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/skbuff_ref.h>
  55#include <linux/splice.h>
  56#include <linux/cache.h>
  57#include <linux/rtnetlink.h>
  58#include <linux/init.h>
  59#include <linux/scatterlist.h>
  60#include <linux/errqueue.h>
  61#include <linux/prefetch.h>
  62#include <linux/bitfield.h>
  63#include <linux/if_vlan.h>
  64#include <linux/mpls.h>
  65#include <linux/kcov.h>
  66#include <linux/iov_iter.h>
  67
  68#include <net/protocol.h>
  69#include <net/dst.h>
  70#include <net/sock.h>
  71#include <net/checksum.h>
  72#include <net/gro.h>
  73#include <net/gso.h>
  74#include <net/hotdata.h>
  75#include <net/ip6_checksum.h>
  76#include <net/xfrm.h>
  77#include <net/mpls.h>
  78#include <net/mptcp.h>
  79#include <net/mctp.h>
  80#include <net/page_pool/helpers.h>
  81#include <net/dropreason.h>
  82
  83#include <linux/uaccess.h>
  84#include <trace/events/skb.h>
  85#include <linux/highmem.h>
  86#include <linux/capability.h>
  87#include <linux/user_namespace.h>
  88#include <linux/indirect_call_wrapper.h>
  89#include <linux/textsearch.h>
  90
  91#include "dev.h"
  92#include "netmem_priv.h"
  93#include "sock_destructor.h"
  94
  95#ifdef CONFIG_SKB_EXTENSIONS
  96static struct kmem_cache *skbuff_ext_cache __ro_after_init;
  97#endif
  98
  99#define GRO_MAX_HEAD_PAD (GRO_MAX_HEAD + NET_SKB_PAD + NET_IP_ALIGN)
 100#define SKB_SMALL_HEAD_SIZE SKB_HEAD_ALIGN(max(MAX_TCP_HEADER, \
 101					       GRO_MAX_HEAD_PAD))
 102
 103/* We want SKB_SMALL_HEAD_CACHE_SIZE to not be a power of two.
 104 * This should ensure that SKB_SMALL_HEAD_HEADROOM is a unique
 105 * size, and we can differentiate heads from skb_small_head_cache
 106 * vs system slabs by looking at their size (skb_end_offset()).
 107 */
 108#define SKB_SMALL_HEAD_CACHE_SIZE					\
 109	(is_power_of_2(SKB_SMALL_HEAD_SIZE) ?			\
 110		(SKB_SMALL_HEAD_SIZE + L1_CACHE_BYTES) :	\
 111		SKB_SMALL_HEAD_SIZE)
 112
 113#define SKB_SMALL_HEAD_HEADROOM						\
 114	SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE)
 115
 116/* kcm_write_msgs() relies on casting paged frags to bio_vec to use
 117 * iov_iter_bvec(). These static asserts ensure the cast is valid is long as the
 118 * netmem is a page.
 119 */
 120static_assert(offsetof(struct bio_vec, bv_page) ==
 121	      offsetof(skb_frag_t, netmem));
 122static_assert(sizeof_field(struct bio_vec, bv_page) ==
 123	      sizeof_field(skb_frag_t, netmem));
 124
 125static_assert(offsetof(struct bio_vec, bv_len) == offsetof(skb_frag_t, len));
 126static_assert(sizeof_field(struct bio_vec, bv_len) ==
 127	      sizeof_field(skb_frag_t, len));
 128
 129static_assert(offsetof(struct bio_vec, bv_offset) ==
 130	      offsetof(skb_frag_t, offset));
 131static_assert(sizeof_field(struct bio_vec, bv_offset) ==
 132	      sizeof_field(skb_frag_t, offset));
 133
 134#undef FN
 135#define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
 136static const char * const drop_reasons[] = {
 137	[SKB_CONSUMED] = "CONSUMED",
 138	DEFINE_DROP_REASON(FN, FN)
 139};
 140
 141static const struct drop_reason_list drop_reasons_core = {
 142	.reasons = drop_reasons,
 143	.n_reasons = ARRAY_SIZE(drop_reasons),
 144};
 145
 146const struct drop_reason_list __rcu *
 147drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_NUM] = {
 148	[SKB_DROP_REASON_SUBSYS_CORE] = RCU_INITIALIZER(&drop_reasons_core),
 149};
 150EXPORT_SYMBOL(drop_reasons_by_subsys);
 151
 152/**
 153 * drop_reasons_register_subsys - register another drop reason subsystem
 154 * @subsys: the subsystem to register, must not be the core
 155 * @list: the list of drop reasons within the subsystem, must point to
 156 *	a statically initialized list
 157 */
 158void drop_reasons_register_subsys(enum skb_drop_reason_subsys subsys,
 159				  const struct drop_reason_list *list)
 160{
 161	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
 162		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
 163		 "invalid subsystem %d\n", subsys))
 164		return;
 165
 166	/* must point to statically allocated memory, so INIT is OK */
 167	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], list);
 168}
 169EXPORT_SYMBOL_GPL(drop_reasons_register_subsys);
 170
 171/**
 172 * drop_reasons_unregister_subsys - unregister a drop reason subsystem
 173 * @subsys: the subsystem to remove, must not be the core
 174 *
 175 * Note: This will synchronize_rcu() to ensure no users when it returns.
 176 */
 177void drop_reasons_unregister_subsys(enum skb_drop_reason_subsys subsys)
 178{
 179	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
 180		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
 181		 "invalid subsystem %d\n", subsys))
 182		return;
 183
 184	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], NULL);
 185
 186	synchronize_rcu();
 187}
 188EXPORT_SYMBOL_GPL(drop_reasons_unregister_subsys);
 189
 190/**
 191 *	skb_panic - private function for out-of-line support
 192 *	@skb:	buffer
 193 *	@sz:	size
 194 *	@addr:	address
 195 *	@msg:	skb_over_panic or skb_under_panic
 196 *
 197 *	Out-of-line support for skb_put() and skb_push().
 198 *	Called via the wrapper skb_over_panic() or skb_under_panic().
 199 *	Keep out of line to prevent kernel bloat.
 200 *	__builtin_return_address is not used because it is not always reliable.
 201 */
 202static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
 203		      const char msg[])
 204{
 205	pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
 206		 msg, addr, skb->len, sz, skb->head, skb->data,
 207		 (unsigned long)skb->tail, (unsigned long)skb->end,
 208		 skb->dev ? skb->dev->name : "<NULL>");
 209	BUG();
 210}
 211
 212static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 
 213{
 214	skb_panic(skb, sz, addr, __func__);
 215}
 216
 217static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
 
 218{
 219	skb_panic(skb, sz, addr, __func__);
 220}
 221
 222#define NAPI_SKB_CACHE_SIZE	64
 223#define NAPI_SKB_CACHE_BULK	16
 224#define NAPI_SKB_CACHE_HALF	(NAPI_SKB_CACHE_SIZE / 2)
 225
 226#if PAGE_SIZE == SZ_4K
 227
 228#define NAPI_HAS_SMALL_PAGE_FRAG	1
 229#define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	((nc).pfmemalloc)
 230
 231/* specialized page frag allocator using a single order 0 page
 232 * and slicing it into 1K sized fragment. Constrained to systems
 233 * with a very limited amount of 1K fragments fitting a single
 234 * page - to avoid excessive truesize underestimation
 235 */
 236
 237struct page_frag_1k {
 238	void *va;
 239	u16 offset;
 240	bool pfmemalloc;
 241};
 242
 243static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
 244{
 245	struct page *page;
 246	int offset;
 247
 248	offset = nc->offset - SZ_1K;
 249	if (likely(offset >= 0))
 250		goto use_frag;
 251
 252	page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
 253	if (!page)
 254		return NULL;
 255
 256	nc->va = page_address(page);
 257	nc->pfmemalloc = page_is_pfmemalloc(page);
 258	offset = PAGE_SIZE - SZ_1K;
 259	page_ref_add(page, offset / SZ_1K);
 260
 261use_frag:
 262	nc->offset = offset;
 263	return nc->va + offset;
 264}
 265#else
 266
 267/* the small page is actually unused in this build; add dummy helpers
 268 * to please the compiler and avoid later preprocessor's conditionals
 269 */
 270#define NAPI_HAS_SMALL_PAGE_FRAG	0
 271#define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	false
 272
 273struct page_frag_1k {
 274};
 275
 276static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
 277{
 278	return NULL;
 279}
 280
 281#endif
 282
 283struct napi_alloc_cache {
 284	local_lock_t bh_lock;
 285	struct page_frag_cache page;
 286	struct page_frag_1k page_small;
 287	unsigned int skb_count;
 288	void *skb_cache[NAPI_SKB_CACHE_SIZE];
 289};
 290
 291static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
 292static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = {
 293	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
 294};
 295
 296/* Double check that napi_get_frags() allocates skbs with
 297 * skb->head being backed by slab, not a page fragment.
 298 * This is to make sure bug fixed in 3226b158e67c
 299 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
 300 * does not accidentally come back.
 301 */
 302void napi_get_frags_check(struct napi_struct *napi)
 303{
 304	struct sk_buff *skb;
 305
 306	local_bh_disable();
 307	skb = napi_get_frags(napi);
 308	WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
 309	napi_free_frags(napi);
 310	local_bh_enable();
 311}
 312
 313void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 314{
 315	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 316	void *data;
 317
 318	fragsz = SKB_DATA_ALIGN(fragsz);
 319
 320	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 321	data = __page_frag_alloc_align(&nc->page, fragsz,
 322				       GFP_ATOMIC | __GFP_NOWARN, align_mask);
 323	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 324	return data;
 325
 326}
 327EXPORT_SYMBOL(__napi_alloc_frag_align);
 328
 329void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 330{
 331	void *data;
 332
 333	if (in_hardirq() || irqs_disabled()) {
 334		struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
 335
 336		fragsz = SKB_DATA_ALIGN(fragsz);
 337		data = __page_frag_alloc_align(nc, fragsz,
 338					       GFP_ATOMIC | __GFP_NOWARN,
 339					       align_mask);
 340	} else {
 341		local_bh_disable();
 342		data = __napi_alloc_frag_align(fragsz, align_mask);
 343		local_bh_enable();
 344	}
 345	return data;
 346}
 347EXPORT_SYMBOL(__netdev_alloc_frag_align);
 348
 349static struct sk_buff *napi_skb_cache_get(void)
 350{
 351	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 352	struct sk_buff *skb;
 353
 354	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 355	if (unlikely(!nc->skb_count)) {
 356		nc->skb_count = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
 357						      GFP_ATOMIC | __GFP_NOWARN,
 358						      NAPI_SKB_CACHE_BULK,
 359						      nc->skb_cache);
 360		if (unlikely(!nc->skb_count)) {
 361			local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 362			return NULL;
 363		}
 364	}
 365
 366	skb = nc->skb_cache[--nc->skb_count];
 367	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 368	kasan_mempool_unpoison_object(skb, kmem_cache_size(net_hotdata.skbuff_cache));
 369
 370	return skb;
 371}
 372
 373static inline void __finalize_skb_around(struct sk_buff *skb, void *data,
 374					 unsigned int size)
 375{
 376	struct skb_shared_info *shinfo;
 377
 378	size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 379
 380	/* Assumes caller memset cleared SKB */
 381	skb->truesize = SKB_TRUESIZE(size);
 382	refcount_set(&skb->users, 1);
 383	skb->head = data;
 384	skb->data = data;
 385	skb_reset_tail_pointer(skb);
 386	skb_set_end_offset(skb, size);
 387	skb->mac_header = (typeof(skb->mac_header))~0U;
 388	skb->transport_header = (typeof(skb->transport_header))~0U;
 389	skb->alloc_cpu = raw_smp_processor_id();
 390	/* make sure we initialize shinfo sequentially */
 391	shinfo = skb_shinfo(skb);
 392	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 393	atomic_set(&shinfo->dataref, 1);
 394
 395	skb_set_kcov_handle(skb, kcov_common_handle());
 396}
 397
 398static inline void *__slab_build_skb(struct sk_buff *skb, void *data,
 399				     unsigned int *size)
 400{
 401	void *resized;
 402
 403	/* Must find the allocation size (and grow it to match). */
 404	*size = ksize(data);
 405	/* krealloc() will immediately return "data" when
 406	 * "ksize(data)" is requested: it is the existing upper
 407	 * bounds. As a result, GFP_ATOMIC will be ignored. Note
 408	 * that this "new" pointer needs to be passed back to the
 409	 * caller for use so the __alloc_size hinting will be
 410	 * tracked correctly.
 411	 */
 412	resized = krealloc(data, *size, GFP_ATOMIC);
 413	WARN_ON_ONCE(resized != data);
 414	return resized;
 415}
 416
 417/* build_skb() variant which can operate on slab buffers.
 418 * Note that this should be used sparingly as slab buffers
 419 * cannot be combined efficiently by GRO!
 420 */
 421struct sk_buff *slab_build_skb(void *data)
 422{
 423	struct sk_buff *skb;
 424	unsigned int size;
 425
 426	skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
 427			       GFP_ATOMIC | __GFP_NOWARN);
 428	if (unlikely(!skb))
 429		return NULL;
 430
 431	memset(skb, 0, offsetof(struct sk_buff, tail));
 432	data = __slab_build_skb(skb, data, &size);
 433	__finalize_skb_around(skb, data, size);
 434
 435	return skb;
 436}
 437EXPORT_SYMBOL(slab_build_skb);
 438
 439/* Caller must provide SKB that is memset cleared */
 440static void __build_skb_around(struct sk_buff *skb, void *data,
 441			       unsigned int frag_size)
 442{
 443	unsigned int size = frag_size;
 444
 445	/* frag_size == 0 is considered deprecated now. Callers
 446	 * using slab buffer should use slab_build_skb() instead.
 447	 */
 448	if (WARN_ONCE(size == 0, "Use slab_build_skb() instead"))
 449		data = __slab_build_skb(skb, data, &size);
 450
 451	__finalize_skb_around(skb, data, size);
 452}
 453
 454/**
 455 * __build_skb - build a network buffer
 456 * @data: data buffer provided by caller
 457 * @frag_size: size of data (must not be 0)
 
 458 *
 459 * Allocate a new &sk_buff. Caller provides space holding head and
 460 * skb_shared_info. @data must have been allocated from the page
 461 * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc()
 462 * allocation is deprecated, and callers should use slab_build_skb()
 463 * instead.)
 464 * The return is the new skb buffer.
 465 * On a failure the return is %NULL, and @data is not freed.
 466 * Notes :
 467 *  Before IO, driver allocates only data buffer where NIC put incoming frame
 468 *  Driver should add room at head (NET_SKB_PAD) and
 469 *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
 470 *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
 471 *  before giving packet to stack.
 472 *  RX rings only contains data buffers, not full skbs.
 473 */
 474struct sk_buff *__build_skb(void *data, unsigned int frag_size)
 475{
 476	struct sk_buff *skb;
 477
 478	skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
 479			       GFP_ATOMIC | __GFP_NOWARN);
 480	if (unlikely(!skb))
 481		return NULL;
 482
 483	memset(skb, 0, offsetof(struct sk_buff, tail));
 484	__build_skb_around(skb, data, frag_size);
 485
 486	return skb;
 487}
 488
 489/* build_skb() is wrapper over __build_skb(), that specifically
 490 * takes care of skb->head and skb->pfmemalloc
 491 */
 492struct sk_buff *build_skb(void *data, unsigned int frag_size)
 493{
 494	struct sk_buff *skb = __build_skb(data, frag_size);
 495
 496	if (likely(skb && frag_size)) {
 497		skb->head_frag = 1;
 498		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 499	}
 500	return skb;
 501}
 502EXPORT_SYMBOL(build_skb);
 503
 504/**
 505 * build_skb_around - build a network buffer around provided skb
 506 * @skb: sk_buff provide by caller, must be memset cleared
 507 * @data: data buffer provided by caller
 508 * @frag_size: size of data
 509 */
 510struct sk_buff *build_skb_around(struct sk_buff *skb,
 511				 void *data, unsigned int frag_size)
 512{
 513	if (unlikely(!skb))
 514		return NULL;
 515
 516	__build_skb_around(skb, data, frag_size);
 517
 518	if (frag_size) {
 519		skb->head_frag = 1;
 520		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 521	}
 522	return skb;
 523}
 524EXPORT_SYMBOL(build_skb_around);
 525
 526/**
 527 * __napi_build_skb - build a network buffer
 528 * @data: data buffer provided by caller
 529 * @frag_size: size of data
 530 *
 531 * Version of __build_skb() that uses NAPI percpu caches to obtain
 532 * skbuff_head instead of inplace allocation.
 533 *
 534 * Returns a new &sk_buff on success, %NULL on allocation failure.
 535 */
 536static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
 537{
 538	struct sk_buff *skb;
 539
 540	skb = napi_skb_cache_get();
 541	if (unlikely(!skb))
 542		return NULL;
 543
 544	memset(skb, 0, offsetof(struct sk_buff, tail));
 545	__build_skb_around(skb, data, frag_size);
 546
 547	return skb;
 548}
 549
 550/**
 551 * napi_build_skb - build a network buffer
 552 * @data: data buffer provided by caller
 553 * @frag_size: size of data
 554 *
 555 * Version of __napi_build_skb() that takes care of skb->head_frag
 556 * and skb->pfmemalloc when the data is a page or page fragment.
 557 *
 558 * Returns a new &sk_buff on success, %NULL on allocation failure.
 559 */
 560struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
 561{
 562	struct sk_buff *skb = __napi_build_skb(data, frag_size);
 563
 564	if (likely(skb) && frag_size) {
 565		skb->head_frag = 1;
 566		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
 567	}
 568
 569	return skb;
 570}
 571EXPORT_SYMBOL(napi_build_skb);
 572
 573/*
 574 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
 575 * the caller if emergency pfmemalloc reserves are being used. If it is and
 576 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
 577 * may be used. Otherwise, the packet data may be discarded until enough
 578 * memory is free
 579 */
 580static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
 581			     bool *pfmemalloc)
 582{
 583	bool ret_pfmemalloc = false;
 584	size_t obj_size;
 585	void *obj;
 586
 587	obj_size = SKB_HEAD_ALIGN(*size);
 588	if (obj_size <= SKB_SMALL_HEAD_CACHE_SIZE &&
 589	    !(flags & KMALLOC_NOT_NORMAL_BITS)) {
 590		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache,
 591				flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 592				node);
 593		*size = SKB_SMALL_HEAD_CACHE_SIZE;
 594		if (obj || !(gfp_pfmemalloc_allowed(flags)))
 595			goto out;
 596		/* Try again but now we are using pfmemalloc reserves */
 597		ret_pfmemalloc = true;
 598		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache, flags, node);
 599		goto out;
 600	}
 601
 602	obj_size = kmalloc_size_roundup(obj_size);
 603	/* The following cast might truncate high-order bits of obj_size, this
 604	 * is harmless because kmalloc(obj_size >= 2^32) will fail anyway.
 605	 */
 606	*size = (unsigned int)obj_size;
 607
 608	/*
 609	 * Try a regular allocation, when that fails and we're not entitled
 610	 * to the reserves, fail.
 611	 */
 612	obj = kmalloc_node_track_caller(obj_size,
 613					flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 614					node);
 615	if (obj || !(gfp_pfmemalloc_allowed(flags)))
 616		goto out;
 617
 618	/* Try again but now we are using pfmemalloc reserves */
 619	ret_pfmemalloc = true;
 620	obj = kmalloc_node_track_caller(obj_size, flags, node);
 621
 622out:
 623	if (pfmemalloc)
 624		*pfmemalloc = ret_pfmemalloc;
 625
 626	return obj;
 627}
 628
 629/* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
 630 *	'private' fields and also do memory statistics to find all the
 631 *	[BEEP] leaks.
 632 *
 633 */
 634
 635/**
 636 *	__alloc_skb	-	allocate a network buffer
 637 *	@size: size to allocate
 638 *	@gfp_mask: allocation mask
 639 *	@flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
 640 *		instead of head cache and allocate a cloned (child) skb.
 641 *		If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
 642 *		allocations in case the data is required for writeback
 643 *	@node: numa node to allocate memory on
 644 *
 645 *	Allocate a new &sk_buff. The returned buffer has no headroom and a
 646 *	tail room of at least size bytes. The object has a reference count
 647 *	of one. The return is the buffer. On a failure the return is %NULL.
 648 *
 649 *	Buffers may only be allocated from interrupts using a @gfp_mask of
 650 *	%GFP_ATOMIC.
 651 */
 652struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 653			    int flags, int node)
 654{
 655	struct kmem_cache *cache;
 
 656	struct sk_buff *skb;
 657	bool pfmemalloc;
 658	u8 *data;
 659
 660	cache = (flags & SKB_ALLOC_FCLONE)
 661		? net_hotdata.skbuff_fclone_cache : net_hotdata.skbuff_cache;
 662
 663	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 664		gfp_mask |= __GFP_MEMALLOC;
 665
 666	/* Get the HEAD */
 667	if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
 668	    likely(node == NUMA_NO_NODE || node == numa_mem_id()))
 669		skb = napi_skb_cache_get();
 670	else
 671		skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
 672	if (unlikely(!skb))
 673		return NULL;
 674	prefetchw(skb);
 675
 676	/* We do our best to align skb_shared_info on a separate cache
 677	 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
 678	 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
 679	 * Both skb->head and skb_shared_info are cache line aligned.
 680	 */
 681	data = kmalloc_reserve(&size, gfp_mask, node, &pfmemalloc);
 682	if (unlikely(!data))
 
 
 683		goto nodata;
 684	/* kmalloc_size_roundup() might give us more room than requested.
 685	 * Put skb_shared_info exactly at the end of allocated zone,
 686	 * to allow max possible filling before reallocation.
 687	 */
 688	prefetchw(data + SKB_WITH_OVERHEAD(size));
 
 689
 690	/*
 691	 * Only clear those fields we need to clear, not those that we will
 692	 * actually initialise below. Hence, don't put any more fields after
 693	 * the tail pointer in struct sk_buff!
 694	 */
 695	memset(skb, 0, offsetof(struct sk_buff, tail));
 696	__build_skb_around(skb, data, size);
 697	skb->pfmemalloc = pfmemalloc;
 
 
 
 
 
 
 
 
 698
 699	if (flags & SKB_ALLOC_FCLONE) {
 700		struct sk_buff_fclones *fclones;
 
 
 
 701
 702		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 
 
 703
 
 
 704		skb->fclone = SKB_FCLONE_ORIG;
 705		refcount_set(&fclones->fclone_ref, 1);
 706	}
 707
 
 
 
 708	return skb;
 709
 710nodata:
 711	kmem_cache_free(cache, skb);
 712	return NULL;
 
 713}
 714EXPORT_SYMBOL(__alloc_skb);
 715
 716/**
 717 *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
 718 *	@dev: network device to receive on
 719 *	@len: length to allocate
 720 *	@gfp_mask: get_free_pages mask, passed to alloc_skb
 721 *
 722 *	Allocate a new &sk_buff and assign it a usage count of one. The
 723 *	buffer has NET_SKB_PAD headroom built in. Users should allocate
 724 *	the headroom they think they need without accounting for the
 725 *	built in space. The built in space is used for optimisations.
 726 *
 727 *	%NULL is returned if there is no free memory.
 
