remove the head of the list. The list lock is taken so the function
3030 *	may be used safely with other locking list functions. The head item is
3031 *	returned or %NULL if the list is empty.
3032 */
3033
3034struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3035{
3036	unsigned long flags;
3037	struct sk_buff *result;
3038
3039	spin_lock_irqsave(&list->lock, flags);
3040	result = __skb_dequeue(list);
3041	spin_unlock_irqrestore(&list->lock, flags);
3042	return result;
3043}
3044EXPORT_SYMBOL(skb_dequeue);
3045
3046/**
3047 *	skb_dequeue_tail - remove from the tail of the queue
3048 *	@list: list to dequeue from
3049 *
3050 *	Remove the tail of the list. The list lock is taken so the function
3051 *	may be used safely with other locking list functions. The tail item is
3052 *	returned or %NULL if the list is empty.
3053 */
3054struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3055{
3056	unsigned long flags;
3057	struct sk_buff *result;
3058
3059	spin_lock_irqsave(&list->lock, flags);
3060	result = __skb_dequeue_tail(list);
3061	spin_unlock_irqrestore(&list->lock, flags);
3062	return result;
3063}
3064EXPORT_SYMBOL(skb_dequeue_tail);
3065
3066/**
3067 *	skb_queue_purge - empty a list
3068 *	@list: list to empty
3069 *
3070 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
3071 *	the list and one reference dropped. This function takes the list
3072 *	lock and is atomic with respect to other list locking functions.
3073 */
3074void skb_queue_purge(struct sk_buff_head *list)
3075{
3076	struct sk_buff *skb;
3077	while ((skb = skb_dequeue(list)) != NULL)
3078		kfree_skb(skb);
3079}
3080EXPORT_SYMBOL(skb_queue_purge);
3081
3082/**
3083 *	skb_rbtree_purge - empty a skb rbtree
3084 *	@root: root of the rbtree to empty
3085 *	Return value: the sum of truesizes of all purged skbs.
3086 *
3087 *	Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3088 *	the list and one reference dropped. This function does not take
3089 *	any lock. Synchronization should be handled by the caller (e.g., TCP
3090 *	out-of-order queue is protected by the socket lock).
3091 */
3092unsigned int skb_rbtree_purge(struct rb_root *root)
3093{
3094	struct rb_node *p = rb_first(root);
3095	unsigned int sum = 0;
3096
3097	while (p) {
3098		struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3099
3100		p = rb_next(p);
3101		rb_erase(&skb->rbnode, root);
3102		sum += skb->truesize;
3103		kfree_skb(skb);
3104	}
3105	return sum;
3106}
3107
3108/**
3109 *	skb_queue_head - queue a buffer at the list head
3110 *	@list: list to use
3111 *	@newsk: buffer to queue
3112 *
3113 *	Queue a buffer at the start of the list. This function takes the
3114 *	list lock and can be used safely with other locking &sk_buff functions
3115 *	safely.
3116 *
3117 *	A buffer cannot be placed on two lists at the same time.
3118 */
3119void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3120{
3121	unsigned long flags;
3122
3123	spin_lock_irqsave(&list->lock, flags);
3124	__skb_queue_head(list, newsk);
3125	spin_unlock_irqrestore(&list->lock, flags);
3126}
3127EXPORT_SYMBOL(skb_queue_head);
3128
3129/**
3130 *	skb_queue_tail - queue a buffer at the list tail
3131 *	@list: list to use
3132 *	@newsk: buffer to queue
3133 *
3134 *	Queue a buffer at the tail of the list. This function takes the
3135 *	list lock and can be used safely with other locking &sk_buff functions
3136 *	safely.
3137 *
3138 *	A buffer cannot be placed on two lists at the same time.
3139 */
3140void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3141{
3142	unsigned long flags;
3143
3144	spin_lock_irqsave(&list->lock, flags);
3145	__skb_queue_tail(list, newsk);
3146	spin_unlock_irqrestore(&list->lock, flags);
3147}
3148EXPORT_SYMBOL(skb_queue_tail);
3149
3150/**
3151 *	skb_unlink	-	remove a buffer from a list
3152 *	@skb: buffer to remove
3153 *	@list: list to use
3154 *
3155 *	Remove a packet from a list. The list locks are taken and this
3156 *	function is atomic with respect to other list locked calls
3157 *
3158 *	You must know what list the SKB is on.
3159 */
3160void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3161{
3162	unsigned long flags;
3163
3164	spin_lock_irqsave(&list->lock, flags);
3165	__skb_unlink(skb, list);
3166	spin_unlock_irqrestore(&list->lock, flags);
3167}
3168EXPORT_SYMBOL(skb_unlink);
3169
3170/**
3171 *	skb_append	-	append a buffer
3172 *	@old: buffer to insert after
3173 *	@newsk: buffer to insert
3174 *	@list: list to use
3175 *
3176 *	Place a packet after a given packet in a list. The list locks are taken
3177 *	and this function is atomic with respect to other list locked calls.
3178 *	A buffer cannot be placed on two lists at the same time.
3179 */
3180void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3181{
3182	unsigned long flags;
3183
3184	spin_lock_irqsave(&list->lock, flags);
3185	__skb_queue_after(list, old, newsk);
3186	spin_unlock_irqrestore(&list->lock, flags);
3187}
3188EXPORT_SYMBOL(skb_append);
3189
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3190static inline void skb_split_inside_header(struct sk_buff *skb,
3191					   struct sk_buff* skb1,
3192					   const u32 len, const int pos)
3193{
3194	int i;
3195
3196	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3197					 pos - len);
3198	/* And move data appendix as is. */
3199	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3200		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3201
3202	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3203	skb_shinfo(skb)->nr_frags  = 0;
3204	skb1->data_len		   = skb->data_len;
3205	skb1->len		   += skb1->data_len;
3206	skb->data_len		   = 0;
3207	skb->len		   = len;
3208	skb_set_tail_pointer(skb, len);
3209}
3210
3211static inline void skb_split_no_header(struct sk_buff *skb,
3212				       struct sk_buff* skb1,
3213				       const u32 len, int pos)
3214{
3215	int i, k = 0;
3216	const int nfrags = skb_shinfo(skb)->nr_frags;
3217
3218	skb_shinfo(skb)->nr_frags = 0;
3219	skb1->len		  = skb1->data_len = skb->len - len;
3220	skb->len		  = len;
3221	skb->data_len		  = len - pos;
3222
3223	for (i = 0; i < nfrags; i++) {
3224		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3225
3226		if (pos + size > len) {
3227			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3228
3229			if (pos < len) {
3230				/* Split frag.
3231				 * We have two variants in this case:
3232				 * 1. Move all the frag to the second
3233				 *    part, if it is possible. F.e.
3234				 *    this approach is mandatory for TUX,
3235				 *    where splitting is expensive.
3236				 * 2. Split is accurately. We make this.
3237				 */
3238				skb_frag_ref(skb, i);
3239				skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3240				skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3241				skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3242				skb_shinfo(skb)->nr_frags++;
3243			}
3244			k++;
3245		} else
3246			skb_shinfo(skb)->nr_frags++;
3247		pos += size;
3248	}
3249	skb_shinfo(skb1)->nr_frags = k;
3250}
3251
3252/**
3253 * skb_split - Split fragmented skb to two parts at length len.
3254 * @skb: the buffer to split
3255 * @skb1: the buffer to receive the second part
3256 * @len: new length for skb
3257 */
3258void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3259{
3260	int pos = skb_headlen(skb);
3261
3262	skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3263				      SKBTX_SHARED_FRAG;
3264	skb_zerocopy_clone(skb1, skb, 0);
3265	if (len < pos)	/* Split line is inside header. */
3266		skb_split_inside_header(skb, skb1, len, pos);
3267	else		/* Second chunk has no header, nothing to copy. */
3268		skb_split_no_header(skb, skb1, len, pos);
3269}
3270EXPORT_SYMBOL(skb_split);
3271
3272/* Shifting from/to a cloned skb is a no-go.
3273 *
3274 * Caller cannot keep skb_shinfo related pointers past calling here!
3275 */
3276static int skb_prepare_for_shift(struct sk_buff *skb)
3277{
3278	return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3279}
3280
3281/**
3282 * skb_shift - Shifts paged data partially from skb to another
3283 * @tgt: buffer into which tail data gets added
3284 * @skb: buffer from which the paged data comes from
3285 * @shiftlen: shift up to this many bytes
3286 *
3287 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3288 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3289 * It's up to caller to free skb if everything was shifted.
3290 *
3291 * If @tgt runs out of frags, the whole operation is aborted.
3292 *
3293 * Skb cannot include anything else but paged data while tgt is allowed
3294 * to have non-paged data as well.
3295 *
3296 * TODO: full sized shift could be optimized but that would need
3297 * specialized skb free'er to handle frags without up-to-date nr_frags.
3298 */
3299int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3300{
3301	int from, to, merge, todo;
3302	skb_frag_t *fragfrom, *fragto;
3303
3304	BUG_ON(shiftlen > skb->len);
3305
3306	if (skb_headlen(skb))
3307		return 0;
3308	if (skb_zcopy(tgt) || skb_zcopy(skb))
3309		return 0;
3310
3311	todo = shiftlen;
3312	from = 0;
3313	to = skb_shinfo(tgt)->nr_frags;
3314	fragfrom = &skb_shinfo(skb)->frags[from];
3315
3316	/* Actual merge is delayed until the point when we know we can
3317	 * commit all, so that we don't have to undo partial changes
3318	 */
3319	if (!to ||
3320	    !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3321			      skb_frag_off(fragfrom))) {
3322		merge = -1;
3323	} else {
3324		merge = to - 1;
3325
3326		todo -= skb_frag_size(fragfrom);
3327		if (todo < 0) {
3328			if (skb_prepare_for_shift(skb) ||
3329			    skb_prepare_for_shift(tgt))
3330				return 0;
3331
3332			/* All previous frag pointers might be stale! */
3333			fragfrom = &skb_shinfo(skb)->frags[from];
3334			fragto = &skb_shinfo(tgt)->frags[merge];
3335
3336			skb_frag_size_add(fragto, shiftlen);
3337			skb_frag_size_sub(fragfrom, shiftlen);
3338			skb_frag_off_add(fragfrom, shiftlen);
3339
3340			goto onlymerged;
3341		}
3342
3343		from++;
3344	}
3345
3346	/* Skip full, not-fitting skb to avoid expensive operations */
3347	if ((shiftlen == skb->len) &&
3348	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3349		return 0;
3350
3351	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3352		return 0;
3353
3354	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3355		if (to == MAX_SKB_FRAGS)
3356			return 0;
3357
3358		fragfrom = &skb_shinfo(skb)->frags[from];
3359		fragto = &skb_shinfo(tgt)->frags[to];
3360
3361		if (todo >= skb_frag_size(fragfrom)) {
3362			*fragto = *fragfrom;
3363			todo -= skb_frag_size(fragfrom);
3364			from++;
3365			to++;
3366
3367		} else {
3368			__skb_frag_ref(fragfrom);
3369			skb_frag_page_copy(fragto, fragfrom);
3370			skb_frag_off_copy(fragto, fragfrom);
3371			skb_frag_size_set(fragto, todo);
3372
3373			skb_frag_off_add(fragfrom, todo);
3374			skb_frag_size_sub(fragfrom, todo);
3375			todo = 0;
3376
3377			to++;
3378			break;
3379		}
3380	}
3381
3382	/* Ready to "commit" this state change to tgt */
3383	skb_shinfo(tgt)->nr_frags = to;
3384
3385	if (merge >= 0) {
3386		fragfrom = &skb_shinfo(skb)->frags[0];
3387		fragto = &skb_shinfo(tgt)->frags[merge];
3388
3389		skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3390		__skb_frag_unref(fragfrom);
3391	}
3392
3393	/* Reposition in the original skb */
3394	to = 0;
3395	while (from < skb_shinfo(skb)->nr_frags)
3396		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3397	skb_shinfo(skb)->nr_frags = to;
3398
3399	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3400
3401onlymerged:
3402	/* Most likely the tgt won't ever need its checksum anymore, skb on
3403	 * the other hand might need it if it needs to be resent
3404	 */
3405	tgt->ip_summed = CHECKSUM_PARTIAL;
3406	skb->ip_summed = CHECKSUM_PARTIAL;
3407
3408	/* Yak, is it really working this way? Some helper please? */
3409	skb->len -= shiftlen;
3410	skb->data_len -= shiftlen;
3411	skb->truesize -= shiftlen;
3412	tgt->len += shiftlen;
3413	tgt->data_len += shiftlen;
3414	tgt->truesize += shiftlen;
3415
3416	return shiftlen;
3417}
3418
3419/**
3420 * skb_prepare_seq_read - Prepare a sequential read of skb data
3421 * @skb: the buffer to read
3422 * @from: lower offset of data to be read
3423 * @to: upper offset of data to be read
3424 * @st: state variable
3425 *
3426 * Initializes the specified state variable. Must be called before
3427 * invoking skb_seq_read() for the first time.
3428 */
3429void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3430			  unsigned int to, struct skb_seq_state *st)
3431{
3432	st->lower_offset = from;
3433	st->upper_offset = to;
3434	st->root_skb = st->cur_skb = skb;
3435	st->frag_idx = st->stepped_offset = 0;
3436	st->frag_data = NULL;
3437}
3438EXPORT_SYMBOL(skb_prepare_seq_read);
3439
3440/**
3441 * skb_seq_read - Sequentially read skb data
3442 * @consumed: number of bytes consumed by the caller so far
3443 * @data: destination pointer for data to be returned
3444 * @st: state variable
3445 *
3446 * Reads a block of skb data at @consumed relative to the
3447 * lower offset specified to skb_prepare_seq_read(). Assigns
3448 * the head of the data block to @data and returns the length
3449 * of the block or 0 if the end of the skb data or the upper
3450 * offset has been reached.
3451 *
3452 * The caller is not required to consume all of the data
3453 * returned, i.e. @consumed is typically set to the number
3454 * of bytes already consumed and the next call to
3455 * skb_seq_read() will return the remaining part of the block.
3456 *
3457 * Note 1: The size of each block of data returned can be arbitrary,
3458 *       this limitation is the cost for zerocopy sequential
3459 *       reads of potentially non linear data.
3460 *
3461 * Note 2: Fragment lists within fragments are not implemented
3462 *       at the moment, state->root_skb could be replaced with
3463 *       a stack for this purpose.
3464 */
3465unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3466			  struct skb_seq_state *st)
3467{
3468	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3469	skb_frag_t *frag;
3470
3471	if (unlikely(abs_offset >= st->upper_offset)) {
3472		if (st->frag_data) {
3473			kunmap_atomic(st->frag_data);
3474			st->frag_data = NULL;
3475		}
3476		return 0;
3477	}
3478
3479next_skb:
3480	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3481
3482	if (abs_offset < block_limit && !st->frag_data) {
3483		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3484		return block_limit - abs_offset;
3485	}
3486
3487	if (st->frag_idx == 0 && !st->frag_data)
3488		st->stepped_offset += skb_headlen(st->cur_skb);
3489
3490	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3491		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3492		block_limit = skb_frag_size(frag) + st->stepped_offset;
3493
3494		if (abs_offset < block_limit) {
3495			if (!st->frag_data)
3496				st->frag_data = kmap_atomic(skb_frag_page(frag));
3497
3498			*data = (u8 *) st->frag_data + skb_frag_off(frag) +
3499				(abs_offset - st->stepped_offset);
3500
3501			return block_limit - abs_offset;
3502		}
3503
3504		if (st->frag_data) {
3505			kunmap_atomic(st->frag_data);
3506			st->frag_data = NULL;
3507		}
3508
3509		st->frag_idx++;
3510		st->stepped_offset += skb_frag_size(frag);
3511	}
3512
3513	if (st->frag_data) {
3514		kunmap_atomic(st->frag_data);
3515		st->frag_data = NULL;
3516	}
3517
3518	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3519		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3520		st->frag_idx = 0;
3521		goto next_skb;
3522	} else if (st->cur_skb->next) {
3523		st->cur_skb = st->cur_skb->next;
3524		st->frag_idx = 0;
3525		goto next_skb;
3526	}
3527
3528	return 0;
3529}
3530EXPORT_SYMBOL(skb_seq_read);
3531
3532/**
3533 * skb_abort_seq_read - Abort a sequential read of skb data
3534 * @st: state variable
3535 *
3536 * Must be called if skb_seq_read() was not called until it
3537 * returned 0.
3538 */
3539void skb_abort_seq_read(struct skb_seq_state *st)
3540{
3541	if (st->frag_data)
3542		kunmap_atomic(st->frag_data);
3543}
3544EXPORT_SYMBOL(skb_abort_seq_read);
3545
3546#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
3547
3548static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3549					  struct ts_config *conf,
3550					  struct ts_state *state)
3551{
3552	return skb_seq_read(offset, text, TS_SKB_CB(state));
3553}
3554
3555static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3556{
3557	skb_abort_seq_read(TS_SKB_CB(state));
3558}
3559
3560/**
3561 * skb_find_text - Find a text pattern in skb data
3562 * @skb: the buffer to look in
3563 * @from: search offset
3564 * @to: search limit
3565 * @config: textsearch configuration
3566 *
3567 * Finds a pattern in the skb data according to the specified
3568 * textsearch configuration. Use textsearch_next() to retrieve
3569 * subsequent occurrences of the pattern. Returns the offset
3570 * to the first occurrence or UINT_MAX if no match was found.
3571 */
3572unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3573			   unsigned int to, struct ts_config *config)
3574{
3575	struct ts_state state;
3576	unsigned int ret;
3577
3578	config->get_next_block = skb_ts_get_next_block;
3579	config->finish = skb_ts_finish;
3580
3581	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3582
3583	ret = textsearch_find(config, &state);
3584	return (ret <= to - from ? ret : UINT_MAX);
3585}
3586EXPORT_SYMBOL(skb_find_text);
3587
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3588int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3589			 int offset, size_t size)
3590{
3591	int i = skb_shinfo(skb)->nr_frags;
3592
3593	if (skb_can_coalesce(skb, i, page, offset)) {
3594		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3595	} else if (i < MAX_SKB_FRAGS) {
3596		get_page(page);
3597		skb_fill_page_desc(skb, i, page, offset, size);
3598	} else {
3599		return -EMSGSIZE;
3600	}
3601
3602	return 0;
3603}
3604EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3605
3606/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3607 *	skb_pull_rcsum - pull skb and update receive checksum
3608 *	@skb: buffer to update
3609 *	@len: length of data pulled
3610 *
3611 *	This function performs an skb_pull on the packet and updates
3612 *	the CHECKSUM_COMPLETE checksum.  It should be used on
3613 *	receive path processing instead of skb_pull unless you know
3614 *	that the checksum difference is zero (e.g., a valid IP header)
3615 *	or you are setting ip_summed to CHECKSUM_NONE.
3616 */
3617void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3618{
3619	unsigned char *data = skb->data;
3620
3621	BUG_ON(len > skb->len);
3622	__skb_pull(skb, len);
3623	skb_postpull_rcsum(skb, data, len);
3624	return skb->data;
3625}
3626EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3627
3628static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3629{
3630	skb_frag_t head_frag;
3631	struct page *page;
3632
3633	page = virt_to_head_page(frag_skb->head);
3634	__skb_frag_set_page(&head_frag, page);
3635	skb_frag_off_set(&head_frag, frag_skb->data -
3636			 (unsigned char *)page_address(page));
3637	skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3638	return head_frag;
3639}
3640
3641/**
3642 *	skb_segment - Perform protocol segmentation on skb.
3643 *	@head_skb: buffer to segment
3644 *	@features: features for the output path (see dev->features)
3645 *
3646 *	This function performs segmentation on the given skb.  It returns
3647 *	a pointer to the first in a list of new skbs for the segments.
3648 *	In case of error it returns ERR_PTR(err).
3649 */
3650struct sk_buff *skb_segment(struct sk_buff *head_skb,
3651			    netdev_features_t features)
3652{
3653	struct sk_buff *segs = NULL;
3654	struct sk_buff *tail = NULL;
3655	struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3656	skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3657	unsigned int mss = skb_shinfo(head_skb)->gso_size;
3658	unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3659	struct sk_buff *frag_skb = head_skb;
3660	unsigned int offset = doffset;
3661	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3662	unsigned int partial_segs = 0;
3663	unsigned int headroom;
3664	unsigned int len = head_skb->len;
3665	__be16 proto;
3666	bool csum, sg;
 
