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/bitfield.h>
  62#include <linux/if_vlan.h>
  63#include <linux/mpls.h>
  64#include <linux/kcov.h>
  65#include <linux/iov_iter.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/gso.h>
  72#include <net/ip6_checksum.h>
  73#include <net/xfrm.h>
  74#include <net/mpls.h>
  75#include <net/mptcp.h>
  76#include <net/mctp.h>
  77#include <net/page_pool/helpers.h>
  78#include <net/dropreason.h>
  79
  80#include <linux/uaccess.h>
  81#include <trace/events/skb.h>
  82#include <linux/highmem.h>
  83#include <linux/capability.h>
  84#include <linux/user_namespace.h>
  85#include <linux/indirect_call_wrapper.h>
  86#include <linux/textsearch.h>
  87
  88#include "dev.h"
  89#include "sock_destructor.h"
  90
  91struct kmem_cache *skbuff_cache __ro_after_init;
  92static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
  93#ifdef CONFIG_SKB_EXTENSIONS
  94static struct kmem_cache *skbuff_ext_cache __ro_after_init;
  95#endif
  96
  97
  98static struct kmem_cache *skb_small_head_cache __ro_after_init;
  99
 100#define SKB_SMALL_HEAD_SIZE SKB_HEAD_ALIGN(MAX_TCP_HEADER)
 101
 102/* We want SKB_SMALL_HEAD_CACHE_SIZE to not be a power of two.
 103 * This should ensure that SKB_SMALL_HEAD_HEADROOM is a unique
 104 * size, and we can differentiate heads from skb_small_head_cache
 105 * vs system slabs by looking at their size (skb_end_offset()).
 106 */
 107#define SKB_SMALL_HEAD_CACHE_SIZE					\
 108	(is_power_of_2(SKB_SMALL_HEAD_SIZE) ?			\
 109		(SKB_SMALL_HEAD_SIZE + L1_CACHE_BYTES) :	\
 110		SKB_SMALL_HEAD_SIZE)
 111
 112#define SKB_SMALL_HEAD_HEADROOM						\
 113	SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE)
 114
 115int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
 116EXPORT_SYMBOL(sysctl_max_skb_frags);
 117
 118#undef FN
 119#define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
 120static const char * const drop_reasons[] = {
 121	[SKB_CONSUMED] = "CONSUMED",
 122	DEFINE_DROP_REASON(FN, FN)
 123};
 124
 125static const struct drop_reason_list drop_reasons_core = {
 126	.reasons = drop_reasons,
 127	.n_reasons = ARRAY_SIZE(drop_reasons),
 128};
 129
 130const struct drop_reason_list __rcu *
 131drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_NUM] = {
 132	[SKB_DROP_REASON_SUBSYS_CORE] = RCU_INITIALIZER(&drop_reasons_core),
 133};
 134EXPORT_SYMBOL(drop_reasons_by_subsys);
 135
 136/**
 137 * drop_reasons_register_subsys - register another drop reason subsystem
 138 * @subsys: the subsystem to register, must not be the core
 139 * @list: the list of drop reasons within the subsystem, must point to
 140 *	a statically initialized list
 141 */
 142void drop_reasons_register_subsys(enum skb_drop_reason_subsys subsys,
 143				  const struct drop_reason_list *list)
 144{
 145	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
 146		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
 147		 "invalid subsystem %d\n", subsys))
 148		return;
 149
 150	/* must point to statically allocated memory, so INIT is OK */
 151	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], list);
 152}
 153EXPORT_SYMBOL_GPL(drop_reasons_register_subsys);
 154
 155/**
 156 * drop_reasons_unregister_subsys - unregister a drop reason subsystem
 157 * @subsys: the subsystem to remove, must not be the core
 158 *
 159 * Note: This will synchronize_rcu() to ensure no users when it returns.
 160 */
 161void drop_reasons_unregister_subsys(enum skb_drop_reason_subsys subsys)
 162{
 163	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
 164		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
 165		 "invalid subsystem %d\n", subsys))
 166		return;
 167
 168	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], NULL);
 169
 170	synchronize_rcu();
 171}
 172EXPORT_SYMBOL_GPL(drop_reasons_unregister_subsys);
 173
 174/**
 175 *	skb_panic - private function for out-of-line support
 176 *	@skb:	buffer
 177 *	@sz:	size
 178 *	@addr:	address
 179 *	@msg:	skb_over_panic or skb_under_panic
 180 *
 181 *	Out-of-line support for skb_put() and skb_push().
 182 *	Called via the wrapper skb_over_panic() or skb_under_panic().
 183 *	Keep out of line to prevent kernel bloat.
 184 *	__builtin_return_address is not used because it is not always reliable.
 185 */
 186static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 187		      const char msg[])
 188{
 189	pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
 190		 msg, addr, skb->len, sz, skb->head, skb->data,
 191		 (unsigned long)skb->tail, (unsigned long)skb->end,
 192		 skb->dev ? skb->dev->name : "<NULL>");
 193	BUG();
 194}
 195
 196static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 197{
 198	skb_panic(skb, sz, addr, __func__);
 199}
 200
 201static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 202{
 203	skb_panic(skb, sz, addr, __func__);
 204}
 205
 206#define NAPI_SKB_CACHE_SIZE	64
 207#define NAPI_SKB_CACHE_BULK	16
 208#define NAPI_SKB_CACHE_HALF	(NAPI_SKB_CACHE_SIZE / 2)
 209
 210#if PAGE_SIZE == SZ_4K
 211
 212#define NAPI_HAS_SMALL_PAGE_FRAG	1
 213#define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	((nc).pfmemalloc)
 214
 215/* specialized page frag allocator using a single order 0 page
 216 * and slicing it into 1K sized fragment. Constrained to systems
 217 * with a very limited amount of 1K fragments fitting a single
 218 * page - to avoid excessive truesize underestimation
 219 */
 220
 221struct page_frag_1k {
 222	void *va;
 223	u16 offset;
 224	bool pfmemalloc;
 225};
 226
 227static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
 228{
 229	struct page *page;
 230	int offset;
 231
 232	offset = nc->offset - SZ_1K;
 233	if (likely(offset >= 0))
 234		goto use_frag;
 235
 236	page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
 237	if (!page)
 238		return NULL;
 239
 240	nc->va = page_address(page);
 241	nc->pfmemalloc = page_is_pfmemalloc(page);
 242	offset = PAGE_SIZE - SZ_1K;
 243	page_ref_add(page, offset / SZ_1K);
 244
 245use_frag:
 246	nc->offset = offset;
 247	return nc->va + offset;
 248}
 249#else
 250
 251/* the small page is actually unused in this build; add dummy helpers
 252 * to please the compiler and avoid later preprocessor's conditionals
 253 */
 254#define NAPI_HAS_SMALL_PAGE_FRAG	0
 255#define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	false
 256
 257struct page_frag_1k {
 258};
 259
 260static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
 261{
 262	return NULL;
 263}
 264
 265#endif
 266
 267struct napi_alloc_cache {
 268	struct page_frag_cache page;
 269	struct page_frag_1k page_small;
 270	unsigned int skb_count;
 271	void *skb_cache[NAPI_SKB_CACHE_SIZE];
 272};
 273
 274static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
 275static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
 276
 277/* Double check that napi_get_frags() allocates skbs with
 278 * skb->head being backed by slab, not a page fragment.
 279 * This is to make sure bug fixed in 3226b158e67c
 280 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
 281 * does not accidentally come back.
 282 */
 283void napi_get_frags_check(struct napi_struct *napi)
 284{
 285	struct sk_buff *skb;
 286
 287	local_bh_disable();
 288	skb = napi_get_frags(napi);
 289	WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
 290	napi_free_frags(napi);
 291	local_bh_enable();
 292}
 293
 294void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 295{
 296	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 297
 298	fragsz = SKB_DATA_ALIGN(fragsz);
 299
 300	return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
 301}
 302EXPORT_SYMBOL(__napi_alloc_frag_align);
 303
 304void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 305{
 306	void *data;
 307
 308	fragsz = SKB_DATA_ALIGN(fragsz);
 309	if (in_hardirq() || irqs_disabled()) {
 310		struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
 311
 312		data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
 313	} else {
 314		struct napi_alloc_cache *nc;
 315
 316		local_bh_disable();
 317		nc = this_cpu_ptr(&napi_alloc_cache);
 318		data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
 319		local_bh_enable();
 320	}
 321	return data;
 322}
 323EXPORT_SYMBOL(__netdev_alloc_frag_align);
 324
 325static struct sk_buff *napi_skb_cache_get(void)
 326{
 327	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 328	struct sk_buff *skb;
 329
 330	if (unlikely(!nc->skb_count)) {
 331		nc->skb_count = kmem_cache_alloc_bulk(skbuff_cache,
 332						      GFP_ATOMIC,
 333						      NAPI_SKB_CACHE_BULK,
 334						      nc->skb_cache);
 335		if (unlikely(!nc->skb_count))
 336			return NULL;
 337	}
 338
 339	skb = nc->skb_cache[--nc->skb_count];
 340	kasan_mempool_unpoison_object(skb, kmem_cache_size(skbuff_cache));
 341
 342	return skb;
 343}
 344
 345static inline void __finalize_skb_around(struct sk_buff *skb, void *data,
 346					 unsigned int size)
 347{
 348	struct skb_shared_info *shinfo;
 349
 350	size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 351
 352	/* Assumes caller memset cleared SKB */
 353	skb->truesize = SKB_TRUESIZE(size);
 354	refcount_set(&skb->users, 1);
 355	skb->head = data;
 356	skb->data = data;
 357	skb_reset_tail_pointer(skb);
 358	skb_set_end_offset(skb, size);
 359	skb->mac_header = (typeof(skb->mac_header))~0U;
 360	skb->transport_header = (typeof(skb->transport_header))~0U;
 361	skb->alloc_cpu = raw_smp_processor_id();
 362	/* make sure we initialize shinfo sequentially */
 363	shinfo = skb_shinfo(skb);
 364	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 365	atomic_set(&shinfo->dataref, 1);
 366
 367	skb_set_kcov_handle(skb, kcov_common_handle());
 368}
 369
 370static inline void *__slab_build_skb(struct sk_buff *skb, void *data,
 371				     unsigned int *size)
 372{
 373	void *resized;
 374
 375	/* Must find the allocation size (and grow it to match). */
 376	*size = ksize(data);
 377	/* krealloc() will immediately return "data" when
 378	 * "ksize(data)" is requested: it is the existing upper
 379	 * bounds. As a result, GFP_ATOMIC will be ignored. Note
 380	 * that this "new" pointer needs to be passed back to the
 381	 * caller for use so the __alloc_size hinting will be
 382	 * tracked correctly.
 383	 */
 384	resized = krealloc(data, *size, GFP_ATOMIC);
 385	WARN_ON_ONCE(resized != data);
 386	return resized;
 387}
 388
 389/* build_skb() variant which can operate on slab buffers.
 390 * Note that this should be used sparingly as slab buffers
 391 * cannot be combined efficiently by GRO!
 392 */
 393struct sk_buff *slab_build_skb(void *data)
 394{
 395	struct sk_buff *skb;
 396	unsigned int size;
 397
 398	skb = kmem_cache_alloc(skbuff_cache, GFP_ATOMIC);
 399	if (unlikely(!skb))
 400		return NULL;
 401
 402	memset(skb, 0, offsetof(struct sk_buff, tail));
 403	data = __slab_build_skb(skb, data, &size);
 404	__finalize_skb_around(skb, data, size);
 405
 406	return skb;
 407}
 408EXPORT_SYMBOL(slab_build_skb);
 409
 410/* Caller must provide SKB that is memset cleared */
 411static void __build_skb_around(struct sk_buff *skb, void *data,
 412			       unsigned int frag_size)
 413{
 414	unsigned int size = frag_size;
 415
 416	/* frag_size == 0 is considered deprecated now. Callers
 417	 * using slab buffer should use slab_build_skb() instead.
 418	 */
 419	if (WARN_ONCE(size == 0, "Use slab_build_skb() instead"))
 420		data = __slab_build_skb(skb, data, &size);
 421
 422	__finalize_skb_around(skb, data, size);
 423}
 424
 425/**
 426 * __build_skb - build a network buffer
 427 * @data: data buffer provided by caller
 428 * @frag_size: size of data (must not be 0)
 429 *
 430 * Allocate a new &sk_buff. Caller provides space holding head and
 431 * skb_shared_info. @data must have been allocated from the page
 432 * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc()
 433 * allocation is deprecated, and callers should use slab_build_skb()
 434 * instead.)
 435 * The return is the new skb buffer.
 436 * On a failure the return is %NULL, and @data is not freed.
 437 * Notes :
 438 *  Before IO, driver allocates only data buffer where NIC put incoming frame
 439 *  Driver should add room at head (NET_SKB_PAD) and
 440 *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
 441 *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
 442 *  before giving packet to stack.
 443 *  RX rings only contains data buffers, not full skbs.
 444 */
 445struct sk_buff *__build_skb(void *data, unsigned int frag_size)
 446{
 447	struct sk_buff *skb;
 448
 449	skb = kmem_cache_alloc(skbuff_cache, GFP_ATOMIC);
 450	if (unlikely(!skb))
 451		return NULL;
 452
 453	memset(skb, 0, offsetof(struct sk_buff, tail));
 454	__build_skb_around(skb, data, frag_size);
 455
 456	return skb;
 457}
 458
 459/* build_skb() is wrapper over __build_skb(), that specifically
 460 * takes care of skb->head and skb->pfmemalloc
 461 */
 462struct sk_buff *build_skb(void *data, unsigned int frag_size)
 463{
 464	struct sk_buff *skb = __build_skb(data, frag_size);
 465
 466	if (likely(skb && frag_size)) {
 467		skb->head_frag = 1;
 468		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 469	}
 470	return skb;
 471}
 472EXPORT_SYMBOL(build_skb);
 473
 474/**
 475 * build_skb_around - build a network buffer around provided skb
 476 * @skb: sk_buff provide by caller, must be memset cleared
 477 * @data: data buffer provided by caller
 478 * @frag_size: size of data
 479 */
 480struct sk_buff *build_skb_around(struct sk_buff *skb,
 481				 void *data, unsigned int frag_size)
 482{
 483	if (unlikely(!skb))
 484		return NULL;
 485
 486	__build_skb_around(skb, data, frag_size);
 487
 488	if (frag_size) {
 489		skb->head_frag = 1;
 490		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 491	}
 492	return skb;
 493}
 494EXPORT_SYMBOL(build_skb_around);
 495
 496/**
 497 * __napi_build_skb - build a network buffer
 498 * @data: data buffer provided by caller
 499 * @frag_size: size of data
 500 *
 501 * Version of __build_skb() that uses NAPI percpu caches to obtain
 502 * skbuff_head instead of inplace allocation.
 503 *
 504 * Returns a new &sk_buff on success, %NULL on allocation failure.
 505 */
 506static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
 507{
 508	struct sk_buff *skb;
 509
 510	skb = napi_skb_cache_get();
 511	if (unlikely(!skb))
 512		return NULL;
 513
 514	memset(skb, 0, offsetof(struct sk_buff, tail));
 515	__build_skb_around(skb, data, frag_size);
 516
 517	return skb;
 518}
 519
 520/**
 521 * napi_build_skb - build a network buffer
 522 * @data: data buffer provided by caller
 523 * @frag_size: size of data
 524 *
 525 * Version of __napi_build_skb() that takes care of skb->head_frag
 526 * and skb->pfmemalloc when the data is a page or page fragment.
 527 *
 528 * Returns a new &sk_buff on success, %NULL on allocation failure.
 529 */
 530struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
 531{
 532	struct sk_buff *skb = __napi_build_skb(data, frag_size);
 533
 534	if (likely(skb) && frag_size) {
 535		skb->head_frag = 1;
 536		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 537	}
 538
 539	return skb;
 540}
 541EXPORT_SYMBOL(napi_build_skb);
 542
 543/*
 544 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
 545 * the caller if emergency pfmemalloc reserves are being used. If it is and
 546 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
 547 * may be used. Otherwise, the packet data may be discarded until enough
 548 * memory is free
 549 */
 550static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
 551			     bool *pfmemalloc)
 552{
 553	bool ret_pfmemalloc = false;
 554	size_t obj_size;
 555	void *obj;
 556
 557	obj_size = SKB_HEAD_ALIGN(*size);
 558	if (obj_size <= SKB_SMALL_HEAD_CACHE_SIZE &&
 559	    !(flags & KMALLOC_NOT_NORMAL_BITS)) {
 560		obj = kmem_cache_alloc_node(skb_small_head_cache,
 561				flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 562				node);
 563		*size = SKB_SMALL_HEAD_CACHE_SIZE;
 564		if (obj || !(gfp_pfmemalloc_allowed(flags)))
 565			goto out;
 566		/* Try again but now we are using pfmemalloc reserves */
 567		ret_pfmemalloc = true;
 568		obj = kmem_cache_alloc_node(skb_small_head_cache, flags, node);
 569		goto out;
 570	}
 571
 572	obj_size = kmalloc_size_roundup(obj_size);
 573	/* The following cast might truncate high-order bits of obj_size, this
 574	 * is harmless because kmalloc(obj_size >= 2^32) will fail anyway.
 575	 */
 576	*size = (unsigned int)obj_size;
 577
 578	/*
 579	 * Try a regular allocation, when that fails and we're not entitled
 580	 * to the reserves, fail.
 581	 */
 582	obj = kmalloc_node_track_caller(obj_size,
 583					flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 584					node);
 585	if (obj || !(gfp_pfmemalloc_allowed(flags)))
 586		goto out;
 587
 588	/* Try again but now we are using pfmemalloc reserves */
 589	ret_pfmemalloc = true;
 590	obj = kmalloc_node_track_caller(obj_size, flags, node);
 591
 592out:
 593	if (pfmemalloc)
 594		*pfmemalloc = ret_pfmemalloc;
 595
 596	return obj;
 597}
 598
 599/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 600 *	'private' fields and also do memory statistics to find all the
 601 *	[BEEP] leaks.
 602 *
 603 */
 604
 605/**
 606 *	__alloc_skb	-	allocate a network buffer
 607 *	@size: size to allocate
 608 *	@gfp_mask: allocation mask
 609 *	@flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
 610 *		instead of head cache and allocate a cloned (child) skb.
 611 *		If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
 612 *		allocations in case the data is required for writeback
 613 *	@node: numa node to allocate memory on
 614 *
 615 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 616 *	tail room of at least size bytes. The object has a reference count
 617 *	of one. The return is the buffer. On a failure the return is %NULL.
 618 *
 619 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 620 *	%GFP_ATOMIC.
 621 */
 622struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 623			    int flags, int node)
 624{
 625	struct kmem_cache *cache;
 626	struct sk_buff *skb;
 627	bool pfmemalloc;
 628	u8 *data;
 629
 630	cache = (flags & SKB_ALLOC_FCLONE)
 631		? skbuff_fclone_cache : skbuff_cache;
 632
 633	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 634		gfp_mask |= __GFP_MEMALLOC;
 635
 636	/* Get the HEAD */
 637	if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
 638	    likely(node == NUMA_NO_NODE || node == numa_mem_id()))
 639		skb = napi_skb_cache_get();
 640	else
 641		skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
 642	if (unlikely(!skb))
 643		return NULL;
 644	prefetchw(skb);
 645
 646	/* We do our best to align skb_shared_info on a separate cache
 647	 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
 648	 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
 649	 * Both skb->head and skb_shared_info are cache line aligned.
 650	 */
 651	data = kmalloc_reserve(&size, gfp_mask, node, &pfmemalloc);
 652	if (unlikely(!data))
 653		goto nodata;
 654	/* kmalloc_size_roundup() might give us more room than requested.
 655	 * Put skb_shared_info exactly at the end of allocated zone,
 656	 * to allow max possible filling before reallocation.
 657	 */
 658	prefetchw(data + SKB_WITH_OVERHEAD(size));
 659
 660	/*
 661	 * Only clear those fields we need to clear, not those that we will
 662	 * actually initialise below. Hence, don't put any more fields after
 663	 * the tail pointer in struct sk_buff!
 664	 */
 665	memset(skb, 0, offsetof(struct sk_buff, tail));
 666	__build_skb_around(skb, data, size);
 667	skb->pfmemalloc = pfmemalloc;
 668
 669	if (flags & SKB_ALLOC_FCLONE) {
 670		struct sk_buff_fclones *fclones;
 671
 672		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 673
 674		skb->fclone = SKB_FCLONE_ORIG;
 675		refcount_set(&fclones->fclone_ref, 1);
 676	}
 677
 678	return skb;
 679
 680nodata:
 681	kmem_cache_free(cache, skb);
 682	return NULL;
 683}
 684EXPORT_SYMBOL(__alloc_skb);
 685
 686/**
 687 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 688 *	@dev: network device to receive on
 689 *	@len: length to allocate
 690 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 691 *
 692 *	Allocate a new &sk_buff and assign it a usage count of one. The
 693 *	buffer has NET_SKB_PAD headroom built in. Users should allocate
 694 *	the headroom they think they need without accounting for the
 695 *	built in space. The built in space is used for optimisations.
 696 *
 697 *	%NULL is returned if there is no free memory.
 698 */
 699struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
 700				   gfp_t gfp_mask)
 701{
 702	struct page_frag_cache *nc;
 703	struct sk_buff *skb;
 704	bool pfmemalloc;
 705	void *data;
 706
 707	len += NET_SKB_PAD;
 708
 709	/* If requested length is either too small or too big,
 710	 * we use kmalloc() for skb->head allocation.
 711	 */
 712	if (len <= SKB_WITH_OVERHEAD(1024) ||
 713	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
 714	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 715		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 716		if (!skb)
 717			goto skb_fail;
 718		goto skb_success;
 719	}
 720
 721	len = SKB_HEAD_ALIGN(len);
 722
 723	if (sk_memalloc_socks())
 724		gfp_mask |= __GFP_MEMALLOC;
 725
 726	if (in_hardirq() || irqs_disabled()) {
 727		nc = this_cpu_ptr(&netdev_alloc_cache);
 728		data = page_frag_alloc(nc, len, gfp_mask);
 729		pfmemalloc = nc->pfmemalloc;
 730	} else {
 731		local_bh_disable();
 732		nc = this_cpu_ptr(&napi_alloc_cache.page);
 733		data = page_frag_alloc(nc, len, gfp_mask);
 734		pfmemalloc = nc->pfmemalloc;
 735		local_bh_enable();
 736	}
 737
 738	if (unlikely(!data))
 739		return NULL;
 740
 741	skb = __build_skb(data, len);
 742	if (unlikely(!skb)) {
 743		skb_free_frag(data);
 744		return NULL;
 745	}
 746
 747	if (pfmemalloc)
 748		skb->pfmemalloc = 1;
 749	skb->head_frag = 1;
 750
 751skb_success:
 752	skb_reserve(skb, NET_SKB_PAD);
 753	skb->dev = dev;
 754
 755skb_fail:
 756	return skb;
 757}
 758EXPORT_SYMBOL(__netdev_alloc_skb);
 759
 760/**
 761 *	__napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
 762 *	@napi: napi instance this buffer was allocated for
 763 *	@len: length to allocate
 764 *	@gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
 765 *
 766 *	Allocate a new sk_buff for use in NAPI receive.  This buffer will
 767 *	attempt to allocate the head from a special reserved region used
 768 *	only for NAPI Rx allocation.  By doing this we can save several
 769 *	CPU cycles by avoiding having to disable and re-enable IRQs.
 770 *
 771 *	%NULL is returned if there is no free memory.
 772 */
 773struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
 774				 gfp_t gfp_mask)
 775{
 776	struct napi_alloc_cache *nc;
 777	struct sk_buff *skb;
 778	bool pfmemalloc;
 779	void *data;
 780
 781	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
 782	len += NET_SKB_PAD + NET_IP_ALIGN;
 783
 784	/* If requested length is either too small or too big,
 785	 * we use kmalloc() for skb->head allocation.
 786	 * When the small frag allocator is available, prefer it over kmalloc
 787	 * for small fragments
 788	 */
 789	if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
 790	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
 791	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 792		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
 793				  NUMA_NO_NODE);
 794		if (!skb)
 795			goto skb_fail;
 796		goto skb_success;
 797	}
 798
 799	nc = this_cpu_ptr(&napi_alloc_cache);
 800
 801	if (sk_memalloc_socks())
 802		gfp_mask |= __GFP_MEMALLOC;
 803
 804	if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
 805		/* we are artificially inflating the allocation size, but
 806		 * that is not as bad as it may look like, as:
 807		 * - 'len' less than GRO_MAX_HEAD makes little sense
 808		 * - On most systems, larger 'len' values lead to fragment
 809		 *   size above 512 bytes
 810		 * - kmalloc would use the kmalloc-1k slab for such values
 811		 * - Builds with smaller GRO_MAX_HEAD will very likely do
 812		 *   little networking, as that implies no WiFi and no
 813		 *   tunnels support, and 32 bits arches.
 814		 */
 815		len = SZ_1K;
 816
 817		data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
 818		pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
 819	} else {
 820		len = SKB_HEAD_ALIGN(len);
 821
 822		data = page_frag_alloc(&nc->page, len, gfp_mask);
 823		pfmemalloc = nc->page.pfmemalloc;
 824	}
 825
 826	if (unlikely(!data))
 827		return NULL;
 828
 829	skb = __napi_build_skb(data, len);
 830	if (unlikely(!skb)) {
 831		skb_free_frag(data);
 832		return NULL;
 833	}
 834
 835	if (pfmemalloc)
 836		skb->pfmemalloc = 1;
 837	skb->head_frag = 1;
 838
 839skb_success:
 840	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
 841	skb->dev = napi->dev;
 842
 843skb_fail:
 844	return skb;
 845}
 846EXPORT_SYMBOL(__napi_alloc_skb);
 847
 848void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 849		     int size, unsigned int truesize)
 850{
 851	DEBUG_NET_WARN_ON_ONCE(size > truesize);
 852
 853	skb_fill_page_desc(skb, i, page, off, size);
 854	skb->len += size;
 855	skb->data_len += size;
 856	skb->truesize += truesize;
 857}
 858EXPORT_SYMBOL(skb_add_rx_frag);
 859
 860void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
 861			  unsigned int truesize)
 862{
 863	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 864
 865	DEBUG_NET_WARN_ON_ONCE(size > truesize);
 866
 867	skb_frag_size_add(frag, size);
 868	skb->len += size;
 869	skb->data_len += size;
 870	skb->truesize += truesize;
 871}
 872EXPORT_SYMBOL(skb_coalesce_rx_frag);
 873
 874static void skb_drop_list(struct sk_buff **listp)
 875{
 876	kfree_skb_list(*listp);
 877	*listp = NULL;
 878}
 879
 880static inline void skb_drop_fraglist(struct sk_buff *skb)
 881{
 882	skb_drop_list(&skb_shinfo(skb)->frag_list);
 883}
 884
 885static void skb_clone_fraglist(struct sk_buff *skb)
 886{
 887	struct sk_buff *list;
 888
 889	skb_walk_frags(skb, list)
 890		skb_get(list);
 891}
 892
 893static bool is_pp_page(struct page *page)
 894{
 895	return (page->pp_magic & ~0x3UL) == PP_SIGNATURE;
 896}
 897
 898#if IS_ENABLED(CONFIG_PAGE_POOL)
 899bool napi_pp_put_page(struct page *page, bool napi_safe)
 900{
 901	bool allow_direct = false;
 902	struct page_pool *pp;
 903
 904	page = compound_head(page);
 905
 906	/* page->pp_magic is OR'ed with PP_SIGNATURE after the allocation
 907	 * in order to preserve any existing bits, such as bit 0 for the
 908	 * head page of compound page and bit 1 for pfmemalloc page, so
 909	 * mask those bits for freeing side when doing below checking,
 910	 * and page_is_pfmemalloc() is checked in __page_pool_put_page()
 911	 * to avoid recycling the pfmemalloc page.