 
 
 
 
 
 
 
 
 
 728 */
 729struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
 730				   gfp_t gfp_mask)
 731{
 732	struct page_frag_cache *nc;
 733	struct sk_buff *skb;
 734	bool pfmemalloc;
 735	void *data;
 736
 737	len += NET_SKB_PAD;
 
 
 738
 739	/* If requested length is either too small or too big,
 740	 * we use kmalloc() for skb->head allocation.
 741	 */
 742	if (len <= SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE) ||
 743	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
 744	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 745		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
 746		if (!skb)
 747			goto skb_fail;
 748		goto skb_success;
 749	}
 750
 751	len = SKB_HEAD_ALIGN(len);
 752
 753	if (sk_memalloc_socks())
 754		gfp_mask |= __GFP_MEMALLOC;
 755
 756	if (in_hardirq() || irqs_disabled()) {
 757		nc = this_cpu_ptr(&netdev_alloc_cache);
 758		data = page_frag_alloc(nc, len, gfp_mask);
 759		pfmemalloc = page_frag_cache_is_pfmemalloc(nc);
 760	} else {
 761		local_bh_disable();
 762		local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 763
 764		nc = this_cpu_ptr(&napi_alloc_cache.page);
 765		data = page_frag_alloc(nc, len, gfp_mask);
 766		pfmemalloc = page_frag_cache_is_pfmemalloc(nc);
 
 
 
 
 
 
 
 
 767
 768		local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 769		local_bh_enable();
 770	}
 
 
 771
 772	if (unlikely(!data))
 773		return NULL;
 
 774
 775	skb = __build_skb(data, len);
 776	if (unlikely(!skb)) {
 777		skb_free_frag(data);
 778		return NULL;
 779	}
 780
 781	if (pfmemalloc)
 782		skb->pfmemalloc = 1;
 783	skb->head_frag = 1;
 784
 785skb_success:
 786	skb_reserve(skb, NET_SKB_PAD);
 787	skb->dev = dev;
 
 
 
 
 
 788
 789skb_fail:
 790	return skb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 791}
 792EXPORT_SYMBOL(__netdev_alloc_skb);
 793
 794/**
 795 *	napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
 796 *	@napi: napi instance this buffer was allocated for
 797 *	@len: length to allocate
 798 *
 799 *	Allocate a new sk_buff for use in NAPI receive.  This buffer will
 800 *	attempt to allocate the head from a special reserved region used
 801 *	only for NAPI Rx allocation.  By doing this we can save several
 802 *	CPU cycles by avoiding having to disable and re-enable IRQs.
 
 803 *
 804 *	%NULL is returned if there is no free memory.
 805 */
 806struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len)
 
 807{
 808	gfp_t gfp_mask = GFP_ATOMIC | __GFP_NOWARN;
 809	struct napi_alloc_cache *nc;
 810	struct sk_buff *skb;
 811	bool pfmemalloc;
 812	void *data;
 813
 814	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
 815	len += NET_SKB_PAD + NET_IP_ALIGN;
 816
 817	/* If requested length is either too small or too big,
 818	 * we use kmalloc() for skb->head allocation.
 819	 * When the small frag allocator is available, prefer it over kmalloc
 820	 * for small fragments
 821	 */
 822	if ((!NAPI_HAS_SMALL_PAGE_FRAG &&
 823	     len <= SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE)) ||
 824	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
 825	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
 826		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
 827				  NUMA_NO_NODE);
 828		if (!skb)
 829			goto skb_fail;
 830		goto skb_success;
 831	}
 832
 833	if (sk_memalloc_socks())
 834		gfp_mask |= __GFP_MEMALLOC;
 835
 836	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 837	nc = this_cpu_ptr(&napi_alloc_cache);
 838	if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
 839		/* we are artificially inflating the allocation size, but
 840		 * that is not as bad as it may look like, as:
 841		 * - 'len' less than GRO_MAX_HEAD makes little sense
 842		 * - On most systems, larger 'len' values lead to fragment
 843		 *   size above 512 bytes
 844		 * - kmalloc would use the kmalloc-1k slab for such values
 845		 * - Builds with smaller GRO_MAX_HEAD will very likely do
 846		 *   little networking, as that implies no WiFi and no
 847		 *   tunnels support, and 32 bits arches.
 848		 */
 849		len = SZ_1K;
 850
 851		data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
 852		pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
 853	} else {
 854		len = SKB_HEAD_ALIGN(len);
 855
 856		data = page_frag_alloc(&nc->page, len, gfp_mask);
 857		pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page);
 858	}
 859	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 860
 861	if (unlikely(!data))
 862		return NULL;
 863
 864	skb = __napi_build_skb(data, len);
 865	if (unlikely(!skb)) {
 866		skb_free_frag(data);
 867		return NULL;
 868	}
 869
 870	if (pfmemalloc)
 871		skb->pfmemalloc = 1;
 872	skb->head_frag = 1;
 873
 874skb_success:
 875	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
 876	skb->dev = napi->dev;
 877
 878skb_fail:
 879	return skb;
 880}
 881EXPORT_SYMBOL(napi_alloc_skb);
 882
 883void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
 884			    int off, int size, unsigned int truesize)
 885{
 886	DEBUG_NET_WARN_ON_ONCE(size > truesize);
 887
 888	skb_fill_netmem_desc(skb, i, netmem, off, size);
 889	skb->len += size;
 890	skb->data_len += size;
 891	skb->truesize += truesize;
 892}
 893EXPORT_SYMBOL(skb_add_rx_frag_netmem);
 894
 895void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
 896			  unsigned int truesize)
 897{
 898	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 899
 900	DEBUG_NET_WARN_ON_ONCE(size > truesize);
 901
 902	skb_frag_size_add(frag, size);
 903	skb->len += size;
 904	skb->data_len += size;
 905	skb->truesize += truesize;
 906}
 907EXPORT_SYMBOL(skb_coalesce_rx_frag);
 908
 909static void skb_drop_list(struct sk_buff **listp)
 910{
 911	kfree_skb_list(*listp);
 
 912	*listp = NULL;
 
 
 
 
 
 
 913}
 914
 915static inline void skb_drop_fraglist(struct sk_buff *skb)
 916{
 917	skb_drop_list(&skb_shinfo(skb)->frag_list);
 918}
 919
 920static void skb_clone_fraglist(struct sk_buff *skb)
 921{
 922	struct sk_buff *list;
 923
 924	skb_walk_frags(skb, list)
 925		skb_get(list);
 926}
 927
 928static bool is_pp_netmem(netmem_ref netmem)
 929{
 930	return (netmem_get_pp_magic(netmem) & ~0x3UL) == PP_SIGNATURE;
 
 
 
 931}
 932
 933int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
 934		    unsigned int headroom)
 935{
 936#if IS_ENABLED(CONFIG_PAGE_POOL)
 937	u32 size, truesize, len, max_head_size, off;
 938	struct sk_buff *skb = *pskb, *nskb;
 939	int err, i, head_off;
 940	void *data;
 941
 942	/* XDP does not support fraglist so we need to linearize
 943	 * the skb.
 944	 */
 945	if (skb_has_frag_list(skb))
 946		return -EOPNOTSUPP;
 947
 948	max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE - headroom);
 949	if (skb->len > max_head_size + MAX_SKB_FRAGS * PAGE_SIZE)
 950		return -ENOMEM;
 951
 952	size = min_t(u32, skb->len, max_head_size);
 953	truesize = SKB_HEAD_ALIGN(size) + headroom;
 954	data = page_pool_dev_alloc_va(pool, &truesize);
 955	if (!data)
 956		return -ENOMEM;
 957
 958	nskb = napi_build_skb(data, truesize);
 959	if (!nskb) {
 960		page_pool_free_va(pool, data, true);
 961		return -ENOMEM;
 962	}
 963
 964	skb_reserve(nskb, headroom);
 965	skb_copy_header(nskb, skb);
 966	skb_mark_for_recycle(nskb);
 967
 968	err = skb_copy_bits(skb, 0, nskb->data, size);
 969	if (err) {
 970		consume_skb(nskb);
 971		return err;
 972	}
 973	skb_put(nskb, size);
 974
 975	head_off = skb_headroom(nskb) - skb_headroom(skb);
 976	skb_headers_offset_update(nskb, head_off);
 977
 978	off = size;
 979	len = skb->len - off;
 980	for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
 981		struct page *page;
 982		u32 page_off;
 983
 984		size = min_t(u32, len, PAGE_SIZE);
 985		truesize = size;
 986
 987		page = page_pool_dev_alloc(pool, &page_off, &truesize);
 988		if (!page) {
 989			consume_skb(nskb);
 990			return -ENOMEM;
 991		}
 992
 993		skb_add_rx_frag(nskb, i, page, page_off, size, truesize);
 994		err = skb_copy_bits(skb, off, page_address(page) + page_off,
 995				    size);
 996		if (err) {
 997			consume_skb(nskb);
 998			return err;
 999		}
1000
1001		len -= size;
1002		off += size;
1003	}
1004
1005	consume_skb(skb);
1006	*pskb = nskb;
1007
1008	return 0;
1009#else
1010	return -EOPNOTSUPP;
1011#endif
1012}
1013EXPORT_SYMBOL(skb_pp_cow_data);
1014
1015int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
1016			 struct bpf_prog *prog)
1017{
1018	if (!prog->aux->xdp_has_frags)
1019		return -EINVAL;
1020
1021	return skb_pp_cow_data(pool, pskb, XDP_PACKET_HEADROOM);
1022}
1023EXPORT_SYMBOL(skb_cow_data_for_xdp);
1024
1025#if IS_ENABLED(CONFIG_PAGE_POOL)
1026bool napi_pp_put_page(netmem_ref netmem)
1027{
1028	netmem = netmem_compound_head(netmem);
1029
1030	/* page->pp_magic is OR'ed with PP_SIGNATURE after the allocation
1031	 * in order to preserve any existing bits, such as bit 0 for the
1032	 * head page of compound page and bit 1 for pfmemalloc page, so
1033	 * mask those bits for freeing side when doing below checking,
1034	 * and page_is_pfmemalloc() is checked in __page_pool_put_page()
1035	 * to avoid recycling the pfmemalloc page.
1036	 */
1037	if (unlikely(!is_pp_netmem(netmem)))
1038		return false;
1039
1040	page_pool_put_full_netmem(netmem_get_pp(netmem), netmem, false);
1041
1042	return true;
1043}
1044EXPORT_SYMBOL(napi_pp_put_page);
1045#endif
1046
1047static bool skb_pp_recycle(struct sk_buff *skb, void *data)
1048{
1049	if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
1050		return false;
1051	return napi_pp_put_page(page_to_netmem(virt_to_page(data)));
1052}
1053
1054/**
1055 * skb_pp_frag_ref() - Increase fragment references of a page pool aware skb
1056 * @skb:	page pool aware skb
1057 *
1058 * Increase the fragment reference count (pp_ref_count) of a skb. This is
1059 * intended to gain fragment references only for page pool aware skbs,
1060 * i.e. when skb->pp_recycle is true, and not for fragments in a
1061 * non-pp-recycling skb. It has a fallback to increase references on normal
1062 * pages, as page pool aware skbs may also have normal page fragments.
1063 */
1064static int skb_pp_frag_ref(struct sk_buff *skb)
1065{
1066	struct skb_shared_info *shinfo;
1067	netmem_ref head_netmem;
1068	int i;
1069
1070	if (!skb->pp_recycle)
1071		return -EINVAL;
1072
1073	shinfo = skb_shinfo(skb);
1074
1075	for (i = 0; i < shinfo->nr_frags; i++) {
1076		head_netmem = netmem_compound_head(shinfo->frags[i].netmem);
1077		if (likely(is_pp_netmem(head_netmem)))
1078			page_pool_ref_netmem(head_netmem);
1079		else
1080			page_ref_inc(netmem_to_page(head_netmem));
1081	}
1082	return 0;
1083}
1084
1085static void skb_kfree_head(void *head, unsigned int end_offset)
1086{
1087	if (end_offset == SKB_SMALL_HEAD_HEADROOM)
1088		kmem_cache_free(net_hotdata.skb_small_head_cache, head);
1089	else
1090		kfree(head);
1091}
1092
1093static void skb_free_head(struct sk_buff *skb)
1094{
1095	unsigned char *head = skb->head;
1096
1097	if (skb->head_frag) {
1098		if (skb_pp_recycle(skb, head))
1099			return;
1100		skb_free_frag(head);
1101	} else {
1102		skb_kfree_head(head, skb_end_offset(skb));
1103	}
1104}
1105
1106static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
1107{
1108	struct skb_shared_info *shinfo = skb_shinfo(skb);
1109	int i;
1110
1111	if (!skb_data_unref(skb, shinfo))
1112		goto exit;
1113
1114	if (skb_zcopy(skb)) {
1115		bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
1116
1117		skb_zcopy_clear(skb, true);
1118		if (skip_unref)
1119			goto free_head;
1120	}
1121
1122	for (i = 0; i < shinfo->nr_frags; i++)
1123		__skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
1124
1125free_head:
1126	if (shinfo->frag_list)
1127		kfree_skb_list_reason(shinfo->frag_list, reason);
1128
1129	skb_free_head(skb);
1130exit:
1131	/* When we clone an SKB we copy the reycling bit. The pp_recycle
1132	 * bit is only set on the head though, so in order to avoid races
1133	 * while trying to recycle fragments on __skb_frag_unref() we need
1134	 * to make one SKB responsible for triggering the recycle path.
1135	 * So disable the recycling bit if an SKB is cloned and we have
1136	 * additional references to the fragmented part of the SKB.
1137	 * Eventually the last SKB will have the recycling bit set and it's
1138	 * dataref set to 0, which will trigger the recycling
1139	 */
1140	skb->pp_recycle = 0;
1141}
1142
1143/*
1144 *	Free an skbuff by memory without cleaning the state.
1145 */
1146static void kfree_skbmem(struct sk_buff *skb)
1147{
1148	struct sk_buff_fclones *fclones;
 
1149
1150	switch (skb->fclone) {
1151	case SKB_FCLONE_UNAVAILABLE:
1152		kmem_cache_free(net_hotdata.skbuff_cache, skb);
1153		return;
1154
1155	case SKB_FCLONE_ORIG:
1156		fclones = container_of(skb, struct sk_buff_fclones, skb1);
 
 
 
1157
1158		/* We usually free the clone (TX completion) before original skb
1159		 * This test would have no chance to be true for the clone,
1160		 * while here, branch prediction will be good.
 
 
 
1161		 */
1162		if (refcount_read(&fclones->fclone_ref) == 1)
1163			goto fastpath;
1164		break;
1165
1166	default: /* SKB_FCLONE_CLONE */
1167		fclones = container_of(skb, struct sk_buff_fclones, skb2);
1168		break;
1169	}
1170	if (!refcount_dec_and_test(&fclones->fclone_ref))
1171		return;
1172fastpath:
1173	kmem_cache_free(net_hotdata.skbuff_fclone_cache, fclones);
1174}
1175
1176void skb_release_head_state(struct sk_buff *skb)
1177{
1178	skb_dst_drop(skb);
 
 
 
1179	if (skb->destructor) {
1180		DEBUG_NET_WARN_ON_ONCE(in_hardirq());
1181		skb->destructor(skb);
1182	}
1183#if IS_ENABLED(CONFIG_NF_CONNTRACK)
1184	nf_conntrack_put(skb_nfct(skb));
 
 
 
 
 
 
 
 
 
 
 
 
 