3667	int nfrags = skb_shinfo(head_skb)->nr_frags;
3668	int err = -ENOMEM;
3669	int i = 0;
3670	int pos;
3671	int dummy;
3672
3673	if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3674	    (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3675		/* gso_size is untrusted, and we have a frag_list with a linear
3676		 * non head_frag head.
3677		 *
3678		 * (we assume checking the first list_skb member suffices;
3679		 * i.e if either of the list_skb members have non head_frag
3680		 * head, then the first one has too).
3681		 *
3682		 * If head_skb's headlen does not fit requested gso_size, it
3683		 * means that the frag_list members do NOT terminate on exact
3684		 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3685		 * sharing. Therefore we must fallback to copying the frag_list
3686		 * skbs; we do so by disabling SG.
3687		 */
3688		if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3689			features &= ~NETIF_F_SG;
3690	}
3691
3692	__skb_push(head_skb, doffset);
3693	proto = skb_network_protocol(head_skb, &dummy);
3694	if (unlikely(!proto))
3695		return ERR_PTR(-EINVAL);
3696
3697	sg = !!(features & NETIF_F_SG);
3698	csum = !!can_checksum_protocol(features, proto);
3699
3700	if (sg && csum && (mss != GSO_BY_FRAGS))  {
3701		if (!(features & NETIF_F_GSO_PARTIAL)) {
3702			struct sk_buff *iter;
3703			unsigned int frag_len;
3704
3705			if (!list_skb ||
3706			    !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3707				goto normal;
3708
3709			/* If we get here then all the required
3710			 * GSO features except frag_list are supported.
3711			 * Try to split the SKB to multiple GSO SKBs
3712			 * with no frag_list.
3713			 * Currently we can do that only when the buffers don't
3714			 * have a linear part and all the buffers except
3715			 * the last are of the same length.
3716			 */
3717			frag_len = list_skb->len;
3718			skb_walk_frags(head_skb, iter) {
3719				if (frag_len != iter->len && iter->next)
3720					goto normal;
3721				if (skb_headlen(iter) && !iter->head_frag)
3722					goto normal;
3723
3724				len -= iter->len;
3725			}
3726
3727			if (len != frag_len)
3728				goto normal;
3729		}
3730
3731		/* GSO partial only requires that we trim off any excess that
3732		 * doesn't fit into an MSS sized block, so take care of that
3733		 * now.
3734		 */
3735		partial_segs = len / mss;
3736		if (partial_segs > 1)
3737			mss *= partial_segs;
3738		else
3739			partial_segs = 0;
3740	}
3741
3742normal:
3743	headroom = skb_headroom(head_skb);
3744	pos = skb_headlen(head_skb);
3745
3746	do {
3747		struct sk_buff *nskb;
3748		skb_frag_t *nskb_frag;
3749		int hsize;
3750		int size;
3751
3752		if (unlikely(mss == GSO_BY_FRAGS)) {
3753			len = list_skb->len;
3754		} else {
3755			len = head_skb->len - offset;
3756			if (len > mss)
3757				len = mss;
3758		}
3759
3760		hsize = skb_headlen(head_skb) - offset;
3761		if (hsize < 0)
3762			hsize = 0;
3763		if (hsize > len || !sg)
3764			hsize = len;
3765
3766		if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3767		    (skb_headlen(list_skb) == len || sg)) {
3768			BUG_ON(skb_headlen(list_skb) > len);
3769
3770			i = 0;
3771			nfrags = skb_shinfo(list_skb)->nr_frags;
3772			frag = skb_shinfo(list_skb)->frags;
3773			frag_skb = list_skb;
3774			pos += skb_headlen(list_skb);
3775
3776			while (pos < offset + len) {
3777				BUG_ON(i >= nfrags);
3778
3779				size = skb_frag_size(frag);
3780				if (pos + size > offset + len)
3781					break;
3782
3783				i++;
3784				pos += size;
3785				frag++;
3786			}
3787
3788			nskb = skb_clone(list_skb, GFP_ATOMIC);
3789			list_skb = list_skb->next;
3790
3791			if (unlikely(!nskb))
3792				goto err;
3793
3794			if (unlikely(pskb_trim(nskb, len))) {
3795				kfree_skb(nskb);
3796				goto err;
3797			}
3798
3799			hsize = skb_end_offset(nskb);
3800			if (skb_cow_head(nskb, doffset + headroom)) {
3801				kfree_skb(nskb);
3802				goto err;
3803			}
3804
3805			nskb->truesize += skb_end_offset(nskb) - hsize;
3806			skb_release_head_state(nskb);
3807			__skb_push(nskb, doffset);
3808		} else {
3809			nskb = __alloc_skb(hsize + doffset + headroom,
3810					   GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3811					   NUMA_NO_NODE);
3812
3813			if (unlikely(!nskb))
3814				goto err;
3815
3816			skb_reserve(nskb, headroom);
3817			__skb_put(nskb, doffset);
3818		}
3819
3820		if (segs)
3821			tail->next = nskb;
3822		else
3823			segs = nskb;
3824		tail = nskb;
3825
3826		__copy_skb_header(nskb, head_skb);
3827
3828		skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3829		skb_reset_mac_len(nskb);
3830
3831		skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3832						 nskb->data - tnl_hlen,
3833						 doffset + tnl_hlen);
3834
3835		if (nskb->len == len + doffset)
3836			goto perform_csum_check;
3837
3838		if (!sg) {
3839			if (!nskb->remcsum_offload)
3840				nskb->ip_summed = CHECKSUM_NONE;
3841			SKB_GSO_CB(nskb)->csum =
3842				skb_copy_and_csum_bits(head_skb, offset,
3843						       skb_put(nskb, len),
3844						       len, 0);
3845			SKB_GSO_CB(nskb)->csum_start =
3846				skb_headroom(nskb) + doffset;
3847			continue;
3848		}
3849
3850		nskb_frag = skb_shinfo(nskb)->frags;
3851
3852		skb_copy_from_linear_data_offset(head_skb, offset,
3853						 skb_put(nskb, hsize), hsize);
3854
3855		skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3856					      SKBTX_SHARED_FRAG;
3857
3858		if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3859		    skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3860			goto err;
3861
3862		while (pos < offset + len) {
3863			if (i >= nfrags) {
 