 912	 */
 913	if (unlikely(!is_pp_page(page)))
 914		return false;
 915
 916	pp = page->pp;
 917
 918	/* Allow direct recycle if we have reasons to believe that we are
 919	 * in the same context as the consumer would run, so there's
 920	 * no possible race.
 921	 * __page_pool_put_page() makes sure we're not in hardirq context
 922	 * and interrupts are enabled prior to accessing the cache.
 923	 */
 924	if (napi_safe || in_softirq()) {
 925		const struct napi_struct *napi = READ_ONCE(pp->p.napi);
 926
 927		allow_direct = napi &&
 928			READ_ONCE(napi->list_owner) == smp_processor_id();
 929	}
 930
 931	/* Driver set this to memory recycling info. Reset it on recycle.
 932	 * This will *not* work for NIC using a split-page memory model.
 933	 * The page will be returned to the pool here regardless of the
 934	 * 'flipped' fragment being in use or not.
 935	 */
 936	page_pool_put_full_page(pp, page, allow_direct);
 937
 938	return true;
 939}
 940EXPORT_SYMBOL(napi_pp_put_page);
 941#endif
 942
 943static bool skb_pp_recycle(struct sk_buff *skb, void *data, bool napi_safe)
 944{
 945	if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
 946		return false;
 947	return napi_pp_put_page(virt_to_page(data), napi_safe);
 948}
 949
 950/**
 951 * skb_pp_frag_ref() - Increase fragment references of a page pool aware skb
 952 * @skb:	page pool aware skb
 953 *
 954 * Increase the fragment reference count (pp_ref_count) of a skb. This is
 955 * intended to gain fragment references only for page pool aware skbs,
 956 * i.e. when skb->pp_recycle is true, and not for fragments in a
 957 * non-pp-recycling skb. It has a fallback to increase references on normal
 958 * pages, as page pool aware skbs may also have normal page fragments.
 959 */
 960static int skb_pp_frag_ref(struct sk_buff *skb)
 961{
 962	struct skb_shared_info *shinfo;
 963	struct page *head_page;
 964	int i;
 965
 966	if (!skb->pp_recycle)
 967		return -EINVAL;
 968
 969	shinfo = skb_shinfo(skb);
 970
 971	for (i = 0; i < shinfo->nr_frags; i++) {
 972		head_page = compound_head(skb_frag_page(&shinfo->frags[i]));
 973		if (likely(is_pp_page(head_page)))
 974			page_pool_ref_page(head_page);
 975		else
 976			page_ref_inc(head_page);
 977	}
 978	return 0;
 979}
 980
 981static void skb_kfree_head(void *head, unsigned int end_offset)
 982{
 983	if (end_offset == SKB_SMALL_HEAD_HEADROOM)
 984		kmem_cache_free(skb_small_head_cache, head);
 985	else
 986		kfree(head);
 987}
 988
 989static void skb_free_head(struct sk_buff *skb, bool napi_safe)
 990{
 991	unsigned char *head = skb->head;
 992
 993	if (skb->head_frag) {
 994		if (skb_pp_recycle(skb, head, napi_safe))
 995			return;
 996		skb_free_frag(head);
 997	} else {
 998		skb_kfree_head(head, skb_end_offset(skb));
 999	}
1000}
1001
1002static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason,
1003			     bool napi_safe)
1004{
1005	struct skb_shared_info *shinfo = skb_shinfo(skb);
1006	int i;
1007
1008	if (skb->cloned &&
1009	    atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
1010			      &shinfo->dataref))
1011		goto exit;
1012
1013	if (skb_zcopy(skb)) {
1014		bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
1015
1016		skb_zcopy_clear(skb, true);
1017		if (skip_unref)
1018			goto free_head;
1019	}
1020
1021	for (i = 0; i < shinfo->nr_frags; i++)
1022		napi_frag_unref(&shinfo->frags[i], skb->pp_recycle, napi_safe);
1023
1024free_head:
1025	if (shinfo->frag_list)
1026		kfree_skb_list_reason(shinfo->frag_list, reason);
1027
1028	skb_free_head(skb, napi_safe);
1029exit:
1030	/* When we clone an SKB we copy the reycling bit. The pp_recycle
1031	 * bit is only set on the head though, so in order to avoid races
1032	 * while trying to recycle fragments on __skb_frag_unref() we need
1033	 * to make one SKB responsible for triggering the recycle path.
1034	 * So disable the recycling bit if an SKB is cloned and we have
1035	 * additional references to the fragmented part of the SKB.
1036	 * Eventually the last SKB will have the recycling bit set and it's
1037	 * dataref set to 0, which will trigger the recycling
1038	 */
1039	skb->pp_recycle = 0;
1040}
1041
1042/*
1043 *	Free an skbuff by memory without cleaning the state.
1044 */
1045static void kfree_skbmem(struct sk_buff *skb)
1046{
1047	struct sk_buff_fclones *fclones;
1048
1049	switch (skb->fclone) {
1050	case SKB_FCLONE_UNAVAILABLE:
1051		kmem_cache_free(skbuff_cache, skb);
1052		return;
1053
1054	case SKB_FCLONE_ORIG:
1055		fclones = container_of(skb, struct sk_buff_fclones, skb1);
1056
1057		/* We usually free the clone (TX completion) before original skb
1058		 * This test would have no chance to be true for the clone,
1059		 * while here, branch prediction will be good.
1060		 */
1061		if (refcount_read(&fclones->fclone_ref) == 1)
1062			goto fastpath;
1063		break;
1064
1065	default: /* SKB_FCLONE_CLONE */
1066		fclones = container_of(skb, struct sk_buff_fclones, skb2);
1067		break;
1068	}
1069	if (!refcount_dec_and_test(&fclones->fclone_ref))
1070		return;
1071fastpath:
1072	kmem_cache_free(skbuff_fclone_cache, fclones);
1073}
1074
1075void skb_release_head_state(struct sk_buff *skb)
1076{
1077	skb_dst_drop(skb);
1078	if (skb->destructor) {
1079		DEBUG_NET_WARN_ON_ONCE(in_hardirq());
1080		skb->destructor(skb);
1081	}
1082#if IS_ENABLED(CONFIG_NF_CONNTRACK)
1083	nf_conntrack_put(skb_nfct(skb));
1084#endif
1085	skb_ext_put(skb);
1086}
1087
1088/* Free everything but the sk_buff shell. */
1089static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason,
1090			    bool napi_safe)
1091{
1092	skb_release_head_state(skb);
1093	if (likely(skb->head))
1094		skb_release_data(skb, reason, napi_safe);
1095}
1096
1097/**
1098 *	__kfree_skb - private function
1099 *	@skb: buffer
1100 *
1101 *	Free an sk_buff. Release anything attached to the buffer.
1102 *	Clean the state. This is an internal helper function. Users should
1103 *	always call kfree_skb
1104 */
1105
1106void __kfree_skb(struct sk_buff *skb)
1107{
1108	skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED, false);
1109	kfree_skbmem(skb);
1110}
1111EXPORT_SYMBOL(__kfree_skb);
1112
1113static __always_inline
1114bool __kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
1115{
1116	if (unlikely(!skb_unref(skb)))
1117		return false;
1118
1119	DEBUG_NET_WARN_ON_ONCE(reason == SKB_NOT_DROPPED_YET ||
1120			       u32_get_bits(reason,
1121					    SKB_DROP_REASON_SUBSYS_MASK) >=
1122				SKB_DROP_REASON_SUBSYS_NUM);
1123
1124	if (reason == SKB_CONSUMED)
1125		trace_consume_skb(skb, __builtin_return_address(0));
1126	else
1127		trace_kfree_skb(skb, __builtin_return_address(0), reason);
1128	return true;
1129}
1130
1131/**
1132 *	kfree_skb_reason - free an sk_buff with special reason
1133 *	@skb: buffer to free
1134 *	@reason: reason why this skb is dropped
1135 *
1136 *	Drop a reference to the buffer and free it if the usage count has
1137 *	hit zero. Meanwhile, pass the drop reason to 'kfree_skb'
1138 *	tracepoint.
1139 */
1140void __fix_address
1141kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
1142{
1143	if (__kfree_skb_reason(skb, reason))
1144		__kfree_skb(skb);
1145}
1146EXPORT_SYMBOL(kfree_skb_reason);
1147
1148#define KFREE_SKB_BULK_SIZE	16
1149
1150struct skb_free_array {
1151	unsigned int skb_count;
1152	void *skb_array[KFREE_SKB_BULK_SIZE];
1153};
1154
1155static void kfree_skb_add_bulk(struct sk_buff *skb,
1156			       struct skb_free_array *sa,
1157			       enum skb_drop_reason reason)
1158{
1159	/* if SKB is a clone, don't handle this case */
1160	if (unlikely(skb->fclone != SKB_FCLONE_UNAVAILABLE)) {
1161		__kfree_skb(skb);
1162		return;
1163	}
1164
1165	skb_release_all(skb, reason, false);
1166	sa->skb_array[sa->skb_count++] = skb;
1167
1168	if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
1169		kmem_cache_free_bulk(skbuff_cache, KFREE_SKB_BULK_SIZE,
1170				     sa->skb_array);
1171		sa->skb_count = 0;
1172	}
1173}
1174
1175void __fix_address
1176kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
1177{
1178	struct skb_free_array sa;
1179
1180	sa.skb_count = 0;
1181
1182	while (segs) {
1183		struct sk_buff *next = segs->next;
1184
1185		if (__kfree_skb_reason(segs, reason)) {
1186			skb_poison_list(segs);
1187			kfree_skb_add_bulk(segs, &sa, reason);
1188		}
1189
1190		segs = next;
1191	}
1192
1193	if (sa.skb_count)
1194		kmem_cache_free_bulk(skbuff_cache, sa.skb_count, sa.skb_array);
1195}
1196EXPORT_SYMBOL(kfree_skb_list_reason);
1197
1198/* Dump skb information and contents.
1199 *
1200 * Must only be called from net_ratelimit()-ed paths.
1201 *
1202 * Dumps whole packets if full_pkt, only headers otherwise.
1203 */
1204void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
1205{
1206	struct skb_shared_info *sh = skb_shinfo(skb);
1207	struct net_device *dev = skb->dev;
1208	struct sock *sk = skb->sk;
1209	struct sk_buff *list_skb;
1210	bool has_mac, has_trans;
1211	int headroom, tailroom;
1212	int i, len, seg_len;
1213
1214	if (full_pkt)
1215		len = skb->len;
1216	else
1217		len = min_t(int, skb->len, MAX_HEADER + 128);
1218
1219	headroom = skb_headroom(skb);
1220	tailroom = skb_tailroom(skb);
1221
1222	has_mac = skb_mac_header_was_set(skb);
1223	has_trans = skb_transport_header_was_set(skb);
1224
1225	printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1226	       "mac=(%d,%d) net=(%d,%d) trans=%d\n"
1227	       "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1228	       "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1229	       "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
1230	       level, skb->len, headroom, skb_headlen(skb), tailroom,
1231	       has_mac ? skb->mac_header : -1,
1232	       has_mac ? skb_mac_header_len(skb) : -1,
1233	       skb->network_header,
1234	       has_trans ? skb_network_header_len(skb) : -1,
1235	       has_trans ? skb->transport_header : -1,
1236	       sh->tx_flags, sh->nr_frags,
1237	       sh->gso_size, sh->gso_type, sh->gso_segs,
1238	       skb->csum, skb->ip_summed, skb->csum_complete_sw,
1239	       skb->csum_valid, skb->csum_level,
1240	       skb->hash, skb->sw_hash, skb->l4_hash,
1241	       ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
1242
1243	if (dev)
1244		printk("%sdev name=%s feat=%pNF\n",
1245		       level, dev->name, &dev->features);
1246	if (sk)
1247		printk("%ssk family=%hu type=%u proto=%u\n",
1248		       level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1249
1250	if (full_pkt && headroom)
1251		print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1252			       16, 1, skb->head, headroom, false);
1253
1254	seg_len = min_t(int, skb_headlen(skb), len);
1255	if (seg_len)
1256		print_hex_dump(level, "skb linear:   ", DUMP_PREFIX_OFFSET,
1257			       16, 1, skb->data, seg_len, false);
1258	len -= seg_len;
1259
1260	if (full_pkt && tailroom)
1261		print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1262			       16, 1, skb_tail_pointer(skb), tailroom, false);
1263
1264	for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1265		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1266		u32 p_off, p_len, copied;
1267		struct page *p;
1268		u8 *vaddr;
1269
1270		skb_frag_foreach_page(frag, skb_frag_off(frag),
1271				      skb_frag_size(frag), p, p_off, p_len,
1272				      copied) {
1273			seg_len = min_t(int, p_len, len);
1274			vaddr = kmap_atomic(p);
1275			print_hex_dump(level, "skb frag:     ",
1276				       DUMP_PREFIX_OFFSET,
1277				       16, 1, vaddr + p_off, seg_len, false);
1278			kunmap_atomic(vaddr);
1279			len -= seg_len;
1280			if (!len)
1281				break;
1282		}
1283	}
1284
1285	if (full_pkt && skb_has_frag_list(skb)) {
1286		printk("skb fraglist:\n");
1287		skb_walk_frags(skb, list_skb)
1288			skb_dump(level, list_skb, true);
1289	}
1290}
1291EXPORT_SYMBOL(skb_dump);
1292
1293/**
1294 *	skb_tx_error - report an sk_buff xmit error
1295 *	@skb: buffer that triggered an error
1296 *
1297 *	Report xmit error if a device callback is tracking this skb.
1298 *	skb must be freed afterwards.
1299 */
1300void skb_tx_error(struct sk_buff *skb)
1301{
1302	if (skb) {
1303		skb_zcopy_downgrade_managed(skb);
1304		skb_zcopy_clear(skb, true);
1305	}
1306}
1307EXPORT_SYMBOL(skb_tx_error);
1308
1309#ifdef CONFIG_TRACEPOINTS
1310/**
1311 *	consume_skb - free an skbuff
1312 *	@skb: buffer to free
1313 *
1314 *	Drop a ref to the buffer and free it if the usage count has hit zero
1315 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
1316 *	is being dropped after a failure and notes that
1317 */
1318void consume_skb(struct sk_buff *skb)
1319{
1320	if (!skb_unref(skb))
1321		return;
1322
1323	trace_consume_skb(skb, __builtin_return_address(0));
1324	__kfree_skb(skb);
1325}
1326EXPORT_SYMBOL(consume_skb);
1327#endif
1328
1329/**
1330 *	__consume_stateless_skb - free an skbuff, assuming it is stateless
1331 *	@skb: buffer to free
1332 *
1333 *	Alike consume_skb(), but this variant assumes that this is the last
1334 *	skb reference and all the head states have been already dropped
1335 */
1336void __consume_stateless_skb(struct sk_buff *skb)
1337{
1338	trace_consume_skb(skb, __builtin_return_address(0));
1339	skb_release_data(skb, SKB_CONSUMED, false);
1340	kfree_skbmem(skb);
1341}
1342
1343static void napi_skb_cache_put(struct sk_buff *skb)
1344{
1345	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1346	u32 i;
1347
1348	if (!kasan_mempool_poison_object(skb))
1349		return;
1350
1351	nc->skb_cache[nc->skb_count++] = skb;
1352
1353	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1354		for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1355			kasan_mempool_unpoison_object(nc->skb_cache[i],
1356						kmem_cache_size(skbuff_cache));
1357
1358		kmem_cache_free_bulk(skbuff_cache, NAPI_SKB_CACHE_HALF,
1359				     nc->skb_cache + NAPI_SKB_CACHE_HALF);
1360		nc->skb_count = NAPI_SKB_CACHE_HALF;
1361	}
1362}
1363
1364void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
1365{
1366	skb_release_all(skb, reason, true);
1367	napi_skb_cache_put(skb);
1368}
1369
1370void napi_skb_free_stolen_head(struct sk_buff *skb)
1371{
1372	if (unlikely(skb->slow_gro)) {
1373		nf_reset_ct(skb);
1374		skb_dst_drop(skb);
1375		skb_ext_put(skb);
1376		skb_orphan(skb);
1377		skb->slow_gro = 0;
1378	}
1379	napi_skb_cache_put(skb);
1380}
1381
1382void napi_consume_skb(struct sk_buff *skb, int budget)
1383{
1384	/* Zero budget indicate non-NAPI context called us, like netpoll */
1385	if (unlikely(!budget)) {
1386		dev_consume_skb_any(skb);
1387		return;
1388	}
1389
1390	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1391
1392	if (!skb_unref(skb))
1393		return;
1394
1395	/* if reaching here SKB is ready to free */
1396	trace_consume_skb(skb, __builtin_return_address(0));
1397
1398	/* if SKB is a clone, don't handle this case */
1399	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1400		__kfree_skb(skb);
1401		return;
1402	}
1403
1404	skb_release_all(skb, SKB_CONSUMED, !!budget);
1405	napi_skb_cache_put(skb);
1406}
1407EXPORT_SYMBOL(napi_consume_skb);
1408
1409/* Make sure a field is contained by headers group */
1410#define CHECK_SKB_FIELD(field) \
1411	BUILD_BUG_ON(offsetof(struct sk_buff, field) !=		\
1412		     offsetof(struct sk_buff, headers.field));	\
1413
1414static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1415{
1416	new->tstamp		= old->tstamp;
1417	/* We do not copy old->sk */
1418	new->dev		= old->dev;
1419	memcpy(new->cb, old->cb, sizeof(old->cb));
1420	skb_dst_copy(new, old);
1421	__skb_ext_copy(new, old);
1422	__nf_copy(new, old, false);
1423
1424	/* Note : this field could be in the headers group.
1425	 * It is not yet because we do not want to have a 16 bit hole
1426	 */
1427	new->queue_mapping = old->queue_mapping;
1428
1429	memcpy(&new->headers, &old->headers, sizeof(new->headers));
1430	CHECK_SKB_FIELD(protocol);
1431	CHECK_SKB_FIELD(csum);
1432	CHECK_SKB_FIELD(hash);
1433	CHECK_SKB_FIELD(priority);
1434	CHECK_SKB_FIELD(skb_iif);
1435	CHECK_SKB_FIELD(vlan_proto);
1436	CHECK_SKB_FIELD(vlan_tci);
1437	CHECK_SKB_FIELD(transport_header);
1438	CHECK_SKB_FIELD(network_header);
1439	CHECK_SKB_FIELD(mac_header);
1440	CHECK_SKB_FIELD(inner_protocol);
1441	CHECK_SKB_FIELD(inner_transport_header);
1442	CHECK_SKB_FIELD(inner_network_header);
1443	CHECK_SKB_FIELD(inner_mac_header);
1444	CHECK_SKB_FIELD(mark);
1445#ifdef CONFIG_NETWORK_SECMARK
1446	CHECK_SKB_FIELD(secmark);
1447#endif
1448#ifdef CONFIG_NET_RX_BUSY_POLL
1449	CHECK_SKB_FIELD(napi_id);
1450#endif
1451	CHECK_SKB_FIELD(alloc_cpu);
1452#ifdef CONFIG_XPS
1453	CHECK_SKB_FIELD(sender_cpu);
1454#endif
1455#ifdef CONFIG_NET_SCHED
1456	CHECK_SKB_FIELD(tc_index);
1457#endif
1458
1459}
1460
1461/*
1462 * You should not add any new code to this function.  Add it to
1463 * __copy_skb_header above instead.
1464 */
1465static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1466{
1467#define C(x) n->x = skb->x
1468
1469	n->next = n->prev = NULL;
1470	n->sk = NULL;
1471	__copy_skb_header(n, skb);
1472
1473	C(len);
1474	C(data_len);
1475	C(mac_len);
1476	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1477	n->cloned = 1;
1478	n->nohdr = 0;
1479	n->peeked = 0;
1480	C(pfmemalloc);
1481	C(pp_recycle);
1482	n->destructor = NULL;
1483	C(tail);
1484	C(end);
1485	C(head);
1486	C(head_frag);
1487	C(data);
1488	C(truesize);
1489	refcount_set(&n->users, 1);
1490
1491	atomic_inc(&(skb_shinfo(skb)->dataref));
1492	skb->cloned = 1;
1493
1494	return n;
1495#undef C
1496}
1497
1498/**
1499 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1500 * @first: first sk_buff of the msg
1501 */
1502struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1503{
1504	struct sk_buff *n;
1505
1506	n = alloc_skb(0, GFP_ATOMIC);
1507	if (!n)
1508		return NULL;
1509
1510	n->len = first->len;
1511	n->data_len = first->len;
1512	n->truesize = first->truesize;
1513
1514	skb_shinfo(n)->frag_list = first;
1515
1516	__copy_skb_header(n, first);
1517	n->destructor = NULL;
1518
1519	return n;
1520}
1521EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1522
1523/**
1524 *	skb_morph	-	morph one skb into another
1525 *	@dst: the skb to receive the contents
1526 *	@src: the skb to supply the contents
1527 *
1528 *	This is identical to skb_clone except that the target skb is
1529 *	supplied by the user.
1530 *
1531 *	The target skb is returned upon exit.
1532 */
1533struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1534{
1535	skb_release_all(dst, SKB_CONSUMED, false);
1536	return __skb_clone(dst, src);
1537}
1538EXPORT_SYMBOL_GPL(skb_morph);
1539
1540int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1541{
1542	unsigned long max_pg, num_pg, new_pg, old_pg, rlim;
1543	struct user_struct *user;
1544
1545	if (capable(CAP_IPC_LOCK) || !size)
1546		return 0;
1547
1548	rlim = rlimit(RLIMIT_MEMLOCK);
1549	if (rlim == RLIM_INFINITY)
1550		return 0;
1551
1552	num_pg = (size >> PAGE_SHIFT) + 2;	/* worst case */
1553	max_pg = rlim >> PAGE_SHIFT;
1554	user = mmp->user ? : current_user();
1555
1556	old_pg = atomic_long_read(&user->locked_vm);
1557	do {
1558		new_pg = old_pg + num_pg;
1559		if (new_pg > max_pg)
1560			return -ENOBUFS;
1561	} while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1562
1563	if (!mmp->user) {
1564		mmp->user = get_uid(user);
1565		mmp->num_pg = num_pg;
1566	} else {
1567		mmp->num_pg += num_pg;
1568	}
1569
1570	return 0;
1571}
1572EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1573
1574void mm_unaccount_pinned_pages(struct mmpin *mmp)
1575{
1576	if (mmp->user) {
1577		atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1578		free_uid(mmp->user);
1579	}
1580}
1581EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1582
1583static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1584{
1585	struct ubuf_info_msgzc *uarg;
1586	struct sk_buff *skb;
1587
1588	WARN_ON_ONCE(!in_task());
1589
1590	skb = sock_omalloc(sk, 0, GFP_KERNEL);
1591	if (!skb)
1592		return NULL;
1593
1594	BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1595	uarg = (void *)skb->cb;
1596	uarg->mmp.user = NULL;
1597
1598	if (mm_account_pinned_pages(&uarg->mmp, size)) {
1599		kfree_skb(skb);
1600		return NULL;
1601	}
1602
1603	uarg->ubuf.callback = msg_zerocopy_callback;
1604	uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1605	uarg->len = 1;
1606	uarg->bytelen = size;
1607	uarg->zerocopy = 1;
1608	uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1609	refcount_set(&uarg->ubuf.refcnt, 1);
1610	sock_hold(sk);
1611
1612	return &uarg->ubuf;
1613}
1614
1615static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1616{
1617	return container_of((void *)uarg, struct sk_buff, cb);
1618}
1619
1620struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1621				       struct ubuf_info *uarg)
1622{
1623	if (uarg) {
1624		struct ubuf_info_msgzc *uarg_zc;
1625		const u32 byte_limit = 1 << 19;		/* limit to a few TSO */
1626		u32 bytelen, next;
1627
1628		/* there might be non MSG_ZEROCOPY users */
1629		if (uarg->callback != msg_zerocopy_callback)
1630			return NULL;
1631
1632		/* realloc only when socket is locked (TCP, UDP cork),
1633		 * so uarg->len and sk_zckey access is serialized
1634		 */
1635		if (!sock_owned_by_user(sk)) {
1636			WARN_ON_ONCE(1);
1637			return NULL;
1638		}
1639
1640		uarg_zc = uarg_to_msgzc(uarg);
1641		bytelen = uarg_zc->bytelen + size;
1642		if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1643			/* TCP can create new skb to attach new uarg */
1644			if (sk->sk_type == SOCK_STREAM)
1645				goto new_alloc;
1646			return NULL;
1647		}
1648
1649		next = (u32)atomic_read(&sk->sk_zckey);
1650		if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1651			if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1652				return NULL;
1653			uarg_zc->len++;
1654			uarg_zc->bytelen = bytelen;
1655			atomic_set(&sk->sk_zckey, ++next);
1656
1657			/* no extra ref when appending to datagram (MSG_MORE) */
1658			if (sk->sk_type == SOCK_STREAM)
1659				net_zcopy_get(uarg);
1660
1661			return uarg;
1662		}
1663	}
1664
1665new_alloc:
1666	return msg_zerocopy_alloc(sk, size);
1667}
1668EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1669
1670static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1671{
1672	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1673	u32 old_lo, old_hi;
1674	u64 sum_len;
1675
1676	old_lo = serr->ee.ee_info;
1677	old_hi = serr->ee.ee_data;
1678	sum_len = old_hi - old_lo + 1ULL + len;
1679
1680	if (sum_len >= (1ULL << 32))
1681		return false;
1682
1683	if (lo != old_hi + 1)
1684		return false;
1685
1686	serr->ee.ee_data += len;
1687	return true;
1688}
1689
1690static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1691{
1692	struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1693	struct sock_exterr_skb *serr;
1694	struct sock *sk = skb->sk;
1695	struct sk_buff_head *q;
1696	unsigned long flags;
1697	bool is_zerocopy;
1698	u32 lo, hi;
1699	u16 len;
1700
1701	mm_unaccount_pinned_pages(&uarg->mmp);
1702
1703	/* if !len, there was only 1 call, and it was aborted
1704	 * so do not queue a completion notification
1705	 */
1706	if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1707		goto release;
1708
1709	len = uarg->len;
1710	lo = uarg->id;
1711	hi = uarg->id + len - 1;
1712	is_zerocopy = uarg->zerocopy;
1713
1714	serr = SKB_EXT_ERR(skb);
1715	memset(serr, 0, sizeof(*serr));
1716	serr->ee.ee_errno = 0;
1717	serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1718	serr->ee.ee_data = hi;
1719	serr->ee.ee_info = lo;
1720	if (!is_zerocopy)
1721		serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1722
1723	q = &sk->sk_error_queue;
1724	spin_lock_irqsave(&q->lock, flags);
1725	tail = skb_peek_tail(q);
1726	if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1727	    !skb_zerocopy_notify_extend(tail, lo, len)) {
1728		__skb_queue_tail(q, skb);
1729		skb = NULL;
1730	}
1731	spin_unlock_irqrestore(&q->lock, flags);
1732
1733	sk_error_report(sk);
1734
1735release:
1736	consume_skb(skb);
1737	sock_put(sk);
1738}
1739
1740void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1741			   bool success)
1742{
1743	struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1744
1745	uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1746
1747	if (refcount_dec_and_test(&uarg->refcnt))
1748		__msg_zerocopy_callback(uarg_zc);
1749}
1750EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1751
1752void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1753{
1754	struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1755
1756	atomic_dec(&sk->sk_zckey);
1757	uarg_to_msgzc(uarg)->len--;
1758
1759	if (have_uref)
1760		msg_zerocopy_callback(NULL, uarg, true);
1761}
1762EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1763
1764int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1765			     struct msghdr *msg, int len,
1766			     struct ubuf_info *uarg)
1767{
1768	struct ubuf_info *orig_uarg = skb_zcopy(skb);
1769	int err, orig_len = skb->len;
1770
1771	/* An skb can only point to one uarg. This edge case happens when
1772	 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1773	 */
1774	if (orig_uarg && uarg != orig_uarg)
1775		return -EEXIST;
1776
1777	err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1778	if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1779		struct sock *save_sk = skb->sk;
1780
1781		/* Streams do not free skb on error. Reset to prev state. */
1782		iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1783		skb->sk = sk;
1784		___pskb_trim(skb, orig_len);
1785		skb->sk = save_sk;
1786		return err;
1787	}
1788
1789	skb_zcopy_set(skb, uarg, NULL);
1790	return skb->len - orig_len;
1791}
1792EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1793
1794void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1795{
1796	int i;
1797
1798	skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1799	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1800		skb_frag_ref(skb, i);
1801}
1802EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1803
1804static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1805			      gfp_t gfp_mask)
1806{
1807	if (skb_zcopy(orig)) {
1808		if (skb_zcopy(nskb)) {
1809			/* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1810			if (!gfp_mask) {
1811				WARN_ON_ONCE(1);
1812				return -ENOMEM;
1813			}
1814			if (skb_uarg(nskb) == skb_uarg(orig))
1815				return 0;
1816			if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1817				return -EIO;
1818		}
1819		skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1820	}
1821	return 0;
1822}
1823
1824/**
1825 *	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
1826 *	@skb: the skb to modify
1827 *	@gfp_mask: allocation priority
1828 *
1829 *	This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1830 *	It will copy all frags into kernel and drop the reference
1831 *	to userspace pages.