1185#endif
1186	skb_ext_put(skb);
1187}
1188
1189/* Free everything but the sk_buff shell. */
1190static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
1191{
1192	skb_release_head_state(skb);
1193	if (likely(skb->head))
1194		skb_release_data(skb, reason);
1195}
1196
1197/**
1198 *	__kfree_skb - private function
1199 *	@skb: buffer
1200 *
1201 *	Free an sk_buff. Release anything attached to the buffer.
1202 *	Clean the state. This is an internal helper function. Users should
1203 *	always call kfree_skb
1204 */
1205
1206void __kfree_skb(struct sk_buff *skb)
1207{
1208	skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1209	kfree_skbmem(skb);
1210}
1211EXPORT_SYMBOL(__kfree_skb);
1212
1213static __always_inline
1214bool __sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1215			  enum skb_drop_reason reason)
1216{
1217	if (unlikely(!skb_unref(skb)))
1218		return false;
1219
1220	DEBUG_NET_WARN_ON_ONCE(reason == SKB_NOT_DROPPED_YET ||
1221			       u32_get_bits(reason,
1222					    SKB_DROP_REASON_SUBSYS_MASK) >=
1223				SKB_DROP_REASON_SUBSYS_NUM);
1224
1225	if (reason == SKB_CONSUMED)
1226		trace_consume_skb(skb, __builtin_return_address(0));
1227	else
1228		trace_kfree_skb(skb, __builtin_return_address(0), reason, sk);
1229	return true;
1230}
1231
1232/**
1233 *	sk_skb_reason_drop - free an sk_buff with special reason
1234 *	@sk: the socket to receive @skb, or NULL if not applicable
1235 *	@skb: buffer to free
1236 *	@reason: reason why this skb is dropped
1237 *
1238 *	Drop a reference to the buffer and free it if the usage count has hit
1239 *	zero. Meanwhile, pass the receiving socket and drop reason to
1240 *	'kfree_skb' tracepoint.
1241 */
1242void __fix_address
1243sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, enum skb_drop_reason reason)
1244{
1245	if (__sk_skb_reason_drop(sk, skb, reason))
1246		__kfree_skb(skb);
1247}
1248EXPORT_SYMBOL(sk_skb_reason_drop);
1249
1250#define KFREE_SKB_BULK_SIZE	16
1251
1252struct skb_free_array {
1253	unsigned int skb_count;
1254	void *skb_array[KFREE_SKB_BULK_SIZE];
1255};
1256
1257static void kfree_skb_add_bulk(struct sk_buff *skb,
1258			       struct skb_free_array *sa,
1259			       enum skb_drop_reason reason)
1260{
1261	/* if SKB is a clone, don't handle this case */
1262	if (unlikely(skb->fclone != SKB_FCLONE_UNAVAILABLE)) {
1263		__kfree_skb(skb);
1264		return;
1265	}
1266
1267	skb_release_all(skb, reason);
1268	sa->skb_array[sa->skb_count++] = skb;
1269
1270	if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
1271		kmem_cache_free_bulk(net_hotdata.skbuff_cache, KFREE_SKB_BULK_SIZE,
1272				     sa->skb_array);
1273		sa->skb_count = 0;
1274	}
1275}
1276
1277void __fix_address
1278kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
1279{
1280	struct skb_free_array sa;
1281
1282	sa.skb_count = 0;
1283
1284	while (segs) {
1285		struct sk_buff *next = segs->next;
1286
1287		if (__sk_skb_reason_drop(NULL, segs, reason)) {
1288			skb_poison_list(segs);
1289			kfree_skb_add_bulk(segs, &sa, reason);
1290		}
1291
1292		segs = next;
1293	}
1294
1295	if (sa.skb_count)
1296		kmem_cache_free_bulk(net_hotdata.skbuff_cache, sa.skb_count, sa.skb_array);
1297}
1298EXPORT_SYMBOL(kfree_skb_list_reason);
1299
1300/* Dump skb information and contents.
1301 *
1302 * Must only be called from net_ratelimit()-ed paths.
1303 *
1304 * Dumps whole packets if full_pkt, only headers otherwise.
1305 */
1306void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
1307{
1308	struct skb_shared_info *sh = skb_shinfo(skb);
1309	struct net_device *dev = skb->dev;
1310	struct sock *sk = skb->sk;
1311	struct sk_buff *list_skb;
1312	bool has_mac, has_trans;
1313	int headroom, tailroom;
1314	int i, len, seg_len;
1315
1316	if (full_pkt)
1317		len = skb->len;
1318	else
1319		len = min_t(int, skb->len, MAX_HEADER + 128);
1320
1321	headroom = skb_headroom(skb);
1322	tailroom = skb_tailroom(skb);
1323
1324	has_mac = skb_mac_header_was_set(skb);
1325	has_trans = skb_transport_header_was_set(skb);
1326
1327	printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1328	       "mac=(%d,%d) mac_len=%u net=(%d,%d) trans=%d\n"
1329	       "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1330	       "csum(0x%x start=%u offset=%u ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1331	       "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n"
1332	       "priority=0x%x mark=0x%x alloc_cpu=%u vlan_all=0x%x\n"
1333	       "encapsulation=%d inner(proto=0x%04x, mac=%u, net=%u, trans=%u)\n",
1334	       level, skb->len, headroom, skb_headlen(skb), tailroom,
1335	       has_mac ? skb->mac_header : -1,
1336	       has_mac ? skb_mac_header_len(skb) : -1,
1337	       skb->mac_len,
1338	       skb->network_header,
1339	       has_trans ? skb_network_header_len(skb) : -1,
1340	       has_trans ? skb->transport_header : -1,
1341	       sh->tx_flags, sh->nr_frags,
1342	       sh->gso_size, sh->gso_type, sh->gso_segs,
1343	       skb->csum, skb->csum_start, skb->csum_offset, skb->ip_summed,
1344	       skb->csum_complete_sw, skb->csum_valid, skb->csum_level,
1345	       skb->hash, skb->sw_hash, skb->l4_hash,
1346	       ntohs(skb->protocol), skb->pkt_type, skb->skb_iif,
1347	       skb->priority, skb->mark, skb->alloc_cpu, skb->vlan_all,
1348	       skb->encapsulation, skb->inner_protocol, skb->inner_mac_header,
1349	       skb->inner_network_header, skb->inner_transport_header);
1350
1351	if (dev)
1352		printk("%sdev name=%s feat=%pNF\n",
1353		       level, dev->name, &dev->features);
1354	if (sk)
1355		printk("%ssk family=%hu type=%u proto=%u\n",
1356		       level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1357
1358	if (full_pkt && headroom)
1359		print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1360			       16, 1, skb->head, headroom, false);
1361
1362	seg_len = min_t(int, skb_headlen(skb), len);
1363	if (seg_len)
1364		print_hex_dump(level, "skb linear:   ", DUMP_PREFIX_OFFSET,
1365			       16, 1, skb->data, seg_len, false);
1366	len -= seg_len;
1367
1368	if (full_pkt && tailroom)
1369		print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1370			       16, 1, skb_tail_pointer(skb), tailroom, false);
1371
1372	for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1373		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1374		u32 p_off, p_len, copied;
1375		struct page *p;
1376		u8 *vaddr;
1377
1378		if (skb_frag_is_net_iov(frag)) {
1379			printk("%sskb frag %d: not readable\n", level, i);
1380			len -= skb_frag_size(frag);
1381			if (!len)
1382				break;
1383			continue;
1384		}
1385
1386		skb_frag_foreach_page(frag, skb_frag_off(frag),
1387				      skb_frag_size(frag), p, p_off, p_len,
1388				      copied) {
1389			seg_len = min_t(int, p_len, len);
1390			vaddr = kmap_atomic(p);
1391			print_hex_dump(level, "skb frag:     ",
1392				       DUMP_PREFIX_OFFSET,
1393				       16, 1, vaddr + p_off, seg_len, false);
1394			kunmap_atomic(vaddr);
1395			len -= seg_len;
1396			if (!len)
1397				break;
1398		}
1399	}
1400
1401	if (full_pkt && skb_has_frag_list(skb)) {
1402		printk("skb fraglist:\n");
1403		skb_walk_frags(skb, list_skb)
1404			skb_dump(level, list_skb, true);
1405	}
1406}
1407EXPORT_SYMBOL(skb_dump);
1408
1409/**
1410 *	skb_tx_error - report an sk_buff xmit error
1411 *	@skb: buffer that triggered an error
1412 *
1413 *	Report xmit error if a device callback is tracking this skb.
1414 *	skb must be freed afterwards.
1415 */
1416void skb_tx_error(struct sk_buff *skb)
1417{
1418	if (skb) {
1419		skb_zcopy_downgrade_managed(skb);
1420		skb_zcopy_clear(skb, true);
1421	}
1422}
1423EXPORT_SYMBOL(skb_tx_error);
1424
1425#ifdef CONFIG_TRACEPOINTS
1426/**
1427 *	consume_skb - free an skbuff
1428 *	@skb: buffer to free
1429 *
1430 *	Drop a ref to the buffer and free it if the usage count has hit zero
1431 *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
1432 *	is being dropped after a failure and notes that
1433 */
1434void consume_skb(struct sk_buff *skb)
1435{
1436	if (!skb_unref(skb))
1437		return;
1438
1439	trace_consume_skb(skb, __builtin_return_address(0));
 
 
 
1440	__kfree_skb(skb);
1441}
1442EXPORT_SYMBOL(consume_skb);
1443#endif
1444
1445/**
1446 *	__consume_stateless_skb - free an skbuff, assuming it is stateless
1447 *	@skb: buffer to free
1448 *
1449 *	Alike consume_skb(), but this variant assumes that this is the last
1450 *	skb reference and all the head states have been already dropped
 
1451 */
1452void __consume_stateless_skb(struct sk_buff *skb)
1453{
1454	trace_consume_skb(skb, __builtin_return_address(0));
1455	skb_release_data(skb, SKB_CONSUMED);
1456	kfree_skbmem(skb);
1457}
1458
1459static void napi_skb_cache_put(struct sk_buff *skb)
1460{
1461	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1462	u32 i;
1463
1464	if (!kasan_mempool_poison_object(skb))
1465		return;
1466
1467	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
1468	nc->skb_cache[nc->skb_count++] = skb;
1469
1470	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1471		for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1472			kasan_mempool_unpoison_object(nc->skb_cache[i],
1473						kmem_cache_size(net_hotdata.skbuff_cache));
1474
1475		kmem_cache_free_bulk(net_hotdata.skbuff_cache, NAPI_SKB_CACHE_HALF,
1476				     nc->skb_cache + NAPI_SKB_CACHE_HALF);
1477		nc->skb_count = NAPI_SKB_CACHE_HALF;
1478	}
1479	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
1480}
1481
1482void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
1483{
1484	skb_release_all(skb, reason);
1485	napi_skb_cache_put(skb);
1486}
1487
1488void napi_skb_free_stolen_head(struct sk_buff *skb)
1489{
1490	if (unlikely(skb->slow_gro)) {
1491		nf_reset_ct(skb);
1492		skb_dst_drop(skb);
1493		skb_ext_put(skb);
1494		skb_orphan(skb);
1495		skb->slow_gro = 0;
1496	}
1497	napi_skb_cache_put(skb);
1498}
 
1499
1500void napi_consume_skb(struct sk_buff *skb, int budget)
 
 
 
 
 
 
 
 
 
 
 
 
1501{
1502	/* Zero budget indicate non-NAPI context called us, like netpoll */
1503	if (unlikely(!budget)) {
1504		dev_consume_skb_any(skb);
1505		return;
1506	}
1507
1508	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1509
1510	if (!skb_unref(skb))
1511		return;
1512
1513	/* if reaching here SKB is ready to free */
1514	trace_consume_skb(skb, __builtin_return_address(0));
1515
1516	/* if SKB is a clone, don't handle this case */
1517	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1518		__kfree_skb(skb);
1519		return;
1520	}
1521
1522	skb_release_all(skb, SKB_CONSUMED);
1523	napi_skb_cache_put(skb);
1524}
1525EXPORT_SYMBOL(napi_consume_skb);
1526
1527/* Make sure a field is contained by headers group */
1528#define CHECK_SKB_FIELD(field) \
1529	BUILD_BUG_ON(offsetof(struct sk_buff, field) !=		\
1530		     offsetof(struct sk_buff, headers.field));	\
1531
1532static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1533{
1534	new->tstamp		= old->tstamp;
1535	/* We do not copy old->sk */
1536	new->dev		= old->dev;
1537	memcpy(new->cb, old->cb, sizeof(old->cb));
 
 
1538	skb_dst_copy(new, old);
1539	__skb_ext_copy(new, old);
1540	__nf_copy(new, old, false);
1541
1542	/* Note : this field could be in the headers group.
1543	 * It is not yet because we do not want to have a 16 bit hole
1544	 */
1545	new->queue_mapping = old->queue_mapping;
1546
1547	memcpy(&new->headers, &old->headers, sizeof(new->headers));
1548	CHECK_SKB_FIELD(protocol);
1549	CHECK_SKB_FIELD(csum);
1550	CHECK_SKB_FIELD(hash);
1551	CHECK_SKB_FIELD(priority);
1552	CHECK_SKB_FIELD(skb_iif);
1553	CHECK_SKB_FIELD(vlan_proto);
1554	CHECK_SKB_FIELD(vlan_tci);
1555	CHECK_SKB_FIELD(transport_header);
1556	CHECK_SKB_FIELD(network_header);
1557	CHECK_SKB_FIELD(mac_header);
1558	CHECK_SKB_FIELD(inner_protocol);
1559	CHECK_SKB_FIELD(inner_transport_header);
1560	CHECK_SKB_FIELD(inner_network_header);
1561	CHECK_SKB_FIELD(inner_mac_header);
1562	CHECK_SKB_FIELD(mark);
1563#ifdef CONFIG_NETWORK_SECMARK
1564	CHECK_SKB_FIELD(secmark);
1565#endif
1566#ifdef CONFIG_NET_RX_BUSY_POLL
1567	CHECK_SKB_FIELD(napi_id);
1568#endif
1569	CHECK_SKB_FIELD(alloc_cpu);
1570#ifdef CONFIG_XPS
1571	CHECK_SKB_FIELD(sender_cpu);
 
 
 
 
 
 
 
 
 
 
 
 
 
1572#endif
1573#ifdef CONFIG_NET_SCHED
1574	CHECK_SKB_FIELD(tc_index);
 
 
 
1575#endif
 
1576
 
1577}
1578
1579/*
1580 * You should not add any new code to this function.  Add it to
1581 * __copy_skb_header above instead.
1582 */
1583static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1584{
1585#define C(x) n->x = skb->x
1586
1587	n->next = n->prev = NULL;
1588	n->sk = NULL;
1589	__copy_skb_header(n, skb);
1590
1591	C(len);
1592	C(data_len);
1593	C(mac_len);
1594	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1595	n->cloned = 1;
1596	n->nohdr = 0;
1597	n->peeked = 0;
1598	C(pfmemalloc);
1599	C(pp_recycle);
1600	n->destructor = NULL;
1601	C(tail);
1602	C(end);
1603	C(head);
1604	C(head_frag);
1605	C(data);
1606	C(truesize);
1607	refcount_set(&n->users, 1);
1608
1609	atomic_inc(&(skb_shinfo(skb)->dataref));
1610	skb->cloned = 1;
1611
1612	return n;
1613#undef C
1614}
1615
1616/**
1617 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1618 * @first: first sk_buff of the msg
1619 */
1620struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1621{
1622	struct sk_buff *n;
1623
1624	n = alloc_skb(0, GFP_ATOMIC);
1625	if (!n)
1626		return NULL;
1627
1628	n->len = first->len;
1629	n->data_len = first->len;
1630	n->truesize = first->truesize;
1631
1632	skb_shinfo(n)->frag_list = first;
1633
1634	__copy_skb_header(n, first);
1635	n->destructor = NULL;
1636
1637	return n;
1638}
1639EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1640
1641/**
1642 *	skb_morph	-	morph one skb into another
1643 *	@dst: the skb to receive the contents
1644 *	@src: the skb to supply the contents
1645 *
1646 *	This is identical to skb_clone except that the target skb is
1647 *	supplied by the user.
1648 *
1649 *	The target skb is returned upon exit.
1650 */
1651struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1652{
1653	skb_release_all(dst, SKB_CONSUMED);
1654	return __skb_clone(dst, src);
1655}
1656EXPORT_SYMBOL_GPL(skb_morph);
1657
1658int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1659{
1660	unsigned long max_pg, num_pg, new_pg, old_pg, rlim;
1661	struct user_struct *user;
1662
1663	if (capable(CAP_IPC_LOCK) || !size)
1664		return 0;
1665
1666	rlim = rlimit(RLIMIT_MEMLOCK);
1667	if (rlim == RLIM_INFINITY)
1668		return 0;
1669
1670	num_pg = (size >> PAGE_SHIFT) + 2;	/* worst case */
1671	max_pg = rlim >> PAGE_SHIFT;
1672	user = mmp->user ? : current_user();
1673
1674	old_pg = atomic_long_read(&user->locked_vm);
1675	do {
1676		new_pg = old_pg + num_pg;
1677		if (new_pg > max_pg)
1678			return -ENOBUFS;
1679	} while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1680
1681	if (!mmp->user) {
1682		mmp->user = get_uid(user);
1683		mmp->num_pg = num_pg;
1684	} else {
1685		mmp->num_pg += num_pg;
1686	}
1687
1688	return 0;
1689}
1690EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1691
1692void mm_unaccount_pinned_pages(struct mmpin *mmp)
1693{
1694	if (mmp->user) {
1695		atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1696		free_uid(mmp->user);
1697	}
1698}
1699EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1700
1701static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1702{
1703	struct ubuf_info_msgzc *uarg;
1704	struct sk_buff *skb;
1705
1706	WARN_ON_ONCE(!in_task());
1707
1708	skb = sock_omalloc(sk, 0, GFP_KERNEL);
1709	if (!skb)
1710		return NULL;
1711
1712	BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1713	uarg = (void *)skb->cb;
1714	uarg->mmp.user = NULL;
1715
1716	if (mm_account_pinned_pages(&uarg->mmp, size)) {
1717		kfree_skb(skb);
1718		return NULL;
1719	}
1720
1721	uarg->ubuf.ops = &msg_zerocopy_ubuf_ops;
1722	uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1723	uarg->len = 1;
1724	uarg->bytelen = size;
1725	uarg->zerocopy = 1;
1726	uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1727	refcount_set(&uarg->ubuf.refcnt, 1);
1728	sock_hold(sk);
1729
1730	return &uarg->ubuf;
1731}
1732
1733static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1734{
1735	return container_of((void *)uarg, struct sk_buff, cb);
1736}
1737
1738struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1739				       struct ubuf_info *uarg)
1740{
1741	if (uarg) {
1742		struct ubuf_info_msgzc *uarg_zc;
1743		const u32 byte_limit = 1 << 19;		/* limit to a few TSO */
1744		u32 bytelen, next;
1745
1746		/* there might be non MSG_ZEROCOPY users */
1747		if (uarg->ops != &msg_zerocopy_ubuf_ops)
1748			return NULL;
1749
1750		/* realloc only when socket is locked (TCP, UDP cork),
1751		 * so uarg->len and sk_zckey access is serialized
1752		 */
1753		if (!sock_owned_by_user(sk)) {
1754			WARN_ON_ONCE(1);
1755			return NULL;
1756		}
1757
1758		uarg_zc = uarg_to_msgzc(uarg);
1759		bytelen = uarg_zc->bytelen + size;
1760		if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1761			/* TCP can create new skb to attach new uarg */
1762			if (sk->sk_type == SOCK_STREAM)
1763				goto new_alloc;
1764			return NULL;
1765		}
1766
1767		next = (u32)atomic_read(&sk->sk_zckey);
1768		if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1769			if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1770				return NULL;
1771			uarg_zc->len++;
1772			uarg_zc->bytelen = bytelen;
1773			atomic_set(&sk->sk_zckey, ++next);
1774
1775			/* no extra ref when appending to datagram (MSG_MORE) */
1776			if (sk->sk_type == SOCK_STREAM)
1777				net_zcopy_get(uarg);
1778
1779			return uarg;
1780		}
1781	}
1782
1783new_alloc:
1784	return msg_zerocopy_alloc(sk, size);
1785}
1786EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1787
1788static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1789{
1790	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1791	u32 old_lo, old_hi;
1792	u64 sum_len;
1793
1794	old_lo = serr->ee.ee_info;
1795	old_hi = serr->ee.ee_data;
1796	sum_len = old_hi - old_lo + 1ULL + len;
1797
1798	if (sum_len >= (1ULL << 32))
1799		return false;
1800
1801	if (lo != old_hi + 1)
1802		return false;
1803
1804	serr->ee.ee_data += len;
1805	return true;
1806}
1807
1808static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1809{
1810	struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1811	struct sock_exterr_skb *serr;
1812	struct sock *sk = skb->sk;
1813	struct sk_buff_head *q;
1814	unsigned long flags;
1815	bool is_zerocopy;
1816	u32 lo, hi;
1817	u16 len;
1818
1819	mm_unaccount_pinned_pages(&uarg->mmp);
1820
1821	/* if !len, there was only 1 call, and it was aborted
1822	 * so do not queue a completion notification
1823	 */
1824	if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1825		goto release;
1826
1827	len = uarg->len;
1828	lo = uarg->id;
1829	hi = uarg->id + len - 1;
1830	is_zerocopy = uarg->zerocopy;
1831
1832	serr = SKB_EXT_ERR(skb);
1833	memset(serr, 0, sizeof(*serr));
1834	serr->ee.ee_errno = 0;
1835	serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1836	serr->ee.ee_data = hi;
1837	serr->ee.ee_info = lo;
1838	if (!is_zerocopy)
1839		serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1840
1841	q = &sk->sk_error_queue;
1842	spin_lock_irqsave(&q->lock, flags);
1843	tail = skb_peek_tail(q);
1844	if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1845	    !skb_zerocopy_notify_extend(tail, lo, len)) {
1846		__skb_queue_tail(q, skb);
1847		skb = NULL;
1848	}
1849	spin_unlock_irqrestore(&q->lock, flags);
1850
1851	sk_error_report(sk);
1852
1853release:
1854	consume_skb(skb);
1855	sock_put(sk);
1856}
1857
1858static void msg_zerocopy_complete(struct sk_buff *skb, struct ubuf_info *uarg,
1859				  bool success)
1860{
1861	struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1862
1863	uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1864
1865	if (refcount_dec_and_test(&uarg->refcnt))
1866		__msg_zerocopy_callback(uarg_zc);
1867}
1868
1869void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1870{
1871	struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1872
1873	atomic_dec(&sk->sk_zckey);
1874	uarg_to_msgzc(uarg)->len--;
1875
1876	if (have_uref)
1877		msg_zerocopy_complete(NULL, uarg, true);
1878}
1879EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1880
1881const struct ubuf_info_ops msg_zerocopy_ubuf_ops = {
1882	.complete = msg_zerocopy_complete,
1883};
1884EXPORT_SYMBOL_GPL(msg_zerocopy_ubuf_ops);
1885
1886int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1887			     struct msghdr *msg, int len,
1888			     struct ubuf_info *uarg)
1889{
1890	int err, orig_len = skb->len;
1891
1892	if (uarg->ops->link_skb) {
1893		err = uarg->ops->link_skb(skb, uarg);
1894		if (err)
1895			return err;
1896	} else {
1897		struct ubuf_info *orig_uarg = skb_zcopy(skb);
1898
1899		/* An skb can only point to one uarg. This edge case happens
1900		 * when TCP appends to an skb, but zerocopy_realloc triggered
1901		 * a new alloc.
1902		 */
1903		if (orig_uarg && uarg != orig_uarg)
1904			return -EEXIST;
1905	}
1906
1907	err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1908	if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1909		struct sock *save_sk = skb->sk;
1910
1911		/* Streams do not free skb on error. Reset to prev state. */
1912		iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1913		skb->sk = sk;
1914		___pskb_trim(skb, orig_len);
1915		skb->sk = save_sk;
1916		return err;
1917	}
1918
1919	skb_zcopy_set(skb, uarg, NULL);
1920	return skb->len - orig_len;
1921}
1922EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1923
1924void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1925{
1926	int i;
1927
1928	skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1929	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1930		skb_frag_ref(skb, i);
1931}
1932EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1933
1934static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1935			      gfp_t gfp_mask)
1936{
1937	if (skb_zcopy(orig)) {
1938		if (skb_zcopy(nskb)) {
1939			/* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1940			if (!gfp_mask) {
1941				WARN_ON_ONCE(1);
1942				return -ENOMEM;
1943			}
1944			if (skb_uarg(nskb) == skb_uarg(orig))
1945				return 0;
1946			if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1947				return -EIO;
1948		}
1949		skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1950	}
1951	return 0;
1952}
1953
1954/**
1955 *	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
1956 *	@skb: the skb to modify
1957 *	@gfp_mask: allocation priority
1958 *
1959 *	This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1960 *	It will copy all frags into kernel and drop the reference
1961 *	to userspace pages.
1962 *
1963 *	If this function is called from an interrupt gfp_mask() must be
1964 *	%GFP_ATOMIC.
1965 *
1966 *	Returns 0 on success or a negative error code on failure
1967 *	to allocate kernel memory to copy to.
1968 */
1969int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1970{
 