 
3864				i = 0;
3865				nfrags = skb_shinfo(list_skb)->nr_frags;
3866				frag = skb_shinfo(list_skb)->frags;
3867				frag_skb = list_skb;
3868				if (!skb_headlen(list_skb)) {
3869					BUG_ON(!nfrags);
3870				} else {
3871					BUG_ON(!list_skb->head_frag);
3872
3873					/* to make room for head_frag. */
3874					i--;
3875					frag--;
3876				}
3877				if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3878				    skb_zerocopy_clone(nskb, frag_skb,
3879						       GFP_ATOMIC))
3880					goto err;
3881
3882				list_skb = list_skb->next;
3883			}
3884
3885			if (unlikely(skb_shinfo(nskb)->nr_frags >=
3886				     MAX_SKB_FRAGS)) {
3887				net_warn_ratelimited(
3888					"skb_segment: too many frags: %u %u\n",
3889					pos, mss);
3890				err = -EINVAL;
3891				goto err;
3892			}
3893
3894			*nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
 
 
 
3895			__skb_frag_ref(nskb_frag);
3896			size = skb_frag_size(nskb_frag);
3897
3898			if (pos < offset) {
3899				skb_frag_off_add(nskb_frag, offset - pos);
3900				skb_frag_size_sub(nskb_frag, offset - pos);
3901			}
3902
3903			skb_shinfo(nskb)->nr_frags++;
3904
3905			if (pos + size <= offset + len) {
3906				i++;
3907				frag++;
3908				pos += size;
3909			} else {
3910				skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3911				goto skip_fraglist;
3912			}
3913
3914			nskb_frag++;
3915		}
3916
3917skip_fraglist:
3918		nskb->data_len = len - hsize;
3919		nskb->len += nskb->data_len;
3920		nskb->truesize += nskb->data_len;
3921
3922perform_csum_check:
3923		if (!csum) {
3924			if (skb_has_shared_frag(nskb) &&
3925			    __skb_linearize(nskb))
3926				goto err;
3927
 
3928			if (!nskb->remcsum_offload)
3929				nskb->ip_summed = CHECKSUM_NONE;
3930			SKB_GSO_CB(nskb)->csum =
3931				skb_checksum(nskb, doffset,
3932					     nskb->len - doffset, 0);
3933			SKB_GSO_CB(nskb)->csum_start =
3934				skb_headroom(nskb) + doffset;
3935		}
3936	} while ((offset += len) < head_skb->len);
3937
3938	/* Some callers want to get the end of the list.
3939	 * Put it in segs->prev to avoid walking the list.
3940	 * (see validate_xmit_skb_list() for example)
3941	 */
3942	segs->prev = tail;
3943
3944	if (partial_segs) {
3945		struct sk_buff *iter;
3946		int type = skb_shinfo(head_skb)->gso_type;
3947		unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3948
3949		/* Update type to add partial and then remove dodgy if set */
3950		type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3951		type &= ~SKB_GSO_DODGY;
3952
3953		/* Update GSO info and prepare to start updating headers on
3954		 * our way back down the stack of protocols.
3955		 */
3956		for (iter = segs; iter; iter = iter->next) {
3957			skb_shinfo(iter)->gso_size = gso_size;
3958			skb_shinfo(iter)->gso_segs = partial_segs;
3959			skb_shinfo(iter)->gso_type = type;
3960			SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3961		}
3962
3963		if (tail->len - doffset <= gso_size)
3964			skb_shinfo(tail)->gso_size = 0;
3965		else if (tail != segs)
3966			skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3967	}
3968
3969	/* Following permits correct backpressure, for protocols
3970	 * using skb_set_owner_w().
3971	 * Idea is to tranfert ownership from head_skb to last segment.
3972	 */
3973	if (head_skb->destructor == sock_wfree) {
3974		swap(tail->truesize, head_skb->truesize);
3975		swap(tail->destructor, head_skb->destructor);
3976		swap(tail->sk, head_skb->sk);
3977	}
3978	return segs;
3979
3980err:
3981	kfree_skb_list(segs);
3982	return ERR_PTR(err);
3983}
3984EXPORT_SYMBOL_GPL(skb_segment);
3985
3986int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
3987{
3988	struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3989	unsigned int offset = skb_gro_offset(skb);
3990	unsigned int headlen = skb_headlen(skb);
3991	unsigned int len = skb_gro_len(skb);
 
3992	unsigned int delta_truesize;
3993	struct sk_buff *lp;
3994
3995	if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
3996		return -E2BIG;
3997
3998	lp = NAPI_GRO_CB(p)->last;
3999	pinfo = skb_shinfo(lp);
4000
4001	if (headlen <= offset) {
4002		skb_frag_t *frag;
4003		skb_frag_t *frag2;
4004		int i = skbinfo->nr_frags;
4005		int nr_frags = pinfo->nr_frags + i;
4006
4007		if (nr_frags > MAX_SKB_FRAGS)
4008			goto merge;
4009
4010		offset -= headlen;
4011		pinfo->nr_frags = nr_frags;
4012		skbinfo->nr_frags = 0;
4013
4014		frag = pinfo->frags + nr_frags;
4015		frag2 = skbinfo->frags + i;
4016		do {
4017			*--frag = *--frag2;
4018		} while (--i);
4019
4020		skb_frag_off_add(frag, offset);
4021		skb_frag_size_sub(frag, offset);
4022
4023		/* all fragments truesize : remove (head size + sk_buff) */
4024		delta_truesize = skb->truesize -
4025				 SKB_TRUESIZE(skb_end_offset(skb));
4026
4027		skb->truesize -= skb->data_len;
4028		skb->len -= skb->data_len;
4029		skb->data_len = 0;
4030
4031		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4032		goto done;
4033	} else if (skb->head_frag) {
4034		int nr_frags = pinfo->nr_frags;
4035		skb_frag_t *frag = pinfo->frags + nr_frags;
4036		struct page *page = virt_to_head_page(skb->head);
4037		unsigned int first_size = headlen - offset;
4038		unsigned int first_offset;
4039
4040		if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4041			goto merge;
4042
4043		first_offset = skb->data -
4044			       (unsigned char *)page_address(page) +
4045			       offset;
4046
4047		pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4048
4049		__skb_frag_set_page(frag, page);
4050		skb_frag_off_set(frag, first_offset);
4051		skb_frag_size_set(frag, first_size);
4052
4053		memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4054		/* We dont need to clear skbinfo->nr_frags here */
4055
4056		delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4057		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4058		goto done;
4059	}
4060
4061merge:
4062	delta_truesize = skb->truesize;
4063	if (offset > headlen) {
4064		unsigned int eat = offset - headlen;
4065
4066		skb_frag_off_add(&skbinfo->frags[0], eat);
4067		skb_frag_size_sub(&skbinfo->frags[0], eat);
4068		skb->data_len -= eat;
4069		skb->len -= eat;
4070		offset = headlen;
4071	}
4072
4073	__skb_pull(skb, offset);
4074
4075	if (NAPI_GRO_CB(p)->last == p)
4076		skb_shinfo(p)->frag_list = skb;
4077	else
4078		NAPI_GRO_CB(p)->last->next = skb;
4079	NAPI_GRO_CB(p)->last = skb;
4080	__skb_header_release(skb);
4081	lp = p;
4082
4083done:
4084	NAPI_GRO_CB(p)->count++;
4085	p->data_len += len;
4086	p->truesize += delta_truesize;
4087	p->len += len;
4088	if (lp != p) {
4089		lp->data_len += len;
4090		lp->truesize += delta_truesize;
4091		lp->len += len;
4092	}
4093	NAPI_GRO_CB(skb)->same_flow = 1;
4094	return 0;
4095}
4096EXPORT_SYMBOL_GPL(skb_gro_receive);
4097
4098#ifdef CONFIG_SKB_EXTENSIONS
4099#define SKB_EXT_ALIGN_VALUE	8
4100#define SKB_EXT_CHUNKSIZEOF(x)	(ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4101
4102static const u8 skb_ext_type_len[] = {
4103#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4104	[SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4105#endif
4106#ifdef CONFIG_XFRM
4107	[SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4108#endif
4109#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4110	[TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4111#endif
4112};
4113
4114static __always_inline unsigned int skb_ext_total_length(void)
4115{
4116	return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4117#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4118		skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4119#endif
4120#ifdef CONFIG_XFRM
4121		skb_ext_type_len[SKB_EXT_SEC_PATH] +
4122#endif
4123#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4124		skb_ext_type_len[TC_SKB_EXT] +
4125#endif
4126		0;
4127}
4128
4129static void skb_extensions_init(void)
4130{
4131	BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4132	BUILD_BUG_ON(skb_ext_total_length() > 255);
4133
4134	skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4135					     SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4136					     0,
4137					     SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4138					     NULL);
4139}
4140#else
4141static void skb_extensions_init(void) {}
4142#endif
4143
4144void __init skb_init(void)
4145{
4146	skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4147					      sizeof(struct sk_buff),
4148					      0,
4149					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4150					      offsetof(struct sk_buff, cb),
4151					      sizeof_field(struct sk_buff, cb),
4152					      NULL);
4153	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4154						sizeof(struct sk_buff_fclones),
4155						0,
4156						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4157						NULL);
4158	skb_extensions_init();
4159}
4160
 