1832 *
1833 *	If this function is called from an interrupt gfp_mask() must be
1834 *	%GFP_ATOMIC.
1835 *
1836 *	Returns 0 on success or a negative error code on failure
1837 *	to allocate kernel memory to copy to.
1838 */
1839int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1840{
1841	int num_frags = skb_shinfo(skb)->nr_frags;
1842	struct page *page, *head = NULL;
1843	int i, order, psize, new_frags;
1844	u32 d_off;
1845
1846	if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1847		return -EINVAL;
1848
1849	if (!num_frags)
1850		goto release;
1851
1852	/* We might have to allocate high order pages, so compute what minimum
1853	 * page order is needed.
1854	 */
1855	order = 0;
1856	while ((PAGE_SIZE << order) * MAX_SKB_FRAGS < __skb_pagelen(skb))
1857		order++;
1858	psize = (PAGE_SIZE << order);
1859
1860	new_frags = (__skb_pagelen(skb) + psize - 1) >> (PAGE_SHIFT + order);
1861	for (i = 0; i < new_frags; i++) {
1862		page = alloc_pages(gfp_mask | __GFP_COMP, order);
1863		if (!page) {
1864			while (head) {
1865				struct page *next = (struct page *)page_private(head);
1866				put_page(head);
1867				head = next;
1868			}
1869			return -ENOMEM;
1870		}
1871		set_page_private(page, (unsigned long)head);
1872		head = page;
1873	}
1874
1875	page = head;
1876	d_off = 0;
1877	for (i = 0; i < num_frags; i++) {
1878		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1879		u32 p_off, p_len, copied;
1880		struct page *p;
1881		u8 *vaddr;
1882
1883		skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1884				      p, p_off, p_len, copied) {
1885			u32 copy, done = 0;
1886			vaddr = kmap_atomic(p);
1887
1888			while (done < p_len) {
1889				if (d_off == psize) {
1890					d_off = 0;
1891					page = (struct page *)page_private(page);
1892				}
1893				copy = min_t(u32, psize - d_off, p_len - done);
1894				memcpy(page_address(page) + d_off,
1895				       vaddr + p_off + done, copy);
1896				done += copy;
1897				d_off += copy;
1898			}
1899			kunmap_atomic(vaddr);
1900		}
1901	}
1902
1903	/* skb frags release userspace buffers */
1904	for (i = 0; i < num_frags; i++)
1905		skb_frag_unref(skb, i);
1906
1907	/* skb frags point to kernel buffers */
1908	for (i = 0; i < new_frags - 1; i++) {
1909		__skb_fill_page_desc(skb, i, head, 0, psize);
1910		head = (struct page *)page_private(head);
1911	}
1912	__skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1913	skb_shinfo(skb)->nr_frags = new_frags;
1914
1915release:
1916	skb_zcopy_clear(skb, false);
1917	return 0;
1918}
1919EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1920
1921/**
1922 *	skb_clone	-	duplicate an sk_buff
1923 *	@skb: buffer to clone
1924 *	@gfp_mask: allocation priority
1925 *
1926 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
1927 *	copies share the same packet data but not structure. The new
1928 *	buffer has a reference count of 1. If the allocation fails the
1929 *	function returns %NULL otherwise the new buffer is returned.
1930 *
1931 *	If this function is called from an interrupt gfp_mask() must be
1932 *	%GFP_ATOMIC.
1933 */
1934
1935struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1936{
1937	struct sk_buff_fclones *fclones = container_of(skb,
1938						       struct sk_buff_fclones,
1939						       skb1);
1940	struct sk_buff *n;
1941
1942	if (skb_orphan_frags(skb, gfp_mask))
1943		return NULL;
1944
1945	if (skb->fclone == SKB_FCLONE_ORIG &&
1946	    refcount_read(&fclones->fclone_ref) == 1) {
1947		n = &fclones->skb2;
1948		refcount_set(&fclones->fclone_ref, 2);
1949		n->fclone = SKB_FCLONE_CLONE;
1950	} else {
1951		if (skb_pfmemalloc(skb))
1952			gfp_mask |= __GFP_MEMALLOC;
1953
1954		n = kmem_cache_alloc(skbuff_cache, gfp_mask);
1955		if (!n)
1956			return NULL;
1957
1958		n->fclone = SKB_FCLONE_UNAVAILABLE;
1959	}
1960
1961	return __skb_clone(n, skb);
1962}
1963EXPORT_SYMBOL(skb_clone);
1964
1965void skb_headers_offset_update(struct sk_buff *skb, int off)
1966{
1967	/* Only adjust this if it actually is csum_start rather than csum */
1968	if (skb->ip_summed == CHECKSUM_PARTIAL)
1969		skb->csum_start += off;
1970	/* {transport,network,mac}_header and tail are relative to skb->head */
1971	skb->transport_header += off;
1972	skb->network_header   += off;
1973	if (skb_mac_header_was_set(skb))
1974		skb->mac_header += off;
1975	skb->inner_transport_header += off;
1976	skb->inner_network_header += off;
1977	skb->inner_mac_header += off;
1978}
1979EXPORT_SYMBOL(skb_headers_offset_update);
1980
1981void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1982{
1983	__copy_skb_header(new, old);
1984
1985	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1986	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1987	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1988}
1989EXPORT_SYMBOL(skb_copy_header);
1990
1991static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1992{
1993	if (skb_pfmemalloc(skb))
1994		return SKB_ALLOC_RX;
1995	return 0;
1996}
1997
1998/**
1999 *	skb_copy	-	create private copy of an sk_buff
2000 *	@skb: buffer to copy
2001 *	@gfp_mask: allocation priority
2002 *
2003 *	Make a copy of both an &sk_buff and its data. This is used when the
2004 *	caller wishes to modify the data and needs a private copy of the
2005 *	data to alter. Returns %NULL on failure or the pointer to the buffer
2006 *	on success. The returned buffer has a reference count of 1.
2007 *
2008 *	As by-product this function converts non-linear &sk_buff to linear
2009 *	one, so that &sk_buff becomes completely private and caller is allowed
2010 *	to modify all the data of returned buffer. This means that this
2011 *	function is not recommended for use in circumstances when only
2012 *	header is going to be modified. Use pskb_copy() instead.
2013 */
2014
2015struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
2016{
2017	int headerlen = skb_headroom(skb);
2018	unsigned int size = skb_end_offset(skb) + skb->data_len;
2019	struct sk_buff *n = __alloc_skb(size, gfp_mask,
2020					skb_alloc_rx_flag(skb), NUMA_NO_NODE);
2021
2022	if (!n)
2023		return NULL;
2024
2025	/* Set the data pointer */
2026	skb_reserve(n, headerlen);
2027	/* Set the tail pointer and length */
2028	skb_put(n, skb->len);
2029
2030	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
2031
2032	skb_copy_header(n, skb);
2033	return n;
2034}
2035EXPORT_SYMBOL(skb_copy);
2036
2037/**
2038 *	__pskb_copy_fclone	-  create copy of an sk_buff with private head.
2039 *	@skb: buffer to copy
2040 *	@headroom: headroom of new skb
2041 *	@gfp_mask: allocation priority
2042 *	@fclone: if true allocate the copy of the skb from the fclone
2043 *	cache instead of the head cache; it is recommended to set this
2044 *	to true for the cases where the copy will likely be cloned
2045 *
2046 *	Make a copy of both an &sk_buff and part of its data, located
2047 *	in header. Fragmented data remain shared. This is used when
2048 *	the caller wishes to modify only header of &sk_buff and needs
2049 *	private copy of the header to alter. Returns %NULL on failure
2050 *	or the pointer to the buffer on success.
2051 *	The returned buffer has a reference count of 1.
2052 */
2053
2054struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
2055				   gfp_t gfp_mask, bool fclone)
2056{
2057	unsigned int size = skb_headlen(skb) + headroom;
2058	int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
2059	struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
2060
2061	if (!n)
2062		goto out;
2063
2064	/* Set the data pointer */
2065	skb_reserve(n, headroom);
2066	/* Set the tail pointer and length */
2067	skb_put(n, skb_headlen(skb));
2068	/* Copy the bytes */
2069	skb_copy_from_linear_data(skb, n->data, n->len);
2070
2071	n->truesize += skb->data_len;
2072	n->data_len  = skb->data_len;
2073	n->len	     = skb->len;
2074
2075	if (skb_shinfo(skb)->nr_frags) {
2076		int i;
2077
2078		if (skb_orphan_frags(skb, gfp_mask) ||
2079		    skb_zerocopy_clone(n, skb, gfp_mask)) {
2080			kfree_skb(n);
2081			n = NULL;
2082			goto out;
2083		}
2084		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2085			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
2086			skb_frag_ref(skb, i);
2087		}
2088		skb_shinfo(n)->nr_frags = i;
2089	}
2090
2091	if (skb_has_frag_list(skb)) {
2092		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
2093		skb_clone_fraglist(n);
2094	}
2095
2096	skb_copy_header(n, skb);
2097out:
2098	return n;
2099}
2100EXPORT_SYMBOL(__pskb_copy_fclone);
2101
2102/**
2103 *	pskb_expand_head - reallocate header of &sk_buff
2104 *	@skb: buffer to reallocate
2105 *	@nhead: room to add at head
2106 *	@ntail: room to add at tail
2107 *	@gfp_mask: allocation priority
2108 *
2109 *	Expands (or creates identical copy, if @nhead and @ntail are zero)
2110 *	header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
2111 *	reference count of 1. Returns zero in the case of success or error,
2112 *	if expansion failed. In the last case, &sk_buff is not changed.
2113 *
2114 *	All the pointers pointing into skb header may change and must be
2115 *	reloaded after call to this function.
2116 */
2117
2118int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
2119		     gfp_t gfp_mask)
2120{
2121	unsigned int osize = skb_end_offset(skb);
2122	unsigned int size = osize + nhead + ntail;
2123	long off;
2124	u8 *data;
2125	int i;
2126
2127	BUG_ON(nhead < 0);
2128
2129	BUG_ON(skb_shared(skb));
2130
2131	skb_zcopy_downgrade_managed(skb);
2132
2133	if (skb_pfmemalloc(skb))
2134		gfp_mask |= __GFP_MEMALLOC;
2135
2136	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
2137	if (!data)
2138		goto nodata;
2139	size = SKB_WITH_OVERHEAD(size);
2140
2141	/* Copy only real data... and, alas, header. This should be
2142	 * optimized for the cases when header is void.
2143	 */
2144	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
2145
2146	memcpy((struct skb_shared_info *)(data + size),
2147	       skb_shinfo(skb),
2148	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
2149
2150	/*
2151	 * if shinfo is shared we must drop the old head gracefully, but if it
2152	 * is not we can just drop the old head and let the existing refcount
2153	 * be since all we did is relocate the values
2154	 */
2155	if (skb_cloned(skb)) {
2156		if (skb_orphan_frags(skb, gfp_mask))
2157			goto nofrags;
2158		if (skb_zcopy(skb))
2159			refcount_inc(&skb_uarg(skb)->refcnt);
2160		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2161			skb_frag_ref(skb, i);
2162
2163		if (skb_has_frag_list(skb))
2164			skb_clone_fraglist(skb);
2165
2166		skb_release_data(skb, SKB_CONSUMED, false);
2167	} else {
2168		skb_free_head(skb, false);
2169	}
2170	off = (data + nhead) - skb->head;
2171
2172	skb->head     = data;
2173	skb->head_frag = 0;
2174	skb->data    += off;
2175
2176	skb_set_end_offset(skb, size);
2177#ifdef NET_SKBUFF_DATA_USES_OFFSET
2178	off           = nhead;
2179#endif
2180	skb->tail	      += off;
2181	skb_headers_offset_update(skb, nhead);
2182	skb->cloned   = 0;
2183	skb->hdr_len  = 0;
2184	skb->nohdr    = 0;
2185	atomic_set(&skb_shinfo(skb)->dataref, 1);
2186
2187	skb_metadata_clear(skb);
2188
2189	/* It is not generally safe to change skb->truesize.
2190	 * For the moment, we really care of rx path, or
2191	 * when skb is orphaned (not attached to a socket).
2192	 */
2193	if (!skb->sk || skb->destructor == sock_edemux)
2194		skb->truesize += size - osize;
2195
2196	return 0;
2197
2198nofrags:
2199	skb_kfree_head(data, size);
2200nodata:
2201	return -ENOMEM;
2202}
2203EXPORT_SYMBOL(pskb_expand_head);
2204
2205/* Make private copy of skb with writable head and some headroom */
2206
2207struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
2208{
2209	struct sk_buff *skb2;
2210	int delta = headroom - skb_headroom(skb);
2211
2212	if (delta <= 0)
2213		skb2 = pskb_copy(skb, GFP_ATOMIC);
2214	else {
2215		skb2 = skb_clone(skb, GFP_ATOMIC);
2216		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
2217					     GFP_ATOMIC)) {
2218			kfree_skb(skb2);
2219			skb2 = NULL;
2220		}
2221	}
2222	return skb2;
2223}
2224EXPORT_SYMBOL(skb_realloc_headroom);
2225
2226/* Note: We plan to rework this in linux-6.4 */
2227int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2228{
2229	unsigned int saved_end_offset, saved_truesize;
2230	struct skb_shared_info *shinfo;
2231	int res;
2232
2233	saved_end_offset = skb_end_offset(skb);
2234	saved_truesize = skb->truesize;
2235
2236	res = pskb_expand_head(skb, 0, 0, pri);
2237	if (res)
2238		return res;
2239
2240	skb->truesize = saved_truesize;
2241
2242	if (likely(skb_end_offset(skb) == saved_end_offset))
2243		return 0;
2244
2245	/* We can not change skb->end if the original or new value
2246	 * is SKB_SMALL_HEAD_HEADROOM, as it might break skb_kfree_head().
2247	 */
2248	if (saved_end_offset == SKB_SMALL_HEAD_HEADROOM ||
2249	    skb_end_offset(skb) == SKB_SMALL_HEAD_HEADROOM) {
2250		/* We think this path should not be taken.
2251		 * Add a temporary trace to warn us just in case.
2252		 */
2253		pr_err_once("__skb_unclone_keeptruesize() skb_end_offset() %u -> %u\n",
2254			    saved_end_offset, skb_end_offset(skb));
2255		WARN_ON_ONCE(1);
2256		return 0;
2257	}
2258
2259	shinfo = skb_shinfo(skb);
2260
2261	/* We are about to change back skb->end,
2262	 * we need to move skb_shinfo() to its new location.
2263	 */
2264	memmove(skb->head + saved_end_offset,
2265		shinfo,
2266		offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2267
2268	skb_set_end_offset(skb, saved_end_offset);
2269
2270	return 0;
2271}
2272
2273/**
2274 *	skb_expand_head - reallocate header of &sk_buff
2275 *	@skb: buffer to reallocate
2276 *	@headroom: needed headroom
2277 *
2278 *	Unlike skb_realloc_headroom, this one does not allocate a new skb
2279 *	if possible; copies skb->sk to new skb as needed
2280 *	and frees original skb in case of failures.
2281 *
2282 *	It expect increased headroom and generates warning otherwise.
2283 */
2284
2285struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2286{
2287	int delta = headroom - skb_headroom(skb);
2288	int osize = skb_end_offset(skb);
2289	struct sock *sk = skb->sk;
2290
2291	if (WARN_ONCE(delta <= 0,
2292		      "%s is expecting an increase in the headroom", __func__))
2293		return skb;
2294
2295	delta = SKB_DATA_ALIGN(delta);
2296	/* pskb_expand_head() might crash, if skb is shared. */
2297	if (skb_shared(skb) || !is_skb_wmem(skb)) {
2298		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2299
2300		if (unlikely(!nskb))
2301			goto fail;
2302
2303		if (sk)
2304			skb_set_owner_w(nskb, sk);
2305		consume_skb(skb);
2306		skb = nskb;
2307	}
2308	if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2309		goto fail;
2310
2311	if (sk && is_skb_wmem(skb)) {
2312		delta = skb_end_offset(skb) - osize;
2313		refcount_add(delta, &sk->sk_wmem_alloc);
2314		skb->truesize += delta;
2315	}
2316	return skb;
2317
2318fail:
2319	kfree_skb(skb);
2320	return NULL;
2321}
2322EXPORT_SYMBOL(skb_expand_head);
2323
2324/**
2325 *	skb_copy_expand	-	copy and expand sk_buff
2326 *	@skb: buffer to copy
2327 *	@newheadroom: new free bytes at head
2328 *	@newtailroom: new free bytes at tail
2329 *	@gfp_mask: allocation priority
2330 *
2331 *	Make a copy of both an &sk_buff and its data and while doing so
2332 *	allocate additional space.
2333 *
2334 *	This is used when the caller wishes to modify the data and needs a
2335 *	private copy of the data to alter as well as more space for new fields.
2336 *	Returns %NULL on failure or the pointer to the buffer
2337 *	on success. The returned buffer has a reference count of 1.
2338 *
2339 *	You must pass %GFP_ATOMIC as the allocation priority if this function
2340 *	is called from an interrupt.
2341 */
2342struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2343				int newheadroom, int newtailroom,
2344				gfp_t gfp_mask)
2345{
2346	/*
2347	 *	Allocate the copy buffer
2348	 */
2349	struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
2350					gfp_mask, skb_alloc_rx_flag(skb),
2351					NUMA_NO_NODE);
2352	int oldheadroom = skb_headroom(skb);
2353	int head_copy_len, head_copy_off;
2354
2355	if (!n)
2356		return NULL;
2357
2358	skb_reserve(n, newheadroom);
2359
2360	/* Set the tail pointer and length */
2361	skb_put(n, skb->len);
2362
2363	head_copy_len = oldheadroom;
2364	head_copy_off = 0;
2365	if (newheadroom <= head_copy_len)
2366		head_copy_len = newheadroom;
2367	else
2368		head_copy_off = newheadroom - head_copy_len;
2369
2370	/* Copy the linear header and data. */
2371	BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2372			     skb->len + head_copy_len));
2373
2374	skb_copy_header(n, skb);
2375
2376	skb_headers_offset_update(n, newheadroom - oldheadroom);
2377
2378	return n;
2379}
2380EXPORT_SYMBOL(skb_copy_expand);
2381
2382/**
2383 *	__skb_pad		-	zero pad the tail of an skb
2384 *	@skb: buffer to pad
2385 *	@pad: space to pad
2386 *	@free_on_error: free buffer on error
2387 *
2388 *	Ensure that a buffer is followed by a padding area that is zero
2389 *	filled. Used by network drivers which may DMA or transfer data
2390 *	beyond the buffer end onto the wire.
2391 *
2392 *	May return error in out of memory cases. The skb is freed on error
2393 *	if @free_on_error is true.
2394 */
2395
2396int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2397{
2398	int err;
2399	int ntail;
2400
2401	/* If the skbuff is non linear tailroom is always zero.. */
2402	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2403		memset(skb->data+skb->len, 0, pad);
2404		return 0;
2405	}
2406
2407	ntail = skb->data_len + pad - (skb->end - skb->tail);
2408	if (likely(skb_cloned(skb) || ntail > 0)) {
2409		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2410		if (unlikely(err))
2411			goto free_skb;
2412	}
2413
2414	/* FIXME: The use of this function with non-linear skb's really needs
2415	 * to be audited.
2416	 */
2417	err = skb_linearize(skb);
2418	if (unlikely(err))
2419		goto free_skb;
2420
2421	memset(skb->data + skb->len, 0, pad);
2422	return 0;
2423
2424free_skb:
2425	if (free_on_error)
2426		kfree_skb(skb);
2427	return err;
2428}
2429EXPORT_SYMBOL(__skb_pad);
2430
2431/**
2432 *	pskb_put - add data to the tail of a potentially fragmented buffer
2433 *	@skb: start of the buffer to use
2434 *	@tail: tail fragment of the buffer to use
2435 *	@len: amount of data to add
2436 *
2437 *	This function extends the used data area of the potentially
2438 *	fragmented buffer. @tail must be the last fragment of @skb -- or
2439 *	@skb itself. If this would exceed the total buffer size the kernel
2440 *	will panic. A pointer to the first byte of the extra data is
2441 *	returned.
2442 */
2443
2444void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2445{
2446	if (tail != skb) {
2447		skb->data_len += len;
2448		skb->len += len;
2449	}
2450	return skb_put(tail, len);
2451}
2452EXPORT_SYMBOL_GPL(pskb_put);
2453
2454/**
2455 *	skb_put - add data to a buffer
2456 *	@skb: buffer to use
2457 *	@len: amount of data to add
2458 *
2459 *	This function extends the used data area of the buffer. If this would
2460 *	exceed the total buffer size the kernel will panic. A pointer to the
2461 *	first byte of the extra data is returned.
2462 */
2463void *skb_put(struct sk_buff *skb, unsigned int len)
2464{
2465	void *tmp = skb_tail_pointer(skb);
2466	SKB_LINEAR_ASSERT(skb);
2467	skb->tail += len;
2468	skb->len  += len;
2469	if (unlikely(skb->tail > skb->end))
2470		skb_over_panic(skb, len, __builtin_return_address(0));
2471	return tmp;
2472}
2473EXPORT_SYMBOL(skb_put);
2474
2475/**
2476 *	skb_push - add data to the start of a buffer
2477 *	@skb: buffer to use
2478 *	@len: amount of data to add
2479 *
2480 *	This function extends the used data area of the buffer at the buffer
2481 *	start. If this would exceed the total buffer headroom the kernel will
2482 *	panic. A pointer to the first byte of the extra data is returned.
2483 */
2484void *skb_push(struct sk_buff *skb, unsigned int len)
2485{
2486	skb->data -= len;
2487	skb->len  += len;
2488	if (unlikely(skb->data < skb->head))
2489		skb_under_panic(skb, len, __builtin_return_address(0));
2490	return skb->data;
2491}
2492EXPORT_SYMBOL(skb_push);
2493
2494/**
2495 *	skb_pull - remove data from the start of a buffer
2496 *	@skb: buffer to use
2497 *	@len: amount of data to remove
2498 *
2499 *	This function removes data from the start of a buffer, returning
2500 *	the memory to the headroom. A pointer to the next data in the buffer
2501 *	is returned. Once the data has been pulled future pushes will overwrite
2502 *	the old data.
2503 */
2504void *skb_pull(struct sk_buff *skb, unsigned int len)
2505{
2506	return skb_pull_inline(skb, len);
2507}
2508EXPORT_SYMBOL(skb_pull);
2509
2510/**
2511 *	skb_pull_data - remove data from the start of a buffer returning its
2512 *	original position.
2513 *	@skb: buffer to use
2514 *	@len: amount of data to remove
2515 *
2516 *	This function removes data from the start of a buffer, returning
2517 *	the memory to the headroom. A pointer to the original data in the buffer
2518 *	is returned after checking if there is enough data to pull. Once the
2519 *	data has been pulled future pushes will overwrite the old data.
2520 */
2521void *skb_pull_data(struct sk_buff *skb, size_t len)
2522{
2523	void *data = skb->data;
2524
2525	if (skb->len < len)
2526		return NULL;
2527
2528	skb_pull(skb, len);
2529
2530	return data;
2531}
2532EXPORT_SYMBOL(skb_pull_data);
2533
2534/**
2535 *	skb_trim - remove end from a buffer
2536 *	@skb: buffer to alter
2537 *	@len: new length
2538 *
2539 *	Cut the length of a buffer down by removing data from the tail. If
2540 *	the buffer is already under the length specified it is not modified.
2541 *	The skb must be linear.
2542 */
2543void skb_trim(struct sk_buff *skb, unsigned int len)
2544{
2545	if (skb->len > len)
2546		__skb_trim(skb, len);
2547}
2548EXPORT_SYMBOL(skb_trim);
2549
2550/* Trims skb to length len. It can change skb pointers.