1971	int num_frags = skb_shinfo(skb)->nr_frags;
1972	struct page *page, *head = NULL;
1973	int i, order, psize, new_frags;
1974	u32 d_off;
1975
1976	if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1977		return -EINVAL;
1978
1979	if (!skb_frags_readable(skb))
1980		return -EFAULT;
1981
1982	if (!num_frags)
1983		goto release;
1984
1985	/* We might have to allocate high order pages, so compute what minimum
1986	 * page order is needed.
1987	 */
1988	order = 0;
1989	while ((PAGE_SIZE << order) * MAX_SKB_FRAGS < __skb_pagelen(skb))
1990		order++;
1991	psize = (PAGE_SIZE << order);
1992
1993	new_frags = (__skb_pagelen(skb) + psize - 1) >> (PAGE_SHIFT + order);
1994	for (i = 0; i < new_frags; i++) {
1995		page = alloc_pages(gfp_mask | __GFP_COMP, order);
1996		if (!page) {
1997			while (head) {
1998				struct page *next = (struct page *)page_private(head);
1999				put_page(head);
2000				head = next;
2001			}
2002			return -ENOMEM;
2003		}
2004		set_page_private(page, (unsigned long)head);
 
 
 
 
2005		head = page;
2006	}
2007
2008	page = head;
2009	d_off = 0;
2010	for (i = 0; i < num_frags; i++) {
2011		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2012		u32 p_off, p_len, copied;
2013		struct page *p;
2014		u8 *vaddr;
2015
2016		skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
2017				      p, p_off, p_len, copied) {
2018			u32 copy, done = 0;
2019			vaddr = kmap_atomic(p);
2020
2021			while (done < p_len) {
2022				if (d_off == psize) {
2023					d_off = 0;
2024					page = (struct page *)page_private(page);
2025				}
2026				copy = min_t(u32, psize - d_off, p_len - done);
2027				memcpy(page_address(page) + d_off,
2028				       vaddr + p_off + done, copy);
2029				done += copy;
2030				d_off += copy;
2031			}
2032			kunmap_atomic(vaddr);
2033		}
2034	}
2035
2036	/* skb frags release userspace buffers */
2037	for (i = 0; i < num_frags; i++)
2038		skb_frag_unref(skb, i);
2039
 
 
2040	/* skb frags point to kernel buffers */
2041	for (i = 0; i < new_frags - 1; i++) {
2042		__skb_fill_netmem_desc(skb, i, page_to_netmem(head), 0, psize);
2043		head = (struct page *)page_private(head);
2044	}
2045	__skb_fill_netmem_desc(skb, new_frags - 1, page_to_netmem(head), 0,
2046			       d_off);
2047	skb_shinfo(skb)->nr_frags = new_frags;
2048
2049release:
2050	skb_zcopy_clear(skb, false);
2051	return 0;
2052}
2053EXPORT_SYMBOL_GPL(skb_copy_ubufs);
2054
2055/**
2056 *	skb_clone	-	duplicate an sk_buff
2057 *	@skb: buffer to clone
2058 *	@gfp_mask: allocation priority
2059 *
2060 *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
2061 *	copies share the same packet data but not structure. The new
2062 *	buffer has a reference count of 1. If the allocation fails the
2063 *	function returns %NULL otherwise the new buffer is returned.
2064 *
2065 *	If this function is called from an interrupt gfp_mask() must be
2066 *	%GFP_ATOMIC.
2067 */
2068
2069struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
2070{
2071	struct sk_buff_fclones *fclones = container_of(skb,
2072						       struct sk_buff_fclones,
2073						       skb1);
2074	struct sk_buff *n;
2075
2076	if (skb_orphan_frags(skb, gfp_mask))
2077		return NULL;
 
 
2078
 
2079	if (skb->fclone == SKB_FCLONE_ORIG &&
2080	    refcount_read(&fclones->fclone_ref) == 1) {
2081		n = &fclones->skb2;
2082		refcount_set(&fclones->fclone_ref, 2);
2083		n->fclone = SKB_FCLONE_CLONE;
 
2084	} else {
2085		if (skb_pfmemalloc(skb))
2086			gfp_mask |= __GFP_MEMALLOC;
2087
2088		n = kmem_cache_alloc(net_hotdata.skbuff_cache, gfp_mask);
2089		if (!n)
2090			return NULL;
2091
 
 
2092		n->fclone = SKB_FCLONE_UNAVAILABLE;
2093	}
2094
2095	return __skb_clone(n, skb);
2096}
2097EXPORT_SYMBOL(skb_clone);
2098
2099void skb_headers_offset_update(struct sk_buff *skb, int off)
2100{
2101	/* Only adjust this if it actually is csum_start rather than csum */
2102	if (skb->ip_summed == CHECKSUM_PARTIAL)
2103		skb->csum_start += off;
2104	/* {transport,network,mac}_header and tail are relative to skb->head */
2105	skb->transport_header += off;
2106	skb->network_header   += off;
2107	if (skb_mac_header_was_set(skb))
2108		skb->mac_header += off;
2109	skb->inner_transport_header += off;
2110	skb->inner_network_header += off;
2111	skb->inner_mac_header += off;
2112}
2113EXPORT_SYMBOL(skb_headers_offset_update);
2114
2115void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
2116{
2117	__copy_skb_header(new, old);
2118
 
 
 
 
 
 
 
2119	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
2120	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
2121	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
2122}
2123EXPORT_SYMBOL(skb_copy_header);
2124
2125static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
2126{
2127	if (skb_pfmemalloc(skb))
2128		return SKB_ALLOC_RX;
2129	return 0;
2130}
2131
2132/**
2133 *	skb_copy	-	create private copy of an sk_buff
2134 *	@skb: buffer to copy
2135 *	@gfp_mask: allocation priority
2136 *
2137 *	Make a copy of both an &sk_buff and its data. This is used when the
2138 *	caller wishes to modify the data and needs a private copy of the
2139 *	data to alter. Returns %NULL on failure or the pointer to the buffer
2140 *	on success. The returned buffer has a reference count of 1.
2141 *
2142 *	As by-product this function converts non-linear &sk_buff to linear
2143 *	one, so that &sk_buff becomes completely private and caller is allowed
2144 *	to modify all the data of returned buffer. This means that this
2145 *	function is not recommended for use in circumstances when only
2146 *	header is going to be modified. Use pskb_copy() instead.
2147 */
2148
2149struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
2150{
2151	struct sk_buff *n;
2152	unsigned int size;
2153	int headerlen;
2154
2155	if (!skb_frags_readable(skb))
2156		return NULL;
2157
2158	if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2159		return NULL;
2160
2161	headerlen = skb_headroom(skb);
2162	size = skb_end_offset(skb) + skb->data_len;
2163	n = __alloc_skb(size, gfp_mask,
2164			skb_alloc_rx_flag(skb), NUMA_NO_NODE);
2165	if (!n)
2166		return NULL;
2167
2168	/* Set the data pointer */
2169	skb_reserve(n, headerlen);
2170	/* Set the tail pointer and length */
2171	skb_put(n, skb->len);
2172
2173	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
 
2174
2175	skb_copy_header(n, skb);
2176	return n;
2177}
2178EXPORT_SYMBOL(skb_copy);
2179
2180/**
2181 *	__pskb_copy_fclone	-  create copy of an sk_buff with private head.
2182 *	@skb: buffer to copy
2183 *	@headroom: headroom of new skb
2184 *	@gfp_mask: allocation priority
2185 *	@fclone: if true allocate the copy of the skb from the fclone
2186 *	cache instead of the head cache; it is recommended to set this
2187 *	to true for the cases where the copy will likely be cloned
2188 *
2189 *	Make a copy of both an &sk_buff and part of its data, located
2190 *	in header. Fragmented data remain shared. This is used when
2191 *	the caller wishes to modify only header of &sk_buff and needs
2192 *	private copy of the header to alter. Returns %NULL on failure
2193 *	or the pointer to the buffer on success.
2194 *	The returned buffer has a reference count of 1.
2195 */
2196
2197struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
2198				   gfp_t gfp_mask, bool fclone)
2199{
2200	unsigned int size = skb_headlen(skb) + headroom;
2201	int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
2202	struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
2203
2204	if (!n)
2205		goto out;
2206
2207	/* Set the data pointer */
2208	skb_reserve(n, headroom);
2209	/* Set the tail pointer and length */
2210	skb_put(n, skb_headlen(skb));
2211	/* Copy the bytes */
2212	skb_copy_from_linear_data(skb, n->data, n->len);
2213
2214	n->truesize += skb->data_len;
2215	n->data_len  = skb->data_len;
2216	n->len	     = skb->len;
2217
2218	if (skb_shinfo(skb)->nr_frags) {
2219		int i;
2220
2221		if (skb_orphan_frags(skb, gfp_mask) ||
2222		    skb_zerocopy_clone(n, skb, gfp_mask)) {
2223			kfree_skb(n);
2224			n = NULL;
2225			goto out;
 
2226		}
2227		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2228			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
2229			skb_frag_ref(skb, i);
2230		}
2231		skb_shinfo(n)->nr_frags = i;
2232	}
2233
2234	if (skb_has_frag_list(skb)) {
2235		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
2236		skb_clone_fraglist(n);
2237	}
2238
2239	skb_copy_header(n, skb);
2240out:
2241	return n;
2242}
2243EXPORT_SYMBOL(__pskb_copy_fclone);
2244
2245/**
2246 *	pskb_expand_head - reallocate header of &sk_buff
2247 *	@skb: buffer to reallocate
2248 *	@nhead: room to add at head
2249 *	@ntail: room to add at tail
2250 *	@gfp_mask: allocation priority
2251 *
2252 *	Expands (or creates identical copy, if @nhead and @ntail are zero)
2253 *	header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
2254 *	reference count of 1. Returns zero in the case of success or error,
2255 *	if expansion failed. In the last case, &sk_buff is not changed.
2256 *
2257 *	All the pointers pointing into skb header may change and must be
2258 *	reloaded after call to this function.
2259 */
2260
2261int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
2262		     gfp_t gfp_mask)
2263{
2264	unsigned int osize = skb_end_offset(skb);
2265	unsigned int size = osize + nhead + ntail;
2266	long off;
2267	u8 *data;
2268	int i;
 
 
 
2269
2270	BUG_ON(nhead < 0);
2271
2272	BUG_ON(skb_shared(skb));
2273
2274	skb_zcopy_downgrade_managed(skb);
2275
2276	if (skb_pfmemalloc(skb))
2277		gfp_mask |= __GFP_MEMALLOC;
2278
2279	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
 
2280	if (!data)
2281		goto nodata;
2282	size = SKB_WITH_OVERHEAD(size);
2283
2284	/* Copy only real data... and, alas, header. This should be
2285	 * optimized for the cases when header is void.
2286	 */
2287	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
2288
2289	memcpy((struct skb_shared_info *)(data + size),
2290	       skb_shinfo(skb),
2291	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
2292
2293	/*
2294	 * if shinfo is shared we must drop the old head gracefully, but if it
2295	 * is not we can just drop the old head and let the existing refcount
2296	 * be since all we did is relocate the values
2297	 */
2298	if (skb_cloned(skb)) {
2299		if (skb_orphan_frags(skb, gfp_mask))
2300			goto nofrags;
2301		if (skb_zcopy(skb))
2302			refcount_inc(&skb_uarg(skb)->refcnt);
 
2303		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2304			skb_frag_ref(skb, i);
2305
2306		if (skb_has_frag_list(skb))
2307			skb_clone_fraglist(skb);
2308
2309		skb_release_data(skb, SKB_CONSUMED);
2310	} else {
2311		skb_free_head(skb);
2312	}
2313	off = (data + nhead) - skb->head;
2314
2315	skb->head     = data;
2316	skb->head_frag = 0;
2317	skb->data    += off;
2318
2319	skb_set_end_offset(skb, size);
2320#ifdef NET_SKBUFF_DATA_USES_OFFSET
 
2321	off           = nhead;
 
 
2322#endif
 
2323	skb->tail	      += off;
2324	skb_headers_offset_update(skb, nhead);
 
 
 
 
 
 
2325	skb->cloned   = 0;
2326	skb->hdr_len  = 0;
2327	skb->nohdr    = 0;
2328	atomic_set(&skb_shinfo(skb)->dataref, 1);
2329
2330	skb_metadata_clear(skb);
2331
2332	/* It is not generally safe to change skb->truesize.
2333	 * For the moment, we really care of rx path, or
2334	 * when skb is orphaned (not attached to a socket).
2335	 */
2336	if (!skb->sk || skb->destructor == sock_edemux)
2337		skb->truesize += size - osize;
2338
2339	return 0;
2340
2341nofrags:
2342	skb_kfree_head(data, size);
2343nodata:
2344	return -ENOMEM;
2345}
2346EXPORT_SYMBOL(pskb_expand_head);
2347
2348/* Make private copy of skb with writable head and some headroom */
2349
2350struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
2351{
2352	struct sk_buff *skb2;
2353	int delta = headroom - skb_headroom(skb);
2354
2355	if (delta <= 0)
2356		skb2 = pskb_copy(skb, GFP_ATOMIC);
2357	else {
2358		skb2 = skb_clone(skb, GFP_ATOMIC);
2359		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
2360					     GFP_ATOMIC)) {
2361			kfree_skb(skb2);
2362			skb2 = NULL;
2363		}
2364	}
2365	return skb2;
2366}
2367EXPORT_SYMBOL(skb_realloc_headroom);
2368
2369/* Note: We plan to rework this in linux-6.4 */
2370int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2371{
2372	unsigned int saved_end_offset, saved_truesize;
2373	struct skb_shared_info *shinfo;
2374	int res;
2375
2376	saved_end_offset = skb_end_offset(skb);
2377	saved_truesize = skb->truesize;
2378
2379	res = pskb_expand_head(skb, 0, 0, pri);
2380	if (res)
2381		return res;
2382
2383	skb->truesize = saved_truesize;
2384
2385	if (likely(skb_end_offset(skb) == saved_end_offset))
2386		return 0;
2387
2388	/* We can not change skb->end if the original or new value
2389	 * is SKB_SMALL_HEAD_HEADROOM, as it might break skb_kfree_head().
2390	 */
2391	if (saved_end_offset == SKB_SMALL_HEAD_HEADROOM ||
2392	    skb_end_offset(skb) == SKB_SMALL_HEAD_HEADROOM) {
2393		/* We think this path should not be taken.
2394		 * Add a temporary trace to warn us just in case.
2395		 */
2396		pr_err_once("__skb_unclone_keeptruesize() skb_end_offset() %u -> %u\n",
2397			    saved_end_offset, skb_end_offset(skb));
2398		WARN_ON_ONCE(1);
2399		return 0;
2400	}
2401
2402	shinfo = skb_shinfo(skb);
2403
2404	/* We are about to change back skb->end,
2405	 * we need to move skb_shinfo() to its new location.
2406	 */
2407	memmove(skb->head + saved_end_offset,
2408		shinfo,
2409		offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2410
2411	skb_set_end_offset(skb, saved_end_offset);
2412
2413	return 0;
2414}
2415
2416/**
2417 *	skb_expand_head - reallocate header of &sk_buff
2418 *	@skb: buffer to reallocate
2419 *	@headroom: needed headroom
2420 *
2421 *	Unlike skb_realloc_headroom, this one does not allocate a new skb
2422 *	if possible; copies skb->sk to new skb as needed
2423 *	and frees original skb in case of failures.
2424 *
2425 *	It expect increased headroom and generates warning otherwise.
2426 */
2427
2428struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2429{
2430	int delta = headroom - skb_headroom(skb);
2431	int osize = skb_end_offset(skb);
2432	struct sock *sk = skb->sk;
2433
2434	if (WARN_ONCE(delta <= 0,
2435		      "%s is expecting an increase in the headroom", __func__))
2436		return skb;
2437
2438	delta = SKB_DATA_ALIGN(delta);
2439	/* pskb_expand_head() might crash, if skb is shared. */
2440	if (skb_shared(skb) || !is_skb_wmem(skb)) {
2441		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2442
2443		if (unlikely(!nskb))
2444			goto fail;
2445
2446		if (sk)
2447			skb_set_owner_w(nskb, sk);
2448		consume_skb(skb);
2449		skb = nskb;
2450	}
2451	if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2452		goto fail;
2453
2454	if (sk && is_skb_wmem(skb)) {
2455		delta = skb_end_offset(skb) - osize;
2456		refcount_add(delta, &sk->sk_wmem_alloc);
2457		skb->truesize += delta;
2458	}
2459	return skb;
2460
2461fail:
2462	kfree_skb(skb);
2463	return NULL;
2464}
2465EXPORT_SYMBOL(skb_expand_head);
2466
2467/**
2468 *	skb_copy_expand	-	copy and expand sk_buff
2469 *	@skb: buffer to copy
2470 *	@newheadroom: new free bytes at head
2471 *	@newtailroom: new free bytes at tail
2472 *	@gfp_mask: allocation priority
2473 *
2474 *	Make a copy of both an &sk_buff and its data and while doing so
2475 *	allocate additional space.
2476 *
2477 *	This is used when the caller wishes to modify the data and needs a
2478 *	private copy of the data to alter as well as more space for new fields.
2479 *	Returns %NULL on failure or the pointer to the buffer
2480 *	on success. The returned buffer has a reference count of 1.
2481 *
2482 *	You must pass %GFP_ATOMIC as the allocation priority if this function
2483 *	is called from an interrupt.
2484 */
2485struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2486				int newheadroom, int newtailroom,
2487				gfp_t gfp_mask)
2488{
2489	/*
2490	 *	Allocate the copy buffer
2491	 */
 
 
 
2492	int head_copy_len, head_copy_off;
2493	struct sk_buff *n;
2494	int oldheadroom;
2495
2496	if (!skb_frags_readable(skb))
2497		return NULL;
2498
2499	if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2500		return NULL;
2501
2502	oldheadroom = skb_headroom(skb);
2503	n = __alloc_skb(newheadroom + skb->len + newtailroom,
2504			gfp_mask, skb_alloc_rx_flag(skb),
2505			NUMA_NO_NODE);
2506	if (!n)
2507		return NULL;
2508
2509	skb_reserve(n, newheadroom);
2510
2511	/* Set the tail pointer and length */
2512	skb_put(n, skb->len);
2513
2514	head_copy_len = oldheadroom;
2515	head_copy_off = 0;
2516	if (newheadroom <= head_copy_len)
2517		head_copy_len = newheadroom;
2518	else
2519		head_copy_off = newheadroom - head_copy_len;
2520
2521	/* Copy the linear header and data. */
2522	BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2523			     skb->len + head_copy_len));
2524
2525	skb_copy_header(n, skb);
2526
2527	skb_headers_offset_update(n, newheadroom - oldheadroom);
 