 
 
 
 
 
 
 
 
 
4161static int
4162__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4163	       unsigned int recursion_level)
4164{
4165	int start = skb_headlen(skb);
4166	int i, copy = start - offset;
4167	struct sk_buff *frag_iter;
4168	int elt = 0;
4169
4170	if (unlikely(recursion_level >= 24))
4171		return -EMSGSIZE;
4172
4173	if (copy > 0) {
4174		if (copy > len)
4175			copy = len;
4176		sg_set_buf(sg, skb->data + offset, copy);
4177		elt++;
4178		if ((len -= copy) == 0)
4179			return elt;
4180		offset += copy;
4181	}
4182
4183	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4184		int end;
4185
4186		WARN_ON(start > offset + len);
4187
4188		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4189		if ((copy = end - offset) > 0) {
4190			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4191			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4192				return -EMSGSIZE;
4193
4194			if (copy > len)
4195				copy = len;
4196			sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4197				    skb_frag_off(frag) + offset - start);
4198			elt++;
4199			if (!(len -= copy))
4200				return elt;
4201			offset += copy;
4202		}
4203		start = end;
4204	}
4205
4206	skb_walk_frags(skb, frag_iter) {
4207		int end, ret;
4208
4209		WARN_ON(start > offset + len);
4210
4211		end = start + frag_iter->len;
4212		if ((copy = end - offset) > 0) {
4213			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4214				return -EMSGSIZE;
4215
4216			if (copy > len)
4217				copy = len;
4218			ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4219					      copy, recursion_level + 1);
4220			if (unlikely(ret < 0))
4221				return ret;
4222			elt += ret;
4223			if ((len -= copy) == 0)
4224				return elt;
4225			offset += copy;
4226		}
4227		start = end;
4228	}
4229	BUG_ON(len);
4230	return elt;
4231}
4232
4233/**
4234 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4235 *	@skb: Socket buffer containing the buffers to be mapped
4236 *	@sg: The scatter-gather list to map into
4237 *	@offset: The offset into the buffer's contents to start mapping
4238 *	@len: Length of buffer space to be mapped
4239 *
4240 *	Fill the specified scatter-gather list with mappings/pointers into a
4241 *	region of the buffer space attached to a socket buffer. Returns either
4242 *	the number of scatterlist items used, or -EMSGSIZE if the contents
4243 *	could not fit.
4244 */
4245int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4246{
4247	int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4248
4249	if (nsg <= 0)
4250		return nsg;
4251
4252	sg_mark_end(&sg[nsg - 1]);
4253
4254	return nsg;
4255}
4256EXPORT_SYMBOL_GPL(skb_to_sgvec);
4257
4258/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4259 * sglist without mark the sg which contain last skb data as the end.
4260 * So the caller can mannipulate sg list as will when padding new data after
4261 * the first call without calling sg_unmark_end to expend sg list.
4262 *
4263 * Scenario to use skb_to_sgvec_nomark:
4264 * 1. sg_init_table
4265 * 2. skb_to_sgvec_nomark(payload1)
4266 * 3. skb_to_sgvec_nomark(payload2)
4267 *
4268 * This is equivalent to:
4269 * 1. sg_init_table
4270 * 2. skb_to_sgvec(payload1)
4271 * 3. sg_unmark_end
4272 * 4. skb_to_sgvec(payload2)
4273 *
4274 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4275 * is more preferable.
4276 */
4277int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4278			int offset, int len)
4279{
4280	return __skb_to_sgvec(skb, sg, offset, len, 0);
4281}
4282EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4283
 
 
 
 
 
4284
 
 
 
4285
4286/**
4287 *	skb_cow_data - Check that a socket buffer's data buffers are writable
4288 *	@skb: The socket buffer to check.
4289 *	@tailbits: Amount of trailing space to be added
4290 *	@trailer: Returned pointer to the skb where the @tailbits space begins
4291 *
4292 *	Make sure that the data buffers attached to a socket buffer are
4293 *	writable. If they are not, private copies are made of the data buffers
4294 *	and the socket buffer is set to use these instead.
4295 *
4296 *	If @tailbits is given, make sure that there is space to write @tailbits
4297 *	bytes of data beyond current end of socket buffer.  @trailer will be
4298 *	set to point to the skb in which this space begins.
4299 *
4300 *	The number of scatterlist elements required to completely map the
4301 *	COW'd and extended socket buffer will be returned.
4302 */
4303int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4304{
4305	int copyflag;
4306	int elt;
4307	struct sk_buff *skb1, **skb_p;
4308
4309	/* If skb is cloned or its head is paged, reallocate
4310	 * head pulling out all the pages (pages are considered not writable
4311	 * at the moment even if they are anonymous).
4312	 */
4313	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4314	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4315		return -ENOMEM;
4316
4317	/* Easy case. Most of packets will go this way. */
4318	if (!skb_has_frag_list(skb)) {
4319		/* A little of trouble, not enough of space for trailer.
4320		 * This should not happen, when stack is tuned to generate
4321		 * good frames. OK, on miss we reallocate and reserve even more
4322		 * space, 128 bytes is fair. */
4323
4324		if (skb_tailroom(skb) < tailbits &&
4325		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4326			return -ENOMEM;
4327
4328		/* Voila! */
4329		*trailer = skb;
4330		return 1;
4331	}
4332
4333	/* Misery. We are in troubles, going to mincer fragments... */
4334
4335	elt = 1;
4336	skb_p = &skb_shinfo(skb)->frag_list;
4337	copyflag = 0;
4338
4339	while ((skb1 = *skb_p) != NULL) {
4340		int ntail = 0;
4341
4342		/* The fragment is partially pulled by someone,
4343		 * this can happen on input. Copy it and everything
4344		 * after it. */
4345
4346		if (skb_shared(skb1))
4347			copyflag = 1;
4348
4349		/* If the skb is the last, worry about trailer. */
4350
4351		if (skb1->next == NULL && tailbits) {
4352			if (skb_shinfo(skb1)->nr_frags ||
4353			    skb_has_frag_list(skb1) ||
4354			    skb_tailroom(skb1) < tailbits)
4355				ntail = tailbits + 128;
4356		}
4357
4358		if (copyflag ||
4359		    skb_cloned(skb1) ||
4360		    ntail ||
4361		    skb_shinfo(skb1)->nr_frags ||
4362		    skb_has_frag_list(skb1)) {
4363			struct sk_buff *skb2;
4364
4365			/* Fuck, we are miserable poor guys... */
4366			if (ntail == 0)
4367				skb2 = skb_copy(skb1, GFP_ATOMIC);
4368			else
4369				skb2 = skb_copy_expand(skb1,
4370						       skb_headroom(skb1),
4371						       ntail,
4372						       GFP_ATOMIC);
4373			if (unlikely(skb2 == NULL))
4374				return -ENOMEM;
4375
4376			if (skb1->sk)
4377				skb_set_owner_w(skb2, skb1->sk);
4378
4379			/* Looking around. Are we still alive?
4380			 * OK, link new skb, drop old one */
4381
4382			skb2->next = skb1->next;
4383			*skb_p = skb2;
4384			kfree_skb(skb1);
4385			skb1 = skb2;
4386		}
4387		elt++;
4388		*trailer = skb1;
4389		skb_p = &skb1->next;
4390	}
4391
4392	return elt;
4393}
4394EXPORT_SYMBOL_GPL(skb_cow_data);
4395
4396static void sock_rmem_free(struct sk_buff *skb)
4397{
4398	struct sock *sk = skb->sk;
4399
4400	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4401}
4402
4403static void skb_set_err_queue(struct sk_buff *skb)
4404{
4405	/* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4406	 * So, it is safe to (mis)use it to mark skbs on the error queue.
4407	 */
4408	skb->pkt_type = PACKET_OUTGOING;
4409	BUILD_BUG_ON(PACKET_OUTGOING == 0);
4410}
4411
4412/*
4413 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4414 */
4415int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4416{
4417	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4418	    (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4419		return -ENOMEM;
4420
4421	skb_orphan(skb);
4422	skb->sk = sk;
4423	skb->destructor = sock_rmem_free;
4424	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4425	skb_set_err_queue(skb);
4426
4427	/* before exiting rcu section, make sure dst is refcounted */
4428	skb_dst_force(skb);
4429
4430	skb_queue_tail(&sk->sk_error_queue, skb);
4431	if (!sock_flag(sk, SOCK_DEAD))
4432		sk->sk_error_report(sk);
4433	return 0;
4434}
4435EXPORT_SYMBOL(sock_queue_err_skb);
4436
4437static bool is_icmp_err_skb(const struct sk_buff *skb)
4438{
4439	return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4440		       SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4441}
4442
4443struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4444{
4445	struct sk_buff_head *q = &sk->sk_error_queue;
4446	struct sk_buff *skb, *skb_next = NULL;
4447	bool icmp_next = false;
4448	unsigned long flags;
 
4449
4450	spin_lock_irqsave(&q->lock, flags);
4451	skb = __skb_dequeue(q);
4452	if (skb && (skb_next = skb_peek(q))) {
4453		icmp_next = is_icmp_err_skb(skb_next);
4454		if (icmp_next)
4455			sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4456	}
4457	spin_unlock_irqrestore(&q->lock, flags);
4458
4459	if (is_icmp_err_skb(skb) && !icmp_next)
4460		sk->sk_err = 0;
4461
4462	if (skb_next)
4463		sk->sk_error_report(sk);
4464
4465	return skb;
4466}
4467EXPORT_SYMBOL(sock_dequeue_err_skb);
4468
4469/**
4470 * skb_clone_sk - create clone of skb, and take reference to socket
4471 * @skb: the skb to clone
4472 *
4473 * This function creates a clone of a buffer that holds a reference on
4474 * sk_refcnt.  Buffers created via this function are meant to be
4475 * returned using sock_queue_err_skb, or free via kfree_skb.
4476 *
4477 * When passing buffers allocated with this function to sock_queue_err_skb
4478 * it is necessary to wrap the call with sock_hold/sock_put in order to
4479 * prevent the socket from being released prior to being enqueued on
4480 * the sk_error_queue.
4481 */
4482struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4483{
4484	struct sock *sk = skb->sk;
4485	struct sk_buff *clone;
4486
4487	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4488		return NULL;
4489
4490	clone = skb_clone(skb, GFP_ATOMIC);
4491	if (!clone) {
4492		sock_put(sk);
4493		return NULL;
4494	}
4495
4496	clone->sk = sk;
4497	clone->destructor = sock_efree;
4498
4499	return clone;
4500}
4501EXPORT_SYMBOL(skb_clone_sk);
4502
4503static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4504					struct sock *sk,
4505					int tstype,
4506					bool opt_stats)
4507{
4508	struct sock_exterr_skb *serr;
4509	int err;
4510
4511	BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4512
4513	serr = SKB_EXT_ERR(skb);
4514	memset(serr, 0, sizeof(*serr));
4515	serr->ee.ee_errno = ENOMSG;
4516	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4517	serr->ee.ee_info = tstype;
4518	serr->opt_stats = opt_stats;
4519	serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4520	if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4521		serr->ee.ee_data = skb_shinfo(skb)->tskey;
4522		if (sk->sk_protocol == IPPROTO_TCP &&
4523		    sk->sk_type == SOCK_STREAM)
4524			serr->ee.ee_data -= sk->sk_tskey;
4525	}
4526
4527	err = sock_queue_err_skb(sk, skb);
4528
4529	if (err)
4530		kfree_skb(skb);
4531}
4532
4533static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4534{
4535	bool ret;
4536
4537	if (likely(sysctl_tstamp_allow_data || tsonly))
4538		return true;
4539
4540	read_lock_bh(&sk->sk_callback_lock);
4541	ret = sk->sk_socket && sk->sk_socket->file &&
4542	      file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4543	read_unlock_bh(&sk->sk_callback_lock);
4544	return ret;
4545}
4546
4547void skb_complete_tx_timestamp(struct sk_buff *skb,
4548			       struct skb_shared_hwtstamps *hwtstamps)
4549{
4550	struct sock *sk = skb->sk;
4551
4552	if (!skb_may_tx_timestamp(sk, false))
4553		goto err;
 