2551 */
2552
2553int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2554{
2555	struct sk_buff **fragp;
2556	struct sk_buff *frag;
2557	int offset = skb_headlen(skb);
2558	int nfrags = skb_shinfo(skb)->nr_frags;
2559	int i;
2560	int err;
2561
2562	if (skb_cloned(skb) &&
2563	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2564		return err;
2565
2566	i = 0;
2567	if (offset >= len)
2568		goto drop_pages;
2569
2570	for (; i < nfrags; i++) {
2571		int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2572
2573		if (end < len) {
2574			offset = end;
2575			continue;
2576		}
2577
2578		skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2579
2580drop_pages:
2581		skb_shinfo(skb)->nr_frags = i;
2582
2583		for (; i < nfrags; i++)
2584			skb_frag_unref(skb, i);
2585
2586		if (skb_has_frag_list(skb))
2587			skb_drop_fraglist(skb);
2588		goto done;
2589	}
2590
2591	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2592	     fragp = &frag->next) {
2593		int end = offset + frag->len;
2594
2595		if (skb_shared(frag)) {
2596			struct sk_buff *nfrag;
2597
2598			nfrag = skb_clone(frag, GFP_ATOMIC);
2599			if (unlikely(!nfrag))
2600				return -ENOMEM;
2601
2602			nfrag->next = frag->next;
2603			consume_skb(frag);
2604			frag = nfrag;
2605			*fragp = frag;
2606		}
2607
2608		if (end < len) {
2609			offset = end;
2610			continue;
2611		}
2612
2613		if (end > len &&
2614		    unlikely((err = pskb_trim(frag, len - offset))))
2615			return err;
2616
2617		if (frag->next)
2618			skb_drop_list(&frag->next);
2619		break;
2620	}
2621
2622done:
2623	if (len > skb_headlen(skb)) {
2624		skb->data_len -= skb->len - len;
2625		skb->len       = len;
2626	} else {
2627		skb->len       = len;
2628		skb->data_len  = 0;
2629		skb_set_tail_pointer(skb, len);
2630	}
2631
2632	if (!skb->sk || skb->destructor == sock_edemux)
2633		skb_condense(skb);
2634	return 0;
2635}
2636EXPORT_SYMBOL(___pskb_trim);
2637
2638/* Note : use pskb_trim_rcsum() instead of calling this directly
2639 */
2640int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2641{
2642	if (skb->ip_summed == CHECKSUM_COMPLETE) {
2643		int delta = skb->len - len;
2644
2645		skb->csum = csum_block_sub(skb->csum,
2646					   skb_checksum(skb, len, delta, 0),
2647					   len);
2648	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2649		int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2650		int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2651
2652		if (offset + sizeof(__sum16) > hdlen)
2653			return -EINVAL;
2654	}
2655	return __pskb_trim(skb, len);
2656}
2657EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2658
2659/**
2660 *	__pskb_pull_tail - advance tail of skb header
2661 *	@skb: buffer to reallocate
2662 *	@delta: number of bytes to advance tail
2663 *
2664 *	The function makes a sense only on a fragmented &sk_buff,
2665 *	it expands header moving its tail forward and copying necessary
2666 *	data from fragmented part.
2667 *
2668 *	&sk_buff MUST have reference count of 1.
2669 *
2670 *	Returns %NULL (and &sk_buff does not change) if pull failed
2671 *	or value of new tail of skb in the case of success.
2672 *
2673 *	All the pointers pointing into skb header may change and must be
2674 *	reloaded after call to this function.
2675 */
2676
2677/* Moves tail of skb head forward, copying data from fragmented part,
2678 * when it is necessary.
2679 * 1. It may fail due to malloc failure.
2680 * 2. It may change skb pointers.
2681 *
2682 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2683 */
2684void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2685{
2686	/* If skb has not enough free space at tail, get new one
2687	 * plus 128 bytes for future expansions. If we have enough
2688	 * room at tail, reallocate without expansion only if skb is cloned.
2689	 */
2690	int i, k, eat = (skb->tail + delta) - skb->end;
2691
2692	if (eat > 0 || skb_cloned(skb)) {
2693		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2694				     GFP_ATOMIC))
2695			return NULL;
2696	}
2697
2698	BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2699			     skb_tail_pointer(skb), delta));
2700
2701	/* Optimization: no fragments, no reasons to preestimate
2702	 * size of pulled pages. Superb.
2703	 */
2704	if (!skb_has_frag_list(skb))
2705		goto pull_pages;
2706
2707	/* Estimate size of pulled pages. */
2708	eat = delta;
2709	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2710		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2711
2712		if (size >= eat)
2713			goto pull_pages;
2714		eat -= size;
2715	}
2716
2717	/* If we need update frag list, we are in troubles.
2718	 * Certainly, it is possible to add an offset to skb data,
2719	 * but taking into account that pulling is expected to
2720	 * be very rare operation, it is worth to fight against
2721	 * further bloating skb head and crucify ourselves here instead.
2722	 * Pure masohism, indeed. 8)8)
2723	 */
2724	if (eat) {
2725		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2726		struct sk_buff *clone = NULL;
2727		struct sk_buff *insp = NULL;
2728
2729		do {
2730			if (list->len <= eat) {
2731				/* Eaten as whole. */
2732				eat -= list->len;
2733				list = list->next;
2734				insp = list;
2735			} else {
2736				/* Eaten partially. */
2737				if (skb_is_gso(skb) && !list->head_frag &&
2738				    skb_headlen(list))
2739					skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2740
2741				if (skb_shared(list)) {
2742					/* Sucks! We need to fork list. :-( */
2743					clone = skb_clone(list, GFP_ATOMIC);
2744					if (!clone)
2745						return NULL;
2746					insp = list->next;
2747					list = clone;
2748				} else {
2749					/* This may be pulled without
2750					 * problems. */
2751					insp = list;
2752				}
2753				if (!pskb_pull(list, eat)) {
2754					kfree_skb(clone);
2755					return NULL;
2756				}
2757				break;
2758			}
2759		} while (eat);
2760
2761		/* Free pulled out fragments. */
2762		while ((list = skb_shinfo(skb)->frag_list) != insp) {
2763			skb_shinfo(skb)->frag_list = list->next;
2764			consume_skb(list);
2765		}
2766		/* And insert new clone at head. */
2767		if (clone) {
2768			clone->next = list;
2769			skb_shinfo(skb)->frag_list = clone;
2770		}
2771	}
2772	/* Success! Now we may commit changes to skb data. */
2773
2774pull_pages:
2775	eat = delta;
2776	k = 0;
2777	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2778		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2779
2780		if (size <= eat) {
2781			skb_frag_unref(skb, i);
2782			eat -= size;
2783		} else {
2784			skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2785
2786			*frag = skb_shinfo(skb)->frags[i];
2787			if (eat) {
2788				skb_frag_off_add(frag, eat);
2789				skb_frag_size_sub(frag, eat);
2790				if (!i)
2791					goto end;
2792				eat = 0;
2793			}
2794			k++;
2795		}
2796	}
2797	skb_shinfo(skb)->nr_frags = k;
2798
2799end:
2800	skb->tail     += delta;
2801	skb->data_len -= delta;
2802
2803	if (!skb->data_len)
2804		skb_zcopy_clear(skb, false);
2805
2806	return skb_tail_pointer(skb);
2807}
2808EXPORT_SYMBOL(__pskb_pull_tail);
2809
2810/**
2811 *	skb_copy_bits - copy bits from skb to kernel buffer
2812 *	@skb: source skb
2813 *	@offset: offset in source
2814 *	@to: destination buffer
2815 *	@len: number of bytes to copy
2816 *
2817 *	Copy the specified number of bytes from the source skb to the
2818 *	destination buffer.
2819 *
2820 *	CAUTION ! :
2821 *		If its prototype is ever changed,
2822 *		check arch/{*}/net/{*}.S files,
2823 *		since it is called from BPF assembly code.
2824 */
2825int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2826{
2827	int start = skb_headlen(skb);
2828	struct sk_buff *frag_iter;
2829	int i, copy;
2830
2831	if (offset > (int)skb->len - len)
2832		goto fault;
2833
2834	/* Copy header. */
2835	if ((copy = start - offset) > 0) {
2836		if (copy > len)
2837			copy = len;
2838		skb_copy_from_linear_data_offset(skb, offset, to, copy);
2839		if ((len -= copy) == 0)
2840			return 0;
2841		offset += copy;
2842		to     += copy;
2843	}
2844
2845	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2846		int end;
2847		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2848
2849		WARN_ON(start > offset + len);
2850
2851		end = start + skb_frag_size(f);
2852		if ((copy = end - offset) > 0) {
2853			u32 p_off, p_len, copied;
2854			struct page *p;
2855			u8 *vaddr;
2856
2857			if (copy > len)
2858				copy = len;
2859
2860			skb_frag_foreach_page(f,
2861					      skb_frag_off(f) + offset - start,
2862					      copy, p, p_off, p_len, copied) {
2863				vaddr = kmap_atomic(p);
2864				memcpy(to + copied, vaddr + p_off, p_len);
2865				kunmap_atomic(vaddr);
2866			}
2867
2868			if ((len -= copy) == 0)
2869				return 0;
2870			offset += copy;
2871			to     += copy;
2872		}
2873		start = end;
2874	}
2875
2876	skb_walk_frags(skb, frag_iter) {
2877		int end;
2878
2879		WARN_ON(start > offset + len);
2880
2881		end = start + frag_iter->len;
2882		if ((copy = end - offset) > 0) {
2883			if (copy > len)
2884				copy = len;
2885			if (skb_copy_bits(frag_iter, offset - start, to, copy))
2886				goto fault;
2887			if ((len -= copy) == 0)
2888				return 0;
2889			offset += copy;
2890			to     += copy;
2891		}
2892		start = end;
2893	}
2894
2895	if (!len)
2896		return 0;
2897
2898fault:
2899	return -EFAULT;
2900}
2901EXPORT_SYMBOL(skb_copy_bits);
2902
2903/*
2904 * Callback from splice_to_pipe(), if we need to release some pages
2905 * at the end of the spd in case we error'ed out in filling the pipe.
2906 */
2907static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2908{
2909	put_page(spd->pages[i]);
2910}
2911
2912static struct page *linear_to_page(struct page *page, unsigned int *len,
2913				   unsigned int *offset,
2914				   struct sock *sk)
2915{
2916	struct page_frag *pfrag = sk_page_frag(sk);
2917
2918	if (!sk_page_frag_refill(sk, pfrag))
2919		return NULL;
2920
2921	*len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2922
2923	memcpy(page_address(pfrag->page) + pfrag->offset,
2924	       page_address(page) + *offset, *len);
2925	*offset = pfrag->offset;
2926	pfrag->offset += *len;
2927
2928	return pfrag->page;
2929}
2930
2931static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2932			     struct page *page,
2933			     unsigned int offset)
2934{
2935	return	spd->nr_pages &&
2936		spd->pages[spd->nr_pages - 1] == page &&
2937		(spd->partial[spd->nr_pages - 1].offset +
2938		 spd->partial[spd->nr_pages - 1].len == offset);
2939}
2940
2941/*
2942 * Fill page/offset/length into spd, if it can hold more pages.
2943 */
2944static bool spd_fill_page(struct splice_pipe_desc *spd,
2945			  struct pipe_inode_info *pipe, struct page *page,
2946			  unsigned int *len, unsigned int offset,
2947			  bool linear,
2948			  struct sock *sk)
2949{
2950	if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2951		return true;
2952
2953	if (linear) {
2954		page = linear_to_page(page, len, &offset, sk);
2955		if (!page)
2956			return true;
2957	}
2958	if (spd_can_coalesce(spd, page, offset)) {
2959		spd->partial[spd->nr_pages - 1].len += *len;
2960		return false;
2961	}
2962	get_page(page);
2963	spd->pages[spd->nr_pages] = page;
2964	spd->partial[spd->nr_pages].len = *len;
2965	spd->partial[spd->nr_pages].offset = offset;
2966	spd->nr_pages++;
2967
2968	return false;
2969}
2970
2971static bool __splice_segment(struct page *page, unsigned int poff,
2972			     unsigned int plen, unsigned int *off,
2973			     unsigned int *len,
2974			     struct splice_pipe_desc *spd, bool linear,
2975			     struct sock *sk,
2976			     struct pipe_inode_info *pipe)
2977{
2978	if (!*len)
2979		return true;
2980
2981	/* skip this segment if already processed */
2982	if (*off >= plen) {
2983		*off -= plen;
2984		return false;
2985	}
2986
2987	/* ignore any bits we already processed */
2988	poff += *off;
2989	plen -= *off;
2990	*off = 0;
2991
2992	do {
2993		unsigned int flen = min(*len, plen);
2994
2995		if (spd_fill_page(spd, pipe, page, &flen, poff,
2996				  linear, sk))
2997			return true;
2998		poff += flen;
2999		plen -= flen;
3000		*len -= flen;
3001	} while (*len && plen);
3002
3003	return false;
3004}
3005
3006/*
3007 * Map linear and fragment data from the skb to spd. It reports true if the
3008 * pipe is full or if we already spliced the requested length.
3009 */
3010static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
3011			      unsigned int *offset, unsigned int *len,
3012			      struct splice_pipe_desc *spd, struct sock *sk)
3013{
3014	int seg;
3015	struct sk_buff *iter;
3016
3017	/* map the linear part :
3018	 * If skb->head_frag is set, this 'linear' part is backed by a
3019	 * fragment, and if the head is not shared with any clones then
3020	 * we can avoid a copy since we own the head portion of this page.
3021	 */
3022	if (__splice_segment(virt_to_page(skb->data),
3023			     (unsigned long) skb->data & (PAGE_SIZE - 1),
3024			     skb_headlen(skb),
3025			     offset, len, spd,
3026			     skb_head_is_locked(skb),
3027			     sk, pipe))
3028		return true;
3029
3030	/*
3031	 * then map the fragments
3032	 */
3033	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
3034		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
3035
3036		if (__splice_segment(skb_frag_page(f),
3037				     skb_frag_off(f), skb_frag_size(f),
3038				     offset, len, spd, false, sk, pipe))
3039			return true;
3040	}
3041
3042	skb_walk_frags(skb, iter) {
3043		if (*offset >= iter->len) {
3044			*offset -= iter->len;
3045			continue;
3046		}
3047		/* __skb_splice_bits() only fails if the output has no room
3048		 * left, so no point in going over the frag_list for the error
3049		 * case.
3050		 */
3051		if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
3052			return true;
3053	}
3054
3055	return false;
3056}
3057
3058/*
3059 * Map data from the skb to a pipe. Should handle both the linear part,
3060 * the fragments, and the frag list.
3061 */
3062int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3063		    struct pipe_inode_info *pipe, unsigned int tlen,
3064		    unsigned int flags)
3065{
3066	struct partial_page partial[MAX_SKB_FRAGS];
3067	struct page *pages[MAX_SKB_FRAGS];
3068	struct splice_pipe_desc spd = {
3069		.pages = pages,
3070		.partial = partial,
3071		.nr_pages_max = MAX_SKB_FRAGS,
3072		.ops = &nosteal_pipe_buf_ops,
3073		.spd_release = sock_spd_release,
3074	};
3075	int ret = 0;
3076
3077	__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
3078
3079	if (spd.nr_pages)
3080		ret = splice_to_pipe(pipe, &spd);
3081
3082	return ret;
3083}
3084EXPORT_SYMBOL_GPL(skb_splice_bits);
3085
3086static int sendmsg_locked(struct sock *sk, struct msghdr *msg)
3087{
3088	struct socket *sock = sk->sk_socket;
3089	size_t size = msg_data_left(msg);
3090
3091	if (!sock)
3092		return -EINVAL;
3093
3094	if (!sock->ops->sendmsg_locked)
3095		return sock_no_sendmsg_locked(sk, msg, size);
3096
3097	return sock->ops->sendmsg_locked(sk, msg, size);
3098}
3099
3100static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg)
3101{
3102	struct socket *sock = sk->sk_socket;
3103
3104	if (!sock)
3105		return -EINVAL;
3106	return sock_sendmsg(sock, msg);
3107}
3108
3109typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg);
3110static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
3111			   int len, sendmsg_func sendmsg)
3112{
3113	unsigned int orig_len = len;
3114	struct sk_buff *head = skb;
3115	unsigned short fragidx;
3116	int slen, ret;
3117
3118do_frag_list:
3119
3120	/* Deal with head data */
3121	while (offset < skb_headlen(skb) && len) {
3122		struct kvec kv;
3123		struct msghdr msg;
3124
3125		slen = min_t(int, len, skb_headlen(skb) - offset);
3126		kv.iov_base = skb->data + offset;
3127		kv.iov_len = slen;
3128		memset(&msg, 0, sizeof(msg));
3129		msg.msg_flags = MSG_DONTWAIT;
3130
3131		iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &kv, 1, slen);
3132		ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3133				      sendmsg_unlocked, sk, &msg);
3134		if (ret <= 0)
3135			goto error;
3136
3137		offset += ret;
3138		len -= ret;
3139	}
3140
3141	/* All the data was skb head? */
3142	if (!len)
3143		goto out;
3144
3145	/* Make offset relative to start of frags */
3146	offset -= skb_headlen(skb);
3147
3148	/* Find where we are in frag list */
3149	for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3150		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3151
3152		if (offset < skb_frag_size(frag))
3153			break;
3154
3155		offset -= skb_frag_size(frag);
3156	}
3157
3158	for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3159		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3160
3161		slen = min_t(size_t, len, skb_frag_size(frag) - offset);
3162
3163		while (slen) {
3164			struct bio_vec bvec;
3165			struct msghdr msg = {
3166				.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT,
3167			};
3168
3169			bvec_set_page(&bvec, skb_frag_page(frag), slen,
3170				      skb_frag_off(frag) + offset);
3171			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1,
3172				      slen);
3173
3174			ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3175					      sendmsg_unlocked, sk, &msg);
3176			if (ret <= 0)
3177				goto error;
3178
3179			len -= ret;
3180			offset += ret;
3181			slen -= ret;
3182		}
3183
3184		offset = 0;
3185	}
3186
3187	if (len) {
3188		/* Process any frag lists */
3189
3190		if (skb == head) {
3191			if (skb_has_frag_list(skb)) {
3192				skb = skb_shinfo(skb)->frag_list;
3193				goto do_frag_list;
3194			}
3195		} else if (skb->next) {
3196			skb = skb->next;
3197			goto do_frag_list;
3198		}
3199	}
3200
3201out:
3202	return orig_len - len;
3203
3204error:
3205	return orig_len == len ? ret : orig_len - len;
3206}
3207
3208/* Send skb data on a socket. Socket must be locked. */
3209int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3210			 int len)
3211{
3212	return __skb_send_sock(sk, skb, offset, len, sendmsg_locked);
3213}
3214EXPORT_SYMBOL_GPL(skb_send_sock_locked);
3215
3216/* Send skb data on a socket. Socket must be unlocked. */
3217int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
3218{
3219	return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked);
3220}
3221
3222/**
3223 *	skb_store_bits - store bits from kernel buffer to skb
3224 *	@skb: destination buffer
3225 *	@offset: offset in destination
3226 *	@from: source buffer
3227 *	@len: number of bytes to copy
3228 *
3229 *	Copy the specified number of bytes from the source buffer to the
3230 *	destination skb.  This function handles all the messy bits of
3231 *	traversing fragment lists and such.
3232 */
3233
3234int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
3235{
3236	int start = skb_headlen(skb);
3237	struct sk_buff *frag_iter;
3238	int i, copy;
3239
3240	if (offset > (int)skb->len - len)
3241		goto fault;
3242
3243	if ((copy = start - offset) > 0) {
3244		if (copy > len)
3245			copy = len;
3246		skb_copy_to_linear_data_offset(skb, offset, from, copy);
3247		if ((len -= copy) == 0)
3248			return 0;
3249		offset += copy;
3250		from += copy;
3251	}
3252
3253	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3254		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3255		int end;
3256
3257		WARN_ON(start > offset + len);
3258
3259		end = start + skb_frag_size(frag);
3260		if ((copy = end - offset) > 0) {
3261			u32 p_off, p_len, copied;
3262			struct page *p;
3263			u8 *vaddr;
3264
3265			if (copy > len)
3266				copy = len;
3267
3268			skb_frag_foreach_page(frag,
3269					      skb_frag_off(frag) + offset - start,
3270					      copy, p, p_off, p_len, copied) {
3271				vaddr = kmap_atomic(p);
3272				memcpy(vaddr + p_off, from + copied, p_len);
3273				kunmap_atomic(vaddr);
3274			}
3275
3276			if ((len -= copy) == 0)
3277				return 0;
3278			offset += copy;
3279			from += copy;
3280		}
3281		start = end;
3282	}
3283
3284	skb_walk_frags(skb, frag_iter) {
3285		int end;
3286
3287		WARN_ON(start > offset + len);
3288
3289		end = start + frag_iter->len;
3290		if ((copy = end - offset) > 0) {
3291			if (copy > len)
3292				copy = len;
3293			if (skb_store_bits(frag_iter, offset - start,
3294					   from, copy))
3295				goto fault;
3296			if ((len -= copy) == 0)
3297				return 0;
3298			offset += copy;
3299			from += copy;
3300		}
3301		start = end;
3302	}
3303	if (!len)
3304		return 0;
3305
3306fault:
3307	return -EFAULT;
3308}
3309EXPORT_SYMBOL(skb_store_bits);
3310
3311/* Checksum skb data. */
3312__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3313		      __wsum csum, const struct skb_checksum_ops *ops)
3314{
3315	int start = skb_headlen(skb);
3316	int i, copy = start - offset;
3317	struct sk_buff *frag_iter;
3318	int pos = 0;
3319
3320	/* Checksum header. */
3321	if (copy > 0) {
3322		if (copy > len)
3323			copy = len;
3324		csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3325				       skb->data + offset, copy, csum);
3326		if ((len -= copy) == 0)
3327			return csum;
3328		offset += copy;
3329		pos	= copy;
3330	}
3331
3332	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3333		int end;
3334		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3335
3336		WARN_ON(start > offset + len);
3337
3338		end = start + skb_frag_size(frag);
3339		if ((copy = end - offset) > 0) {
3340			u32 p_off, p_len, copied;
3341			struct page *p;
3342			__wsum csum2;
3343			u8 *vaddr;
3344
3345			if (copy > len)
3346				copy = len;
3347
3348			skb_frag_foreach_page(frag,
3349					      skb_frag_off(frag) + offset - start,
3350					      copy, p, p_off, p_len, copied) {
3351				vaddr = kmap_atomic(p);
3352				csum2 = INDIRECT_CALL_1(ops->update,
3353							csum_partial_ext,
3354							vaddr + p_off, p_len, 0);
3355				kunmap_atomic(vaddr);
3356				csum = INDIRECT_CALL_1(ops->combine,
3357						       csum_block_add_ext, csum,
3358						       csum2, pos, p_len);
3359				pos += p_len;
3360			}
3361
3362			if (!(len -= copy))
3363				return csum;
3364			offset += copy;
3365		}
3366		start = end;
3367	}
3368
3369	skb_walk_frags(skb, frag_iter) {
3370		int end;
3371
3372		WARN_ON(start > offset + len);
3373
3374		end = start + frag_iter->len;
3375		if ((copy = end - offset) > 0) {
3376			__wsum csum2;
3377			if (copy > len)
3378				copy = len;
3379			csum2 = __skb_checksum(frag_iter, offset - start,
3380					       copy, 0, ops);
3381			csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3382					       csum, csum2, pos, copy);
3383			if ((len -= copy) == 0)
3384				return csum;
3385			offset += copy;
3386			pos    += copy;
3387		}
3388		start = end;
3389	}
3390	BUG_ON(len);
3391
3392	return csum;
3393}
3394EXPORT_SYMBOL(__skb_checksum);
3395
3396__wsum skb_checksum(const struct sk_buff *skb, int offset,
3397		    int len, __wsum csum)
3398{
3399	const struct skb_checksum_ops ops = {
3400		.update  = csum_partial_ext,
3401		.combine = csum_block_add_ext,
3402	};
3403
3404	return __skb_checksum(skb, offset, len, csum, &ops);
3405}
3406EXPORT_SYMBOL(skb_checksum);
3407
3408/* Both of above in one bottle. */
3409
3410__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3411				    u8 *to, int len)
3412{
3413	int start = skb_headlen(skb);
3414	int i, copy = start - offset;
3415	struct sk_buff *frag_iter;
3416	int pos = 0;
3417	__wsum csum = 0;
3418
3419	/* Copy header. */
3420	if (copy > 0) {
3421		if (copy > len)
3422			copy = len;
3423		csum = csum_partial_copy_nocheck(skb->data + offset, to,
3424						 copy);
3425		if ((len -= copy) == 0)
3426			return csum;
3427		offset += copy;
3428		to     += copy;
3429		pos	= copy;
3430	}
3431
3432	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3433		int end;
3434
3435		WARN_ON(start > offset + len);
3436
3437		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3438		if ((copy = end - offset) > 0) {
3439			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3440			u32 p_off, p_len, copied;
3441			struct page *p;
3442			__wsum csum2;
3443			u8 *vaddr;
3444
3445			if (copy > len)
3446				copy = len;
3447
3448			skb_frag_foreach_page(frag,
3449					      skb_frag_off(frag) + offset - start,
3450					      copy, p, p_off, p_len, copied) {
3451				vaddr = kmap_atomic(p);
3452				csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3453								  to + copied,
3454								  p_len);
3455				kunmap_atomic(vaddr);
3456				csum = csum_block_add(csum, csum2, pos);
3457				pos += p_len;
3458			}
3459
3460			if (!(len -= copy))
3461				return csum;
3462			offset += copy;
3463			to     += copy;
3464		}
3465		start = end;
3466	}
3467
3468	skb_walk_frags(skb, frag_iter) {
3469		__wsum csum2;
3470		int end;
3471
3472		WARN_ON(start > offset + len);
3473
3474		end = start + frag_iter->len;
3475		if ((copy = end - offset) > 0) {
3476			if (copy > len)
3477				copy = len;
3478			csum2 = skb_copy_and_csum_bits(frag_iter,
3479						       offset - start,
3480						       to, copy);
3481			csum = csum_block_add(csum, csum2, pos);
3482			if ((len -= copy) == 0)
3483				return csum;
3484			offset += copy;
3485			to     += copy;
3486			pos    += copy;
3487		}
3488		start = end;
3489	}
3490	BUG_ON(len);
3491	return csum;
3492}
3493EXPORT_SYMBOL(skb_copy_and_csum_bits);
3494
3495__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3496{
3497	__sum16 sum;
3498
3499	sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3500	/* See comments in __skb_checksum_complete(). */
3501	if (likely(!sum)) {
3502		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3503		    !skb->csum_complete_sw)
3504			netdev_rx_csum_fault(skb->dev, skb);
3505	}
3506	if (!skb_shared(skb))
3507		skb->csum_valid = !sum;
3508	return sum;
3509}
3510EXPORT_SYMBOL(__skb_checksum_complete_head);
3511
3512/* This function assumes skb->csum already holds pseudo header's checksum,
3513 * which has been changed from the hardware checksum, for example, by
3514 * __skb_checksum_validate_complete(). And, the original skb->csum must
3515 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3516 *
3517 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3518 * zero. The new checksum is stored back into skb->csum unless the skb is
3519 * shared.