 
 
 
 
 
 
 
 
2528
2529	return n;
2530}
2531EXPORT_SYMBOL(skb_copy_expand);
2532
2533/**
2534 *	__skb_pad		-	zero pad the tail of an skb
2535 *	@skb: buffer to pad
2536 *	@pad: space to pad
2537 *	@free_on_error: free buffer on error
2538 *
2539 *	Ensure that a buffer is followed by a padding area that is zero
2540 *	filled. Used by network drivers which may DMA or transfer data
2541 *	beyond the buffer end onto the wire.
2542 *
2543 *	May return error in out of memory cases. The skb is freed on error
2544 *	if @free_on_error is true.
2545 */
2546
2547int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2548{
2549	int err;
2550	int ntail;
2551
2552	/* If the skbuff is non linear tailroom is always zero.. */
2553	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2554		memset(skb->data+skb->len, 0, pad);
2555		return 0;
2556	}
2557
2558	ntail = skb->data_len + pad - (skb->end - skb->tail);
2559	if (likely(skb_cloned(skb) || ntail > 0)) {
2560		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2561		if (unlikely(err))
2562			goto free_skb;
2563	}
2564
2565	/* FIXME: The use of this function with non-linear skb's really needs
2566	 * to be audited.
2567	 */
2568	err = skb_linearize(skb);
2569	if (unlikely(err))
2570		goto free_skb;
2571
2572	memset(skb->data + skb->len, 0, pad);
2573	return 0;
2574
2575free_skb:
2576	if (free_on_error)
2577		kfree_skb(skb);
2578	return err;
2579}
2580EXPORT_SYMBOL(__skb_pad);
2581
2582/**
2583 *	pskb_put - add data to the tail of a potentially fragmented buffer
2584 *	@skb: start of the buffer to use
2585 *	@tail: tail fragment of the buffer to use
2586 *	@len: amount of data to add
2587 *
2588 *	This function extends the used data area of the potentially
2589 *	fragmented buffer. @tail must be the last fragment of @skb -- or
2590 *	@skb itself. If this would exceed the total buffer size the kernel
2591 *	will panic. A pointer to the first byte of the extra data is
2592 *	returned.
2593 */
2594
2595void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2596{
2597	if (tail != skb) {
2598		skb->data_len += len;
2599		skb->len += len;
2600	}
2601	return skb_put(tail, len);
2602}
2603EXPORT_SYMBOL_GPL(pskb_put);
2604
2605/**
2606 *	skb_put - add data to a buffer
2607 *	@skb: buffer to use
2608 *	@len: amount of data to add
2609 *
2610 *	This function extends the used data area of the buffer. If this would
2611 *	exceed the total buffer size the kernel will panic. A pointer to the
2612 *	first byte of the extra data is returned.
2613 */
2614void *skb_put(struct sk_buff *skb, unsigned int len)
2615{
2616	void *tmp = skb_tail_pointer(skb);
2617	SKB_LINEAR_ASSERT(skb);
2618	skb->tail += len;
2619	skb->len  += len;
2620	if (unlikely(skb->tail > skb->end))
2621		skb_over_panic(skb, len, __builtin_return_address(0));
2622	return tmp;
2623}
2624EXPORT_SYMBOL(skb_put);
2625
2626/**
2627 *	skb_push - add data to the start of a buffer
2628 *	@skb: buffer to use
2629 *	@len: amount of data to add
2630 *
2631 *	This function extends the used data area of the buffer at the buffer
2632 *	start. If this would exceed the total buffer headroom the kernel will
2633 *	panic. A pointer to the first byte of the extra data is returned.
2634 */
2635void *skb_push(struct sk_buff *skb, unsigned int len)
2636{
2637	skb->data -= len;
2638	skb->len  += len;
2639	if (unlikely(skb->data < skb->head))
2640		skb_under_panic(skb, len, __builtin_return_address(0));
2641	return skb->data;
2642}
2643EXPORT_SYMBOL(skb_push);
2644
2645/**
2646 *	skb_pull - remove data from the start of a buffer
2647 *	@skb: buffer to use
2648 *	@len: amount of data to remove
2649 *
2650 *	This function removes data from the start of a buffer, returning
2651 *	the memory to the headroom. A pointer to the next data in the buffer
2652 *	is returned. Once the data has been pulled future pushes will overwrite
2653 *	the old data.
2654 */
2655void *skb_pull(struct sk_buff *skb, unsigned int len)
2656{
2657	return skb_pull_inline(skb, len);
2658}
2659EXPORT_SYMBOL(skb_pull);
2660
2661/**
2662 *	skb_pull_data - remove data from the start of a buffer returning its
2663 *	original position.
2664 *	@skb: buffer to use
2665 *	@len: amount of data to remove
2666 *
2667 *	This function removes data from the start of a buffer, returning
2668 *	the memory to the headroom. A pointer to the original data in the buffer
2669 *	is returned after checking if there is enough data to pull. Once the
2670 *	data has been pulled future pushes will overwrite the old data.
2671 */
2672void *skb_pull_data(struct sk_buff *skb, size_t len)
2673{
2674	void *data = skb->data;
2675
2676	if (skb->len < len)
2677		return NULL;
2678
2679	skb_pull(skb, len);
2680
2681	return data;
2682}
2683EXPORT_SYMBOL(skb_pull_data);
2684
2685/**
2686 *	skb_trim - remove end from a buffer
2687 *	@skb: buffer to alter
2688 *	@len: new length
2689 *
2690 *	Cut the length of a buffer down by removing data from the tail. If
2691 *	the buffer is already under the length specified it is not modified.
2692 *	The skb must be linear.
2693 */
2694void skb_trim(struct sk_buff *skb, unsigned int len)
2695{
2696	if (skb->len > len)
2697		__skb_trim(skb, len);
2698}
2699EXPORT_SYMBOL(skb_trim);
2700
2701/* Trims skb to length len. It can change skb pointers.
2702 */
2703
2704int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2705{
2706	struct sk_buff **fragp;
2707	struct sk_buff *frag;
2708	int offset = skb_headlen(skb);
2709	int nfrags = skb_shinfo(skb)->nr_frags;
2710	int i;
2711	int err;
2712
2713	if (skb_cloned(skb) &&
2714	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2715		return err;
2716
2717	i = 0;
2718	if (offset >= len)
2719		goto drop_pages;
2720
2721	for (; i < nfrags; i++) {
2722		int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2723
2724		if (end < len) {
2725			offset = end;
2726			continue;
2727		}
2728
2729		skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2730
2731drop_pages:
2732		skb_shinfo(skb)->nr_frags = i;
2733
2734		for (; i < nfrags; i++)
2735			skb_frag_unref(skb, i);
2736
2737		if (skb_has_frag_list(skb))
2738			skb_drop_fraglist(skb);
2739		goto done;
2740	}
2741
2742	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2743	     fragp = &frag->next) {
2744		int end = offset + frag->len;
2745
2746		if (skb_shared(frag)) {
2747			struct sk_buff *nfrag;
2748
2749			nfrag = skb_clone(frag, GFP_ATOMIC);
2750			if (unlikely(!nfrag))
2751				return -ENOMEM;
2752
2753			nfrag->next = frag->next;
2754			consume_skb(frag);
2755			frag = nfrag;
2756			*fragp = frag;
2757		}
2758
2759		if (end < len) {
2760			offset = end;
2761			continue;
2762		}
2763
2764		if (end > len &&
2765		    unlikely((err = pskb_trim(frag, len - offset))))
2766			return err;
2767
2768		if (frag->next)
2769			skb_drop_list(&frag->next);
2770		break;
2771	}
2772
2773done:
2774	if (len > skb_headlen(skb)) {
2775		skb->data_len -= skb->len - len;
2776		skb->len       = len;
2777	} else {
2778		skb->len       = len;
2779		skb->data_len  = 0;
2780		skb_set_tail_pointer(skb, len);
2781	}
2782
2783	if (!skb->sk || skb->destructor == sock_edemux)
2784		skb_condense(skb);
2785	return 0;
2786}
2787EXPORT_SYMBOL(___pskb_trim);
2788
2789/* Note : use pskb_trim_rcsum() instead of calling this directly
2790 */
2791int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2792{
2793	if (skb->ip_summed == CHECKSUM_COMPLETE) {
2794		int delta = skb->len - len;
2795
2796		skb->csum = csum_block_sub(skb->csum,
2797					   skb_checksum(skb, len, delta, 0),
2798					   len);
2799	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2800		int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2801		int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2802
2803		if (offset + sizeof(__sum16) > hdlen)
2804			return -EINVAL;
2805	}
2806	return __pskb_trim(skb, len);
2807}
2808EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2809
2810/**
2811 *	__pskb_pull_tail - advance tail of skb header
2812 *	@skb: buffer to reallocate
2813 *	@delta: number of bytes to advance tail
2814 *
2815 *	The function makes a sense only on a fragmented &sk_buff,
2816 *	it expands header moving its tail forward and copying necessary
2817 *	data from fragmented part.
2818 *
2819 *	&sk_buff MUST have reference count of 1.
2820 *
2821 *	Returns %NULL (and &sk_buff does not change) if pull failed
2822 *	or value of new tail of skb in the case of success.
2823 *
2824 *	All the pointers pointing into skb header may change and must be
2825 *	reloaded after call to this function.
2826 */
2827
2828/* Moves tail of skb head forward, copying data from fragmented part,
2829 * when it is necessary.
2830 * 1. It may fail due to malloc failure.
2831 * 2. It may change skb pointers.
2832 *
2833 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2834 */
2835void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2836{
2837	/* If skb has not enough free space at tail, get new one
2838	 * plus 128 bytes for future expansions. If we have enough
2839	 * room at tail, reallocate without expansion only if skb is cloned.
2840	 */
2841	int i, k, eat = (skb->tail + delta) - skb->end;
2842
2843	if (!skb_frags_readable(skb))
2844		return NULL;
2845
2846	if (eat > 0 || skb_cloned(skb)) {
2847		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2848				     GFP_ATOMIC))
2849			return NULL;
2850	}
2851
2852	BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2853			     skb_tail_pointer(skb), delta));
2854
2855	/* Optimization: no fragments, no reasons to preestimate
2856	 * size of pulled pages. Superb.
2857	 */
2858	if (!skb_has_frag_list(skb))
2859		goto pull_pages;
2860
2861	/* Estimate size of pulled pages. */
2862	eat = delta;
2863	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2864		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2865
2866		if (size >= eat)
2867			goto pull_pages;
2868		eat -= size;
2869	}
2870
2871	/* If we need update frag list, we are in troubles.
2872	 * Certainly, it is possible to add an offset to skb data,
2873	 * but taking into account that pulling is expected to
2874	 * be very rare operation, it is worth to fight against
2875	 * further bloating skb head and crucify ourselves here instead.
2876	 * Pure masohism, indeed. 8)8)
2877	 */
2878	if (eat) {
2879		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2880		struct sk_buff *clone = NULL;
2881		struct sk_buff *insp = NULL;
2882
2883		do {
 
 
2884			if (list->len <= eat) {
2885				/* Eaten as whole. */
2886				eat -= list->len;
2887				list = list->next;
2888				insp = list;
2889			} else {
2890				/* Eaten partially. */
2891				if (skb_is_gso(skb) && !list->head_frag &&
2892				    skb_headlen(list))
2893					skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2894
2895				if (skb_shared(list)) {
2896					/* Sucks! We need to fork list. :-( */
2897					clone = skb_clone(list, GFP_ATOMIC);
2898					if (!clone)
2899						return NULL;
2900					insp = list->next;
2901					list = clone;
2902				} else {
2903					/* This may be pulled without
2904					 * problems. */
2905					insp = list;
2906				}
2907				if (!pskb_pull(list, eat)) {
2908					kfree_skb(clone);
2909					return NULL;
2910				}
2911				break;
2912			}
2913		} while (eat);
2914
2915		/* Free pulled out fragments. */
2916		while ((list = skb_shinfo(skb)->frag_list) != insp) {
2917			skb_shinfo(skb)->frag_list = list->next;
2918			consume_skb(list);
2919		}
2920		/* And insert new clone at head. */
2921		if (clone) {
2922			clone->next = list;
2923			skb_shinfo(skb)->frag_list = clone;
2924		}
2925	}
2926	/* Success! Now we may commit changes to skb data. */
2927
2928pull_pages:
2929	eat = delta;
2930	k = 0;
2931	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2932		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2933
2934		if (size <= eat) {
2935			skb_frag_unref(skb, i);
2936			eat -= size;
2937		} else {
2938			skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2939
2940			*frag = skb_shinfo(skb)->frags[i];
2941			if (eat) {
2942				skb_frag_off_add(frag, eat);
2943				skb_frag_size_sub(frag, eat);
2944				if (!i)
2945					goto end;
2946				eat = 0;
2947			}
2948			k++;
2949		}
2950	}
2951	skb_shinfo(skb)->nr_frags = k;
2952
2953end:
2954	skb->tail     += delta;
2955	skb->data_len -= delta;
2956
2957	if (!skb->data_len)
2958		skb_zcopy_clear(skb, false);
2959
2960	return skb_tail_pointer(skb);
2961}
2962EXPORT_SYMBOL(__pskb_pull_tail);
2963
2964/**
2965 *	skb_copy_bits - copy bits from skb to kernel buffer
2966 *	@skb: source skb
2967 *	@offset: offset in source
2968 *	@to: destination buffer
2969 *	@len: number of bytes to copy
2970 *
2971 *	Copy the specified number of bytes from the source skb to the
2972 *	destination buffer.
2973 *
2974 *	CAUTION ! :
2975 *		If its prototype is ever changed,
2976 *		check arch/{*}/net/{*}.S files,
2977 *		since it is called from BPF assembly code.
2978 */
2979int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2980{
2981	int start = skb_headlen(skb);
2982	struct sk_buff *frag_iter;
2983	int i, copy;
2984
2985	if (offset > (int)skb->len - len)
2986		goto fault;
2987
2988	/* Copy header. */
2989	if ((copy = start - offset) > 0) {
2990		if (copy > len)
2991			copy = len;
2992		skb_copy_from_linear_data_offset(skb, offset, to, copy);
2993		if ((len -= copy) == 0)
2994			return 0;
2995		offset += copy;
2996		to     += copy;
2997	}
2998
2999	if (!skb_frags_readable(skb))
3000		goto fault;
3001
3002	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3003		int end;
3004		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
3005
3006		WARN_ON(start > offset + len);
3007
3008		end = start + skb_frag_size(f);
3009		if ((copy = end - offset) > 0) {
3010			u32 p_off, p_len, copied;
3011			struct page *p;
3012			u8 *vaddr;
3013
3014			if (copy > len)
3015				copy = len;
3016
3017			skb_frag_foreach_page(f,
3018					      skb_frag_off(f) + offset - start,
3019					      copy, p, p_off, p_len, copied) {
3020				vaddr = kmap_atomic(p);
3021				memcpy(to + copied, vaddr + p_off, p_len);
3022				kunmap_atomic(vaddr);
3023			}
3024
3025			if ((len -= copy) == 0)
3026				return 0;
3027			offset += copy;
3028			to     += copy;
3029		}
3030		start = end;
3031	}
3032
3033	skb_walk_frags(skb, frag_iter) {
3034		int end;
3035
3036		WARN_ON(start > offset + len);
3037
3038		end = start + frag_iter->len;
3039		if ((copy = end - offset) > 0) {
3040			if (copy > len)
3041				copy = len;
3042			if (skb_copy_bits(frag_iter, offset - start, to, copy))
3043				goto fault;
3044			if ((len -= copy) == 0)
3045				return 0;
3046			offset += copy;
3047			to     += copy;
3048		}
3049		start = end;
3050	}
3051
3052	if (!len)
3053		return 0;
3054
3055fault:
3056	return -EFAULT;
3057}
3058EXPORT_SYMBOL(skb_copy_bits);
3059
3060/*
3061 * Callback from splice_to_pipe(), if we need to release some pages
3062 * at the end of the spd in case we error'ed out in filling the pipe.
3063 */
3064static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3065{
3066	put_page(spd->pages[i]);
3067}
3068
3069static struct page *linear_to_page(struct page *page, unsigned int *len,
3070				   unsigned int *offset,
3071				   struct sock *sk)
3072{
3073	struct page_frag *pfrag = sk_page_frag(sk);
 
3074
3075	if (!sk_page_frag_refill(sk, pfrag))
3076		return NULL;
 
 
 
3077
3078	*len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
 
 
 
 
 
 
 
 
 
 
 
 
 
3079
3080	memcpy(page_address(pfrag->page) + pfrag->offset,
3081	       page_address(page) + *offset, *len);
3082	*offset = pfrag->offset;
3083	pfrag->offset += *len;
3084
3085	return pfrag->page;
 
 
 
 
3086}
3087
3088static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
3089			     struct page *page,
3090			     unsigned int offset)
3091{
3092	return	spd->nr_pages &&
3093		spd->pages[spd->nr_pages - 1] == page &&
3094		(spd->partial[spd->nr_pages - 1].offset +
3095		 spd->partial[spd->nr_pages - 1].len == offset);
3096}
3097
3098/*
3099 * Fill page/offset/length into spd, if it can hold more pages.
3100 */
3101static bool spd_fill_page(struct splice_pipe_desc *spd,
3102			  struct pipe_inode_info *pipe, struct page *page,
3103			  unsigned int *len, unsigned int offset,
3104			  bool linear,
3105			  struct sock *sk)
3106{
3107	if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
3108		return true;
3109
3110	if (linear) {
3111		page = linear_to_page(page, len, &offset, sk);
3112		if (!page)
3113			return true;
3114	}
3115	if (spd_can_coalesce(spd, page, offset)) {
3116		spd->partial[spd->nr_pages - 1].len += *len;
3117		return false;
3118	}
3119	get_page(page);
3120	spd->pages[spd->nr_pages] = page;
3121	spd->partial[spd->nr_pages].len = *len;
3122	spd->partial[spd->nr_pages].offset = offset;
3123	spd->nr_pages++;
3124
3125	return false;
3126}
3127
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3128static bool __splice_segment(struct page *page, unsigned int poff,
3129			     unsigned int plen, unsigned int *off,
3130			     unsigned int *len,
3131			     struct splice_pipe_desc *spd, bool linear,
3132			     struct sock *sk,
3133			     struct pipe_inode_info *pipe)
3134{
3135	if (!*len)
3136		return true;
3137
3138	/* skip this segment if already processed */
3139	if (*off >= plen) {
3140		*off -= plen;
3141		return false;
3142	}
3143
3144	/* ignore any bits we already processed */
3145	poff += *off;
3146	plen -= *off;
3147	*off = 0;
 
3148
3149	do {
3150		unsigned int flen = min(*len, plen);
3151
3152		if (spd_fill_page(spd, pipe, page, &flen, poff,
3153				  linear, sk))
 
 
3154			return true;
3155		poff += flen;
3156		plen -= flen;
3157		*len -= flen;
 
3158	} while (*len && plen);
3159
3160	return false;
3161}
3162
3163/*
3164 * Map linear and fragment data from the skb to spd. It reports true if the
3165 * pipe is full or if we already spliced the requested length.
3166 */
3167static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
3168			      unsigned int *offset, unsigned int *len,
3169			      struct splice_pipe_desc *spd, struct sock *sk)
3170{
3171	int seg;
3172	struct sk_buff *iter;
3173
3174	/* map the linear part :
3175	 * If skb->head_frag is set, this 'linear' part is backed by a
3176	 * fragment, and if the head is not shared with any clones then
3177	 * we can avoid a copy since we own the head portion of this page.
3178	 */
3179	if (__splice_segment(virt_to_page(skb->data),
3180			     (unsigned long) skb->data & (PAGE_SIZE - 1),
3181			     skb_headlen(skb),
3182			     offset, len, spd,
3183			     skb_head_is_locked(skb),
3184			     sk, pipe))
3185		return true;
3186
3187	/*
3188	 * then map the fragments
3189	 */
3190	if (!skb_frags_readable(skb))
3191		return false;
3192
3193	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
3194		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
3195
3196		if (WARN_ON_ONCE(!skb_frag_page(f)))
3197			return false;
3198
3199		if (__splice_segment(skb_frag_page(f),
3200				     skb_frag_off(f), skb_frag_size(f),
3201				     offset, len, spd, false, sk, pipe))
3202			return true;
3203	}
3204
3205	skb_walk_frags(skb, iter) {
3206		if (*offset >= iter->len) {
3207			*offset -= iter->len;
3208			continue;
3209		}
3210		/* __skb_splice_bits() only fails if the output has no room
3211		 * left, so no point in going over the frag_list for the error
3212		 * case.
3213		 */
3214		if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
3215			return true;
3216	}
3217
3218	return false;
3219}
3220
3221/*
3222 * Map data from the skb to a pipe. Should handle both the linear part,
3223 * the fragments, and the frag list.
 