 
 
4554
4555	/* Take a reference to prevent skb_orphan() from freeing the socket,
4556	 * but only if the socket refcount is not zero.
4557	 */
4558	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4559		*skb_hwtstamps(skb) = *hwtstamps;
4560		__skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4561		sock_put(sk);
4562		return;
4563	}
4564
4565err:
4566	kfree_skb(skb);
4567}
4568EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4569
4570void __skb_tstamp_tx(struct sk_buff *orig_skb,
4571		     struct skb_shared_hwtstamps *hwtstamps,
4572		     struct sock *sk, int tstype)
4573{
4574	struct sk_buff *skb;
4575	bool tsonly, opt_stats = false;
4576
4577	if (!sk)
4578		return;
4579
4580	if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4581	    skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4582		return;
4583
4584	tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4585	if (!skb_may_tx_timestamp(sk, tsonly))
4586		return;
4587
4588	if (tsonly) {
4589#ifdef CONFIG_INET
4590		if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4591		    sk->sk_protocol == IPPROTO_TCP &&
4592		    sk->sk_type == SOCK_STREAM) {
4593			skb = tcp_get_timestamping_opt_stats(sk);
4594			opt_stats = true;
4595		} else
4596#endif
4597			skb = alloc_skb(0, GFP_ATOMIC);
4598	} else {
4599		skb = skb_clone(orig_skb, GFP_ATOMIC);
4600	}
4601	if (!skb)
4602		return;
4603
4604	if (tsonly) {
4605		skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4606					     SKBTX_ANY_TSTAMP;
4607		skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4608	}
4609
4610	if (hwtstamps)
4611		*skb_hwtstamps(skb) = *hwtstamps;
4612	else
4613		skb->tstamp = ktime_get_real();
4614
4615	__skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4616}
4617EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4618
4619void skb_tstamp_tx(struct sk_buff *orig_skb,
4620		   struct skb_shared_hwtstamps *hwtstamps)
4621{
4622	return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4623			       SCM_TSTAMP_SND);
4624}
4625EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4626
4627void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4628{
4629	struct sock *sk = skb->sk;
4630	struct sock_exterr_skb *serr;
4631	int err = 1;
4632
4633	skb->wifi_acked_valid = 1;
4634	skb->wifi_acked = acked;
4635
4636	serr = SKB_EXT_ERR(skb);
4637	memset(serr, 0, sizeof(*serr));
4638	serr->ee.ee_errno = ENOMSG;
4639	serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4640
4641	/* Take a reference to prevent skb_orphan() from freeing the socket,
4642	 * but only if the socket refcount is not zero.
4643	 */
4644	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4645		err = sock_queue_err_skb(sk, skb);
4646		sock_put(sk);
4647	}
4648	if (err)
4649		kfree_skb(skb);
 
 
4650}
4651EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4652
4653/**
4654 * skb_partial_csum_set - set up and verify partial csum values for packet
4655 * @skb: the skb to set
4656 * @start: the number of bytes after skb->data to start checksumming.
4657 * @off: the offset from start to place the checksum.
4658 *
4659 * For untrusted partially-checksummed packets, we need to make sure the values
4660 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4661 *
4662 * This function checks and sets those values and skb->ip_summed: if this
4663 * returns false you should drop the packet.
4664 */
4665bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4666{
4667	u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4668	u32 csum_start = skb_headroom(skb) + (u32)start;
4669
4670	if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4671		net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4672				     start, off, skb_headroom(skb), skb_headlen(skb));
4673		return false;
4674	}
4675	skb->ip_summed = CHECKSUM_PARTIAL;
4676	skb->csum_start = csum_start;
4677	skb->csum_offset = off;
4678	skb_set_transport_header(skb, start);
4679	return true;
4680}
4681EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4682
4683static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4684			       unsigned int max)
4685{
4686	if (skb_headlen(skb) >= len)
4687		return 0;
4688
4689	/* If we need to pullup then pullup to the max, so we
4690	 * won't need to do it again.
4691	 */
4692	if (max > skb->len)
4693		max = skb->len;
4694
4695	if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4696		return -ENOMEM;
4697
4698	if (skb_headlen(skb) < len)
4699		return -EPROTO;
4700
4701	return 0;
4702}
4703
4704#define MAX_TCP_HDR_LEN (15 * 4)
4705
4706static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4707				      typeof(IPPROTO_IP) proto,
4708				      unsigned int off)
4709{
4710	switch (proto) {
4711		int err;
4712
4713	case IPPROTO_TCP:
4714		err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4715					  off + MAX_TCP_HDR_LEN);
4716		if (!err && !skb_partial_csum_set(skb, off,
4717						  offsetof(struct tcphdr,
4718							   check)))
4719			err = -EPROTO;
4720		return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4721
4722	case IPPROTO_UDP:
4723		err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4724					  off + sizeof(struct udphdr));
4725		if (!err && !skb_partial_csum_set(skb, off,
4726						  offsetof(struct udphdr,
4727							   check)))
4728			err = -EPROTO;
4729		return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4730	}
4731
4732	return ERR_PTR(-EPROTO);
4733}
4734
4735/* This value should be large enough to cover a tagged ethernet header plus
4736 * maximally sized IP and TCP or UDP headers.
4737 */
4738#define MAX_IP_HDR_LEN 128
4739
4740static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4741{
4742	unsigned int off;
4743	bool fragment;
4744	__sum16 *csum;
4745	int err;
4746
4747	fragment = false;
4748
4749	err = skb_maybe_pull_tail(skb,
4750				  sizeof(struct iphdr),
4751				  MAX_IP_HDR_LEN);
4752	if (err < 0)
4753		goto out;
4754
4755	if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4756		fragment = true;
4757
4758	off = ip_hdrlen(skb);
4759
4760	err = -EPROTO;
4761
4762	if (fragment)
4763		goto out;
4764
4765	csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4766	if (IS_ERR(csum))
4767		return PTR_ERR(csum);
4768
4769	if (recalculate)
4770		*csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4771					   ip_hdr(skb)->daddr,
4772					   skb->len - off,
4773					   ip_hdr(skb)->protocol, 0);
4774	err = 0;
4775
4776out:
4777	return err;
4778}
4779
4780/* This value should be large enough to cover a tagged ethernet header plus
4781 * an IPv6 header, all options, and a maximal TCP or UDP header.
4782 */
4783#define MAX_IPV6_HDR_LEN 256
4784
4785#define OPT_HDR(type, skb, off) \
4786	(type *)(skb_network_header(skb) + (off))
4787
4788static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4789{
4790	int err;
4791	u8 nexthdr;
4792	unsigned int off;
4793	unsigned int len;
4794	bool fragment;
4795	bool done;
4796	__sum16 *csum;
4797
4798	fragment = false;
4799	done = false;
4800
4801	off = sizeof(struct ipv6hdr);
4802
4803	err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4804	if (err < 0)
4805		goto out;
4806
4807	nexthdr = ipv6_hdr(skb)->nexthdr;
4808
4809	len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4810	while (off <= len && !done) {
4811		switch (nexthdr) {
4812		case IPPROTO_DSTOPTS:
4813		case IPPROTO_HOPOPTS:
4814		case IPPROTO_ROUTING: {
4815			struct ipv6_opt_hdr *hp;
4816
4817			err = skb_maybe_pull_tail(skb,
4818						  off +
4819						  sizeof(struct ipv6_opt_hdr),
4820						  MAX_IPV6_HDR_LEN);
4821			if (err < 0)
4822				goto out;
4823
4824			hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4825			nexthdr = hp->nexthdr;
4826			off += ipv6_optlen(hp);
4827			break;
4828		}
4829		case IPPROTO_AH: {
4830			struct ip_auth_hdr *hp;
4831
4832			err = skb_maybe_pull_tail(skb,
4833						  off +
4834						  sizeof(struct ip_auth_hdr),
4835						  MAX_IPV6_HDR_LEN);
4836			if (err < 0)
4837				goto out;
4838
4839			hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4840			nexthdr = hp->nexthdr;
4841			off += ipv6_authlen(hp);
4842			break;
4843		}
4844		case IPPROTO_FRAGMENT: {
4845			struct frag_hdr *hp;
4846
4847			err = skb_maybe_pull_tail(skb,
4848						  off +
4849						  sizeof(struct frag_hdr),
4850						  MAX_IPV6_HDR_LEN);
4851			if (err < 0)
4852				goto out;
4853
4854			hp = OPT_HDR(struct frag_hdr, skb, off);
4855
4856			if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4857				fragment = true;
4858
4859			nexthdr = hp->nexthdr;
4860			off += sizeof(struct frag_hdr);
4861			break;
4862		}
4863		default:
4864			done = true;
4865			break;
4866		}
4867	}
4868
4869	err = -EPROTO;
4870
4871	if (!done || fragment)
4872		goto out;
4873
4874	csum = skb_checksum_setup_ip(skb, nexthdr, off);
4875	if (IS_ERR(csum))
4876		return PTR_ERR(csum);
4877
4878	if (recalculate)
4879		*csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4880					 &ipv6_hdr(skb)->daddr,
4881					 skb->len - off, nexthdr, 0);
4882	err = 0;
4883
4884out:
4885	return err;
4886}
4887
4888/**
4889 * skb_checksum_setup - set up partial checksum offset
4890 * @skb: the skb to set up
4891 * @recalculate: if true the pseudo-header checksum will be recalculated
4892 */
4893int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4894{
4895	int err;
4896
4897	switch (skb->protocol) {
4898	case htons(ETH_P_IP):
4899		err = skb_checksum_setup_ipv4(skb, recalculate);
4900		break;
4901
4902	case htons(ETH_P_IPV6):
4903		err = skb_checksum_setup_ipv6(skb, recalculate);
4904		break;
4905
4906	default:
4907		err = -EPROTO;
4908		break;
4909	}
4910
4911	return err;
4912}
4913EXPORT_SYMBOL(skb_checksum_setup);
4914
4915/**
4916 * skb_checksum_maybe_trim - maybe trims the given skb
4917 * @skb: the skb to check
4918 * @transport_len: the data length beyond the network header
4919 *
4920 * Checks whether the given skb has data beyond the given transport length.
4921 * If so, returns a cloned skb trimmed to this transport length.
4922 * Otherwise returns the provided skb. Returns NULL in error cases
4923 * (e.g. transport_len exceeds skb length or out-of-memory).
4924 *
4925 * Caller needs to set the skb transport header and free any returned skb if it
4926 * differs from the provided skb.
4927 */
4928static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4929					       unsigned int transport_len)
4930{
4931	struct sk_buff *skb_chk;
4932	unsigned int len = skb_transport_offset(skb) + transport_len;
4933	int ret;
4934
4935	if (skb->len < len)
4936		return NULL;
4937	else if (skb->len == len)
4938		return skb;
4939
4940	skb_chk = skb_clone(skb, GFP_ATOMIC);
4941	if (!skb_chk)
4942		return NULL;
4943
4944	ret = pskb_trim_rcsum(skb_chk, len);
4945	if (ret) {
4946		kfree_skb(skb_chk);
4947		return NULL;
4948	}
4949
4950	return skb_chk;
4951}
4952
4953/**
4954 * skb_checksum_trimmed - validate checksum of an skb
4955 * @skb: the skb to check
4956 * @transport_len: the data length beyond the network header
4957 * @skb_chkf: checksum function to use
4958 *
4959 * Applies the given checksum function skb_chkf to the provided skb.
4960 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4961 *
4962 * If the skb has data beyond the given transport length, then a
4963 * trimmed & cloned skb is checked and returned.
4964 *
4965 * Caller needs to set the skb transport header and free any returned skb if it
4966 * differs from the provided skb.
4967 */
4968struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4969				     unsigned int transport_len,
4970				     __sum16(*skb_chkf)(struct sk_buff *skb))
4971{
4972	struct sk_buff *skb_chk;
4973	unsigned int offset = skb_transport_offset(skb);
4974	__sum16 ret;
4975
4976	skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4977	if (!skb_chk)
4978		goto err;
4979
4980	if (!pskb_may_pull(skb_chk, offset))
4981		goto err;
4982
4983	skb_pull_rcsum(skb_chk, offset);
4984	ret = skb_chkf(skb_chk);
4985	skb_push_rcsum(skb_chk, offset);
4986
4987	if (ret)
4988		goto err;
4989
4990	return skb_chk;
4991
4992err:
4993	if (skb_chk && skb_chk != skb)
4994		kfree_skb(skb_chk);
4995
4996	return NULL;
4997
4998}
4999EXPORT_SYMBOL(skb_checksum_trimmed);
5000
5001void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5002{
5003	net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5004			     skb->dev->name);
5005}
5006EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5007
5008void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5009{
5010	if (head_stolen) {
5011		skb_release_head_state(skb);
5012		kmem_cache_free(skbuff_head_cache, skb);
5013	} else {
5014		__kfree_skb(skb);
5015	}
5016}
5017EXPORT_SYMBOL(kfree_skb_partial);
5018
5019/**
5020 * skb_try_coalesce - try to merge skb to prior one
5021 * @to: prior buffer
5022 * @from: buffer to add
5023 * @fragstolen: pointer to boolean
5024 * @delta_truesize: how much more was allocated than was requested
5025 */
5026bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5027		      bool *fragstolen, int *delta_truesize)
5028{
5029	struct skb_shared_info *to_shinfo, *from_shinfo;
5030	int i, delta, len = from->len;
5031
5032	*fragstolen = false;
5033
5034	if (skb_cloned(to))
5035		return false;
5036
5037	if (len <= skb_tailroom(to)) {
5038		if (len)
5039			BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5040		*delta_truesize = 0;
5041		return true;
5042	}
5043
5044	to_shinfo = skb_shinfo(to);
5045	from_shinfo = skb_shinfo(from);
5046	if (to_shinfo->frag_list || from_shinfo->frag_list)
5047		return false;
5048	if (skb_zcopy(to) || skb_zcopy(from))
5049		return false;
5050
5051	if (skb_headlen(from) != 0) {
5052		struct page *page;
5053		unsigned int offset;
5054
5055		if (to_shinfo->nr_frags +
5056		    from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5057			return false;
5058
5059		if (skb_head_is_locked(from))
5060			return false;
5061
5062		delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5063
5064		page = virt_to_head_page(from->head);
5065		offset = from->data - (unsigned char *)page_address(page);
5066
5067		skb_fill_page_desc(to, to_shinfo->nr_frags,
5068				   page, offset, skb_headlen(from));
5069		*fragstolen = true;
5070	} else {
5071		if (to_shinfo->nr_frags +
5072		    from_shinfo->nr_frags > MAX_SKB_FRAGS)
5073			return false;
5074
5075		delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5076	}
5077
5078	WARN_ON_ONCE(delta < len);
5079
5080	memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5081	       from_shinfo->frags,
5082	       from_shinfo->nr_frags * sizeof(skb_frag_t));
5083	to_shinfo->nr_frags += from_shinfo->nr_frags;
5084
5085	if (!skb_cloned(from))
5086		from_shinfo->nr_frags = 0;
5087
5088	/* if the skb is not cloned this does nothing
5089	 * since we set nr_frags to 0.
5090	 */
5091	for (i = 0; i < from_shinfo->nr_frags; i++)
5092		__skb_frag_ref(&from_shinfo->frags[i]);
5093
5094	to->truesize += delta;
5095	to->len += len;
5096	to->data_len += len;
5097
5098	*delta_truesize = delta;
5099	return true;
5100}
5101EXPORT_SYMBOL(skb_try_coalesce);
5102
5103/**
5104 * skb_scrub_packet - scrub an skb
5105 *
5106 * @skb: buffer to clean
5107 * @xnet: packet is crossing netns
5108 *
5109 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5110 * into/from a tunnel. Some information have to be cleared during these
5111 * operations.
5112 * skb_scrub_packet can also be used to clean a skb before injecting it in
5113 * another namespace (@xnet == true). We have to clear all information in the
5114 * skb that could impact namespace isolation.
5115 */
5116void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5117{
 