3520 */
3521__sum16 __skb_checksum_complete(struct sk_buff *skb)
3522{
3523	__wsum csum;
3524	__sum16 sum;
3525
3526	csum = skb_checksum(skb, 0, skb->len, 0);
3527
3528	sum = csum_fold(csum_add(skb->csum, csum));
3529	/* This check is inverted, because we already knew the hardware
3530	 * checksum is invalid before calling this function. So, if the
3531	 * re-computed checksum is valid instead, then we have a mismatch
3532	 * between the original skb->csum and skb_checksum(). This means either
3533	 * the original hardware checksum is incorrect or we screw up skb->csum
3534	 * when moving skb->data around.
3535	 */
3536	if (likely(!sum)) {
3537		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3538		    !skb->csum_complete_sw)
3539			netdev_rx_csum_fault(skb->dev, skb);
3540	}
3541
3542	if (!skb_shared(skb)) {
3543		/* Save full packet checksum */
3544		skb->csum = csum;
3545		skb->ip_summed = CHECKSUM_COMPLETE;
3546		skb->csum_complete_sw = 1;
3547		skb->csum_valid = !sum;
3548	}
3549
3550	return sum;
3551}
3552EXPORT_SYMBOL(__skb_checksum_complete);
3553
3554static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3555{
3556	net_warn_ratelimited(
3557		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3558		__func__);
3559	return 0;
3560}
3561
3562static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3563				       int offset, int len)
3564{
3565	net_warn_ratelimited(
3566		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3567		__func__);
3568	return 0;
3569}
3570
3571static const struct skb_checksum_ops default_crc32c_ops = {
3572	.update  = warn_crc32c_csum_update,
3573	.combine = warn_crc32c_csum_combine,
3574};
3575
3576const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3577	&default_crc32c_ops;
3578EXPORT_SYMBOL(crc32c_csum_stub);
3579
3580 /**
3581 *	skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3582 *	@from: source buffer
3583 *
3584 *	Calculates the amount of linear headroom needed in the 'to' skb passed
3585 *	into skb_zerocopy().
3586 */
3587unsigned int
3588skb_zerocopy_headlen(const struct sk_buff *from)
3589{
3590	unsigned int hlen = 0;
3591
3592	if (!from->head_frag ||
3593	    skb_headlen(from) < L1_CACHE_BYTES ||
3594	    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3595		hlen = skb_headlen(from);
3596		if (!hlen)
3597			hlen = from->len;
3598	}
3599
3600	if (skb_has_frag_list(from))
3601		hlen = from->len;
3602
3603	return hlen;
3604}
3605EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3606
3607/**
3608 *	skb_zerocopy - Zero copy skb to skb
3609 *	@to: destination buffer
3610 *	@from: source buffer
3611 *	@len: number of bytes to copy from source buffer
3612 *	@hlen: size of linear headroom in destination buffer
3613 *
3614 *	Copies up to `len` bytes from `from` to `to` by creating references
3615 *	to the frags in the source buffer.
3616 *
3617 *	The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3618 *	headroom in the `to` buffer.
3619 *
3620 *	Return value:
3621 *	0: everything is OK
3622 *	-ENOMEM: couldn't orphan frags of @from due to lack of memory
3623 *	-EFAULT: skb_copy_bits() found some problem with skb geometry
3624 */
3625int
3626skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3627{
3628	int i, j = 0;
3629	int plen = 0; /* length of skb->head fragment */
3630	int ret;
3631	struct page *page;
3632	unsigned int offset;
3633
3634	BUG_ON(!from->head_frag && !hlen);
3635
3636	/* dont bother with small payloads */
3637	if (len <= skb_tailroom(to))
3638		return skb_copy_bits(from, 0, skb_put(to, len), len);
3639
3640	if (hlen) {
3641		ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3642		if (unlikely(ret))
3643			return ret;
3644		len -= hlen;
3645	} else {
3646		plen = min_t(int, skb_headlen(from), len);
3647		if (plen) {
3648			page = virt_to_head_page(from->head);
3649			offset = from->data - (unsigned char *)page_address(page);
3650			__skb_fill_page_desc(to, 0, page, offset, plen);
3651			get_page(page);
3652			j = 1;
3653			len -= plen;
3654		}
3655	}
3656
3657	skb_len_add(to, len + plen);
3658
3659	if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3660		skb_tx_error(from);
3661		return -ENOMEM;
3662	}
3663	skb_zerocopy_clone(to, from, GFP_ATOMIC);
3664
3665	for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3666		int size;
3667
3668		if (!len)
3669			break;
3670		skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3671		size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3672					len);
3673		skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3674		len -= size;
3675		skb_frag_ref(to, j);
3676		j++;
3677	}
3678	skb_shinfo(to)->nr_frags = j;
3679
3680	return 0;
3681}
3682EXPORT_SYMBOL_GPL(skb_zerocopy);
3683
3684void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3685{
3686	__wsum csum;
3687	long csstart;
3688
3689	if (skb->ip_summed == CHECKSUM_PARTIAL)
3690		csstart = skb_checksum_start_offset(skb);
3691	else
3692		csstart = skb_headlen(skb);
3693
3694	BUG_ON(csstart > skb_headlen(skb));
3695
3696	skb_copy_from_linear_data(skb, to, csstart);
3697
3698	csum = 0;
3699	if (csstart != skb->len)
3700		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3701					      skb->len - csstart);
3702
3703	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3704		long csstuff = csstart + skb->csum_offset;
3705
3706		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
3707	}
3708}
3709EXPORT_SYMBOL(skb_copy_and_csum_dev);
3710
3711/**
3712 *	skb_dequeue - remove from the head of the queue
3713 *	@list: list to dequeue from
3714 *
3715 *	Remove the head of the list. The list lock is taken so the function
3716 *	may be used safely with other locking list functions. The head item is
3717 *	returned or %NULL if the list is empty.
3718 */
3719
3720struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3721{
3722	unsigned long flags;
3723	struct sk_buff *result;
3724
3725	spin_lock_irqsave(&list->lock, flags);
3726	result = __skb_dequeue(list);
3727	spin_unlock_irqrestore(&list->lock, flags);
3728	return result;
3729}
3730EXPORT_SYMBOL(skb_dequeue);
3731
3732/**
3733 *	skb_dequeue_tail - remove from the tail of the queue
3734 *	@list: list to dequeue from
3735 *
3736 *	Remove the tail of the list. The list lock is taken so the function
3737 *	may be used safely with other locking list functions. The tail item is
3738 *	returned or %NULL if the list is empty.
3739 */
3740struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3741{
3742	unsigned long flags;
3743	struct sk_buff *result;
3744
3745	spin_lock_irqsave(&list->lock, flags);
3746	result = __skb_dequeue_tail(list);
3747	spin_unlock_irqrestore(&list->lock, flags);
3748	return result;
3749}
3750EXPORT_SYMBOL(skb_dequeue_tail);
3751
3752/**
3753 *	skb_queue_purge_reason - empty a list
3754 *	@list: list to empty
3755 *	@reason: drop reason
3756 *
3757 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
3758 *	the list and one reference dropped. This function takes the list
3759 *	lock and is atomic with respect to other list locking functions.
3760 */
3761void skb_queue_purge_reason(struct sk_buff_head *list,
3762			    enum skb_drop_reason reason)
3763{
3764	struct sk_buff_head tmp;
3765	unsigned long flags;
3766
3767	if (skb_queue_empty_lockless(list))
3768		return;
3769
3770	__skb_queue_head_init(&tmp);
3771
3772	spin_lock_irqsave(&list->lock, flags);
3773	skb_queue_splice_init(list, &tmp);
3774	spin_unlock_irqrestore(&list->lock, flags);
3775
3776	__skb_queue_purge_reason(&tmp, reason);
3777}
3778EXPORT_SYMBOL(skb_queue_purge_reason);
3779
3780/**
3781 *	skb_rbtree_purge - empty a skb rbtree
3782 *	@root: root of the rbtree to empty
3783 *	Return value: the sum of truesizes of all purged skbs.
3784 *
3785 *	Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3786 *	the list and one reference dropped. This function does not take
3787 *	any lock. Synchronization should be handled by the caller (e.g., TCP
3788 *	out-of-order queue is protected by the socket lock).
3789 */
3790unsigned int skb_rbtree_purge(struct rb_root *root)
3791{
3792	struct rb_node *p = rb_first(root);
3793	unsigned int sum = 0;
3794
3795	while (p) {
3796		struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3797
3798		p = rb_next(p);
3799		rb_erase(&skb->rbnode, root);
3800		sum += skb->truesize;
3801		kfree_skb(skb);
3802	}
3803	return sum;
3804}
3805
3806void skb_errqueue_purge(struct sk_buff_head *list)
3807{
3808	struct sk_buff *skb, *next;
3809	struct sk_buff_head kill;
3810	unsigned long flags;
3811
3812	__skb_queue_head_init(&kill);
3813
3814	spin_lock_irqsave(&list->lock, flags);
3815	skb_queue_walk_safe(list, skb, next) {
3816		if (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ZEROCOPY ||
3817		    SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING)
3818			continue;
3819		__skb_unlink(skb, list);
3820		__skb_queue_tail(&kill, skb);
3821	}
3822	spin_unlock_irqrestore(&list->lock, flags);
3823	__skb_queue_purge(&kill);
3824}
3825EXPORT_SYMBOL(skb_errqueue_purge);
3826
3827/**
3828 *	skb_queue_head - queue a buffer at the list head
3829 *	@list: list to use
3830 *	@newsk: buffer to queue
3831 *
3832 *	Queue a buffer at the start of the list. This function takes the
3833 *	list lock and can be used safely with other locking &sk_buff functions
3834 *	safely.
3835 *
3836 *	A buffer cannot be placed on two lists at the same time.
3837 */
3838void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3839{
3840	unsigned long flags;
3841
3842	spin_lock_irqsave(&list->lock, flags);
3843	__skb_queue_head(list, newsk);
3844	spin_unlock_irqrestore(&list->lock, flags);
3845}
3846EXPORT_SYMBOL(skb_queue_head);
3847
3848/**
3849 *	skb_queue_tail - queue a buffer at the list tail
3850 *	@list: list to use
3851 *	@newsk: buffer to queue
3852 *
3853 *	Queue a buffer at the tail of the list. This function takes the
3854 *	list lock and can be used safely with other locking &sk_buff functions
3855 *	safely.
3856 *
3857 *	A buffer cannot be placed on two lists at the same time.
3858 */
3859void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3860{
3861	unsigned long flags;
3862
3863	spin_lock_irqsave(&list->lock, flags);
3864	__skb_queue_tail(list, newsk);
3865	spin_unlock_irqrestore(&list->lock, flags);
3866}
3867EXPORT_SYMBOL(skb_queue_tail);
3868
3869/**
3870 *	skb_unlink	-	remove a buffer from a list
3871 *	@skb: buffer to remove
3872 *	@list: list to use
3873 *
3874 *	Remove a packet from a list. The list locks are taken and this
3875 *	function is atomic with respect to other list locked calls
3876 *
3877 *	You must know what list the SKB is on.
3878 */
3879void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3880{
3881	unsigned long flags;
3882
3883	spin_lock_irqsave(&list->lock, flags);
3884	__skb_unlink(skb, list);
3885	spin_unlock_irqrestore(&list->lock, flags);
3886}
3887EXPORT_SYMBOL(skb_unlink);
3888
3889/**
3890 *	skb_append	-	append a buffer
3891 *	@old: buffer to insert after
3892 *	@newsk: buffer to insert
3893 *	@list: list to use
3894 *
3895 *	Place a packet after a given packet in a list. The list locks are taken
3896 *	and this function is atomic with respect to other list locked calls.
3897 *	A buffer cannot be placed on two lists at the same time.
3898 */
3899void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3900{
3901	unsigned long flags;
3902
3903	spin_lock_irqsave(&list->lock, flags);
3904	__skb_queue_after(list, old, newsk);
3905	spin_unlock_irqrestore(&list->lock, flags);
3906}
3907EXPORT_SYMBOL(skb_append);
3908
3909static inline void skb_split_inside_header(struct sk_buff *skb,
3910					   struct sk_buff* skb1,
3911					   const u32 len, const int pos)
3912{
3913	int i;
3914
3915	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3916					 pos - len);
3917	/* And move data appendix as is. */
3918	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3919		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3920
3921	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3922	skb_shinfo(skb)->nr_frags  = 0;
3923	skb1->data_len		   = skb->data_len;
3924	skb1->len		   += skb1->data_len;
3925	skb->data_len		   = 0;
3926	skb->len		   = len;
3927	skb_set_tail_pointer(skb, len);
3928}
3929
3930static inline void skb_split_no_header(struct sk_buff *skb,
3931				       struct sk_buff* skb1,
3932				       const u32 len, int pos)
3933{
3934	int i, k = 0;
3935	const int nfrags = skb_shinfo(skb)->nr_frags;
3936
3937	skb_shinfo(skb)->nr_frags = 0;
3938	skb1->len		  = skb1->data_len = skb->len - len;
3939	skb->len		  = len;
3940	skb->data_len		  = len - pos;
3941
3942	for (i = 0; i < nfrags; i++) {
3943		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3944
3945		if (pos + size > len) {
3946			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3947
3948			if (pos < len) {
3949				/* Split frag.
3950				 * We have two variants in this case:
3951				 * 1. Move all the frag to the second
3952				 *    part, if it is possible. F.e.
3953				 *    this approach is mandatory for TUX,
3954				 *    where splitting is expensive.
3955				 * 2. Split is accurately. We make this.
3956				 */
3957				skb_frag_ref(skb, i);
3958				skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3959				skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3960				skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3961				skb_shinfo(skb)->nr_frags++;
3962			}
3963			k++;
3964		} else
3965			skb_shinfo(skb)->nr_frags++;
3966		pos += size;
3967	}
3968	skb_shinfo(skb1)->nr_frags = k;
3969}
3970
3971/**
3972 * skb_split - Split fragmented skb to two parts at length len.
3973 * @skb: the buffer to split
3974 * @skb1: the buffer to receive the second part
3975 * @len: new length for skb
3976 */
3977void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3978{
3979	int pos = skb_headlen(skb);
3980	const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
3981
3982	skb_zcopy_downgrade_managed(skb);
3983
3984	skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
3985	skb_zerocopy_clone(skb1, skb, 0);
3986	if (len < pos)	/* Split line is inside header. */
3987		skb_split_inside_header(skb, skb1, len, pos);
3988	else		/* Second chunk has no header, nothing to copy. */
3989		skb_split_no_header(skb, skb1, len, pos);
3990}
3991EXPORT_SYMBOL(skb_split);
3992
3993/* Shifting from/to a cloned skb is a no-go.
3994 *
3995 * Caller cannot keep skb_shinfo related pointers past calling here!
3996 */
3997static int skb_prepare_for_shift(struct sk_buff *skb)
3998{
3999	return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
4000}
4001
4002/**
4003 * skb_shift - Shifts paged data partially from skb to another
4004 * @tgt: buffer into which tail data gets added
4005 * @skb: buffer from which the paged data comes from
4006 * @shiftlen: shift up to this many bytes
4007 *
4008 * Attempts to shift up to shiftlen worth of bytes, which may be less than
4009 * the length of the skb, from skb to tgt. Returns number bytes shifted.
4010 * It's up to caller to free skb if everything was shifted.
4011 *
4012 * If @tgt runs out of frags, the whole operation is aborted.
4013 *
4014 * Skb cannot include anything else but paged data while tgt is allowed
4015 * to have non-paged data as well.
4016 *
4017 * TODO: full sized shift could be optimized but that would need
4018 * specialized skb free'er to handle frags without up-to-date nr_frags.
4019 */
4020int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
4021{
4022	int from, to, merge, todo;
4023	skb_frag_t *fragfrom, *fragto;
4024
4025	BUG_ON(shiftlen > skb->len);
4026
4027	if (skb_headlen(skb))
4028		return 0;
4029	if (skb_zcopy(tgt) || skb_zcopy(skb))
4030		return 0;
4031
4032	todo = shiftlen;
4033	from = 0;
4034	to = skb_shinfo(tgt)->nr_frags;
4035	fragfrom = &skb_shinfo(skb)->frags[from];
4036
4037	/* Actual merge is delayed until the point when we know we can
4038	 * commit all, so that we don't have to undo partial changes
4039	 */
4040	if (!to ||
4041	    !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
4042			      skb_frag_off(fragfrom))) {
4043		merge = -1;
4044	} else {
4045		merge = to - 1;
4046
4047		todo -= skb_frag_size(fragfrom);
4048		if (todo < 0) {
4049			if (skb_prepare_for_shift(skb) ||
4050			    skb_prepare_for_shift(tgt))
4051				return 0;
4052
4053			/* All previous frag pointers might be stale! */
4054			fragfrom = &skb_shinfo(skb)->frags[from];
4055			fragto = &skb_shinfo(tgt)->frags[merge];
4056
4057			skb_frag_size_add(fragto, shiftlen);
4058			skb_frag_size_sub(fragfrom, shiftlen);
4059			skb_frag_off_add(fragfrom, shiftlen);
4060
4061			goto onlymerged;
4062		}
4063
4064		from++;
4065	}
4066
4067	/* Skip full, not-fitting skb to avoid expensive operations */
4068	if ((shiftlen == skb->len) &&
4069	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
4070		return 0;
4071
4072	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
4073		return 0;
4074
4075	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
4076		if (to == MAX_SKB_FRAGS)
4077			return 0;
4078
4079		fragfrom = &skb_shinfo(skb)->frags[from];
4080		fragto = &skb_shinfo(tgt)->frags[to];
4081
4082		if (todo >= skb_frag_size(fragfrom)) {
4083			*fragto = *fragfrom;
4084			todo -= skb_frag_size(fragfrom);
4085			from++;
4086			to++;
4087
4088		} else {
4089			__skb_frag_ref(fragfrom);
4090			skb_frag_page_copy(fragto, fragfrom);
4091			skb_frag_off_copy(fragto, fragfrom);
4092			skb_frag_size_set(fragto, todo);
4093
4094			skb_frag_off_add(fragfrom, todo);
4095			skb_frag_size_sub(fragfrom, todo);
4096			todo = 0;
4097
4098			to++;
4099			break;
4100		}
4101	}
4102
4103	/* Ready to "commit" this state change to tgt */
4104	skb_shinfo(tgt)->nr_frags = to;
4105
4106	if (merge >= 0) {
4107		fragfrom = &skb_shinfo(skb)->frags[0];
4108		fragto = &skb_shinfo(tgt)->frags[merge];
4109
4110		skb_frag_size_add(fragto, skb_frag_size(fragfrom));
4111		__skb_frag_unref(fragfrom, skb->pp_recycle);
4112	}
4113
4114	/* Reposition in the original skb */
4115	to = 0;
4116	while (from < skb_shinfo(skb)->nr_frags)
4117		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
4118	skb_shinfo(skb)->nr_frags = to;
4119
4120	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
4121
4122onlymerged:
4123	/* Most likely the tgt won't ever need its checksum anymore, skb on
4124	 * the other hand might need it if it needs to be resent
4125	 */
4126	tgt->ip_summed = CHECKSUM_PARTIAL;
4127	skb->ip_summed = CHECKSUM_PARTIAL;
4128
4129	skb_len_add(skb, -shiftlen);
4130	skb_len_add(tgt, shiftlen);
4131
4132	return shiftlen;
4133}
4134
4135/**
4136 * skb_prepare_seq_read - Prepare a sequential read of skb data
4137 * @skb: the buffer to read
4138 * @from: lower offset of data to be read
4139 * @to: upper offset of data to be read
4140 * @st: state variable
4141 *
4142 * Initializes the specified state variable. Must be called before
4143 * invoking skb_seq_read() for the first time.
4144 */
4145void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
4146			  unsigned int to, struct skb_seq_state *st)
4147{
4148	st->lower_offset = from;
4149	st->upper_offset = to;
4150	st->root_skb = st->cur_skb = skb;
4151	st->frag_idx = st->stepped_offset = 0;
4152	st->frag_data = NULL;
4153	st->frag_off = 0;
4154}
4155EXPORT_SYMBOL(skb_prepare_seq_read);
4156
4157/**
4158 * skb_seq_read - Sequentially read skb data
4159 * @consumed: number of bytes consumed by the caller so far
4160 * @data: destination pointer for data to be returned
4161 * @st: state variable
4162 *
4163 * Reads a block of skb data at @consumed relative to the
4164 * lower offset specified to skb_prepare_seq_read(). Assigns
4165 * the head of the data block to @data and returns the length
4166 * of the block or 0 if the end of the skb data or the upper
4167 * offset has been reached.
4168 *
4169 * The caller is not required to consume all of the data
4170 * returned, i.e. @consumed is typically set to the number
4171 * of bytes already consumed and the next call to
4172 * skb_seq_read() will return the remaining part of the block.
4173 *
4174 * Note 1: The size of each block of data returned can be arbitrary,
4175 *       this limitation is the cost for zerocopy sequential
4176 *       reads of potentially non linear data.
4177 *
4178 * Note 2: Fragment lists within fragments are not implemented
4179 *       at the moment, state->root_skb could be replaced with
4180 *       a stack for this purpose.
4181 */
4182unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
4183			  struct skb_seq_state *st)
4184{
4185	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
4186	skb_frag_t *frag;
4187
4188	if (unlikely(abs_offset >= st->upper_offset)) {
4189		if (st->frag_data) {
4190			kunmap_atomic(st->frag_data);
4191			st->frag_data = NULL;
4192		}
4193		return 0;
4194	}
4195
4196next_skb:
4197	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
4198
4199	if (abs_offset < block_limit && !st->frag_data) {
4200		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
4201		return block_limit - abs_offset;
4202	}
4203
4204	if (st->frag_idx == 0 && !st->frag_data)
4205		st->stepped_offset += skb_headlen(st->cur_skb);
4206
4207	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
4208		unsigned int pg_idx, pg_off, pg_sz;
4209
4210		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
4211
4212		pg_idx = 0;
4213		pg_off = skb_frag_off(frag);
4214		pg_sz = skb_frag_size(frag);
4215
4216		if (skb_frag_must_loop(skb_frag_page(frag))) {
4217			pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
4218			pg_off = offset_in_page(pg_off + st->frag_off);
4219			pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
4220						    PAGE_SIZE - pg_off);
4221		}
4222
4223		block_limit = pg_sz + st->stepped_offset;
4224		if (abs_offset < block_limit) {
4225			if (!st->frag_data)
4226				st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
4227
4228			*data = (u8 *)st->frag_data + pg_off +
4229				(abs_offset - st->stepped_offset);
4230
4231			return block_limit - abs_offset;
4232		}
4233
4234		if (st->frag_data) {
4235			kunmap_atomic(st->frag_data);
4236			st->frag_data = NULL;
4237		}
4238
4239		st->stepped_offset += pg_sz;
4240		st->frag_off += pg_sz;
4241		if (st->frag_off == skb_frag_size(frag)) {
4242			st->frag_off = 0;
4243			st->frag_idx++;
4244		}
4245	}
4246
4247	if (st->frag_data) {
4248		kunmap_atomic(st->frag_data);
4249		st->frag_data = NULL;
4250	}
4251
4252	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
4253		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
4254		st->frag_idx = 0;
4255		goto next_skb;
4256	} else if (st->cur_skb->next) {
4257		st->cur_skb = st->cur_skb->next;
4258		st->frag_idx = 0;
4259		goto next_skb;
4260	}
4261
4262	return 0;
4263}
4264EXPORT_SYMBOL(skb_seq_read);
4265
4266/**
4267 * skb_abort_seq_read - Abort a sequential read of skb data
4268 * @st: state variable
4269 *
4270 * Must be called if skb_seq_read() was not called until it
4271 * returned 0.
4272 */
4273void skb_abort_seq_read(struct skb_seq_state *st)
4274{
4275	if (st->frag_data)
4276		kunmap_atomic(st->frag_data);
4277}
4278EXPORT_SYMBOL(skb_abort_seq_read);
4279
4280#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
4281
4282static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
4283					  struct ts_config *conf,
4284					  struct ts_state *state)
4285{
4286	return skb_seq_read(offset, text, TS_SKB_CB(state));
4287}
4288
4289static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
4290{
4291	skb_abort_seq_read(TS_SKB_CB(state));
4292}
4293
4294/**
4295 * skb_find_text - Find a text pattern in skb data
4296 * @skb: the buffer to look in
4297 * @from: search offset
4298 * @to: search limit
4299 * @config: textsearch configuration
4300 *
4301 * Finds a pattern in the skb data according to the specified
4302 * textsearch configuration. Use textsearch_next() to retrieve
4303 * subsequent occurrences of the pattern. Returns the offset
4304 * to the first occurrence or UINT_MAX if no match was found.
4305 */
4306unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4307			   unsigned int to, struct ts_config *config)
4308{
4309	unsigned int patlen = config->ops->get_pattern_len(config);
4310	struct ts_state state;
4311	unsigned int ret;
4312
4313	BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4314
4315	config->get_next_block = skb_ts_get_next_block;
4316	config->finish = skb_ts_finish;
4317
4318	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4319
4320	ret = textsearch_find(config, &state);
4321	return (ret + patlen <= to - from ? ret : UINT_MAX);
4322}
4323EXPORT_SYMBOL(skb_find_text);
4324
4325int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4326			 int offset, size_t size, size_t max_frags)
4327{
4328	int i = skb_shinfo(skb)->nr_frags;
4329
4330	if (skb_can_coalesce(skb, i, page, offset)) {
4331		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4332	} else if (i < max_frags) {
4333		skb_zcopy_downgrade_managed(skb);
4334		get_page(page);
4335		skb_fill_page_desc_noacc(skb, i, page, offset, size);
4336	} else {
4337		return -EMSGSIZE;
4338	}
4339
4340	return 0;
4341}
4342EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4343
4344/**
4345 *	skb_pull_rcsum - pull skb and update receive checksum
4346 *	@skb: buffer to update
4347 *	@len: length of data pulled
4348 *
4349 *	This function performs an skb_pull on the packet and updates
4350 *	the CHECKSUM_COMPLETE checksum.  It should be used on
4351 *	receive path processing instead of skb_pull unless you know
4352 *	that the checksum difference is zero (e.g., a valid IP header)
4353 *	or you are setting ip_summed to CHECKSUM_NONE.