 
3224 */
3225int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3226		    struct pipe_inode_info *pipe, unsigned int tlen,
3227		    unsigned int flags)
3228{
3229	struct partial_page partial[MAX_SKB_FRAGS];
3230	struct page *pages[MAX_SKB_FRAGS];
3231	struct splice_pipe_desc spd = {
3232		.pages = pages,
3233		.partial = partial,
3234		.nr_pages_max = MAX_SKB_FRAGS,
3235		.ops = &nosteal_pipe_buf_ops,
 
3236		.spd_release = sock_spd_release,
3237	};
 
 
3238	int ret = 0;
3239
3240	__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
3241
3242	if (spd.nr_pages)
3243		ret = splice_to_pipe(pipe, &spd);
3244
3245	return ret;
3246}
3247EXPORT_SYMBOL_GPL(skb_splice_bits);
3248
3249static int sendmsg_locked(struct sock *sk, struct msghdr *msg)
3250{
3251	struct socket *sock = sk->sk_socket;
3252	size_t size = msg_data_left(msg);
3253
3254	if (!sock)
3255		return -EINVAL;
3256
3257	if (!sock->ops->sendmsg_locked)
3258		return sock_no_sendmsg_locked(sk, msg, size);
3259
3260	return sock->ops->sendmsg_locked(sk, msg, size);
3261}
3262
3263static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg)
3264{
3265	struct socket *sock = sk->sk_socket;
3266
3267	if (!sock)
3268		return -EINVAL;
3269	return sock_sendmsg(sock, msg);
3270}
3271
3272typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg);
3273static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
3274			   int len, sendmsg_func sendmsg)
3275{
3276	unsigned int orig_len = len;
3277	struct sk_buff *head = skb;
3278	unsigned short fragidx;
3279	int slen, ret;
3280
3281do_frag_list:
3282
3283	/* Deal with head data */
3284	while (offset < skb_headlen(skb) && len) {
3285		struct kvec kv;
3286		struct msghdr msg;
3287
3288		slen = min_t(int, len, skb_headlen(skb) - offset);
3289		kv.iov_base = skb->data + offset;
3290		kv.iov_len = slen;
3291		memset(&msg, 0, sizeof(msg));
3292		msg.msg_flags = MSG_DONTWAIT;
3293
3294		iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &kv, 1, slen);
3295		ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3296				      sendmsg_unlocked, sk, &msg);
3297		if (ret <= 0)
3298			goto error;
3299
3300		offset += ret;
3301		len -= ret;
3302	}
3303
3304	/* All the data was skb head? */
3305	if (!len)
3306		goto out;
3307
3308	/* Make offset relative to start of frags */
3309	offset -= skb_headlen(skb);
3310
3311	/* Find where we are in frag list */
3312	for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3313		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3314
3315		if (offset < skb_frag_size(frag))
 
 
 
 
 
 
3316			break;
3317
3318		offset -= skb_frag_size(frag);
3319	}
3320
3321	for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3322		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3323
3324		slen = min_t(size_t, len, skb_frag_size(frag) - offset);
3325
3326		while (slen) {
3327			struct bio_vec bvec;
3328			struct msghdr msg = {
3329				.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT,
3330			};
3331
3332			bvec_set_page(&bvec, skb_frag_page(frag), slen,
3333				      skb_frag_off(frag) + offset);
3334			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1,
3335				      slen);
3336
3337			ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3338					      sendmsg_unlocked, sk, &msg);
3339			if (ret <= 0)
3340				goto error;
3341
3342			len -= ret;
3343			offset += ret;
3344			slen -= ret;
3345		}
3346
3347		offset = 0;
3348	}
3349
3350	if (len) {
3351		/* Process any frag lists */
3352
3353		if (skb == head) {
3354			if (skb_has_frag_list(skb)) {
3355				skb = skb_shinfo(skb)->frag_list;
3356				goto do_frag_list;
3357			}
3358		} else if (skb->next) {
3359			skb = skb->next;
3360			goto do_frag_list;
3361		}
 
 
3362	}
3363
3364out:
3365	return orig_len - len;
3366
3367error:
3368	return orig_len == len ? ret : orig_len - len;
3369}
3370
3371/* Send skb data on a socket. Socket must be locked. */
3372int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3373			 int len)
3374{
3375	return __skb_send_sock(sk, skb, offset, len, sendmsg_locked);
3376}
3377EXPORT_SYMBOL_GPL(skb_send_sock_locked);
3378
3379/* Send skb data on a socket. Socket must be unlocked. */
3380int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
3381{
3382	return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked);
3383}
3384
3385/**
3386 *	skb_store_bits - store bits from kernel buffer to skb
3387 *	@skb: destination buffer
3388 *	@offset: offset in destination
3389 *	@from: source buffer
3390 *	@len: number of bytes to copy
3391 *
3392 *	Copy the specified number of bytes from the source buffer to the
3393 *	destination skb.  This function handles all the messy bits of
3394 *	traversing fragment lists and such.
3395 */
3396
3397int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
3398{
3399	int start = skb_headlen(skb);
3400	struct sk_buff *frag_iter;
3401	int i, copy;
3402
3403	if (offset > (int)skb->len - len)
3404		goto fault;
3405
3406	if ((copy = start - offset) > 0) {
3407		if (copy > len)
3408			copy = len;
3409		skb_copy_to_linear_data_offset(skb, offset, from, copy);
3410		if ((len -= copy) == 0)
3411			return 0;
3412		offset += copy;
3413		from += copy;
3414	}
3415
3416	if (!skb_frags_readable(skb))
3417		goto fault;
3418
3419	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3420		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3421		int end;
3422
3423		WARN_ON(start > offset + len);
3424
3425		end = start + skb_frag_size(frag);
3426		if ((copy = end - offset) > 0) {
3427			u32 p_off, p_len, copied;
3428			struct page *p;
3429			u8 *vaddr;
3430
3431			if (copy > len)
3432				copy = len;
3433
3434			skb_frag_foreach_page(frag,
3435					      skb_frag_off(frag) + offset - start,
3436					      copy, p, p_off, p_len, copied) {
3437				vaddr = kmap_atomic(p);
3438				memcpy(vaddr + p_off, from + copied, p_len);
3439				kunmap_atomic(vaddr);
3440			}
3441
3442			if ((len -= copy) == 0)
3443				return 0;
3444			offset += copy;
3445			from += copy;
3446		}
3447		start = end;
3448	}
3449
3450	skb_walk_frags(skb, frag_iter) {
3451		int end;
3452
3453		WARN_ON(start > offset + len);
3454
3455		end = start + frag_iter->len;
3456		if ((copy = end - offset) > 0) {
3457			if (copy > len)
3458				copy = len;
3459			if (skb_store_bits(frag_iter, offset - start,
3460					   from, copy))
3461				goto fault;
3462			if ((len -= copy) == 0)
3463				return 0;
3464			offset += copy;
3465			from += copy;
3466		}
3467		start = end;
3468	}
3469	if (!len)
3470		return 0;
3471
3472fault:
3473	return -EFAULT;
3474}
3475EXPORT_SYMBOL(skb_store_bits);
3476
3477/* Checksum skb data. */
3478__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3479		      __wsum csum, const struct skb_checksum_ops *ops)
 
3480{
3481	int start = skb_headlen(skb);
3482	int i, copy = start - offset;
3483	struct sk_buff *frag_iter;
3484	int pos = 0;
3485
3486	/* Checksum header. */
3487	if (copy > 0) {
3488		if (copy > len)
3489			copy = len;
3490		csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3491				       skb->data + offset, copy, csum);
3492		if ((len -= copy) == 0)
3493			return csum;
3494		offset += copy;
3495		pos	= copy;
3496	}
3497
3498	if (WARN_ON_ONCE(!skb_frags_readable(skb)))
3499		return 0;
3500
3501	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3502		int end;
3503		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3504
3505		WARN_ON(start > offset + len);
3506
3507		end = start + skb_frag_size(frag);
3508		if ((copy = end - offset) > 0) {
3509			u32 p_off, p_len, copied;
3510			struct page *p;
3511			__wsum csum2;
3512			u8 *vaddr;
3513
3514			if (copy > len)
3515				copy = len;
3516
3517			skb_frag_foreach_page(frag,
3518					      skb_frag_off(frag) + offset - start,
3519					      copy, p, p_off, p_len, copied) {
3520				vaddr = kmap_atomic(p);
3521				csum2 = INDIRECT_CALL_1(ops->update,
3522							csum_partial_ext,
3523							vaddr + p_off, p_len, 0);
3524				kunmap_atomic(vaddr);
3525				csum = INDIRECT_CALL_1(ops->combine,
3526						       csum_block_add_ext, csum,
3527						       csum2, pos, p_len);
3528				pos += p_len;
3529			}
3530
3531			if (!(len -= copy))
3532				return csum;
3533			offset += copy;
 
3534		}
3535		start = end;
3536	}
3537
3538	skb_walk_frags(skb, frag_iter) {
3539		int end;
3540
3541		WARN_ON(start > offset + len);
3542
3543		end = start + frag_iter->len;
3544		if ((copy = end - offset) > 0) {
3545			__wsum csum2;
3546			if (copy > len)
3547				copy = len;
3548			csum2 = __skb_checksum(frag_iter, offset - start,
3549					       copy, 0, ops);
3550			csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3551					       csum, csum2, pos, copy);
3552			if ((len -= copy) == 0)
3553				return csum;
3554			offset += copy;
3555			pos    += copy;
3556		}
3557		start = end;
3558	}
3559	BUG_ON(len);
3560
3561	return csum;
3562}
3563EXPORT_SYMBOL(__skb_checksum);
3564
3565__wsum skb_checksum(const struct sk_buff *skb, int offset,
3566		    int len, __wsum csum)
3567{
3568	const struct skb_checksum_ops ops = {
3569		.update  = csum_partial_ext,
3570		.combine = csum_block_add_ext,
3571	};
3572
3573	return __skb_checksum(skb, offset, len, csum, &ops);
3574}
3575EXPORT_SYMBOL(skb_checksum);
3576
3577/* Both of above in one bottle. */
3578
3579__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3580				    u8 *to, int len)
3581{
3582	int start = skb_headlen(skb);
3583	int i, copy = start - offset;
3584	struct sk_buff *frag_iter;
3585	int pos = 0;
3586	__wsum csum = 0;
3587
3588	/* Copy header. */
3589	if (copy > 0) {
3590		if (copy > len)
3591			copy = len;
3592		csum = csum_partial_copy_nocheck(skb->data + offset, to,
3593						 copy);
3594		if ((len -= copy) == 0)
3595			return csum;
3596		offset += copy;
3597		to     += copy;
3598		pos	= copy;
3599	}
3600
3601	if (!skb_frags_readable(skb))
3602		return 0;
3603
3604	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3605		int end;
3606
3607		WARN_ON(start > offset + len);
3608
3609		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3610		if ((copy = end - offset) > 0) {
3611			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3612			u32 p_off, p_len, copied;
3613			struct page *p;
3614			__wsum csum2;
3615			u8 *vaddr;
 
3616
3617			if (copy > len)
3618				copy = len;
3619
3620			skb_frag_foreach_page(frag,
3621					      skb_frag_off(frag) + offset - start,
3622					      copy, p, p_off, p_len, copied) {
3623				vaddr = kmap_atomic(p);
3624				csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3625								  to + copied,
3626								  p_len);
3627				kunmap_atomic(vaddr);
3628				csum = csum_block_add(csum, csum2, pos);
3629				pos += p_len;
3630			}
3631
3632			if (!(len -= copy))
3633				return csum;
3634			offset += copy;
3635			to     += copy;
 