5118	skb->pkt_type = PACKET_HOST;
5119	skb->skb_iif = 0;
5120	skb->ignore_df = 0;
5121	skb_dst_drop(skb);
5122	skb_ext_reset(skb);
5123	nf_reset_ct(skb);
5124	nf_reset_trace(skb);
5125
5126#ifdef CONFIG_NET_SWITCHDEV
5127	skb->offload_fwd_mark = 0;
5128	skb->offload_l3_fwd_mark = 0;
5129#endif
5130
5131	if (!xnet)
5132		return;
5133
5134	ipvs_reset(skb);
5135	skb->mark = 0;
5136	skb->tstamp = 0;
5137}
5138EXPORT_SYMBOL_GPL(skb_scrub_packet);
5139
5140/**
5141 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5142 *
5143 * @skb: GSO skb
5144 *
5145 * skb_gso_transport_seglen is used to determine the real size of the
5146 * individual segments, including Layer4 headers (TCP/UDP).
5147 *
5148 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5149 */
5150static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5151{
5152	const struct skb_shared_info *shinfo = skb_shinfo(skb);
5153	unsigned int thlen = 0;
5154
5155	if (skb->encapsulation) {
5156		thlen = skb_inner_transport_header(skb) -
5157			skb_transport_header(skb);
5158
5159		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5160			thlen += inner_tcp_hdrlen(skb);
5161	} else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5162		thlen = tcp_hdrlen(skb);
5163	} else if (unlikely(skb_is_gso_sctp(skb))) {
5164		thlen = sizeof(struct sctphdr);
5165	} else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5166		thlen = sizeof(struct udphdr);
5167	}
5168	/* UFO sets gso_size to the size of the fragmentation
5169	 * payload, i.e. the size of the L4 (UDP) header is already
5170	 * accounted for.
5171	 */
5172	return thlen + shinfo->gso_size;
5173}
5174
5175/**
5176 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5177 *
5178 * @skb: GSO skb
5179 *
5180 * skb_gso_network_seglen is used to determine the real size of the
5181 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5182 *
5183 * The MAC/L2 header is not accounted for.
5184 */
5185static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5186{
5187	unsigned int hdr_len = skb_transport_header(skb) -
5188			       skb_network_header(skb);
5189
5190	return hdr_len + skb_gso_transport_seglen(skb);
5191}
5192
5193/**
5194 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5195 *
5196 * @skb: GSO skb
5197 *
5198 * skb_gso_mac_seglen is used to determine the real size of the
5199 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5200 * headers (TCP/UDP).
5201 */
5202static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5203{
5204	unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5205
5206	return hdr_len + skb_gso_transport_seglen(skb);
5207}
5208
5209/**
5210 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5211 *
5212 * There are a couple of instances where we have a GSO skb, and we
5213 * want to determine what size it would be after it is segmented.
5214 *
5215 * We might want to check:
5216 * -    L3+L4+payload size (e.g. IP forwarding)
5217 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5218 *
5219 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5220 *
5221 * @skb: GSO skb
5222 *
5223 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5224 *           GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5225 *
5226 * @max_len: The maximum permissible length.
5227 *
5228 * Returns true if the segmented length <= max length.
5229 */
5230static inline bool skb_gso_size_check(const struct sk_buff *skb,
5231				      unsigned int seg_len,
5232				      unsigned int max_len) {
5233	const struct skb_shared_info *shinfo = skb_shinfo(skb);
5234	const struct sk_buff *iter;
5235
5236	if (shinfo->gso_size != GSO_BY_FRAGS)
5237		return seg_len <= max_len;
5238
5239	/* Undo this so we can re-use header sizes */
5240	seg_len -= GSO_BY_FRAGS;
5241
5242	skb_walk_frags(skb, iter) {
5243		if (seg_len + skb_headlen(iter) > max_len)
5244			return false;
5245	}
5246
5247	return true;
5248}
5249
5250/**
5251 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5252 *
5253 * @skb: GSO skb
5254 * @mtu: MTU to validate against
5255 *
5256 * skb_gso_validate_network_len validates if a given skb will fit a
5257 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5258 * payload.
5259 */
5260bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5261{
5262	return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5263}
5264EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5265
5266/**
5267 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5268 *
5269 * @skb: GSO skb
5270 * @len: length to validate against
5271 *
5272 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5273 * length once split, including L2, L3 and L4 headers and the payload.
5274 */
5275bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5276{
5277	return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5278}
5279EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5280
5281static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5282{
5283	int mac_len, meta_len;
5284	void *meta;
5285
5286	if (skb_cow(skb, skb_headroom(skb)) < 0) {
5287		kfree_skb(skb);
5288		return NULL;
5289	}
5290
5291	mac_len = skb->data - skb_mac_header(skb);
5292	if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5293		memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5294			mac_len - VLAN_HLEN - ETH_TLEN);
5295	}
5296
5297	meta_len = skb_metadata_len(skb);
5298	if (meta_len) {
5299		meta = skb_metadata_end(skb) - meta_len;
5300		memmove(meta + VLAN_HLEN, meta, meta_len);
5301	}
5302
5303	skb->mac_header += VLAN_HLEN;
5304	return skb;
5305}
5306
5307struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5308{
5309	struct vlan_hdr *vhdr;
5310	u16 vlan_tci;
5311
5312	if (unlikely(skb_vlan_tag_present(skb))) {
5313		/* vlan_tci is already set-up so leave this for another time */
5314		return skb;
5315	}
5316
5317	skb = skb_share_check(skb, GFP_ATOMIC);
5318	if (unlikely(!skb))
5319		goto err_free;
5320
5321	if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5322		goto err_free;
5323
5324	vhdr = (struct vlan_hdr *)skb->data;
5325	vlan_tci = ntohs(vhdr->h_vlan_TCI);
5326	__vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5327
5328	skb_pull_rcsum(skb, VLAN_HLEN);
5329	vlan_set_encap_proto(skb, vhdr);
5330
5331	skb = skb_reorder_vlan_header(skb);
5332	if (unlikely(!skb))
5333		goto err_free;
5334
5335	skb_reset_network_header(skb);
5336	skb_reset_transport_header(skb);
5337	skb_reset_mac_len(skb);
5338
5339	return skb;
5340
5341err_free:
5342	kfree_skb(skb);
5343	return NULL;
5344}
5345EXPORT_SYMBOL(skb_vlan_untag);
5346
5347int skb_ensure_writable(struct sk_buff *skb, int write_len)
5348{
5349	if (!pskb_may_pull(skb, write_len))
5350		return -ENOMEM;
5351
5352	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5353		return 0;
5354
5355	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5356}
5357EXPORT_SYMBOL(skb_ensure_writable);
5358
5359/* remove VLAN header from packet and update csum accordingly.
5360 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5361 */
5362int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5363{
5364	struct vlan_hdr *vhdr;
5365	int offset = skb->data - skb_mac_header(skb);
5366	int err;
5367
5368	if (WARN_ONCE(offset,
5369		      "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5370		      offset)) {
5371		return -EINVAL;
5372	}
5373
5374	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5375	if (unlikely(err))
5376		return err;
5377
5378	skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5379
5380	vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5381	*vlan_tci = ntohs(vhdr->h_vlan_TCI);
5382
5383	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5384	__skb_pull(skb, VLAN_HLEN);
5385
5386	vlan_set_encap_proto(skb, vhdr);
5387	skb->mac_header += VLAN_HLEN;
5388
5389	if (skb_network_offset(skb) < ETH_HLEN)
5390		skb_set_network_header(skb, ETH_HLEN);
5391
5392	skb_reset_mac_len(skb);
 