4354 */
4355void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4356{
4357	unsigned char *data = skb->data;
4358
4359	BUG_ON(len > skb->len);
4360	__skb_pull(skb, len);
4361	skb_postpull_rcsum(skb, data, len);
4362	return skb->data;
4363}
4364EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4365
4366static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4367{
4368	skb_frag_t head_frag;
4369	struct page *page;
4370
4371	page = virt_to_head_page(frag_skb->head);
4372	skb_frag_fill_page_desc(&head_frag, page, frag_skb->data -
4373				(unsigned char *)page_address(page),
4374				skb_headlen(frag_skb));
4375	return head_frag;
4376}
4377
4378struct sk_buff *skb_segment_list(struct sk_buff *skb,
4379				 netdev_features_t features,
4380				 unsigned int offset)
4381{
4382	struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4383	unsigned int tnl_hlen = skb_tnl_header_len(skb);
4384	unsigned int delta_truesize = 0;
4385	unsigned int delta_len = 0;
4386	struct sk_buff *tail = NULL;
4387	struct sk_buff *nskb, *tmp;
4388	int len_diff, err;
4389
4390	skb_push(skb, -skb_network_offset(skb) + offset);
4391
4392	/* Ensure the head is writeable before touching the shared info */
4393	err = skb_unclone(skb, GFP_ATOMIC);
4394	if (err)
4395		goto err_linearize;
4396
4397	skb_shinfo(skb)->frag_list = NULL;
4398
4399	while (list_skb) {
4400		nskb = list_skb;
4401		list_skb = list_skb->next;
4402
4403		err = 0;
4404		delta_truesize += nskb->truesize;
4405		if (skb_shared(nskb)) {
4406			tmp = skb_clone(nskb, GFP_ATOMIC);
4407			if (tmp) {
4408				consume_skb(nskb);
4409				nskb = tmp;
4410				err = skb_unclone(nskb, GFP_ATOMIC);
4411			} else {
4412				err = -ENOMEM;
4413			}
4414		}
4415
4416		if (!tail)
4417			skb->next = nskb;
4418		else
4419			tail->next = nskb;
4420
4421		if (unlikely(err)) {
4422			nskb->next = list_skb;
4423			goto err_linearize;
4424		}
4425
4426		tail = nskb;
4427
4428		delta_len += nskb->len;
4429
4430		skb_push(nskb, -skb_network_offset(nskb) + offset);
4431
4432		skb_release_head_state(nskb);
4433		len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4434		__copy_skb_header(nskb, skb);
4435
4436		skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4437		nskb->transport_header += len_diff;
4438		skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4439						 nskb->data - tnl_hlen,
4440						 offset + tnl_hlen);
4441
4442		if (skb_needs_linearize(nskb, features) &&
4443		    __skb_linearize(nskb))
4444			goto err_linearize;
4445	}
4446
4447	skb->truesize = skb->truesize - delta_truesize;
4448	skb->data_len = skb->data_len - delta_len;
4449	skb->len = skb->len - delta_len;
4450
4451	skb_gso_reset(skb);
4452
4453	skb->prev = tail;
4454
4455	if (skb_needs_linearize(skb, features) &&
4456	    __skb_linearize(skb))
4457		goto err_linearize;
4458
4459	skb_get(skb);
4460
4461	return skb;
4462
4463err_linearize:
4464	kfree_skb_list(skb->next);
4465	skb->next = NULL;
4466	return ERR_PTR(-ENOMEM);
4467}
4468EXPORT_SYMBOL_GPL(skb_segment_list);
4469
4470/**
4471 *	skb_segment - Perform protocol segmentation on skb.
4472 *	@head_skb: buffer to segment
4473 *	@features: features for the output path (see dev->features)
4474 *
4475 *	This function performs segmentation on the given skb.  It returns
4476 *	a pointer to the first in a list of new skbs for the segments.
4477 *	In case of error it returns ERR_PTR(err).
4478 */
4479struct sk_buff *skb_segment(struct sk_buff *head_skb,
4480			    netdev_features_t features)
4481{
4482	struct sk_buff *segs = NULL;
4483	struct sk_buff *tail = NULL;
4484	struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4485	unsigned int mss = skb_shinfo(head_skb)->gso_size;
4486	unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4487	unsigned int offset = doffset;
4488	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4489	unsigned int partial_segs = 0;
4490	unsigned int headroom;
4491	unsigned int len = head_skb->len;
4492	struct sk_buff *frag_skb;
4493	skb_frag_t *frag;
4494	__be16 proto;
4495	bool csum, sg;
4496	int err = -ENOMEM;
4497	int i = 0;
4498	int nfrags, pos;
4499
4500	if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4501	    mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4502		struct sk_buff *check_skb;
4503
4504		for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4505			if (skb_headlen(check_skb) && !check_skb->head_frag) {
4506				/* gso_size is untrusted, and we have a frag_list with
4507				 * a linear non head_frag item.
4508				 *
4509				 * If head_skb's headlen does not fit requested gso_size,
4510				 * it means that the frag_list members do NOT terminate
4511				 * on exact gso_size boundaries. Hence we cannot perform
4512				 * skb_frag_t page sharing. Therefore we must fallback to
4513				 * copying the frag_list skbs; we do so by disabling SG.
4514				 */
4515				features &= ~NETIF_F_SG;
4516				break;
4517			}
4518		}
4519	}
4520
4521	__skb_push(head_skb, doffset);
4522	proto = skb_network_protocol(head_skb, NULL);
4523	if (unlikely(!proto))
4524		return ERR_PTR(-EINVAL);
4525
4526	sg = !!(features & NETIF_F_SG);
4527	csum = !!can_checksum_protocol(features, proto);
4528
4529	if (sg && csum && (mss != GSO_BY_FRAGS))  {
4530		if (!(features & NETIF_F_GSO_PARTIAL)) {
4531			struct sk_buff *iter;
4532			unsigned int frag_len;
4533
4534			if (!list_skb ||
4535			    !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4536				goto normal;
4537
4538			/* If we get here then all the required
4539			 * GSO features except frag_list are supported.
4540			 * Try to split the SKB to multiple GSO SKBs
4541			 * with no frag_list.
4542			 * Currently we can do that only when the buffers don't
4543			 * have a linear part and all the buffers except
4544			 * the last are of the same length.
4545			 */
4546			frag_len = list_skb->len;
4547			skb_walk_frags(head_skb, iter) {
4548				if (frag_len != iter->len && iter->next)
4549					goto normal;
4550				if (skb_headlen(iter) && !iter->head_frag)
4551					goto normal;
4552
4553				len -= iter->len;
4554			}
4555
4556			if (len != frag_len)
4557				goto normal;
4558		}
4559
4560		/* GSO partial only requires that we trim off any excess that
4561		 * doesn't fit into an MSS sized block, so take care of that
4562		 * now.
4563		 * Cap len to not accidentally hit GSO_BY_FRAGS.
4564		 */
4565		partial_segs = min(len, GSO_BY_FRAGS - 1) / mss;
4566		if (partial_segs > 1)
4567			mss *= partial_segs;
4568		else
4569			partial_segs = 0;
4570	}
4571
4572normal:
4573	headroom = skb_headroom(head_skb);
4574	pos = skb_headlen(head_skb);
4575
4576	if (skb_orphan_frags(head_skb, GFP_ATOMIC))
4577		return ERR_PTR(-ENOMEM);
4578
4579	nfrags = skb_shinfo(head_skb)->nr_frags;
4580	frag = skb_shinfo(head_skb)->frags;
4581	frag_skb = head_skb;
4582
4583	do {
4584		struct sk_buff *nskb;
4585		skb_frag_t *nskb_frag;
4586		int hsize;
4587		int size;
4588
4589		if (unlikely(mss == GSO_BY_FRAGS)) {
4590			len = list_skb->len;
4591		} else {
4592			len = head_skb->len - offset;
4593			if (len > mss)
4594				len = mss;
4595		}
4596
4597		hsize = skb_headlen(head_skb) - offset;
4598
4599		if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4600		    (skb_headlen(list_skb) == len || sg)) {
4601			BUG_ON(skb_headlen(list_skb) > len);
4602
4603			nskb = skb_clone(list_skb, GFP_ATOMIC);
4604			if (unlikely(!nskb))
4605				goto err;
4606
4607			i = 0;
4608			nfrags = skb_shinfo(list_skb)->nr_frags;
4609			frag = skb_shinfo(list_skb)->frags;
4610			frag_skb = list_skb;
4611			pos += skb_headlen(list_skb);
4612
4613			while (pos < offset + len) {
4614				BUG_ON(i >= nfrags);
4615
4616				size = skb_frag_size(frag);
4617				if (pos + size > offset + len)
4618					break;
4619
4620				i++;
4621				pos += size;
4622				frag++;
4623			}
4624
4625			list_skb = list_skb->next;
4626
4627			if (unlikely(pskb_trim(nskb, len))) {
4628				kfree_skb(nskb);
4629				goto err;
4630			}
4631
4632			hsize = skb_end_offset(nskb);
4633			if (skb_cow_head(nskb, doffset + headroom)) {
4634				kfree_skb(nskb);
4635				goto err;
4636			}
4637
4638			nskb->truesize += skb_end_offset(nskb) - hsize;
4639			skb_release_head_state(nskb);
4640			__skb_push(nskb, doffset);
4641		} else {
4642			if (hsize < 0)
4643				hsize = 0;
4644			if (hsize > len || !sg)
4645				hsize = len;
4646
4647			nskb = __alloc_skb(hsize + doffset + headroom,
4648					   GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4649					   NUMA_NO_NODE);
4650
4651			if (unlikely(!nskb))
4652				goto err;
4653
4654			skb_reserve(nskb, headroom);
4655			__skb_put(nskb, doffset);
4656		}
4657
4658		if (segs)
4659			tail->next = nskb;
4660		else
4661			segs = nskb;
4662		tail = nskb;
4663
4664		__copy_skb_header(nskb, head_skb);
4665
4666		skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4667		skb_reset_mac_len(nskb);
4668
4669		skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4670						 nskb->data - tnl_hlen,
4671						 doffset + tnl_hlen);
4672
4673		if (nskb->len == len + doffset)
4674			goto perform_csum_check;
4675
4676		if (!sg) {
4677			if (!csum) {
4678				if (!nskb->remcsum_offload)
4679					nskb->ip_summed = CHECKSUM_NONE;
4680				SKB_GSO_CB(nskb)->csum =
4681					skb_copy_and_csum_bits(head_skb, offset,
4682							       skb_put(nskb,
4683								       len),
4684							       len);
4685				SKB_GSO_CB(nskb)->csum_start =
4686					skb_headroom(nskb) + doffset;
4687			} else {
4688				if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4689					goto err;
4690			}
4691			continue;
4692		}
4693
4694		nskb_frag = skb_shinfo(nskb)->frags;
4695
4696		skb_copy_from_linear_data_offset(head_skb, offset,
4697						 skb_put(nskb, hsize), hsize);
4698
4699		skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4700					   SKBFL_SHARED_FRAG;
4701
4702		if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4703			goto err;
4704
4705		while (pos < offset + len) {
4706			if (i >= nfrags) {
4707				if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
4708				    skb_zerocopy_clone(nskb, list_skb,
4709						       GFP_ATOMIC))
4710					goto err;
4711
4712				i = 0;
4713				nfrags = skb_shinfo(list_skb)->nr_frags;
4714				frag = skb_shinfo(list_skb)->frags;
4715				frag_skb = list_skb;
4716				if (!skb_headlen(list_skb)) {
4717					BUG_ON(!nfrags);
4718				} else {
4719					BUG_ON(!list_skb->head_frag);
4720
4721					/* to make room for head_frag. */
4722					i--;
4723					frag--;
4724				}
4725
4726				list_skb = list_skb->next;
4727			}
4728
4729			if (unlikely(skb_shinfo(nskb)->nr_frags >=
4730				     MAX_SKB_FRAGS)) {
4731				net_warn_ratelimited(
4732					"skb_segment: too many frags: %u %u\n",
4733					pos, mss);
4734				err = -EINVAL;
4735				goto err;
4736			}
4737
4738			*nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4739			__skb_frag_ref(nskb_frag);
4740			size = skb_frag_size(nskb_frag);
4741
4742			if (pos < offset) {
4743				skb_frag_off_add(nskb_frag, offset - pos);
4744				skb_frag_size_sub(nskb_frag, offset - pos);
4745			}
4746
4747			skb_shinfo(nskb)->nr_frags++;
4748
4749			if (pos + size <= offset + len) {
4750				i++;
4751				frag++;
4752				pos += size;
4753			} else {
4754				skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4755				goto skip_fraglist;
4756			}
4757
4758			nskb_frag++;
4759		}
4760
4761skip_fraglist:
4762		nskb->data_len = len - hsize;
4763		nskb->len += nskb->data_len;
4764		nskb->truesize += nskb->data_len;
4765
4766perform_csum_check:
4767		if (!csum) {
4768			if (skb_has_shared_frag(nskb) &&
4769			    __skb_linearize(nskb))
4770				goto err;
4771
4772			if (!nskb->remcsum_offload)
4773				nskb->ip_summed = CHECKSUM_NONE;
4774			SKB_GSO_CB(nskb)->csum =
4775				skb_checksum(nskb, doffset,
4776					     nskb->len - doffset, 0);
4777			SKB_GSO_CB(nskb)->csum_start =
4778				skb_headroom(nskb) + doffset;
4779		}
4780	} while ((offset += len) < head_skb->len);
4781
4782	/* Some callers want to get the end of the list.
4783	 * Put it in segs->prev to avoid walking the list.
4784	 * (see validate_xmit_skb_list() for example)
4785	 */
4786	segs->prev = tail;
4787
4788	if (partial_segs) {
4789		struct sk_buff *iter;
4790		int type = skb_shinfo(head_skb)->gso_type;
4791		unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4792
4793		/* Update type to add partial and then remove dodgy if set */
4794		type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4795		type &= ~SKB_GSO_DODGY;
4796
4797		/* Update GSO info and prepare to start updating headers on
4798		 * our way back down the stack of protocols.
4799		 */
4800		for (iter = segs; iter; iter = iter->next) {
4801			skb_shinfo(iter)->gso_size = gso_size;
4802			skb_shinfo(iter)->gso_segs = partial_segs;
4803			skb_shinfo(iter)->gso_type = type;
4804			SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4805		}
4806
4807		if (tail->len - doffset <= gso_size)
4808			skb_shinfo(tail)->gso_size = 0;
4809		else if (tail != segs)
4810			skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4811	}
4812
4813	/* Following permits correct backpressure, for protocols
4814	 * using skb_set_owner_w().
4815	 * Idea is to tranfert ownership from head_skb to last segment.
4816	 */
4817	if (head_skb->destructor == sock_wfree) {
4818		swap(tail->truesize, head_skb->truesize);
4819		swap(tail->destructor, head_skb->destructor);
4820		swap(tail->sk, head_skb->sk);
4821	}
4822	return segs;
4823
4824err:
4825	kfree_skb_list(segs);
4826	return ERR_PTR(err);
4827}
4828EXPORT_SYMBOL_GPL(skb_segment);
4829
4830#ifdef CONFIG_SKB_EXTENSIONS
4831#define SKB_EXT_ALIGN_VALUE	8
4832#define SKB_EXT_CHUNKSIZEOF(x)	(ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4833
4834static const u8 skb_ext_type_len[] = {
4835#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4836	[SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4837#endif
4838#ifdef CONFIG_XFRM
4839	[SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4840#endif
4841#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4842	[TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4843#endif
4844#if IS_ENABLED(CONFIG_MPTCP)
4845	[SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4846#endif
4847#if IS_ENABLED(CONFIG_MCTP_FLOWS)
4848	[SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
4849#endif
4850};
4851
4852static __always_inline unsigned int skb_ext_total_length(void)
4853{
4854	unsigned int l = SKB_EXT_CHUNKSIZEOF(struct skb_ext);
4855	int i;
4856
4857	for (i = 0; i < ARRAY_SIZE(skb_ext_type_len); i++)
4858		l += skb_ext_type_len[i];
4859
4860	return l;
4861}
4862
4863static void skb_extensions_init(void)
4864{
4865	BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4866#if !IS_ENABLED(CONFIG_KCOV_INSTRUMENT_ALL)
4867	BUILD_BUG_ON(skb_ext_total_length() > 255);
4868#endif
4869
4870	skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4871					     SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4872					     0,
4873					     SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4874					     NULL);
4875}
4876#else
4877static void skb_extensions_init(void) {}
4878#endif
4879
4880/* The SKB kmem_cache slab is critical for network performance.  Never
4881 * merge/alias the slab with similar sized objects.  This avoids fragmentation
4882 * that hurts performance of kmem_cache_{alloc,free}_bulk APIs.
4883 */
4884#ifndef CONFIG_SLUB_TINY
4885#define FLAG_SKB_NO_MERGE	SLAB_NO_MERGE
4886#else /* CONFIG_SLUB_TINY - simple loop in kmem_cache_alloc_bulk */
4887#define FLAG_SKB_NO_MERGE	0
4888#endif
4889
4890void __init skb_init(void)
4891{
4892	skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4893					      sizeof(struct sk_buff),
4894					      0,
4895					      SLAB_HWCACHE_ALIGN|SLAB_PANIC|
4896						FLAG_SKB_NO_MERGE,
4897					      offsetof(struct sk_buff, cb),
4898					      sizeof_field(struct sk_buff, cb),
4899					      NULL);
4900	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4901						sizeof(struct sk_buff_fclones),
4902						0,
4903						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4904						NULL);
4905	/* usercopy should only access first SKB_SMALL_HEAD_HEADROOM bytes.
4906	 * struct skb_shared_info is located at the end of skb->head,
4907	 * and should not be copied to/from user.
4908	 */
4909	skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
4910						SKB_SMALL_HEAD_CACHE_SIZE,
4911						0,
4912						SLAB_HWCACHE_ALIGN | SLAB_PANIC,
4913						0,
4914						SKB_SMALL_HEAD_HEADROOM,
4915						NULL);
4916	skb_extensions_init();
4917}
4918
4919static int
4920__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4921	       unsigned int recursion_level)
4922{
4923	int start = skb_headlen(skb);
4924	int i, copy = start - offset;
4925	struct sk_buff *frag_iter;
4926	int elt = 0;
4927
4928	if (unlikely(recursion_level >= 24))
4929		return -EMSGSIZE;
4930
4931	if (copy > 0) {
4932		if (copy > len)
4933			copy = len;
4934		sg_set_buf(sg, skb->data + offset, copy);
4935		elt++;
4936		if ((len -= copy) == 0)
4937			return elt;
4938		offset += copy;
4939	}
4940
4941	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4942		int end;
4943
4944		WARN_ON(start > offset + len);
4945
4946		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4947		if ((copy = end - offset) > 0) {
4948			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4949			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4950				return -EMSGSIZE;
4951
4952			if (copy > len)
4953				copy = len;
4954			sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4955				    skb_frag_off(frag) + offset - start);
4956			elt++;
4957			if (!(len -= copy))
4958				return elt;
4959			offset += copy;
4960		}
4961		start = end;
4962	}
4963
4964	skb_walk_frags(skb, frag_iter) {
4965		int end, ret;
4966
4967		WARN_ON(start > offset + len);
4968
4969		end = start + frag_iter->len;
4970		if ((copy = end - offset) > 0) {
4971			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4972				return -EMSGSIZE;
4973
4974			if (copy > len)
4975				copy = len;
4976			ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4977					      copy, recursion_level + 1);
4978			if (unlikely(ret < 0))
4979				return ret;
4980			elt += ret;
4981			if ((len -= copy) == 0)
4982				return elt;
4983			offset += copy;
4984		}
4985		start = end;
4986	}
4987	BUG_ON(len);
4988	return elt;
4989}
4990
4991/**
4992 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4993 *	@skb: Socket buffer containing the buffers to be mapped
4994 *	@sg: The scatter-gather list to map into
4995 *	@offset: The offset into the buffer's contents to start mapping
4996 *	@len: Length of buffer space to be mapped
4997 *
4998 *	Fill the specified scatter-gather list with mappings/pointers into a
4999 *	region of the buffer space attached to a socket buffer. Returns either
5000 *	the number of scatterlist items used, or -EMSGSIZE if the contents
5001 *	could not fit.
5002 */
5003int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
5004{
5005	int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
5006
5007	if (nsg <= 0)
5008		return nsg;
5009
5010	sg_mark_end(&sg[nsg - 1]);
5011
5012	return nsg;
5013}
5014EXPORT_SYMBOL_GPL(skb_to_sgvec);
5015
5016/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
5017 * sglist without mark the sg which contain last skb data as the end.
5018 * So the caller can mannipulate sg list as will when padding new data after
5019 * the first call without calling sg_unmark_end to expend sg list.
5020 *
5021 * Scenario to use skb_to_sgvec_nomark:
5022 * 1. sg_init_table
5023 * 2. skb_to_sgvec_nomark(payload1)
5024 * 3. skb_to_sgvec_nomark(payload2)
5025 *
5026 * This is equivalent to:
5027 * 1. sg_init_table
5028 * 2. skb_to_sgvec(payload1)
5029 * 3. sg_unmark_end
5030 * 4. skb_to_sgvec(payload2)
5031 *
5032 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
5033 * is more preferable.
5034 */
5035int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
5036			int offset, int len)
5037{
5038	return __skb_to_sgvec(skb, sg, offset, len, 0);
5039}
5040EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
5041
5042
5043
5044/**
5045 *	skb_cow_data - Check that a socket buffer's data buffers are writable
5046 *	@skb: The socket buffer to check.
5047 *	@tailbits: Amount of trailing space to be added
5048 *	@trailer: Returned pointer to the skb where the @tailbits space begins
5049 *
5050 *	Make sure that the data buffers attached to a socket buffer are
5051 *	writable. If they are not, private copies are made of the data buffers
5052 *	and the socket buffer is set to use these instead.
5053 *
5054 *	If @tailbits is given, make sure that there is space to write @tailbits
5055 *	bytes of data beyond current end of socket buffer.  @trailer will be
5056 *	set to point to the skb in which this space begins.
5057 *
5058 *	The number of scatterlist elements required to completely map the
5059 *	COW'd and extended socket buffer will be returned.
5060 */
5061int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
5062{
5063	int copyflag;
5064	int elt;
5065	struct sk_buff *skb1, **skb_p;
5066
5067	/* If skb is cloned or its head is paged, reallocate
5068	 * head pulling out all the pages (pages are considered not writable
5069	 * at the moment even if they are anonymous).
5070	 */
5071	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
5072	    !__pskb_pull_tail(skb, __skb_pagelen(skb)))
5073		return -ENOMEM;
5074
5075	/* Easy case. Most of packets will go this way. */
5076	if (!skb_has_frag_list(skb)) {
5077		/* A little of trouble, not enough of space for trailer.
5078		 * This should not happen, when stack is tuned to generate
5079		 * good frames. OK, on miss we reallocate and reserve even more
5080		 * space, 128 bytes is fair. */
5081
5082		if (skb_tailroom(skb) < tailbits &&
5083		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
5084			return -ENOMEM;
5085
5086		/* Voila! */
5087		*trailer = skb;
5088		return 1;
5089	}
5090
5091	/* Misery. We are in troubles, going to mincer fragments... */
5092
5093	elt = 1;
5094	skb_p = &skb_shinfo(skb)->frag_list;
5095	copyflag = 0;
5096
5097	while ((skb1 = *skb_p) != NULL) {
5098		int ntail = 0;
5099
5100		/* The fragment is partially pulled by someone,
5101		 * this can happen on input. Copy it and everything
5102		 * after it. */
5103
5104		if (skb_shared(skb1))
5105			copyflag = 1;
5106
5107		/* If the skb is the last, worry about trailer. */
5108
5109		if (skb1->next == NULL && tailbits) {
5110			if (skb_shinfo(skb1)->nr_frags ||
5111			    skb_has_frag_list(skb1) ||
5112			    skb_tailroom(skb1) < tailbits)
5113				ntail = tailbits + 128;
5114		}
5115
5116		if (copyflag ||
5117		    skb_cloned(skb1) ||
5118		    ntail ||
5119		    skb_shinfo(skb1)->nr_frags ||
5120		    skb_has_frag_list(skb1)) {
5121			struct sk_buff *skb2;
5122
5123			/* Fuck, we are miserable poor guys... */
5124			if (ntail == 0)
5125				skb2 = skb_copy(skb1, GFP_ATOMIC);
5126			else
5127				skb2 = skb_copy_expand(skb1,
5128						       skb_headroom(skb1),
5129						       ntail,
5130						       GFP_ATOMIC);
5131			if (unlikely(skb2 == NULL))
5132				return -ENOMEM;
5133
5134			if (skb1->sk)
5135				skb_set_owner_w(skb2, skb1->sk);
5136
5137			/* Looking around. Are we still alive?
5138			 * OK, link new skb, drop old one */
5139
5140			skb2->next = skb1->next;
5141			*skb_p = skb2;
5142			kfree_skb(skb1);
5143			skb1 = skb2;
5144		}
5145		elt++;
5146		*trailer = skb1;
5147		skb_p = &skb1->next;
5148	}
5149
5150	return elt;
5151}
5152EXPORT_SYMBOL_GPL(skb_cow_data);
5153
5154static void sock_rmem_free(struct sk_buff *skb)
5155{
5156	struct sock *sk = skb->sk;
5157
5158	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
5159}
5160
5161static void skb_set_err_queue(struct sk_buff *skb)
5162{
5163	/* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
5164	 * So, it is safe to (mis)use it to mark skbs on the error queue.
5165	 */
5166	skb->pkt_type = PACKET_OUTGOING;
5167	BUILD_BUG_ON(PACKET_OUTGOING == 0);
5168}
5169
5170/*
5171 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
5172 */
5173int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
5174{
5175	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
5176	    (unsigned int)READ_ONCE(sk->sk_rcvbuf))
5177		return -ENOMEM;
5178
5179	skb_orphan(skb);
5180	skb->sk = sk;
5181	skb->destructor = sock_rmem_free;
5182	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
5183	skb_set_err_queue(skb);
5184
5185	/* before exiting rcu section, make sure dst is refcounted */
5186	skb_dst_force(skb);
5187
5188	skb_queue_tail(&sk->sk_error_queue, skb);
5189	if (!sock_flag(sk, SOCK_DEAD))
5190		sk_error_report(sk);
5191	return 0;
5192}
5193EXPORT_SYMBOL(sock_queue_err_skb);
5194
5195static bool is_icmp_err_skb(const struct sk_buff *skb)
5196{
5197	return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
5198		       SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
5199}
5200
5201struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
5202{
5203	struct sk_buff_head *q = &sk->sk_error_queue;
5204	struct sk_buff *skb, *skb_next = NULL;
5205	bool icmp_next = false;
5206	unsigned long flags;
5207
5208	if (skb_queue_empty_lockless(q))
5209		return NULL;
5210
5211	spin_lock_irqsave(&q->lock, flags);
5212	skb = __skb_dequeue(q);
5213	if (skb && (skb_next = skb_peek(q))) {
5214		icmp_next = is_icmp_err_skb(skb_next);
5215		if (icmp_next)
5216			sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
5217	}
5218	spin_unlock_irqrestore(&q->lock, flags);
5219
5220	if (is_icmp_err_skb(skb) && !icmp_next)
5221		sk->sk_err = 0;
5222
5223	if (skb_next)
5224		sk_error_report(sk);
5225
5226	return skb;
5227}
5228EXPORT_SYMBOL(sock_dequeue_err_skb);
5229
5230/**
5231 * skb_clone_sk - create clone of skb, and take reference to socket
5232 * @skb: the skb to clone
5233 *
5234 * This function creates a clone of a buffer that holds a reference on
5235 * sk_refcnt.  Buffers created via this function are meant to be
5236 * returned using sock_queue_err_skb, or free via kfree_skb.
5237 *
5238 * When passing buffers allocated with this function to sock_queue_err_skb
5239 * it is necessary to wrap the call with sock_hold/sock_put in order to
5240 * prevent the socket from being released prior to being enqueued on
5241 * the sk_error_queue.