3636		}
3637		start = end;
3638	}
3639
3640	skb_walk_frags(skb, frag_iter) {
3641		__wsum csum2;
3642		int end;
3643
3644		WARN_ON(start > offset + len);
3645
3646		end = start + frag_iter->len;
3647		if ((copy = end - offset) > 0) {
3648			if (copy > len)
3649				copy = len;
3650			csum2 = skb_copy_and_csum_bits(frag_iter,
3651						       offset - start,
3652						       to, copy);
3653			csum = csum_block_add(csum, csum2, pos);
3654			if ((len -= copy) == 0)
3655				return csum;
3656			offset += copy;
3657			to     += copy;
3658			pos    += copy;
3659		}
3660		start = end;
3661	}
3662	BUG_ON(len);
3663	return csum;
3664}
3665EXPORT_SYMBOL(skb_copy_and_csum_bits);
3666
3667__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3668{
3669	__sum16 sum;
3670
3671	sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3672	/* See comments in __skb_checksum_complete(). */
3673	if (likely(!sum)) {
3674		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3675		    !skb->csum_complete_sw)
3676			netdev_rx_csum_fault(skb->dev, skb);
3677	}
3678	if (!skb_shared(skb))
3679		skb->csum_valid = !sum;
3680	return sum;
3681}
3682EXPORT_SYMBOL(__skb_checksum_complete_head);
3683
3684/* This function assumes skb->csum already holds pseudo header's checksum,
3685 * which has been changed from the hardware checksum, for example, by
3686 * __skb_checksum_validate_complete(). And, the original skb->csum must
3687 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3688 *
3689 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3690 * zero. The new checksum is stored back into skb->csum unless the skb is
3691 * shared.
3692 */
3693__sum16 __skb_checksum_complete(struct sk_buff *skb)
3694{
3695	__wsum csum;
3696	__sum16 sum;
3697
3698	csum = skb_checksum(skb, 0, skb->len, 0);
3699
3700	sum = csum_fold(csum_add(skb->csum, csum));
3701	/* This check is inverted, because we already knew the hardware
3702	 * checksum is invalid before calling this function. So, if the
3703	 * re-computed checksum is valid instead, then we have a mismatch
3704	 * between the original skb->csum and skb_checksum(). This means either
3705	 * the original hardware checksum is incorrect or we screw up skb->csum
3706	 * when moving skb->data around.
3707	 */
3708	if (likely(!sum)) {
3709		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3710		    !skb->csum_complete_sw)
3711			netdev_rx_csum_fault(skb->dev, skb);
3712	}
3713
3714	if (!skb_shared(skb)) {
3715		/* Save full packet checksum */
3716		skb->csum = csum;
3717		skb->ip_summed = CHECKSUM_COMPLETE;
3718		skb->csum_complete_sw = 1;
3719		skb->csum_valid = !sum;
3720	}
3721
3722	return sum;
3723}
3724EXPORT_SYMBOL(__skb_checksum_complete);
3725
3726static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3727{
3728	net_warn_ratelimited(
3729		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3730		__func__);
3731	return 0;
3732}
3733
3734static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3735				       int offset, int len)
3736{
3737	net_warn_ratelimited(
3738		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3739		__func__);
3740	return 0;
3741}
3742
3743static const struct skb_checksum_ops default_crc32c_ops = {
3744	.update  = warn_crc32c_csum_update,
3745	.combine = warn_crc32c_csum_combine,
3746};
3747
3748const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3749	&default_crc32c_ops;
3750EXPORT_SYMBOL(crc32c_csum_stub);
3751
3752 /**
3753 *	skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3754 *	@from: source buffer
3755 *
3756 *	Calculates the amount of linear headroom needed in the 'to' skb passed
3757 *	into skb_zerocopy().
3758 */
3759unsigned int
3760skb_zerocopy_headlen(const struct sk_buff *from)
3761{
3762	unsigned int hlen = 0;
3763
3764	if (!from->head_frag ||
3765	    skb_headlen(from) < L1_CACHE_BYTES ||
3766	    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3767		hlen = skb_headlen(from);
3768		if (!hlen)
3769			hlen = from->len;
3770	}
3771
3772	if (skb_has_frag_list(from))
3773		hlen = from->len;
3774
3775	return hlen;
3776}
3777EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3778
3779/**
3780 *	skb_zerocopy - Zero copy skb to skb
3781 *	@to: destination buffer
3782 *	@from: source buffer
3783 *	@len: number of bytes to copy from source buffer
3784 *	@hlen: size of linear headroom in destination buffer
3785 *
3786 *	Copies up to `len` bytes from `from` to `to` by creating references
3787 *	to the frags in the source buffer.
3788 *
3789 *	The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3790 *	headroom in the `to` buffer.
3791 *
3792 *	Return value:
3793 *	0: everything is OK
3794 *	-ENOMEM: couldn't orphan frags of @from due to lack of memory
3795 *	-EFAULT: skb_copy_bits() found some problem with skb geometry
3796 */
3797int
3798skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3799{
3800	int i, j = 0;
3801	int plen = 0; /* length of skb->head fragment */
3802	int ret;
3803	struct page *page;
3804	unsigned int offset;
3805
3806	BUG_ON(!from->head_frag && !hlen);
3807
3808	/* dont bother with small payloads */
3809	if (len <= skb_tailroom(to))
3810		return skb_copy_bits(from, 0, skb_put(to, len), len);
3811
3812	if (hlen) {
3813		ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3814		if (unlikely(ret))
3815			return ret;
3816		len -= hlen;
3817	} else {
3818		plen = min_t(int, skb_headlen(from), len);
3819		if (plen) {
3820			page = virt_to_head_page(from->head);
3821			offset = from->data - (unsigned char *)page_address(page);
3822			__skb_fill_netmem_desc(to, 0, page_to_netmem(page),
3823					       offset, plen);
3824			get_page(page);
3825			j = 1;
3826			len -= plen;
3827		}
3828	}
3829
3830	skb_len_add(to, len + plen);
3831
3832	if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3833		skb_tx_error(from);
3834		return -ENOMEM;
3835	}
3836	skb_zerocopy_clone(to, from, GFP_ATOMIC);
3837
3838	for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3839		int size;
3840
3841		if (!len)
3842			break;
3843		skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3844		size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3845					len);
3846		skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3847		len -= size;
3848		skb_frag_ref(to, j);
3849		j++;
3850	}
3851	skb_shinfo(to)->nr_frags = j;
3852
3853	return 0;
3854}
3855EXPORT_SYMBOL_GPL(skb_zerocopy);
3856
3857void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3858{
3859	__wsum csum;
3860	long csstart;
3861
3862	if (skb->ip_summed == CHECKSUM_PARTIAL)
3863		csstart = skb_checksum_start_offset(skb);
3864	else
3865		csstart = skb_headlen(skb);
3866
3867	BUG_ON(csstart > skb_headlen(skb));
3868
3869	skb_copy_from_linear_data(skb, to, csstart);
3870
3871	csum = 0;
3872	if (csstart != skb->len)
3873		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3874					      skb->len - csstart);
3875
3876	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3877		long csstuff = csstart + skb->csum_offset;
3878
3879		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
3880	}
3881}
3882EXPORT_SYMBOL(skb_copy_and_csum_dev);
3883
3884/**
3885 *	skb_dequeue - remove from the head of the queue
3886 *	@list: list to dequeue from
3887 *
3888 *	Remove the head of the list. The list lock is taken so the function
3889 *	may be used safely with other locking list functions. The head item is
3890 *	returned or %NULL if the list is empty.
3891 */
3892
3893struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3894{
3895	unsigned long flags;
3896	struct sk_buff *result;
3897
3898	spin_lock_irqsave(&list->lock, flags);
3899	result = __skb_dequeue(list);
3900	spin_unlock_irqrestore(&list->lock, flags);
3901	return result;
3902}
3903EXPORT_SYMBOL(skb_dequeue);
3904
3905/**
3906 *	skb_dequeue_tail - remove from the tail of the queue
3907 *	@list: list to dequeue from
3908 *
3909 *	Remove the tail of the list. The list lock is taken so the function
3910 *	may be used safely with other locking list functions. The tail item is
3911 *	returned or %NULL if the list is empty.
3912 */
3913struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3914{
3915	unsigned long flags;
3916	struct sk_buff *result;
3917
3918	spin_lock_irqsave(&list->lock, flags);
3919	result = __skb_dequeue_tail(list);
3920	spin_unlock_irqrestore(&list->lock, flags);
3921	return result;
3922}
3923EXPORT_SYMBOL(skb_dequeue_tail);
3924
3925/**
3926 *	skb_queue_purge_reason - empty a list
3927 *	@list: list to empty
3928 *	@reason: drop reason
3929 *
3930 *	Delete all buffers on an &sk_buff list. Each buffer is removed from
3931 *	the list and one reference dropped. This function takes the list
3932 *	lock and is atomic with respect to other list locking functions.
3933 */
3934void skb_queue_purge_reason(struct sk_buff_head *list,
3935			    enum skb_drop_reason reason)
3936{
3937	struct sk_buff_head tmp;
3938	unsigned long flags;
3939
3940	if (skb_queue_empty_lockless(list))
3941		return;
3942
3943	__skb_queue_head_init(&tmp);
3944
3945	spin_lock_irqsave(&list->lock, flags);
3946	skb_queue_splice_init(list, &tmp);
3947	spin_unlock_irqrestore(&list->lock, flags);
3948
3949	__skb_queue_purge_reason(&tmp, reason);
3950}
3951EXPORT_SYMBOL(skb_queue_purge_reason);
3952
3953/**
3954 *	skb_rbtree_purge - empty a skb rbtree
3955 *	@root: root of the rbtree to empty
3956 *	Return value: the sum of truesizes of all purged skbs.
3957 *
3958 *	Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3959 *	the list and one reference dropped. This function does not take
3960 *	any lock. Synchronization should be handled by the caller (e.g., TCP
3961 *	out-of-order queue is protected by the socket lock).
3962 */
3963unsigned int skb_rbtree_purge(struct rb_root *root)
3964{
3965	struct rb_node *p = rb_first(root);
3966	unsigned int sum = 0;
3967
3968	while (p) {
3969		struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3970
3971		p = rb_next(p);
3972		rb_erase(&skb->rbnode, root);
3973		sum += skb->truesize;
3974		kfree_skb(skb);
3975	}
3976	return sum;
3977}
3978
3979void skb_errqueue_purge(struct sk_buff_head *list)
3980{
3981	struct sk_buff *skb, *next;
3982	struct sk_buff_head kill;
3983	unsigned long flags;
3984
3985	__skb_queue_head_init(&kill);
3986
3987	spin_lock_irqsave(&list->lock, flags);
3988	skb_queue_walk_safe(list, skb, next) {
3989		if (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ZEROCOPY ||
3990		    SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING)
3991			continue;
3992		__skb_unlink(skb, list);
3993		__skb_queue_tail(&kill, skb);
3994	}
3995	spin_unlock_irqrestore(&list->lock, flags);
3996	__skb_queue_purge(&kill);
3997}
3998EXPORT_SYMBOL(skb_errqueue_purge);
3999
4000/**
4001 *	skb_queue_head - queue a buffer at the list head
4002 *	@list: list to use
4003 *	@newsk: buffer to queue
4004 *
4005 *	Queue a buffer at the start of the list. This function takes the
4006 *	list lock and can be used safely with other locking &sk_buff functions
4007 *	safely.
4008 *
4009 *	A buffer cannot be placed on two lists at the same time.
4010 */
4011void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
4012{
4013	unsigned long flags;
4014
4015	spin_lock_irqsave(&list->lock, flags);
4016	__skb_queue_head(list, newsk);
4017	spin_unlock_irqrestore(&list->lock, flags);
4018}
4019EXPORT_SYMBOL(skb_queue_head);
4020
4021/**
4022 *	skb_queue_tail - queue a buffer at the list tail
4023 *	@list: list to use
4024 *	@newsk: buffer to queue
4025 *
4026 *	Queue a buffer at the tail of the list. This function takes the
4027 *	list lock and can be used safely with other locking &sk_buff functions
4028 *	safely.
4029 *
4030 *	A buffer cannot be placed on two lists at the same time.
4031 */
4032void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
4033{
4034	unsigned long flags;
4035
4036	spin_lock_irqsave(&list->lock, flags);
4037	__skb_queue_tail(list, newsk);
4038	spin_unlock_irqrestore(&list->lock, flags);
4039}
4040EXPORT_SYMBOL(skb_queue_tail);
4041
4042/**
4043 *	skb_unlink	-	remove a buffer from a list
4044 *	@skb: buffer to remove
4045 *	@list: list to use
4046 *
4047 *	Remove a packet from a list. The list locks are taken and this
4048 *	function is atomic with respect to other list locked calls
4049 *
4050 *	You must know what list the SKB is on.
4051 */
4052void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
4053{
4054	unsigned long flags;
4055
4056	spin_lock_irqsave(&list->lock, flags);
4057	__skb_unlink(skb, list);
4058	spin_unlock_irqrestore(&list->lock, flags);
4059}
4060EXPORT_SYMBOL(skb_unlink);
4061
4062/**
4063 *	skb_append	-	append a buffer
4064 *	@old: buffer to insert after
4065 *	@newsk: buffer to insert
4066 *	@list: list to use
4067 *
4068 *	Place a packet after a given packet in a list. The list locks are taken
4069 *	and this function is atomic with respect to other list locked calls.
4070 *	A buffer cannot be placed on two lists at the same time.
4071 */
4072void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
4073{
4074	unsigned long flags;
4075
4076	spin_lock_irqsave(&list->lock, flags);
4077	__skb_queue_after(list, old, newsk);
4078	spin_unlock_irqrestore(&list->lock, flags);
4079}
4080EXPORT_SYMBOL(skb_append);
4081
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4082static inline void skb_split_inside_header(struct sk_buff *skb,
4083					   struct sk_buff* skb1,
4084					   const u32 len, const int pos)
4085{
4086	int i;
4087
4088	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
4089					 pos - len);
4090	/* And move data appendix as is. */
4091	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4092		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
4093
4094	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
4095	skb1->unreadable	   = skb->unreadable;
4096	skb_shinfo(skb)->nr_frags  = 0;
4097	skb1->data_len		   = skb->data_len;
4098	skb1->len		   += skb1->data_len;
4099	skb->data_len		   = 0;
4100	skb->len		   = len;
4101	skb_set_tail_pointer(skb, len);
4102}
4103
4104static inline void skb_split_no_header(struct sk_buff *skb,
4105				       struct sk_buff* skb1,
4106				       const u32 len, int pos)
4107{
4108	int i, k = 0;
4109	const int nfrags = skb_shinfo(skb)->nr_frags;
4110
4111	skb_shinfo(skb)->nr_frags = 0;
4112	skb1->len		  = skb1->data_len = skb->len - len;
4113	skb->len		  = len;
4114	skb->data_len		  = len - pos;
4115
4116	for (i = 0; i < nfrags; i++) {
4117		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4118
4119		if (pos + size > len) {
4120			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
4121
4122			if (pos < len) {
4123				/* Split frag.
4124				 * We have two variants in this case:
4125				 * 1. Move all the frag to the second
4126				 *    part, if it is possible. F.e.
4127				 *    this approach is mandatory for TUX,
4128				 *    where splitting is expensive.
4129				 * 2. Split is accurately. We make this.
4130				 */
4131				skb_frag_ref(skb, i);
4132				skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
4133				skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
4134				skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
4135				skb_shinfo(skb)->nr_frags++;
4136			}
4137			k++;
4138		} else
4139			skb_shinfo(skb)->nr_frags++;
4140		pos += size;
4141	}
4142	skb_shinfo(skb1)->nr_frags = k;
4143
4144	skb1->unreadable = skb->unreadable;
4145}
4146
4147/**
4148 * skb_split - Split fragmented skb to two parts at length len.
4149 * @skb: the buffer to split
4150 * @skb1: the buffer to receive the second part
4151 * @len: new length for skb
4152 */
4153void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
4154{
4155	int pos = skb_headlen(skb);
4156	const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
4157
4158	skb_zcopy_downgrade_managed(skb);
4159
4160	skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
4161	skb_zerocopy_clone(skb1, skb, 0);
4162	if (len < pos)	/* Split line is inside header. */
4163		skb_split_inside_header(skb, skb1, len, pos);
4164	else		/* Second chunk has no header, nothing to copy. */
4165		skb_split_no_header(skb, skb1, len, pos);
4166}
4167EXPORT_SYMBOL(skb_split);
4168
4169/* Shifting from/to a cloned skb is a no-go.
4170 *
4171 * Caller cannot keep skb_shinfo related pointers past calling here!
4172 */
4173static int skb_prepare_for_shift(struct sk_buff *skb)
4174{
4175	return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
4176}
4177
4178/**
4179 * skb_shift - Shifts paged data partially from skb to another
4180 * @tgt: buffer into which tail data gets added
4181 * @skb: buffer from which the paged data comes from
4182 * @shiftlen: shift up to this many bytes
4183 *
4184 * Attempts to shift up to shiftlen worth of bytes, which may be less than
4185 * the length of the skb, from skb to tgt. Returns number bytes shifted.
4186 * It's up to caller to free skb if everything was shifted.
4187 *
4188 * If @tgt runs out of frags, the whole operation is aborted.
4189 *
4190 * Skb cannot include anything else but paged data while tgt is allowed
4191 * to have non-paged data as well.
4192 *
4193 * TODO: full sized shift could be optimized but that would need
4194 * specialized skb free'er to handle frags without up-to-date nr_frags.
4195 */
4196int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
4197{
4198	int from, to, merge, todo;
4199	skb_frag_t *fragfrom, *fragto;
4200
4201	BUG_ON(shiftlen > skb->len);
4202
4203	if (skb_headlen(skb))
4204		return 0;
4205	if (skb_zcopy(tgt) || skb_zcopy(skb))
4206		return 0;
4207
4208	DEBUG_NET_WARN_ON_ONCE(tgt->pp_recycle != skb->pp_recycle);
4209	DEBUG_NET_WARN_ON_ONCE(skb_cmp_decrypted(tgt, skb));
4210
4211	todo = shiftlen;
4212	from = 0;
4213	to = skb_shinfo(tgt)->nr_frags;
4214	fragfrom = &skb_shinfo(skb)->frags[from];
4215
4216	/* Actual merge is delayed until the point when we know we can
4217	 * commit all, so that we don't have to undo partial changes
4218	 */
4219	if (!skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
4220			      skb_frag_off(fragfrom))) {
 
4221		merge = -1;
4222	} else {
4223		merge = to - 1;
4224
4225		todo -= skb_frag_size(fragfrom);
4226		if (todo < 0) {
4227			if (skb_prepare_for_shift(skb) ||
4228			    skb_prepare_for_shift(tgt))
4229				return 0;
4230
4231			/* All previous frag pointers might be stale! */
4232			fragfrom = &skb_shinfo(skb)->frags[from];
4233			fragto = &skb_shinfo(tgt)->frags[merge];
4234
4235			skb_frag_size_add(fragto, shiftlen);
4236			skb_frag_size_sub(fragfrom, shiftlen);
4237			skb_frag_off_add(fragfrom, shiftlen);
4238
4239			goto onlymerged;
4240		}
4241
4242		from++;
4243	}
4244
4245	/* Skip full, not-fitting skb to avoid expensive operations */
4246	if ((shiftlen == skb->len) &&
4247	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
4248		return 0;
4249
4250	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
4251		return 0;
4252
4253	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
4254		if (to == MAX_SKB_FRAGS)
4255			return 0;
4256
4257		fragfrom = &skb_shinfo(skb)->frags[from];
4258		fragto = &skb_shinfo(tgt)->frags[to];
4259
4260		if (todo >= skb_frag_size(fragfrom)) {
4261			*fragto = *fragfrom;
4262			todo -= skb_frag_size(fragfrom);
4263			from++;
4264			to++;
4265
4266		} else {
4267			__skb_frag_ref(fragfrom);
4268			skb_frag_page_copy(fragto, fragfrom);
4269			skb_frag_off_copy(fragto, fragfrom);
4270			skb_frag_size_set(fragto, todo);
4271
4272			skb_frag_off_add(fragfrom, todo);
4273			skb_frag_size_sub(fragfrom, todo);
4274			todo = 0;
4275
4276			to++;
4277			break;
4278		}
4279	}
4280
4281	/* Ready to "commit" this state change to tgt */
4282	skb_shinfo(tgt)->nr_frags = to;
4283
4284	if (merge >= 0) {
4285		fragfrom = &skb_shinfo(skb)->frags[0];
4286		fragto = &skb_shinfo(tgt)->frags[merge];
4287
4288		skb_frag_size_add(fragto, skb_frag_size(fragfrom));
4289		__skb_frag_unref(fragfrom, skb->pp_recycle);
4290	}
4291
4292	/* Reposition in the original skb */
4293	to = 0;
4294	while (from < skb_shinfo(skb)->nr_frags)
4295		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
4296	skb_shinfo(skb)->nr_frags = to;
4297
4298	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
4299
4300onlymerged:
4301	/* Most likely the tgt won't ever need its checksum anymore, skb on
4302	 * the other hand might need it if it needs to be resent
4303	 */
4304	tgt->ip_summed = CHECKSUM_PARTIAL;
4305	skb->ip_summed = CHECKSUM_PARTIAL;
4306
4307	skb_len_add(skb, -shiftlen);
4308	skb_len_add(tgt, shiftlen);
 
 
 
 
 
4309
4310	return shiftlen;
4311}
4312
4313/**
4314 * skb_prepare_seq_read - Prepare a sequential read of skb data
4315 * @skb: the buffer to read
4316 * @from: lower offset of data to be read
4317 * @to: upper offset of data to be read
4318 * @st: state variable
4319 *
4320 * Initializes the specified state variable. Must be called before
4321 * invoking skb_seq_read() for the first time.
4322 */
4323void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
4324			  unsigned int to, struct skb_seq_state *st)
4325{
4326	st->lower_offset = from;
4327	st->upper_offset = to;
4328	st->root_skb = st->cur_skb = skb;
4329	st->frag_idx = st->stepped_offset = 0;
4330	st->frag_data = NULL;
4331	st->frag_off = 0;
4332}
4333EXPORT_SYMBOL(skb_prepare_seq_read);
4334
4335/**
4336 * skb_seq_read - Sequentially read skb data
4337 * @consumed: number of bytes consumed by the caller so far
4338 * @data: destination pointer for data to be returned
4339 * @st: state variable
4340 *
4341 * Reads a block of skb data at @consumed relative to the
4342 * lower offset specified to skb_prepare_seq_read(). Assigns
4343 * the head of the data block to @data and returns the length
4344 * of the block or 0 if the end of the skb data or the upper
4345 * offset has been reached.
4346 *
4347 * The caller is not required to consume all of the data
4348 * returned, i.e. @consumed is typically set to the number
4349 * of bytes already consumed and the next call to
4350 * skb_seq_read() will return the remaining part of the block.
4351 *
4352 * Note 1: The size of each block of data returned can be arbitrary,
4353 *       this limitation is the cost for zerocopy sequential
4354 *       reads of potentially non linear data.
4355 *
4356 * Note 2: Fragment lists within fragments are not implemented
4357 *       at the moment, state->root_skb could be replaced with
4358 *       a stack for this purpose.
4359 */
4360unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
4361			  struct skb_seq_state *st)
4362{
4363	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
4364	skb_frag_t *frag;
4365
4366	if (unlikely(abs_offset >= st->upper_offset)) {
4367		if (st->frag_data) {
4368			kunmap_atomic(st->frag_data);
4369			st->frag_data = NULL;
4370		}
4371		return 0;
4372	}
4373
4374next_skb:
4375	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
4376
4377	if (abs_offset < block_limit && !st->frag_data) {
4378		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
4379		return block_limit - abs_offset;
4380	}
4381
4382	if (!skb_frags_readable(st->cur_skb))
4383		return 0;
4384
4385	if (st->frag_idx == 0 && !st->frag_data)
4386		st->stepped_offset += skb_headlen(st->cur_skb);
4387
4388	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
4389		unsigned int pg_idx, pg_off, pg_sz;
4390
4391		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
 
4392
4393		pg_idx = 0;
4394		pg_off = skb_frag_off(frag);
4395		pg_sz = skb_frag_size(frag);
4396
4397		if (skb_frag_must_loop(skb_frag_page(frag))) {
4398			pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
4399			pg_off = offset_in_page(pg_off + st->frag_off);
4400			pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
4401						    PAGE_SIZE - pg_off);
4402		}
4403
4404		block_limit = pg_sz + st->stepped_offset;
4405		if (abs_offset < block_limit) {
4406			if (!st->frag_data)
4407				st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
4408
4409			*data = (u8 *)st->frag_data + pg_off +
4410				(abs_offset - st->stepped_offset);
4411
4412			return block_limit - abs_offset;
4413		}
4414
4415		if (st->frag_data) {
4416			kunmap_atomic(st->frag_data);
4417			st->frag_data = NULL;
4418		}
4419
4420		st->stepped_offset += pg_sz;
4421		st->frag_off += pg_sz;
4422		if (st->frag_off == skb_frag_size(frag)) {
4423			st->frag_off = 0;
4424			st->frag_idx++;
4425		}
4426	}
4427
4428	if (st->frag_data) {
4429		kunmap_atomic(st->frag_data);
4430		st->frag_data = NULL;
4431	}
4432
4433	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
4434		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
4435		st->frag_idx = 0;
4436		goto next_skb;
4437	} else if (st->cur_skb->next) {
4438		st->cur_skb = st->cur_skb->next;
4439		st->frag_idx = 0;
4440		goto next_skb;
4441	}
4442
4443	return 0;
4444}
4445EXPORT_SYMBOL(skb_seq_read);
4446
4447/**
4448 * skb_abort_seq_read - Abort a sequential read of skb data
4449 * @st: state variable
4450 *
4451 * Must be called if skb_seq_read() was not called until it
4452 * returned 0.
4453 */
4454void skb_abort_seq_read(struct skb_seq_state *st)
4455{
4456	if (st->frag_data)
4457		kunmap_atomic(st->frag_data);
4458}
4459EXPORT_SYMBOL(skb_abort_seq_read);
4460
4461/**
4462 * skb_copy_seq_read() - copy from a skb_seq_state to a buffer
4463 * @st: source skb_seq_state
4464 * @offset: offset in source
4465 * @to: destination buffer
4466 * @len: number of bytes to copy
4467 *
4468 * Copy @len bytes from @offset bytes into the source @st to the destination
4469 * buffer @to. `offset` should increase (or be unchanged) with each subsequent
4470 * call to this function. If offset needs to decrease from the previous use `st`
4471 * should be reset first.
4472 *
4473 * Return: 0 on success or -EINVAL if the copy ended early
4474 */
4475int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len)
4476{
4477	const u8 *data;
4478	u32 sqlen;
4479
4480	for (;;) {
4481		sqlen = skb_seq_read(offset, &data, st);
4482		if (sqlen == 0)
4483			return -EINVAL;
4484		if (sqlen >= len) {
4485			memcpy(to, data, len);
4486			return 0;
4487		}
4488		memcpy(to, data, sqlen);
4489		to += sqlen;
4490		offset += sqlen;
4491		len -= sqlen;
4492	}
4493}
4494EXPORT_SYMBOL(skb_copy_seq_read);
4495
4496#define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
4497
4498static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
4499					  struct ts_config *conf,
4500					  struct ts_state *state)
4501{
4502	return skb_seq_read(offset, text, TS_SKB_CB(state));
4503}
4504
4505static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
4506{
4507	skb_abort_seq_read(TS_SKB_CB(state));
4508}
4509
4510/**
4511 * skb_find_text - Find a text pattern in skb data
4512 * @skb: the buffer to look in
4513 * @from: search offset
4514 * @to: search limit
4515 * @config: textsearch configuration
 
4516 *
4517 * Finds a pattern in the skb data according to the specified
4518 * textsearch configuration. Use textsearch_next() to retrieve
4519 * subsequent occurrences of the pattern. Returns the offset
4520 * to the first occurrence or UINT_MAX if no match was found.
4521 */
4522unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4523			   unsigned int to, struct ts_config *config)
 
4524{
4525	unsigned int patlen = config->ops->get_pattern_len(config);
4526	struct ts_state state;
4527	unsigned int ret;
4528
4529	BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4530
4531	config->get_next_block = skb_ts_get_next_block;
4532	config->finish = skb_ts_finish;
4533
4534	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4535
4536	ret = textsearch_find(config, &state);
4537	return (ret + patlen <= to - from ? ret : UINT_MAX);
4538}
4539EXPORT_SYMBOL(skb_find_text);
4540
4541int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4542			 int offset, size_t size, size_t max_frags)
4543{
4544	int i = skb_shinfo(skb)->nr_frags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4545
4546	if (skb_can_coalesce(skb, i, page, offset)) {
4547		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4548	} else if (i < max_frags) {
4549		skb_zcopy_downgrade_managed(skb);
4550		get_page(page);
4551		skb_fill_page_desc_noacc(skb, i, page, offset, size);
4552	} else {
4553		return -EMSGSIZE;
4554	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4555
4556	return 0;
4557}
4558EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4559
4560/**
4561 *	skb_pull_rcsum - pull skb and update receive checksum
4562 *	@skb: buffer to update
4563 *	@len: length of data pulled
4564 *
4565 *	This function performs an skb_pull on the packet and updates
4566 *	the CHECKSUM_COMPLETE checksum.  It should be used on
4567 *	receive path processing instead of skb_pull unless you know
4568 *	that the checksum difference is zero (e.g., a valid IP header)
4569 *	or you are setting ip_summed to CHECKSUM_NONE.
4570 */
4571void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4572{
4573	unsigned char *data = skb->data;
4574
4575	BUG_ON(len > skb->len);
4576	__skb_pull(skb, len);
4577	skb_postpull_rcsum(skb, data, len);
4578	return skb->data;
 