 
5393
5394	return err;
5395}
5396EXPORT_SYMBOL(__skb_vlan_pop);
5397
5398/* Pop a vlan tag either from hwaccel or from payload.
5399 * Expects skb->data at mac header.
5400 */
5401int skb_vlan_pop(struct sk_buff *skb)
5402{
5403	u16 vlan_tci;
5404	__be16 vlan_proto;
5405	int err;
5406
5407	if (likely(skb_vlan_tag_present(skb))) {
5408		__vlan_hwaccel_clear_tag(skb);
5409	} else {
5410		if (unlikely(!eth_type_vlan(skb->protocol)))
 
 
5411			return 0;
5412
5413		err = __skb_vlan_pop(skb, &vlan_tci);
5414		if (err)
5415			return err;
5416	}
5417	/* move next vlan tag to hw accel tag */
5418	if (likely(!eth_type_vlan(skb->protocol)))
 
 
5419		return 0;
5420
5421	vlan_proto = skb->protocol;
5422	err = __skb_vlan_pop(skb, &vlan_tci);
5423	if (unlikely(err))
5424		return err;
5425
5426	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5427	return 0;
5428}
5429EXPORT_SYMBOL(skb_vlan_pop);
5430
5431/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5432 * Expects skb->data at mac header.
5433 */
5434int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5435{
5436	if (skb_vlan_tag_present(skb)) {
5437		int offset = skb->data - skb_mac_header(skb);
5438		int err;
5439
5440		if (WARN_ONCE(offset,
5441			      "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5442			      offset)) {
5443			return -EINVAL;
5444		}
5445
5446		err = __vlan_insert_tag(skb, skb->vlan_proto,
5447					skb_vlan_tag_get(skb));
5448		if (err)
 
5449			return err;
 
5450
5451		skb->protocol = skb->vlan_proto;
5452		skb->mac_len += VLAN_HLEN;
5453
5454		skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
 
5455	}
5456	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5457	return 0;
5458}
5459EXPORT_SYMBOL(skb_vlan_push);
5460
5461/* Update the ethertype of hdr and the skb csum value if required. */
5462static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5463			     __be16 ethertype)
5464{
5465	if (skb->ip_summed == CHECKSUM_COMPLETE) {
5466		__be16 diff[] = { ~hdr->h_proto, ethertype };
5467
5468		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5469	}
5470
5471	hdr->h_proto = ethertype;
5472}
5473
5474/**
5475 * skb_mpls_push() - push a new MPLS header after the mac header
5476 *
5477 * @skb: buffer
5478 * @mpls_lse: MPLS label stack entry to push
5479 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5480 * @mac_len: length of the MAC header
5481 *
5482 * Expects skb->data at mac header.
5483 *
5484 * Returns 0 on success, -errno otherwise.
5485 */
5486int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5487		  int mac_len)
5488{
5489	struct mpls_shim_hdr *lse;
5490	int err;
5491
5492	if (unlikely(!eth_p_mpls(mpls_proto)))
5493		return -EINVAL;
5494
5495	/* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5496	if (skb->encapsulation)
5497		return -EINVAL;
5498
5499	err = skb_cow_head(skb, MPLS_HLEN);
5500	if (unlikely(err))
5501		return err;
5502
5503	if (!skb->inner_protocol) {
5504		skb_set_inner_network_header(skb, mac_len);
5505		skb_set_inner_protocol(skb, skb->protocol);
5506	}
5507
5508	skb_push(skb, MPLS_HLEN);
5509	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5510		mac_len);
5511	skb_reset_mac_header(skb);
5512	skb_set_network_header(skb, mac_len);
5513
5514	lse = mpls_hdr(skb);
5515	lse->label_stack_entry = mpls_lse;
5516	skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5517
5518	if (skb->dev && skb->dev->type == ARPHRD_ETHER)
5519		skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5520	skb->protocol = mpls_proto;
5521
5522	return 0;
5523}
5524EXPORT_SYMBOL_GPL(skb_mpls_push);
5525
5526/**
5527 * skb_mpls_pop() - pop the outermost MPLS header
5528 *
5529 * @skb: buffer
5530 * @next_proto: ethertype of header after popped MPLS header
5531 * @mac_len: length of the MAC header
5532 *
5533 * Expects skb->data at mac header.
5534 *
5535 * Returns 0 on success, -errno otherwise.
5536 */
5537int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len)
5538{
5539	int err;
5540
5541	if (unlikely(!eth_p_mpls(skb->protocol)))
5542		return 0;
5543
5544	err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5545	if (unlikely(err))
5546		return err;
5547
5548	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5549	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5550		mac_len);
5551
5552	__skb_pull(skb, MPLS_HLEN);
5553	skb_reset_mac_header(skb);
5554	skb_set_network_header(skb, mac_len);
5555
5556	if (skb->dev && skb->dev->type == ARPHRD_ETHER) {
5557		struct ethhdr *hdr;
5558
5559		/* use mpls_hdr() to get ethertype to account for VLANs. */
5560		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5561		skb_mod_eth_type(skb, hdr, next_proto);
5562	}
5563	skb->protocol = next_proto;
5564
5565	return 0;
5566}
5567EXPORT_SYMBOL_GPL(skb_mpls_pop);
5568
5569/**
5570 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5571 *
5572 * @skb: buffer
5573 * @mpls_lse: new MPLS label stack entry to update to
5574 *
5575 * Expects skb->data at mac header.
5576 *
5577 * Returns 0 on success, -errno otherwise.
5578 */
5579int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5580{
5581	int err;
5582
5583	if (unlikely(!eth_p_mpls(skb->protocol)))
5584		return -EINVAL;
5585
5586	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5587	if (unlikely(err))
5588		return err;
5589
5590	if (skb->ip_summed == CHECKSUM_COMPLETE) {
5591		__be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5592
5593		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5594	}
5595
5596	mpls_hdr(skb)->label_stack_entry = mpls_lse;
5597
5598	return 0;
5599}
5600EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5601
5602/**
5603 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5604 *
5605 * @skb: buffer
5606 *
5607 * Expects skb->data at mac header.
5608 *
5609 * Returns 0 on success, -errno otherwise.
5610 */
5611int skb_mpls_dec_ttl(struct sk_buff *skb)
5612{
5613	u32 lse;
5614	u8 ttl;
5615
5616	if (unlikely(!eth_p_mpls(skb->protocol)))
5617		return -EINVAL;
5618
5619	lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5620	ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5621	if (!--ttl)
5622		return -EINVAL;
5623
5624	lse &= ~MPLS_LS_TTL_MASK;
5625	lse |= ttl << MPLS_LS_TTL_SHIFT;
5626
5627	return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5628}
5629EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5630
5631/**
5632 * alloc_skb_with_frags - allocate skb with page frags
5633 *
5634 * @header_len: size of linear part
5635 * @data_len: needed length in frags
5636 * @max_page_order: max page order desired.
5637 * @errcode: pointer to error code if any
5638 * @gfp_mask: allocation mask
5639 *
5640 * This can be used to allocate a paged skb, given a maximal order for frags.
5641 */
5642struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5643				     unsigned long data_len,
5644				     int max_page_order,
5645				     int *errcode,
5646				     gfp_t gfp_mask)
5647{
5648	int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5649	unsigned long chunk;
5650	struct sk_buff *skb;
5651	struct page *page;
 
5652	int i;
5653
5654	*errcode = -EMSGSIZE;
5655	/* Note this test could be relaxed, if we succeed to allocate
5656	 * high order pages...
5657	 */
5658	if (npages > MAX_SKB_FRAGS)
5659		return NULL;
5660
 
 
 
 
5661	*errcode = -ENOBUFS;
5662	skb = alloc_skb(header_len, gfp_mask);
5663	if (!skb)
5664		return NULL;
5665
5666	skb->truesize += npages << PAGE_SHIFT;
5667
5668	for (i = 0; npages > 0; i++) {
5669		int order = max_page_order;
5670
5671		while (order) {
5672			if (npages >= 1 << order) {
5673				page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5674						   __GFP_COMP |
5675						   __GFP_NOWARN,
 