5242 */
5243struct sk_buff *skb_clone_sk(struct sk_buff *skb)
5244{
5245	struct sock *sk = skb->sk;
5246	struct sk_buff *clone;
5247
5248	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
5249		return NULL;
5250
5251	clone = skb_clone(skb, GFP_ATOMIC);
5252	if (!clone) {
5253		sock_put(sk);
5254		return NULL;
5255	}
5256
5257	clone->sk = sk;
5258	clone->destructor = sock_efree;
5259
5260	return clone;
5261}
5262EXPORT_SYMBOL(skb_clone_sk);
5263
5264static void __skb_complete_tx_timestamp(struct sk_buff *skb,
5265					struct sock *sk,
5266					int tstype,
5267					bool opt_stats)
5268{
5269	struct sock_exterr_skb *serr;
5270	int err;
5271
5272	BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
5273
5274	serr = SKB_EXT_ERR(skb);
5275	memset(serr, 0, sizeof(*serr));
5276	serr->ee.ee_errno = ENOMSG;
5277	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
5278	serr->ee.ee_info = tstype;
5279	serr->opt_stats = opt_stats;
5280	serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
5281	if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
5282		serr->ee.ee_data = skb_shinfo(skb)->tskey;
5283		if (sk_is_tcp(sk))
5284			serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
5285	}
5286
5287	err = sock_queue_err_skb(sk, skb);
5288
5289	if (err)
5290		kfree_skb(skb);
5291}
5292
5293static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
5294{
5295	bool ret;
5296
5297	if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
5298		return true;
5299
5300	read_lock_bh(&sk->sk_callback_lock);
5301	ret = sk->sk_socket && sk->sk_socket->file &&
5302	      file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
5303	read_unlock_bh(&sk->sk_callback_lock);
5304	return ret;
5305}
5306
5307void skb_complete_tx_timestamp(struct sk_buff *skb,
5308			       struct skb_shared_hwtstamps *hwtstamps)
5309{
5310	struct sock *sk = skb->sk;
5311
5312	if (!skb_may_tx_timestamp(sk, false))
5313		goto err;
5314
5315	/* Take a reference to prevent skb_orphan() from freeing the socket,
5316	 * but only if the socket refcount is not zero.
5317	 */
5318	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5319		*skb_hwtstamps(skb) = *hwtstamps;
5320		__skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5321		sock_put(sk);
5322		return;
5323	}
5324
5325err:
5326	kfree_skb(skb);
5327}
5328EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5329
5330void __skb_tstamp_tx(struct sk_buff *orig_skb,
5331		     const struct sk_buff *ack_skb,
5332		     struct skb_shared_hwtstamps *hwtstamps,
5333		     struct sock *sk, int tstype)
5334{
5335	struct sk_buff *skb;
5336	bool tsonly, opt_stats = false;
5337	u32 tsflags;
5338
5339	if (!sk)
5340		return;
5341
5342	tsflags = READ_ONCE(sk->sk_tsflags);
5343	if (!hwtstamps && !(tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5344	    skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5345		return;
5346
5347	tsonly = tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5348	if (!skb_may_tx_timestamp(sk, tsonly))
5349		return;
5350
5351	if (tsonly) {
5352#ifdef CONFIG_INET
5353		if ((tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5354		    sk_is_tcp(sk)) {
5355			skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5356							     ack_skb);
5357			opt_stats = true;
5358		} else
5359#endif
5360			skb = alloc_skb(0, GFP_ATOMIC);
5361	} else {
5362		skb = skb_clone(orig_skb, GFP_ATOMIC);
5363
5364		if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
5365			kfree_skb(skb);
5366			return;
5367		}
5368	}
5369	if (!skb)
5370		return;
5371
5372	if (tsonly) {
5373		skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5374					     SKBTX_ANY_TSTAMP;
5375		skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5376	}
5377
5378	if (hwtstamps)
5379		*skb_hwtstamps(skb) = *hwtstamps;
5380	else
5381		__net_timestamp(skb);
5382
5383	__skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5384}
5385EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5386
5387void skb_tstamp_tx(struct sk_buff *orig_skb,
5388		   struct skb_shared_hwtstamps *hwtstamps)
5389{
5390	return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5391			       SCM_TSTAMP_SND);
5392}
5393EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5394
5395#ifdef CONFIG_WIRELESS
5396void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5397{
5398	struct sock *sk = skb->sk;
5399	struct sock_exterr_skb *serr;
5400	int err = 1;
5401
5402	skb->wifi_acked_valid = 1;
5403	skb->wifi_acked = acked;
5404
5405	serr = SKB_EXT_ERR(skb);
5406	memset(serr, 0, sizeof(*serr));
5407	serr->ee.ee_errno = ENOMSG;
5408	serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5409
5410	/* Take a reference to prevent skb_orphan() from freeing the socket,
5411	 * but only if the socket refcount is not zero.
5412	 */
5413	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5414		err = sock_queue_err_skb(sk, skb);
5415		sock_put(sk);
5416	}
5417	if (err)
5418		kfree_skb(skb);
5419}
5420EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5421#endif /* CONFIG_WIRELESS */
5422
5423/**
5424 * skb_partial_csum_set - set up and verify partial csum values for packet
5425 * @skb: the skb to set
5426 * @start: the number of bytes after skb->data to start checksumming.
5427 * @off: the offset from start to place the checksum.
5428 *
5429 * For untrusted partially-checksummed packets, we need to make sure the values
5430 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5431 *
5432 * This function checks and sets those values and skb->ip_summed: if this
5433 * returns false you should drop the packet.
5434 */
5435bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5436{
5437	u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5438	u32 csum_start = skb_headroom(skb) + (u32)start;
5439
5440	if (unlikely(csum_start >= U16_MAX || csum_end > skb_headlen(skb))) {
5441		net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5442				     start, off, skb_headroom(skb), skb_headlen(skb));
5443		return false;
5444	}
5445	skb->ip_summed = CHECKSUM_PARTIAL;
5446	skb->csum_start = csum_start;
5447	skb->csum_offset = off;
5448	skb->transport_header = csum_start;
5449	return true;
5450}
5451EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5452
5453static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5454			       unsigned int max)
5455{
5456	if (skb_headlen(skb) >= len)
5457		return 0;
5458
5459	/* If we need to pullup then pullup to the max, so we
5460	 * won't need to do it again.
5461	 */
5462	if (max > skb->len)
5463		max = skb->len;
5464
5465	if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5466		return -ENOMEM;
5467
5468	if (skb_headlen(skb) < len)
5469		return -EPROTO;
5470
5471	return 0;
5472}
5473
5474#define MAX_TCP_HDR_LEN (15 * 4)
5475
5476static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5477				      typeof(IPPROTO_IP) proto,
5478				      unsigned int off)
5479{
5480	int err;
5481
5482	switch (proto) {
5483	case IPPROTO_TCP:
5484		err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5485					  off + MAX_TCP_HDR_LEN);
5486		if (!err && !skb_partial_csum_set(skb, off,
5487						  offsetof(struct tcphdr,
5488							   check)))
5489			err = -EPROTO;
5490		return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5491
5492	case IPPROTO_UDP:
5493		err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5494					  off + sizeof(struct udphdr));
5495		if (!err && !skb_partial_csum_set(skb, off,
5496						  offsetof(struct udphdr,
5497							   check)))
5498			err = -EPROTO;
5499		return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5500	}
5501
5502	return ERR_PTR(-EPROTO);
5503}
5504
5505/* This value should be large enough to cover a tagged ethernet header plus
5506 * maximally sized IP and TCP or UDP headers.
5507 */
5508#define MAX_IP_HDR_LEN 128
5509
5510static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5511{
5512	unsigned int off;
5513	bool fragment;
5514	__sum16 *csum;
5515	int err;
5516
5517	fragment = false;
5518
5519	err = skb_maybe_pull_tail(skb,
5520				  sizeof(struct iphdr),
5521				  MAX_IP_HDR_LEN);
5522	if (err < 0)
5523		goto out;
5524
5525	if (ip_is_fragment(ip_hdr(skb)))
5526		fragment = true;
5527
5528	off = ip_hdrlen(skb);
5529
5530	err = -EPROTO;
5531
5532	if (fragment)
5533		goto out;
5534
5535	csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5536	if (IS_ERR(csum))
5537		return PTR_ERR(csum);
5538
5539	if (recalculate)
5540		*csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5541					   ip_hdr(skb)->daddr,
5542					   skb->len - off,
5543					   ip_hdr(skb)->protocol, 0);
5544	err = 0;
5545
5546out:
5547	return err;
5548}
5549
5550/* This value should be large enough to cover a tagged ethernet header plus
5551 * an IPv6 header, all options, and a maximal TCP or UDP header.
5552 */
5553#define MAX_IPV6_HDR_LEN 256
5554
5555#define OPT_HDR(type, skb, off) \
5556	(type *)(skb_network_header(skb) + (off))
5557
5558static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5559{
5560	int err;
5561	u8 nexthdr;
5562	unsigned int off;
5563	unsigned int len;
5564	bool fragment;
5565	bool done;
5566	__sum16 *csum;
5567
5568	fragment = false;
5569	done = false;
5570
5571	off = sizeof(struct ipv6hdr);
5572
5573	err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5574	if (err < 0)
5575		goto out;
5576
5577	nexthdr = ipv6_hdr(skb)->nexthdr;
5578
5579	len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5580	while (off <= len && !done) {
5581		switch (nexthdr) {
5582		case IPPROTO_DSTOPTS:
5583		case IPPROTO_HOPOPTS:
5584		case IPPROTO_ROUTING: {
5585			struct ipv6_opt_hdr *hp;
5586
5587			err = skb_maybe_pull_tail(skb,
5588						  off +
5589						  sizeof(struct ipv6_opt_hdr),
5590						  MAX_IPV6_HDR_LEN);
5591			if (err < 0)
5592				goto out;
5593
5594			hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5595			nexthdr = hp->nexthdr;
5596			off += ipv6_optlen(hp);
5597			break;
5598		}
5599		case IPPROTO_AH: {
5600			struct ip_auth_hdr *hp;
5601
5602			err = skb_maybe_pull_tail(skb,
5603						  off +
5604						  sizeof(struct ip_auth_hdr),
5605						  MAX_IPV6_HDR_LEN);
5606			if (err < 0)
5607				goto out;
5608
5609			hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5610			nexthdr = hp->nexthdr;
5611			off += ipv6_authlen(hp);
5612			break;
5613		}
5614		case IPPROTO_FRAGMENT: {
5615			struct frag_hdr *hp;
5616
5617			err = skb_maybe_pull_tail(skb,
5618						  off +
5619						  sizeof(struct frag_hdr),
5620						  MAX_IPV6_HDR_LEN);
5621			if (err < 0)
5622				goto out;
5623
5624			hp = OPT_HDR(struct frag_hdr, skb, off);
5625
5626			if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5627				fragment = true;
5628
5629			nexthdr = hp->nexthdr;
5630			off += sizeof(struct frag_hdr);
5631			break;
5632		}
5633		default:
5634			done = true;
5635			break;
5636		}
5637	}
5638
5639	err = -EPROTO;
5640
5641	if (!done || fragment)
5642		goto out;
5643
5644	csum = skb_checksum_setup_ip(skb, nexthdr, off);
5645	if (IS_ERR(csum))
5646		return PTR_ERR(csum);
5647
5648	if (recalculate)
5649		*csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5650					 &ipv6_hdr(skb)->daddr,
5651					 skb->len - off, nexthdr, 0);
5652	err = 0;
5653
5654out:
5655	return err;
5656}
5657
5658/**
5659 * skb_checksum_setup - set up partial checksum offset
5660 * @skb: the skb to set up
5661 * @recalculate: if true the pseudo-header checksum will be recalculated
5662 */
5663int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5664{
5665	int err;
5666
5667	switch (skb->protocol) {
5668	case htons(ETH_P_IP):
5669		err = skb_checksum_setup_ipv4(skb, recalculate);
5670		break;
5671
5672	case htons(ETH_P_IPV6):
5673		err = skb_checksum_setup_ipv6(skb, recalculate);
5674		break;
5675
5676	default:
5677		err = -EPROTO;
5678		break;
5679	}
5680
5681	return err;
5682}
5683EXPORT_SYMBOL(skb_checksum_setup);
5684
5685/**
5686 * skb_checksum_maybe_trim - maybe trims the given skb
5687 * @skb: the skb to check
5688 * @transport_len: the data length beyond the network header
5689 *
5690 * Checks whether the given skb has data beyond the given transport length.
5691 * If so, returns a cloned skb trimmed to this transport length.
5692 * Otherwise returns the provided skb. Returns NULL in error cases
5693 * (e.g. transport_len exceeds skb length or out-of-memory).
5694 *
5695 * Caller needs to set the skb transport header and free any returned skb if it
5696 * differs from the provided skb.
5697 */
5698static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5699					       unsigned int transport_len)
5700{
5701	struct sk_buff *skb_chk;
5702	unsigned int len = skb_transport_offset(skb) + transport_len;
5703	int ret;
5704
5705	if (skb->len < len)
5706		return NULL;
5707	else if (skb->len == len)
5708		return skb;
5709
5710	skb_chk = skb_clone(skb, GFP_ATOMIC);
5711	if (!skb_chk)
5712		return NULL;
5713
5714	ret = pskb_trim_rcsum(skb_chk, len);
5715	if (ret) {
5716		kfree_skb(skb_chk);
5717		return NULL;
5718	}
5719
5720	return skb_chk;
5721}
5722
5723/**
5724 * skb_checksum_trimmed - validate checksum of an skb
5725 * @skb: the skb to check
5726 * @transport_len: the data length beyond the network header
5727 * @skb_chkf: checksum function to use
5728 *
5729 * Applies the given checksum function skb_chkf to the provided skb.
5730 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5731 *
5732 * If the skb has data beyond the given transport length, then a
5733 * trimmed & cloned skb is checked and returned.
5734 *
5735 * Caller needs to set the skb transport header and free any returned skb if it
5736 * differs from the provided skb.
5737 */
5738struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5739				     unsigned int transport_len,
5740				     __sum16(*skb_chkf)(struct sk_buff *skb))
5741{
5742	struct sk_buff *skb_chk;
5743	unsigned int offset = skb_transport_offset(skb);
5744	__sum16 ret;
5745
5746	skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5747	if (!skb_chk)
5748		goto err;
5749
5750	if (!pskb_may_pull(skb_chk, offset))
5751		goto err;
5752
5753	skb_pull_rcsum(skb_chk, offset);
5754	ret = skb_chkf(skb_chk);
5755	skb_push_rcsum(skb_chk, offset);
5756
5757	if (ret)
5758		goto err;
5759
5760	return skb_chk;
5761
5762err:
5763	if (skb_chk && skb_chk != skb)
5764		kfree_skb(skb_chk);
5765
5766	return NULL;
5767
5768}
5769EXPORT_SYMBOL(skb_checksum_trimmed);
5770
5771void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5772{
5773	net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5774			     skb->dev->name);
5775}
5776EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5777
5778void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5779{
5780	if (head_stolen) {
5781		skb_release_head_state(skb);
5782		kmem_cache_free(skbuff_cache, skb);
5783	} else {
5784		__kfree_skb(skb);
5785	}
5786}
5787EXPORT_SYMBOL(kfree_skb_partial);
5788
5789/**
5790 * skb_try_coalesce - try to merge skb to prior one
5791 * @to: prior buffer
5792 * @from: buffer to add
5793 * @fragstolen: pointer to boolean
5794 * @delta_truesize: how much more was allocated than was requested
5795 */
5796bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5797		      bool *fragstolen, int *delta_truesize)
5798{
5799	struct skb_shared_info *to_shinfo, *from_shinfo;
5800	int i, delta, len = from->len;
5801
5802	*fragstolen = false;
5803
5804	if (skb_cloned(to))
5805		return false;
5806
5807	/* In general, avoid mixing page_pool and non-page_pool allocated
5808	 * pages within the same SKB. In theory we could take full
5809	 * references if @from is cloned and !@to->pp_recycle but its
5810	 * tricky (due to potential race with the clone disappearing) and
5811	 * rare, so not worth dealing with.
5812	 */
5813	if (to->pp_recycle != from->pp_recycle)
5814		return false;
5815
5816	if (len <= skb_tailroom(to)) {
5817		if (len)
5818			BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5819		*delta_truesize = 0;
5820		return true;
5821	}
5822
5823	to_shinfo = skb_shinfo(to);
5824	from_shinfo = skb_shinfo(from);
5825	if (to_shinfo->frag_list || from_shinfo->frag_list)
5826		return false;
5827	if (skb_zcopy(to) || skb_zcopy(from))
5828		return false;
5829
5830	if (skb_headlen(from) != 0) {
5831		struct page *page;
5832		unsigned int offset;
5833
5834		if (to_shinfo->nr_frags +
5835		    from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5836			return false;
5837
5838		if (skb_head_is_locked(from))
5839			return false;
5840
5841		delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5842
5843		page = virt_to_head_page(from->head);
5844		offset = from->data - (unsigned char *)page_address(page);
5845
5846		skb_fill_page_desc(to, to_shinfo->nr_frags,
5847				   page, offset, skb_headlen(from));
5848		*fragstolen = true;
5849	} else {
5850		if (to_shinfo->nr_frags +
5851		    from_shinfo->nr_frags > MAX_SKB_FRAGS)
5852			return false;
5853
5854		delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5855	}
5856
5857	WARN_ON_ONCE(delta < len);
5858
5859	memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5860	       from_shinfo->frags,
5861	       from_shinfo->nr_frags * sizeof(skb_frag_t));
5862	to_shinfo->nr_frags += from_shinfo->nr_frags;
5863
5864	if (!skb_cloned(from))
5865		from_shinfo->nr_frags = 0;
5866
5867	/* if the skb is not cloned this does nothing
5868	 * since we set nr_frags to 0.
5869	 */
5870	if (skb_pp_frag_ref(from)) {
5871		for (i = 0; i < from_shinfo->nr_frags; i++)
5872			__skb_frag_ref(&from_shinfo->frags[i]);
5873	}
5874
5875	to->truesize += delta;
5876	to->len += len;
5877	to->data_len += len;
5878
5879	*delta_truesize = delta;
5880	return true;
5881}
5882EXPORT_SYMBOL(skb_try_coalesce);
5883
5884/**
5885 * skb_scrub_packet - scrub an skb
5886 *
5887 * @skb: buffer to clean
5888 * @xnet: packet is crossing netns
5889 *
5890 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5891 * into/from a tunnel. Some information have to be cleared during these
5892 * operations.
5893 * skb_scrub_packet can also be used to clean a skb before injecting it in
5894 * another namespace (@xnet == true). We have to clear all information in the
5895 * skb that could impact namespace isolation.
5896 */
5897void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5898{
5899	skb->pkt_type = PACKET_HOST;
5900	skb->skb_iif = 0;
5901	skb->ignore_df = 0;
5902	skb_dst_drop(skb);
5903	skb_ext_reset(skb);
5904	nf_reset_ct(skb);
5905	nf_reset_trace(skb);
5906
5907#ifdef CONFIG_NET_SWITCHDEV
5908	skb->offload_fwd_mark = 0;
5909	skb->offload_l3_fwd_mark = 0;
5910#endif
5911
5912	if (!xnet)
5913		return;
5914
5915	ipvs_reset(skb);
5916	skb->mark = 0;
5917	skb_clear_tstamp(skb);
5918}
5919EXPORT_SYMBOL_GPL(skb_scrub_packet);
5920
5921static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5922{
5923	int mac_len, meta_len;
5924	void *meta;
5925
5926	if (skb_cow(skb, skb_headroom(skb)) < 0) {
5927		kfree_skb(skb);
5928		return NULL;
5929	}
5930
5931	mac_len = skb->data - skb_mac_header(skb);
5932	if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5933		memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5934			mac_len - VLAN_HLEN - ETH_TLEN);
5935	}
5936
5937	meta_len = skb_metadata_len(skb);
5938	if (meta_len) {
5939		meta = skb_metadata_end(skb) - meta_len;
5940		memmove(meta + VLAN_HLEN, meta, meta_len);
5941	}
5942
5943	skb->mac_header += VLAN_HLEN;
5944	return skb;
5945}
5946
5947struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5948{
5949	struct vlan_hdr *vhdr;
5950	u16 vlan_tci;
5951
5952	if (unlikely(skb_vlan_tag_present(skb))) {
5953		/* vlan_tci is already set-up so leave this for another time */
5954		return skb;
5955	}
5956
5957	skb = skb_share_check(skb, GFP_ATOMIC);
5958	if (unlikely(!skb))
5959		goto err_free;
5960	/* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5961	if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5962		goto err_free;
5963
5964	vhdr = (struct vlan_hdr *)skb->data;
5965	vlan_tci = ntohs(vhdr->h_vlan_TCI);
5966	__vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5967
5968	skb_pull_rcsum(skb, VLAN_HLEN);
5969	vlan_set_encap_proto(skb, vhdr);
5970
5971	skb = skb_reorder_vlan_header(skb);
5972	if (unlikely(!skb))
5973		goto err_free;
5974
5975	skb_reset_network_header(skb);
5976	if (!skb_transport_header_was_set(skb))
5977		skb_reset_transport_header(skb);
5978	skb_reset_mac_len(skb);
5979
5980	return skb;
5981
5982err_free:
5983	kfree_skb(skb);
5984	return NULL;
5985}
5986EXPORT_SYMBOL(skb_vlan_untag);
5987
5988int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
5989{
5990	if (!pskb_may_pull(skb, write_len))
5991		return -ENOMEM;
5992
5993	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5994		return 0;
5995
5996	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5997}
5998EXPORT_SYMBOL(skb_ensure_writable);
5999
6000int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev)
6001{
6002	int needed_headroom = dev->needed_headroom;
6003	int needed_tailroom = dev->needed_tailroom;
6004
6005	/* For tail taggers, we need to pad short frames ourselves, to ensure
6006	 * that the tail tag does not fail at its role of being at the end of
6007	 * the packet, once the conduit interface pads the frame. Account for
6008	 * that pad length here, and pad later.
6009	 */
6010	if (unlikely(needed_tailroom && skb->len < ETH_ZLEN))
6011		needed_tailroom += ETH_ZLEN - skb->len;
6012	/* skb_headroom() returns unsigned int... */
6013	needed_headroom = max_t(int, needed_headroom - skb_headroom(skb), 0);
6014	needed_tailroom = max_t(int, needed_tailroom - skb_tailroom(skb), 0);
6015
6016	if (likely(!needed_headroom && !needed_tailroom && !skb_cloned(skb)))
6017		/* No reallocation needed, yay! */
6018		return 0;
6019
6020	return pskb_expand_head(skb, needed_headroom, needed_tailroom,
6021				GFP_ATOMIC);
6022}
6023EXPORT_SYMBOL(skb_ensure_writable_head_tail);
6024
6025/* remove VLAN header from packet and update csum accordingly.
6026 * expects a non skb_vlan_tag_present skb with a vlan tag payload
6027 */
6028int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
6029{
6030	int offset = skb->data - skb_mac_header(skb);
6031	int err;
6032
6033	if (WARN_ONCE(offset,
6034		      "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
6035		      offset)) {
6036		return -EINVAL;
6037	}
6038
6039	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
6040	if (unlikely(err))
6041		return err;
6042
6043	skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6044
6045	vlan_remove_tag(skb, vlan_tci);
6046
6047	skb->mac_header += VLAN_HLEN;
6048
6049	if (skb_network_offset(skb) < ETH_HLEN)
6050		skb_set_network_header(skb, ETH_HLEN);
6051
6052	skb_reset_mac_len(skb);
6053
6054	return err;
6055}
6056EXPORT_SYMBOL(__skb_vlan_pop);
6057
6058/* Pop a vlan tag either from hwaccel or from payload.
6059 * Expects skb->data at mac header.
6060 */
6061int skb_vlan_pop(struct sk_buff *skb)
6062{
6063	u16 vlan_tci;
6064	__be16 vlan_proto;
6065	int err;
6066
6067	if (likely(skb_vlan_tag_present(skb))) {
6068		__vlan_hwaccel_clear_tag(skb);
6069	} else {
6070		if (unlikely(!eth_type_vlan(skb->protocol)))
6071			return 0;
6072
6073		err = __skb_vlan_pop(skb, &vlan_tci);
6074		if (err)
6075			return err;
6076	}
6077	/* move next vlan tag to hw accel tag */
6078	if (likely(!eth_type_vlan(skb->protocol)))
6079		return 0;
6080
6081	vlan_proto = skb->protocol;
6082	err = __skb_vlan_pop(skb, &vlan_tci);
6083	if (unlikely(err))
6084		return err;
6085
6086	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6087	return 0;
6088}
6089EXPORT_SYMBOL(skb_vlan_pop);
6090
6091/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
6092 * Expects skb->data at mac header.
6093 */
6094int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
6095{
6096	if (skb_vlan_tag_present(skb)) {
6097		int offset = skb->data - skb_mac_header(skb);
6098		int err;
6099
6100		if (WARN_ONCE(offset,
6101			      "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
6102			      offset)) {
6103			return -EINVAL;
6104		}
6105
6106		err = __vlan_insert_tag(skb, skb->vlan_proto,
6107					skb_vlan_tag_get(skb));
6108		if (err)
6109			return err;
6110
6111		skb->protocol = skb->vlan_proto;
6112		skb->mac_len += VLAN_HLEN;
6113
6114		skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6115	}
6116	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6117	return 0;
6118}
6119EXPORT_SYMBOL(skb_vlan_push);
6120
6121/**
6122 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
6123 *
6124 * @skb: Socket buffer to modify
6125 *
6126 * Drop the Ethernet header of @skb.
6127 *
6128 * Expects that skb->data points to the mac header and that no VLAN tags are
6129 * present.
6130 *
6131 * Returns 0 on success, -errno otherwise.
6132 */
6133int skb_eth_pop(struct sk_buff *skb)
6134{
6135	if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
6136	    skb_network_offset(skb) < ETH_HLEN)
6137		return -EPROTO;
6138
6139	skb_pull_rcsum(skb, ETH_HLEN);
6140	skb_reset_mac_header(skb);
6141	skb_reset_mac_len(skb);
6142
6143	return 0;
6144}
6145EXPORT_SYMBOL(skb_eth_pop);
6146
6147/**
6148 * skb_eth_push() - Add a new Ethernet header at the head of a packet
6149 *
6150 * @skb: Socket buffer to modify
6151 * @dst: Destination MAC address of the new header
6152 * @src: Source MAC address of the new header
6153 *
6154 * Prepend @skb with a new Ethernet header.
6155 *
6156 * Expects that skb->data points to the mac header, which must be empty.
6157 *
6158 * Returns 0 on success, -errno otherwise.