5676						   order);
5677				if (page)
5678					goto fill_page;
5679				/* Do not retry other high order allocations */
5680				order = 1;
5681				max_page_order = 0;
5682			}
5683			order--;
5684		}
5685		page = alloc_page(gfp_mask);
5686		if (!page)
5687			goto failure;
5688fill_page:
5689		chunk = min_t(unsigned long, data_len,
5690			      PAGE_SIZE << order);
5691		skb_fill_page_desc(skb, i, page, 0, chunk);
5692		data_len -= chunk;
5693		npages -= 1 << order;
5694	}
5695	return skb;
5696
5697failure:
5698	kfree_skb(skb);
5699	return NULL;
5700}
5701EXPORT_SYMBOL(alloc_skb_with_frags);
5702
5703/* carve out the first off bytes from skb when off < headlen */
5704static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5705				    const int headlen, gfp_t gfp_mask)
5706{
5707	int i;
5708	int size = skb_end_offset(skb);
5709	int new_hlen = headlen - off;
5710	u8 *data;
5711
5712	size = SKB_DATA_ALIGN(size);
5713
5714	if (skb_pfmemalloc(skb))
5715		gfp_mask |= __GFP_MEMALLOC;
5716	data = kmalloc_reserve(size +
5717			       SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5718			       gfp_mask, NUMA_NO_NODE, NULL);
5719	if (!data)
5720		return -ENOMEM;
5721
5722	size = SKB_WITH_OVERHEAD(ksize(data));
5723
5724	/* Copy real data, and all frags */
5725	skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5726	skb->len -= off;
5727
5728	memcpy((struct skb_shared_info *)(data + size),
5729	       skb_shinfo(skb),
5730	       offsetof(struct skb_shared_info,
5731			frags[skb_shinfo(skb)->nr_frags]));
5732	if (skb_cloned(skb)) {
5733		/* drop the old head gracefully */
5734		if (skb_orphan_frags(skb, gfp_mask)) {
5735			kfree(data);
5736			return -ENOMEM;
5737		}
5738		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5739			skb_frag_ref(skb, i);
5740		if (skb_has_frag_list(skb))
5741			skb_clone_fraglist(skb);
5742		skb_release_data(skb);
5743	} else {
5744		/* we can reuse existing recount- all we did was
5745		 * relocate values
5746		 */
5747		skb_free_head(skb);
5748	}
5749
5750	skb->head = data;
5751	skb->data = data;
5752	skb->head_frag = 0;
5753#ifdef NET_SKBUFF_DATA_USES_OFFSET
5754	skb->end = size;
5755#else
5756	skb->end = skb->head + size;
5757#endif
5758	skb_set_tail_pointer(skb, skb_headlen(skb));
5759	skb_headers_offset_update(skb, 0);
5760	skb->cloned = 0;
5761	skb->hdr_len = 0;
5762	skb->nohdr = 0;
5763	atomic_set(&skb_shinfo(skb)->dataref, 1);
5764
5765	return 0;
5766}
5767
5768static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5769
5770/* carve out the first eat bytes from skb's frag_list. May recurse into
5771 * pskb_carve()
5772 */
5773static int pskb_carve_frag_list(struct sk_buff *skb,
5774				struct skb_shared_info *shinfo, int eat,
5775				gfp_t gfp_mask)
5776{
5777	struct sk_buff *list = shinfo->frag_list;
5778	struct sk_buff *clone = NULL;
5779	struct sk_buff *insp = NULL;
5780
5781	do {
5782		if (!list) {
5783			pr_err("Not enough bytes to eat. Want %d\n", eat);
5784			return -EFAULT;
5785		}
5786		if (list->len <= eat) {
5787			/* Eaten as whole. */
5788			eat -= list->len;
5789			list = list->next;
5790			insp = list;
5791		} else {
5792			/* Eaten partially. */
5793			if (skb_shared(list)) {
5794				clone = skb_clone(list, gfp_mask);
5795				if (!clone)
5796					return -ENOMEM;
5797				insp = list->next;
5798				list = clone;
5799			} else {
5800				/* This may be pulled without problems. */
5801				insp = list;
5802			}
5803			if (pskb_carve(list, eat, gfp_mask) < 0) {
5804				kfree_skb(clone);
5805				return -ENOMEM;
5806			}
5807			break;
5808		}
5809	} while (eat);
5810
5811	/* Free pulled out fragments. */
5812	while ((list = shinfo->frag_list) != insp) {
5813		shinfo->frag_list = list->next;
5814		kfree_skb(list);
5815	}
5816	/* And insert new clone at head. */
5817	if (clone) {
5818		clone->next = list;
5819		shinfo->frag_list = clone;
5820	}
5821	return 0;
5822}
5823
5824/* carve off first len bytes from skb. Split line (off) is in the
5825 * non-linear part of skb
5826 */
5827static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5828				       int pos, gfp_t gfp_mask)
5829{
5830	int i, k = 0;
5831	int size = skb_end_offset(skb);
5832	u8 *data;
5833	const int nfrags = skb_shinfo(skb)->nr_frags;
5834	struct skb_shared_info *shinfo;
5835
5836	size = SKB_DATA_ALIGN(size);
5837
5838	if (skb_pfmemalloc(skb))
5839		gfp_mask |= __GFP_MEMALLOC;
5840	data = kmalloc_reserve(size +
5841			       SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5842			       gfp_mask, NUMA_NO_NODE, NULL);
5843	if (!data)
5844		return -ENOMEM;
5845
5846	size = SKB_WITH_OVERHEAD(ksize(data));
5847
5848	memcpy((struct skb_shared_info *)(data + size),
5849	       skb_shinfo(skb), offsetof(struct skb_shared_info,
5850					 frags[skb_shinfo(skb)->nr_frags]));
5851	if (skb_orphan_frags(skb, gfp_mask)) {
5852		kfree(data);
5853		return -ENOMEM;
5854	}
5855	shinfo = (struct skb_shared_info *)(data + size);
5856	for (i = 0; i < nfrags; i++) {
5857		int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5858
5859		if (pos + fsize > off) {
5860			shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5861
5862			if (pos < off) {
5863				/* Split frag.
5864				 * We have two variants in this case:
5865				 * 1. Move all the frag to the second
5866				 *    part, if it is possible. F.e.
5867				 *    this approach is mandatory for TUX,
5868				 *    where splitting is expensive.
5869				 * 2. Split is accurately. We make this.
5870				 */
5871				skb_frag_off_add(&shinfo->frags[0], off - pos);
5872				skb_frag_size_sub(&shinfo->frags[0], off - pos);
5873			}
5874			skb_frag_ref(skb, i);
5875			k++;
5876		}
5877		pos += fsize;
5878	}
5879	shinfo->nr_frags = k;
5880	if (skb_has_frag_list(skb))
5881		skb_clone_fraglist(skb);
5882
5883	if (k == 0) {
5884		/* split line is in frag list */
5885		pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5886	}
5887	skb_release_data(skb);
5888
5889	skb->head = data;
5890	skb->head_frag = 0;
5891	skb->data = data;
5892#ifdef NET_SKBUFF_DATA_USES_OFFSET
5893	skb->end = size;
5894#else
5895	skb->end = skb->head + size;
5896#endif
5897	skb_reset_tail_pointer(skb);
5898	skb_headers_offset_update(skb, 0);
5899	skb->cloned   = 0;
5900	skb->hdr_len  = 0;
5901	skb->nohdr    = 0;
5902	skb->len -= off;
5903	skb->data_len = skb->len;
5904	atomic_set(&skb_shinfo(skb)->dataref, 1);
5905	return 0;
5906}
5907
5908/* remove len bytes from the beginning of the skb */
5909static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5910{
5911	int headlen = skb_headlen(skb);
5912
5913	if (len < headlen)
5914		return pskb_carve_inside_header(skb, len, headlen, gfp);
5915	else
5916		return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5917}
5918
5919/* Extract to_copy bytes starting at off from skb, and return this in
5920 * a new skb
5921 */
5922struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5923			     int to_copy, gfp_t gfp)
5924{
5925	struct sk_buff  *clone = skb_clone(skb, gfp);
5926
5927	if (!clone)
5928		return NULL;
5929
5930	if (pskb_carve(clone, off, gfp) < 0 ||
5931	    pskb_trim(clone, to_copy)) {
5932		kfree_skb(clone);
5933		return NULL;
5934	}
5935	return clone;
5936}
5937EXPORT_SYMBOL(pskb_extract);
5938
5939/**
5940 * skb_condense - try to get rid of fragments/frag_list if possible
5941 * @skb: buffer
5942 *
5943 * Can be used to save memory before skb is added to a busy queue.
5944 * If packet has bytes in frags and enough tail room in skb->head,
5945 * pull all of them, so that we can free the frags right now and adjust
5946 * truesize.
5947 * Notes:
5948 *	We do not reallocate skb->head thus can not fail.
5949 *	Caller must re-evaluate skb->truesize if needed.
5950 */
5951void skb_condense(struct sk_buff *skb)
5952{
5953	if (skb->data_len) {
5954		if (skb->data_len > skb->end - skb->tail ||
5955		    skb_cloned(skb))
5956			return;
5957
5958		/* Nice, we can free page frag(s) right now */
5959		__pskb_pull_tail(skb, skb->data_len);
5960	}
5961	/* At this point, skb->truesize might be over estimated,
5962	 * because skb had a fragment, and fragments do not tell
5963	 * their truesize.
5964	 * When we pulled its content into skb->head, fragment
5965	 * was freed, but __pskb_pull_tail() could not possibly
5966	 * adjust skb->truesize, not knowing the frag truesize.
5967	 */
5968	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5969}
5970
5971#ifdef CONFIG_SKB_EXTENSIONS
5972static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
5973{
5974	return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
5975}
5976
5977static struct skb_ext *skb_ext_alloc(void)
5978{
5979	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5980
5981	if (new) {
5982		memset(new->offset, 0, sizeof(new->offset));
5983		refcount_set(&new->refcnt, 1);
5984	}
5985
5986	return new;
5987}
5988
5989static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
5990					 unsigned int old_active)
5991{
5992	struct skb_ext *new;
5993
5994	if (refcount_read(&old->refcnt) == 1)
5995		return old;
5996
5997	new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5998	if (!new)
5999		return NULL;
6000
6001	memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6002	refcount_set(&new->refcnt, 1);
6003
6004#ifdef CONFIG_XFRM
6005	if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6006		struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6007		unsigned int i;
6008
6009		for (i = 0; i < sp->len; i++)
6010			xfrm_state_hold(sp->xvec[i]);
6011	}
6012#endif
6013	__skb_ext_put(old);
6014	return new;
6015}
6016
6017/**
6018 * skb_ext_add - allocate space for given extension, COW if needed
6019 * @skb: buffer
6020 * @id: extension to allocate space for
6021 *
6022 * Allocates enough space for the given extension.
6023 * If the extension is already present, a pointer to that extension
6024 * is returned.
6025 *
6026 * If the skb was cloned, COW applies and the returned memory can be
6027 * modified without changing the extension space of clones buffers.
6028 *
6029 * Returns pointer to the extension or NULL on allocation failure.
6030 */
6031void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6032{
6033	struct skb_ext *new, *old = NULL;
6034	unsigned int newlen, newoff;
6035
6036	if (skb->active_extensions) {
6037		old = skb->extensions;
6038
6039		new = skb_ext_maybe_cow(old, skb->active_extensions);
6040		if (!new)
6041			return NULL;
6042
6043		if (__skb_ext_exist(new, id))
6044			goto set_active;
6045
6046		newoff = new->chunks;
6047	} else {
6048		newoff = SKB_EXT_CHUNKSIZEOF(*new);
6049
6050		new = skb_ext_alloc();
6051		if (!new)
6052			return NULL;
6053	}
6054
6055	newlen = newoff + skb_ext_type_len[id];
6056	new->chunks = newlen;
6057	new->offset[id] = newoff;
6058set_active:
6059	skb->extensions = new;
6060	skb->active_extensions |= 1 << id;
6061	return skb_ext_get_ptr(new, id);
6062}
6063EXPORT_SYMBOL(skb_ext_add);
6064
6065#ifdef CONFIG_XFRM
6066static void skb_ext_put_sp(struct sec_path *sp)
6067{
6068	unsigned int i;
6069
6070	for (i = 0; i < sp->len; i++)
6071		xfrm_state_put(sp->xvec[i]);
6072}
6073#endif
6074
6075void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6076{
6077	struct skb_ext *ext = skb->extensions;
6078
6079	skb->active_extensions &= ~(1 << id);
6080	if (skb->active_extensions == 0) {
6081		skb->extensions = NULL;
6082		__skb_ext_put(ext);
6083#ifdef CONFIG_XFRM
6084	} else if (id == SKB_EXT_SEC_PATH &&
6085		   refcount_read(&ext->refcnt) == 1) {
6086		struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6087
6088		skb_ext_put_sp(sp);
6089		sp->len = 0;
6090#endif
6091	}
6092}
6093EXPORT_SYMBOL(__skb_ext_del);
6094
6095void __skb_ext_put(struct skb_ext *ext)
6096{
6097	/* If this is last clone, nothing can increment
6098	 * it after check passes.  Avoids one atomic op.
6099	 */
6100	if (refcount_read(&ext->refcnt) == 1)
6101		goto free_now;
6102
6103	if (!refcount_dec_and_test(&ext->refcnt))
6104		return;
6105free_now:
6106#ifdef CONFIG_XFRM
6107	if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6108		skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6109#endif
6110
6111	kmem_cache_free(skbuff_ext_cache, ext);
6112}
6113EXPORT_SYMBOL(__skb_ext_put);
6114#endif /* CONFIG_SKB_EXTENSIONS */