6159 */
6160int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
6161		 const unsigned char *src)
6162{
6163	struct ethhdr *eth;
6164	int err;
6165
6166	if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
6167		return -EPROTO;
6168
6169	err = skb_cow_head(skb, sizeof(*eth));
6170	if (err < 0)
6171		return err;
6172
6173	skb_push(skb, sizeof(*eth));
6174	skb_reset_mac_header(skb);
6175	skb_reset_mac_len(skb);
6176
6177	eth = eth_hdr(skb);
6178	ether_addr_copy(eth->h_dest, dst);
6179	ether_addr_copy(eth->h_source, src);
6180	eth->h_proto = skb->protocol;
6181
6182	skb_postpush_rcsum(skb, eth, sizeof(*eth));
6183
6184	return 0;
6185}
6186EXPORT_SYMBOL(skb_eth_push);
6187
6188/* Update the ethertype of hdr and the skb csum value if required. */
6189static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
6190			     __be16 ethertype)
6191{
6192	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6193		__be16 diff[] = { ~hdr->h_proto, ethertype };
6194
6195		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6196	}
6197
6198	hdr->h_proto = ethertype;
6199}
6200
6201/**
6202 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6203 *                   the packet
6204 *
6205 * @skb: buffer
6206 * @mpls_lse: MPLS label stack entry to push
6207 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6208 * @mac_len: length of the MAC header
6209 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6210 *            ethernet
6211 *
6212 * Expects skb->data at mac header.
6213 *
6214 * Returns 0 on success, -errno otherwise.
6215 */
6216int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6217		  int mac_len, bool ethernet)
6218{
6219	struct mpls_shim_hdr *lse;
6220	int err;
6221
6222	if (unlikely(!eth_p_mpls(mpls_proto)))
6223		return -EINVAL;
6224
6225	/* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6226	if (skb->encapsulation)
6227		return -EINVAL;
6228
6229	err = skb_cow_head(skb, MPLS_HLEN);
6230	if (unlikely(err))
6231		return err;
6232
6233	if (!skb->inner_protocol) {
6234		skb_set_inner_network_header(skb, skb_network_offset(skb));
6235		skb_set_inner_protocol(skb, skb->protocol);
6236	}
6237
6238	skb_push(skb, MPLS_HLEN);
6239	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6240		mac_len);
6241	skb_reset_mac_header(skb);
6242	skb_set_network_header(skb, mac_len);
6243	skb_reset_mac_len(skb);
6244
6245	lse = mpls_hdr(skb);
6246	lse->label_stack_entry = mpls_lse;
6247	skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6248
6249	if (ethernet && mac_len >= ETH_HLEN)
6250		skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6251	skb->protocol = mpls_proto;
6252
6253	return 0;
6254}
6255EXPORT_SYMBOL_GPL(skb_mpls_push);
6256
6257/**
6258 * skb_mpls_pop() - pop the outermost MPLS header
6259 *
6260 * @skb: buffer
6261 * @next_proto: ethertype of header after popped MPLS header
6262 * @mac_len: length of the MAC header
6263 * @ethernet: flag to indicate if the packet is ethernet
6264 *
6265 * Expects skb->data at mac header.
6266 *
6267 * Returns 0 on success, -errno otherwise.
6268 */
6269int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6270		 bool ethernet)
6271{
6272	int err;
6273
6274	if (unlikely(!eth_p_mpls(skb->protocol)))
6275		return 0;
6276
6277	err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6278	if (unlikely(err))
6279		return err;
6280
6281	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6282	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6283		mac_len);
6284
6285	__skb_pull(skb, MPLS_HLEN);
6286	skb_reset_mac_header(skb);
6287	skb_set_network_header(skb, mac_len);
6288
6289	if (ethernet && mac_len >= ETH_HLEN) {
6290		struct ethhdr *hdr;
6291
6292		/* use mpls_hdr() to get ethertype to account for VLANs. */
6293		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6294		skb_mod_eth_type(skb, hdr, next_proto);
6295	}
6296	skb->protocol = next_proto;
6297
6298	return 0;
6299}
6300EXPORT_SYMBOL_GPL(skb_mpls_pop);
6301
6302/**
6303 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6304 *
6305 * @skb: buffer
6306 * @mpls_lse: new MPLS label stack entry to update to
6307 *
6308 * Expects skb->data at mac header.
6309 *
6310 * Returns 0 on success, -errno otherwise.
6311 */
6312int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6313{
6314	int err;
6315
6316	if (unlikely(!eth_p_mpls(skb->protocol)))
6317		return -EINVAL;
6318
6319	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6320	if (unlikely(err))
6321		return err;
6322
6323	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6324		__be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6325
6326		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6327	}
6328
6329	mpls_hdr(skb)->label_stack_entry = mpls_lse;
6330
6331	return 0;
6332}
6333EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6334
6335/**
6336 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6337 *
6338 * @skb: buffer
6339 *
6340 * Expects skb->data at mac header.
6341 *
6342 * Returns 0 on success, -errno otherwise.
6343 */
6344int skb_mpls_dec_ttl(struct sk_buff *skb)
6345{
6346	u32 lse;
6347	u8 ttl;
6348
6349	if (unlikely(!eth_p_mpls(skb->protocol)))
6350		return -EINVAL;
6351
6352	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6353		return -ENOMEM;
6354
6355	lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6356	ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6357	if (!--ttl)
6358		return -EINVAL;
6359
6360	lse &= ~MPLS_LS_TTL_MASK;
6361	lse |= ttl << MPLS_LS_TTL_SHIFT;
6362
6363	return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6364}
6365EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6366
6367/**
6368 * alloc_skb_with_frags - allocate skb with page frags
6369 *
6370 * @header_len: size of linear part
6371 * @data_len: needed length in frags
6372 * @order: max page order desired.
6373 * @errcode: pointer to error code if any
6374 * @gfp_mask: allocation mask
6375 *
6376 * This can be used to allocate a paged skb, given a maximal order for frags.
6377 */
6378struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6379				     unsigned long data_len,
6380				     int order,
6381				     int *errcode,
6382				     gfp_t gfp_mask)
6383{
6384	unsigned long chunk;
6385	struct sk_buff *skb;
6386	struct page *page;
6387	int nr_frags = 0;
6388
6389	*errcode = -EMSGSIZE;
6390	if (unlikely(data_len > MAX_SKB_FRAGS * (PAGE_SIZE << order)))
6391		return NULL;
6392
6393	*errcode = -ENOBUFS;
6394	skb = alloc_skb(header_len, gfp_mask);
6395	if (!skb)
6396		return NULL;
6397
6398	while (data_len) {
6399		if (nr_frags == MAX_SKB_FRAGS - 1)
6400			goto failure;
6401		while (order && PAGE_ALIGN(data_len) < (PAGE_SIZE << order))
6402			order--;
6403
6404		if (order) {
6405			page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6406					   __GFP_COMP |
6407					   __GFP_NOWARN,
6408					   order);
6409			if (!page) {
6410				order--;
6411				continue;
6412			}
6413		} else {
6414			page = alloc_page(gfp_mask);
6415			if (!page)
6416				goto failure;
6417		}
6418		chunk = min_t(unsigned long, data_len,
6419			      PAGE_SIZE << order);
6420		skb_fill_page_desc(skb, nr_frags, page, 0, chunk);
6421		nr_frags++;
6422		skb->truesize += (PAGE_SIZE << order);
6423		data_len -= chunk;
6424	}
6425	return skb;
6426
6427failure:
6428	kfree_skb(skb);
6429	return NULL;
6430}
6431EXPORT_SYMBOL(alloc_skb_with_frags);
6432
6433/* carve out the first off bytes from skb when off < headlen */
6434static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6435				    const int headlen, gfp_t gfp_mask)
6436{
6437	int i;
6438	unsigned int size = skb_end_offset(skb);
6439	int new_hlen = headlen - off;
6440	u8 *data;
6441
6442	if (skb_pfmemalloc(skb))
6443		gfp_mask |= __GFP_MEMALLOC;
6444
6445	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6446	if (!data)
6447		return -ENOMEM;
6448	size = SKB_WITH_OVERHEAD(size);
6449
6450	/* Copy real data, and all frags */
6451	skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6452	skb->len -= off;
6453
6454	memcpy((struct skb_shared_info *)(data + size),
6455	       skb_shinfo(skb),
6456	       offsetof(struct skb_shared_info,
6457			frags[skb_shinfo(skb)->nr_frags]));
6458	if (skb_cloned(skb)) {
6459		/* drop the old head gracefully */
6460		if (skb_orphan_frags(skb, gfp_mask)) {
6461			skb_kfree_head(data, size);
6462			return -ENOMEM;
6463		}
6464		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6465			skb_frag_ref(skb, i);
6466		if (skb_has_frag_list(skb))
6467			skb_clone_fraglist(skb);
6468		skb_release_data(skb, SKB_CONSUMED, false);
6469	} else {
6470		/* we can reuse existing recount- all we did was
6471		 * relocate values
6472		 */
6473		skb_free_head(skb, false);
6474	}
6475
6476	skb->head = data;
6477	skb->data = data;
6478	skb->head_frag = 0;
6479	skb_set_end_offset(skb, size);
6480	skb_set_tail_pointer(skb, skb_headlen(skb));
6481	skb_headers_offset_update(skb, 0);
6482	skb->cloned = 0;
6483	skb->hdr_len = 0;
6484	skb->nohdr = 0;
6485	atomic_set(&skb_shinfo(skb)->dataref, 1);
6486
6487	return 0;
6488}
6489
6490static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6491
6492/* carve out the first eat bytes from skb's frag_list. May recurse into
6493 * pskb_carve()
6494 */
6495static int pskb_carve_frag_list(struct sk_buff *skb,
6496				struct skb_shared_info *shinfo, int eat,
6497				gfp_t gfp_mask)
6498{
6499	struct sk_buff *list = shinfo->frag_list;
6500	struct sk_buff *clone = NULL;
6501	struct sk_buff *insp = NULL;
6502
6503	do {
6504		if (!list) {
6505			pr_err("Not enough bytes to eat. Want %d\n", eat);
6506			return -EFAULT;
6507		}
6508		if (list->len <= eat) {
6509			/* Eaten as whole. */
6510			eat -= list->len;
6511			list = list->next;
6512			insp = list;
6513		} else {
6514			/* Eaten partially. */
6515			if (skb_shared(list)) {
6516				clone = skb_clone(list, gfp_mask);
6517				if (!clone)
6518					return -ENOMEM;
6519				insp = list->next;
6520				list = clone;
6521			} else {
6522				/* This may be pulled without problems. */
6523				insp = list;
6524			}
6525			if (pskb_carve(list, eat, gfp_mask) < 0) {
6526				kfree_skb(clone);
6527				return -ENOMEM;
6528			}
6529			break;
6530		}
6531	} while (eat);
6532
6533	/* Free pulled out fragments. */
6534	while ((list = shinfo->frag_list) != insp) {
6535		shinfo->frag_list = list->next;
6536		consume_skb(list);
6537	}
6538	/* And insert new clone at head. */
6539	if (clone) {
6540		clone->next = list;
6541		shinfo->frag_list = clone;
6542	}
6543	return 0;
6544}
6545
6546/* carve off first len bytes from skb. Split line (off) is in the
6547 * non-linear part of skb
6548 */
6549static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6550				       int pos, gfp_t gfp_mask)
6551{
6552	int i, k = 0;
6553	unsigned int size = skb_end_offset(skb);
6554	u8 *data;
6555	const int nfrags = skb_shinfo(skb)->nr_frags;
6556	struct skb_shared_info *shinfo;
6557
6558	if (skb_pfmemalloc(skb))
6559		gfp_mask |= __GFP_MEMALLOC;
6560
6561	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6562	if (!data)
6563		return -ENOMEM;
6564	size = SKB_WITH_OVERHEAD(size);
6565
6566	memcpy((struct skb_shared_info *)(data + size),
6567	       skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6568	if (skb_orphan_frags(skb, gfp_mask)) {
6569		skb_kfree_head(data, size);
6570		return -ENOMEM;
6571	}
6572	shinfo = (struct skb_shared_info *)(data + size);
6573	for (i = 0; i < nfrags; i++) {
6574		int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6575
6576		if (pos + fsize > off) {
6577			shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6578
6579			if (pos < off) {
6580				/* Split frag.
6581				 * We have two variants in this case:
6582				 * 1. Move all the frag to the second
6583				 *    part, if it is possible. F.e.
6584				 *    this approach is mandatory for TUX,
6585				 *    where splitting is expensive.
6586				 * 2. Split is accurately. We make this.
6587				 */
6588				skb_frag_off_add(&shinfo->frags[0], off - pos);
6589				skb_frag_size_sub(&shinfo->frags[0], off - pos);
6590			}
6591			skb_frag_ref(skb, i);
6592			k++;
6593		}
6594		pos += fsize;
6595	}
6596	shinfo->nr_frags = k;
6597	if (skb_has_frag_list(skb))
6598		skb_clone_fraglist(skb);
6599
6600	/* split line is in frag list */
6601	if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6602		/* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6603		if (skb_has_frag_list(skb))
6604			kfree_skb_list(skb_shinfo(skb)->frag_list);
6605		skb_kfree_head(data, size);
6606		return -ENOMEM;
6607	}
6608	skb_release_data(skb, SKB_CONSUMED, false);
6609
6610	skb->head = data;
6611	skb->head_frag = 0;
6612	skb->data = data;
6613	skb_set_end_offset(skb, size);
6614	skb_reset_tail_pointer(skb);
6615	skb_headers_offset_update(skb, 0);
6616	skb->cloned   = 0;
6617	skb->hdr_len  = 0;
6618	skb->nohdr    = 0;
6619	skb->len -= off;
6620	skb->data_len = skb->len;
6621	atomic_set(&skb_shinfo(skb)->dataref, 1);
6622	return 0;
6623}
6624
6625/* remove len bytes from the beginning of the skb */
6626static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6627{
6628	int headlen = skb_headlen(skb);
6629
6630	if (len < headlen)
6631		return pskb_carve_inside_header(skb, len, headlen, gfp);
6632	else
6633		return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6634}
6635
6636/* Extract to_copy bytes starting at off from skb, and return this in
6637 * a new skb
6638 */
6639struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6640			     int to_copy, gfp_t gfp)
6641{
6642	struct sk_buff  *clone = skb_clone(skb, gfp);
6643
6644	if (!clone)
6645		return NULL;
6646
6647	if (pskb_carve(clone, off, gfp) < 0 ||
6648	    pskb_trim(clone, to_copy)) {
6649		kfree_skb(clone);
6650		return NULL;
6651	}
6652	return clone;
6653}
6654EXPORT_SYMBOL(pskb_extract);
6655
6656/**
6657 * skb_condense - try to get rid of fragments/frag_list if possible
6658 * @skb: buffer
6659 *
6660 * Can be used to save memory before skb is added to a busy queue.
6661 * If packet has bytes in frags and enough tail room in skb->head,
6662 * pull all of them, so that we can free the frags right now and adjust
6663 * truesize.
6664 * Notes:
6665 *	We do not reallocate skb->head thus can not fail.
6666 *	Caller must re-evaluate skb->truesize if needed.
6667 */
6668void skb_condense(struct sk_buff *skb)
6669{
6670	if (skb->data_len) {
6671		if (skb->data_len > skb->end - skb->tail ||
6672		    skb_cloned(skb))
6673			return;
6674
6675		/* Nice, we can free page frag(s) right now */
6676		__pskb_pull_tail(skb, skb->data_len);
6677	}
6678	/* At this point, skb->truesize might be over estimated,
6679	 * because skb had a fragment, and fragments do not tell
6680	 * their truesize.
6681	 * When we pulled its content into skb->head, fragment
6682	 * was freed, but __pskb_pull_tail() could not possibly
6683	 * adjust skb->truesize, not knowing the frag truesize.
6684	 */
6685	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6686}
6687EXPORT_SYMBOL(skb_condense);
6688
6689#ifdef CONFIG_SKB_EXTENSIONS
6690static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6691{
6692	return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6693}
6694
6695/**
6696 * __skb_ext_alloc - allocate a new skb extensions storage
6697 *
6698 * @flags: See kmalloc().
6699 *
6700 * Returns the newly allocated pointer. The pointer can later attached to a
6701 * skb via __skb_ext_set().
6702 * Note: caller must handle the skb_ext as an opaque data.
6703 */
6704struct skb_ext *__skb_ext_alloc(gfp_t flags)
6705{
6706	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6707
6708	if (new) {
6709		memset(new->offset, 0, sizeof(new->offset));
6710		refcount_set(&new->refcnt, 1);
6711	}
6712
6713	return new;
6714}
6715
6716static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6717					 unsigned int old_active)
6718{
6719	struct skb_ext *new;
6720
6721	if (refcount_read(&old->refcnt) == 1)
6722		return old;
6723
6724	new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6725	if (!new)
6726		return NULL;
6727
6728	memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6729	refcount_set(&new->refcnt, 1);
6730
6731#ifdef CONFIG_XFRM
6732	if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6733		struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6734		unsigned int i;
6735
6736		for (i = 0; i < sp->len; i++)
6737			xfrm_state_hold(sp->xvec[i]);
6738	}
6739#endif
6740	__skb_ext_put(old);
6741	return new;
6742}
6743
6744/**
6745 * __skb_ext_set - attach the specified extension storage to this skb
6746 * @skb: buffer
6747 * @id: extension id
6748 * @ext: extension storage previously allocated via __skb_ext_alloc()
6749 *
6750 * Existing extensions, if any, are cleared.
6751 *
6752 * Returns the pointer to the extension.
6753 */
6754void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6755		    struct skb_ext *ext)
6756{
6757	unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6758
6759	skb_ext_put(skb);
6760	newlen = newoff + skb_ext_type_len[id];
6761	ext->chunks = newlen;
6762	ext->offset[id] = newoff;
6763	skb->extensions = ext;
6764	skb->active_extensions = 1 << id;
6765	return skb_ext_get_ptr(ext, id);
6766}
6767
6768/**
6769 * skb_ext_add - allocate space for given extension, COW if needed
6770 * @skb: buffer
6771 * @id: extension to allocate space for
6772 *
6773 * Allocates enough space for the given extension.
6774 * If the extension is already present, a pointer to that extension
6775 * is returned.
6776 *
6777 * If the skb was cloned, COW applies and the returned memory can be
6778 * modified without changing the extension space of clones buffers.
6779 *
6780 * Returns pointer to the extension or NULL on allocation failure.
6781 */
6782void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6783{
6784	struct skb_ext *new, *old = NULL;
6785	unsigned int newlen, newoff;
6786
6787	if (skb->active_extensions) {
6788		old = skb->extensions;
6789
6790		new = skb_ext_maybe_cow(old, skb->active_extensions);
6791		if (!new)
6792			return NULL;
6793
6794		if (__skb_ext_exist(new, id))
6795			goto set_active;
6796
6797		newoff = new->chunks;
6798	} else {
6799		newoff = SKB_EXT_CHUNKSIZEOF(*new);
6800
6801		new = __skb_ext_alloc(GFP_ATOMIC);
6802		if (!new)
6803			return NULL;
6804	}
6805
6806	newlen = newoff + skb_ext_type_len[id];
6807	new->chunks = newlen;
6808	new->offset[id] = newoff;
6809set_active:
6810	skb->slow_gro = 1;
6811	skb->extensions = new;
6812	skb->active_extensions |= 1 << id;
6813	return skb_ext_get_ptr(new, id);
6814}
6815EXPORT_SYMBOL(skb_ext_add);
6816
6817#ifdef CONFIG_XFRM
6818static void skb_ext_put_sp(struct sec_path *sp)
6819{
6820	unsigned int i;
6821
6822	for (i = 0; i < sp->len; i++)
6823		xfrm_state_put(sp->xvec[i]);
6824}
6825#endif
6826
6827#ifdef CONFIG_MCTP_FLOWS
6828static void skb_ext_put_mctp(struct mctp_flow *flow)
6829{
6830	if (flow->key)
6831		mctp_key_unref(flow->key);
6832}
6833#endif
6834
6835void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6836{
6837	struct skb_ext *ext = skb->extensions;
6838
6839	skb->active_extensions &= ~(1 << id);
6840	if (skb->active_extensions == 0) {
6841		skb->extensions = NULL;
6842		__skb_ext_put(ext);
6843#ifdef CONFIG_XFRM
6844	} else if (id == SKB_EXT_SEC_PATH &&
6845		   refcount_read(&ext->refcnt) == 1) {
6846		struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6847
6848		skb_ext_put_sp(sp);
6849		sp->len = 0;
6850#endif
6851	}
6852}
6853EXPORT_SYMBOL(__skb_ext_del);
6854
6855void __skb_ext_put(struct skb_ext *ext)
6856{
6857	/* If this is last clone, nothing can increment
6858	 * it after check passes.  Avoids one atomic op.
6859	 */
6860	if (refcount_read(&ext->refcnt) == 1)
6861		goto free_now;
6862
6863	if (!refcount_dec_and_test(&ext->refcnt))
6864		return;
6865free_now:
6866#ifdef CONFIG_XFRM
6867	if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6868		skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6869#endif
6870#ifdef CONFIG_MCTP_FLOWS
6871	if (__skb_ext_exist(ext, SKB_EXT_MCTP))
6872		skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
6873#endif
6874
6875	kmem_cache_free(skbuff_ext_cache, ext);
6876}
6877EXPORT_SYMBOL(__skb_ext_put);
6878#endif /* CONFIG_SKB_EXTENSIONS */
6879
6880/**
6881 * skb_attempt_defer_free - queue skb for remote freeing
6882 * @skb: buffer
6883 *
6884 * Put @skb in a per-cpu list, using the cpu which
6885 * allocated the skb/pages to reduce false sharing
6886 * and memory zone spinlock contention.
6887 */
6888void skb_attempt_defer_free(struct sk_buff *skb)
6889{
6890	int cpu = skb->alloc_cpu;
6891	struct softnet_data *sd;
6892	unsigned int defer_max;
6893	bool kick;
6894
6895	if (WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
6896	    !cpu_online(cpu) ||
6897	    cpu == raw_smp_processor_id()) {
6898nodefer:	__kfree_skb(skb);
6899		return;
6900	}
6901
6902	DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
6903	DEBUG_NET_WARN_ON_ONCE(skb->destructor);
6904
6905	sd = &per_cpu(softnet_data, cpu);
6906	defer_max = READ_ONCE(sysctl_skb_defer_max);
6907	if (READ_ONCE(sd->defer_count) >= defer_max)
6908		goto nodefer;
6909
6910	spin_lock_bh(&sd->defer_lock);
6911	/* Send an IPI every time queue reaches half capacity. */
6912	kick = sd->defer_count == (defer_max >> 1);
6913	/* Paired with the READ_ONCE() few lines above */
6914	WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
6915
6916	skb->next = sd->defer_list;
6917	/* Paired with READ_ONCE() in skb_defer_free_flush() */
6918	WRITE_ONCE(sd->defer_list, skb);
6919	spin_unlock_bh(&sd->defer_lock);
6920
6921	/* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
6922	 * if we are unlucky enough (this seems very unlikely).
6923	 */
6924	if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1))
6925		smp_call_function_single_async(cpu, &sd->defer_csd);
6926}
6927
6928static void skb_splice_csum_page(struct sk_buff *skb, struct page *page,
6929				 size_t offset, size_t len)
6930{
6931	const char *kaddr;
6932	__wsum csum;
6933
6934	kaddr = kmap_local_page(page);
6935	csum = csum_partial(kaddr + offset, len, 0);
6936	kunmap_local(kaddr);
6937	skb->csum = csum_block_add(skb->csum, csum, skb->len);
6938}
6939
6940/**
6941 * skb_splice_from_iter - Splice (or copy) pages to skbuff
6942 * @skb: The buffer to add pages to
6943 * @iter: Iterator representing the pages to be added
6944 * @maxsize: Maximum amount of pages to be added
6945 * @gfp: Allocation flags
6946 *
6947 * This is a common helper function for supporting MSG_SPLICE_PAGES.  It
6948 * extracts pages from an iterator and adds them to the socket buffer if
6949 * possible, copying them to fragments if not possible (such as if they're slab
6950 * pages).
6951 *
6952 * Returns the amount of data spliced/copied or -EMSGSIZE if there's
6953 * insufficient space in the buffer to transfer anything.
6954 */
6955ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
6956			     ssize_t maxsize, gfp_t gfp)
6957{
6958	size_t frag_limit = READ_ONCE(sysctl_max_skb_frags);
6959	struct page *pages[8], **ppages = pages;
6960	ssize_t spliced = 0, ret = 0;
6961	unsigned int i;
6962
6963	while (iter->count > 0) {
6964		ssize_t space, nr, len;
6965		size_t off;
6966
6967		ret = -EMSGSIZE;
6968		space = frag_limit - skb_shinfo(skb)->nr_frags;
6969		if (space < 0)
6970			break;
6971
6972		/* We might be able to coalesce without increasing nr_frags */
6973		nr = clamp_t(size_t, space, 1, ARRAY_SIZE(pages));
6974
6975		len = iov_iter_extract_pages(iter, &ppages, maxsize, nr, 0, &off);
6976		if (len <= 0) {
6977			ret = len ?: -EIO;
6978			break;
6979		}
6980
6981		i = 0;
6982		do {
6983			struct page *page = pages[i++];
6984			size_t part = min_t(size_t, PAGE_SIZE - off, len);
6985
6986			ret = -EIO;
6987			if (WARN_ON_ONCE(!sendpage_ok(page)))
6988				goto out;
6989
6990			ret = skb_append_pagefrags(skb, page, off, part,
6991						   frag_limit);
6992			if (ret < 0) {
6993				iov_iter_revert(iter, len);
6994				goto out;
6995			}
6996
6997			if (skb->ip_summed == CHECKSUM_NONE)
6998				skb_splice_csum_page(skb, page, off, part);
6999
7000			off = 0;
7001			spliced += part;
7002			maxsize -= part;
7003			len -= part;
7004		} while (len > 0);
7005
7006		if (maxsize <= 0)
7007			break;
7008	}
7009
7010out:
7011	skb_len_add(skb, spliced);
7012	return spliced ?: ret;
7013}
7014EXPORT_SYMBOL(skb_splice_from_iter);
7015
7016static __always_inline
7017size_t memcpy_from_iter_csum(void *iter_from, size_t progress,
7018			     size_t len, void *to, void *priv2)
7019{
7020	__wsum *csum = priv2;
7021	__wsum next = csum_partial_copy_nocheck(iter_from, to + progress, len);
7022
7023	*csum = csum_block_add(*csum, next, progress);
7024	return 0;
7025}
7026
7027static __always_inline
7028size_t copy_from_user_iter_csum(void __user *iter_from, size_t progress,
7029				size_t len, void *to, void *priv2)
7030{
7031	__wsum next, *csum = priv2;
7032
7033	next = csum_and_copy_from_user(iter_from, to + progress, len);
7034	*csum = csum_block_add(*csum, next, progress);
7035	return next ? 0 : len;
7036}
7037
7038bool csum_and_copy_from_iter_full(void *addr, size_t bytes,
7039				  __wsum *csum, struct iov_iter *i)
7040{
7041	size_t copied;
7042
7043	if (WARN_ON_ONCE(!i->data_source))
7044		return false;
7045	copied = iterate_and_advance2(i, bytes, addr, csum,
7046				      copy_from_user_iter_csum,
7047				      memcpy_from_iter_csum);
7048	if (likely(copied == bytes))
7049		return true;
7050	iov_iter_revert(i, copied);
7051	return false;
7052}
7053EXPORT_SYMBOL(csum_and_copy_from_iter_full);