4579}
4580EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4581
4582static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4583{
4584	skb_frag_t head_frag;
4585	struct page *page;
4586
4587	page = virt_to_head_page(frag_skb->head);
4588	skb_frag_fill_page_desc(&head_frag, page, frag_skb->data -
4589				(unsigned char *)page_address(page),
4590				skb_headlen(frag_skb));
4591	return head_frag;
4592}
4593
4594struct sk_buff *skb_segment_list(struct sk_buff *skb,
4595				 netdev_features_t features,
4596				 unsigned int offset)
4597{
4598	struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4599	unsigned int tnl_hlen = skb_tnl_header_len(skb);
4600	unsigned int delta_truesize = 0;
4601	unsigned int delta_len = 0;
4602	struct sk_buff *tail = NULL;
4603	struct sk_buff *nskb, *tmp;
4604	int len_diff, err;
4605
4606	skb_push(skb, -skb_network_offset(skb) + offset);
4607
4608	/* Ensure the head is writeable before touching the shared info */
4609	err = skb_unclone(skb, GFP_ATOMIC);
4610	if (err)
4611		goto err_linearize;
4612
4613	skb_shinfo(skb)->frag_list = NULL;
4614
4615	while (list_skb) {
4616		nskb = list_skb;
4617		list_skb = list_skb->next;
4618
4619		err = 0;
4620		delta_truesize += nskb->truesize;
4621		if (skb_shared(nskb)) {
4622			tmp = skb_clone(nskb, GFP_ATOMIC);
4623			if (tmp) {
4624				consume_skb(nskb);
4625				nskb = tmp;
4626				err = skb_unclone(nskb, GFP_ATOMIC);
4627			} else {
4628				err = -ENOMEM;
4629			}
4630		}
4631
4632		if (!tail)
4633			skb->next = nskb;
4634		else
4635			tail->next = nskb;
4636
4637		if (unlikely(err)) {
4638			nskb->next = list_skb;
4639			goto err_linearize;
4640		}
4641
4642		tail = nskb;
4643
4644		delta_len += nskb->len;
4645
4646		skb_push(nskb, -skb_network_offset(nskb) + offset);
4647
4648		skb_release_head_state(nskb);
4649		len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4650		__copy_skb_header(nskb, skb);
4651
4652		skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4653		nskb->transport_header += len_diff;
4654		skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4655						 nskb->data - tnl_hlen,
4656						 offset + tnl_hlen);
4657
4658		if (skb_needs_linearize(nskb, features) &&
4659		    __skb_linearize(nskb))
4660			goto err_linearize;
4661	}
4662
4663	skb->truesize = skb->truesize - delta_truesize;
4664	skb->data_len = skb->data_len - delta_len;
4665	skb->len = skb->len - delta_len;
4666
4667	skb_gso_reset(skb);
4668
4669	skb->prev = tail;
4670
4671	if (skb_needs_linearize(skb, features) &&
4672	    __skb_linearize(skb))
4673		goto err_linearize;
4674
4675	skb_get(skb);
4676
4677	return skb;
4678
4679err_linearize:
4680	kfree_skb_list(skb->next);
4681	skb->next = NULL;
4682	return ERR_PTR(-ENOMEM);
4683}
4684EXPORT_SYMBOL_GPL(skb_segment_list);
4685
4686/**
4687 *	skb_segment - Perform protocol segmentation on skb.
4688 *	@head_skb: buffer to segment
4689 *	@features: features for the output path (see dev->features)
4690 *
4691 *	This function performs segmentation on the given skb.  It returns
4692 *	a pointer to the first in a list of new skbs for the segments.
4693 *	In case of error it returns ERR_PTR(err).
4694 */
4695struct sk_buff *skb_segment(struct sk_buff *head_skb,
4696			    netdev_features_t features)
4697{
4698	struct sk_buff *segs = NULL;
4699	struct sk_buff *tail = NULL;
4700	struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4701	unsigned int mss = skb_shinfo(head_skb)->gso_size;
4702	unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4703	unsigned int offset = doffset;
4704	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4705	unsigned int partial_segs = 0;
4706	unsigned int headroom;
4707	unsigned int len = head_skb->len;
4708	struct sk_buff *frag_skb;
4709	skb_frag_t *frag;
4710	__be16 proto;
4711	bool csum, sg;
4712	int err = -ENOMEM;
4713	int i = 0;
4714	int nfrags, pos;
4715
4716	if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4717	    mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4718		struct sk_buff *check_skb;
4719
4720		for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4721			if (skb_headlen(check_skb) && !check_skb->head_frag) {
4722				/* gso_size is untrusted, and we have a frag_list with
4723				 * a linear non head_frag item.
4724				 *
4725				 * If head_skb's headlen does not fit requested gso_size,
4726				 * it means that the frag_list members do NOT terminate
4727				 * on exact gso_size boundaries. Hence we cannot perform
4728				 * skb_frag_t page sharing. Therefore we must fallback to
4729				 * copying the frag_list skbs; we do so by disabling SG.
4730				 */
4731				features &= ~NETIF_F_SG;
4732				break;
4733			}
4734		}
4735	}
4736
4737	__skb_push(head_skb, doffset);
4738	proto = skb_network_protocol(head_skb, NULL);
4739	if (unlikely(!proto))
4740		return ERR_PTR(-EINVAL);
4741
4742	sg = !!(features & NETIF_F_SG);
4743	csum = !!can_checksum_protocol(features, proto);
4744
4745	if (sg && csum && (mss != GSO_BY_FRAGS))  {
4746		if (!(features & NETIF_F_GSO_PARTIAL)) {
4747			struct sk_buff *iter;
4748			unsigned int frag_len;
4749
4750			if (!list_skb ||
4751			    !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4752				goto normal;
4753
4754			/* If we get here then all the required
4755			 * GSO features except frag_list are supported.
4756			 * Try to split the SKB to multiple GSO SKBs
4757			 * with no frag_list.
4758			 * Currently we can do that only when the buffers don't
4759			 * have a linear part and all the buffers except
4760			 * the last are of the same length.
4761			 */
4762			frag_len = list_skb->len;
4763			skb_walk_frags(head_skb, iter) {
4764				if (frag_len != iter->len && iter->next)
4765					goto normal;
4766				if (skb_headlen(iter) && !iter->head_frag)
4767					goto normal;
4768
4769				len -= iter->len;
4770			}
4771
4772			if (len != frag_len)
4773				goto normal;
4774		}
4775
4776		/* GSO partial only requires that we trim off any excess that
4777		 * doesn't fit into an MSS sized block, so take care of that
4778		 * now.
4779		 * Cap len to not accidentally hit GSO_BY_FRAGS.
4780		 */
4781		partial_segs = min(len, GSO_BY_FRAGS - 1) / mss;
4782		if (partial_segs > 1)
4783			mss *= partial_segs;
4784		else
4785			partial_segs = 0;
4786	}
4787
4788normal:
4789	headroom = skb_headroom(head_skb);
4790	pos = skb_headlen(head_skb);
4791
4792	if (skb_orphan_frags(head_skb, GFP_ATOMIC))
4793		return ERR_PTR(-ENOMEM);
4794
4795	nfrags = skb_shinfo(head_skb)->nr_frags;
4796	frag = skb_shinfo(head_skb)->frags;
4797	frag_skb = head_skb;
4798
4799	do {
4800		struct sk_buff *nskb;
4801		skb_frag_t *nskb_frag;
4802		int hsize;
4803		int size;
4804
4805		if (unlikely(mss == GSO_BY_FRAGS)) {
4806			len = list_skb->len;
4807		} else {
4808			len = head_skb->len - offset;
4809			if (len > mss)
4810				len = mss;
4811		}
4812
4813		hsize = skb_headlen(head_skb) - offset;
4814
4815		if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4816		    (skb_headlen(list_skb) == len || sg)) {
4817			BUG_ON(skb_headlen(list_skb) > len);
 
 
 
4818
4819			nskb = skb_clone(list_skb, GFP_ATOMIC);
4820			if (unlikely(!nskb))
4821				goto err;
4822
4823			i = 0;
4824			nfrags = skb_shinfo(list_skb)->nr_frags;
4825			frag = skb_shinfo(list_skb)->frags;
4826			frag_skb = list_skb;
4827			pos += skb_headlen(list_skb);
4828
4829			while (pos < offset + len) {
4830				BUG_ON(i >= nfrags);
4831
4832				size = skb_frag_size(frag);
4833				if (pos + size > offset + len)
4834					break;
4835
4836				i++;
4837				pos += size;
4838				frag++;
4839			}
4840
4841			list_skb = list_skb->next;
4842
4843			if (unlikely(pskb_trim(nskb, len))) {
4844				kfree_skb(nskb);
4845				goto err;
4846			}
4847
4848			hsize = skb_end_offset(nskb);
4849			if (skb_cow_head(nskb, doffset + headroom)) {
4850				kfree_skb(nskb);
4851				goto err;
4852			}
4853
4854			nskb->truesize += skb_end_offset(nskb) - hsize;
4855			skb_release_head_state(nskb);
4856			__skb_push(nskb, doffset);
4857		} else {
4858			if (hsize < 0)
4859				hsize = 0;
4860			if (hsize > len || !sg)
4861				hsize = len;
4862
4863			nskb = __alloc_skb(hsize + doffset + headroom,
4864					   GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4865					   NUMA_NO_NODE);
4866
4867			if (unlikely(!nskb))
4868				goto err;
4869
4870			skb_reserve(nskb, headroom);
4871			__skb_put(nskb, doffset);
4872		}
4873
4874		if (segs)
4875			tail->next = nskb;
4876		else
4877			segs = nskb;
4878		tail = nskb;
4879
4880		__copy_skb_header(nskb, head_skb);
4881
4882		skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4883		skb_reset_mac_len(nskb);
4884
4885		skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4886						 nskb->data - tnl_hlen,
4887						 doffset + tnl_hlen);
 
 
 
 
 
 
4888
4889		if (nskb->len == len + doffset)
4890			goto perform_csum_check;
4891
4892		if (!sg) {
4893			if (!csum) {
4894				if (!nskb->remcsum_offload)
4895					nskb->ip_summed = CHECKSUM_NONE;
4896				SKB_GSO_CB(nskb)->csum =
4897					skb_copy_and_csum_bits(head_skb, offset,
4898							       skb_put(nskb,
4899								       len),
4900							       len);
4901				SKB_GSO_CB(nskb)->csum_start =
4902					skb_headroom(nskb) + doffset;
4903			} else {
4904				if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4905					goto err;
4906			}
4907			continue;
4908		}
4909
4910		nskb_frag = skb_shinfo(nskb)->frags;
4911
4912		skb_copy_from_linear_data_offset(head_skb, offset,
4913						 skb_put(nskb, hsize), hsize);
4914
4915		skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4916					   SKBFL_SHARED_FRAG;
4917
4918		if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4919			goto err;
4920
4921		while (pos < offset + len) {
4922			if (i >= nfrags) {
4923				if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
4924				    skb_zerocopy_clone(nskb, list_skb,
4925						       GFP_ATOMIC))
4926					goto err;
4927
4928				i = 0;
4929				nfrags = skb_shinfo(list_skb)->nr_frags;
4930				frag = skb_shinfo(list_skb)->frags;
4931				frag_skb = list_skb;
4932				if (!skb_headlen(list_skb)) {
4933					BUG_ON(!nfrags);
4934				} else {
4935					BUG_ON(!list_skb->head_frag);
4936
4937					/* to make room for head_frag. */
4938					i--;
4939					frag--;
4940				}
4941
4942				list_skb = list_skb->next;
4943			}
4944
4945			if (unlikely(skb_shinfo(nskb)->nr_frags >=
4946				     MAX_SKB_FRAGS)) {
4947				net_warn_ratelimited(
4948					"skb_segment: too many frags: %u %u\n",
4949					pos, mss);
4950				err = -EINVAL;
4951				goto err;
4952			}
4953
4954			*nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4955			__skb_frag_ref(nskb_frag);
4956			size = skb_frag_size(nskb_frag);
4957
4958			if (pos < offset) {
4959				skb_frag_off_add(nskb_frag, offset - pos);
4960				skb_frag_size_sub(nskb_frag, offset - pos);
4961			}
4962
4963			skb_shinfo(nskb)->nr_frags++;
4964
4965			if (pos + size <= offset + len) {
4966				i++;
4967				frag++;
4968				pos += size;
4969			} else {
4970				skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4971				goto skip_fraglist;
4972			}
4973
4974			nskb_frag++;
4975		}
4976
4977skip_fraglist:
4978		nskb->data_len = len - hsize;
4979		nskb->len += nskb->data_len;
4980		nskb->truesize += nskb->data_len;
4981
4982perform_csum_check:
4983		if (!csum) {
4984			if (skb_has_shared_frag(nskb) &&
4985			    __skb_linearize(nskb))
4986				goto err;
4987
4988			if (!nskb->remcsum_offload)
4989				nskb->ip_summed = CHECKSUM_NONE;
4990			SKB_GSO_CB(nskb)->csum =
4991				skb_checksum(nskb, doffset,
4992					     nskb->len - doffset, 0);
4993			SKB_GSO_CB(nskb)->csum_start =
4994				skb_headroom(nskb) + doffset;
4995		}
4996	} while ((offset += len) < head_skb->len);
4997
4998	/* Some callers want to get the end of the list.
4999	 * Put it in segs->prev to avoid walking the list.
5000	 * (see validate_xmit_skb_list() for example)
5001	 */
5002	segs->prev = tail;
5003
5004	if (partial_segs) {
5005		struct sk_buff *iter;
5006		int type = skb_shinfo(head_skb)->gso_type;
5007		unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
5008
5009		/* Update type to add partial and then remove dodgy if set */
5010		type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
5011		type &= ~SKB_GSO_DODGY;
5012
5013		/* Update GSO info and prepare to start updating headers on
5014		 * our way back down the stack of protocols.
5015		 */
5016		for (iter = segs; iter; iter = iter->next) {
5017			skb_shinfo(iter)->gso_size = gso_size;
5018			skb_shinfo(iter)->gso_segs = partial_segs;
5019			skb_shinfo(iter)->gso_type = type;
5020			SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
5021		}
5022
5023		if (tail->len - doffset <= gso_size)
5024			skb_shinfo(tail)->gso_size = 0;
5025		else if (tail != segs)
5026			skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
5027	}
5028
5029	/* Following permits correct backpressure, for protocols
5030	 * using skb_set_owner_w().
5031	 * Idea is to tranfert ownership from head_skb to last segment.
5032	 */
5033	if (head_skb->destructor == sock_wfree) {
5034		swap(tail->truesize, head_skb->truesize);
5035		swap(tail->destructor, head_skb->destructor);
5036		swap(tail->sk, head_skb->sk);
5037	}
5038	return segs;
5039
5040err:
5041	kfree_skb_list(segs);
 
 
 
5042	return ERR_PTR(err);
5043}
5044EXPORT_SYMBOL_GPL(skb_segment);
5045
5046#ifdef CONFIG_SKB_EXTENSIONS
5047#define SKB_EXT_ALIGN_VALUE	8
5048#define SKB_EXT_CHUNKSIZEOF(x)	(ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
5049
5050static const u8 skb_ext_type_len[] = {
5051#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
5052	[SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
5053#endif
5054#ifdef CONFIG_XFRM
5055	[SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
5056#endif
5057#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5058	[TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
5059#endif
5060#if IS_ENABLED(CONFIG_MPTCP)
5061	[SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
5062#endif
5063#if IS_ENABLED(CONFIG_MCTP_FLOWS)
5064	[SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
5065#endif
5066};
5067
5068static __always_inline unsigned int skb_ext_total_length(void)
5069{
5070	unsigned int l = SKB_EXT_CHUNKSIZEOF(struct skb_ext);
5071	int i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5072
5073	for (i = 0; i < ARRAY_SIZE(skb_ext_type_len); i++)
5074		l += skb_ext_type_len[i];
5075
5076	return l;
5077}
5078
5079static void skb_extensions_init(void)
5080{
5081	BUILD_BUG_ON(SKB_EXT_NUM >= 8);
5082#if !IS_ENABLED(CONFIG_KCOV_INSTRUMENT_ALL)
5083	BUILD_BUG_ON(skb_ext_total_length() > 255);
5084#endif
5085
5086	skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
5087					     SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
5088					     0,
5089					     SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5090					     NULL);
5091}
5092#else
5093static void skb_extensions_init(void) {}
5094#endif
5095
5096/* The SKB kmem_cache slab is critical for network performance.  Never
5097 * merge/alias the slab with similar sized objects.  This avoids fragmentation
5098 * that hurts performance of kmem_cache_{alloc,free}_bulk APIs.
5099 */
5100#ifndef CONFIG_SLUB_TINY
5101#define FLAG_SKB_NO_MERGE	SLAB_NO_MERGE
5102#else /* CONFIG_SLUB_TINY - simple loop in kmem_cache_alloc_bulk */
5103#define FLAG_SKB_NO_MERGE	0
5104#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5105
5106void __init skb_init(void)
5107{
5108	net_hotdata.skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
5109					      sizeof(struct sk_buff),
5110					      0,
5111					      SLAB_HWCACHE_ALIGN|SLAB_PANIC|
5112						FLAG_SKB_NO_MERGE,
5113					      offsetof(struct sk_buff, cb),
5114					      sizeof_field(struct sk_buff, cb),
5115					      NULL);
5116	net_hotdata.skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
5117						sizeof(struct sk_buff_fclones),
 
5118						0,
5119						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5120						NULL);
5121	/* usercopy should only access first SKB_SMALL_HEAD_HEADROOM bytes.
5122	 * struct skb_shared_info is located at the end of skb->head,
5123	 * and should not be copied to/from user.
5124	 */
5125	net_hotdata.skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
5126						SKB_SMALL_HEAD_CACHE_SIZE,
5127						0,
5128						SLAB_HWCACHE_ALIGN | SLAB_PANIC,
5129						0,
5130						SKB_SMALL_HEAD_HEADROOM,
5131						NULL);
5132	skb_extensions_init();
5133}
5134
 
 
 
 
 
 
 
 
 
 
5135static int
5136__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
5137	       unsigned int recursion_level)
5138{
5139	int start = skb_headlen(skb);
5140	int i, copy = start - offset;
5141	struct sk_buff *frag_iter;
5142	int elt = 0;
5143
5144	if (unlikely(recursion_level >= 24))
5145		return -EMSGSIZE;
5146
5147	if (copy > 0) {
5148		if (copy > len)
5149			copy = len;
5150		sg_set_buf(sg, skb->data + offset, copy);
5151		elt++;
5152		if ((len -= copy) == 0)
5153			return elt;
5154		offset += copy;
5155	}
5156
5157	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
5158		int end;
5159
5160		WARN_ON(start > offset + len);
5161
5162		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
5163		if ((copy = end - offset) > 0) {
5164			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
5165			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5166				return -EMSGSIZE;
5167
5168			if (copy > len)
5169				copy = len;
5170			sg_set_page(&sg[elt], skb_frag_page(frag), copy,
5171				    skb_frag_off(frag) + offset - start);
5172			elt++;
5173			if (!(len -= copy))
5174				return elt;
5175			offset += copy;
5176		}
5177		start = end;
5178	}
5179
5180	skb_walk_frags(skb, frag_iter) {
5181		int end, ret;
5182
5183		WARN_ON(start > offset + len);
5184
5185		end = start + frag_iter->len;
5186		if ((copy = end - offset) > 0) {
5187			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5188				return -EMSGSIZE;
5189
5190			if (copy > len)
5191				copy = len;
5192			ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
5193					      copy, recursion_level + 1);
5194			if (unlikely(ret < 0))
5195				return ret;
5196			elt += ret;
5197			if ((len -= copy) == 0)
5198				return elt;
5199			offset += copy;
5200		}
5201		start = end;
5202	}
5203	BUG_ON(len);
5204	return elt;
5205}
5206
5207/**
5208 *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
5209 *	@skb: Socket buffer containing the buffers to be mapped
5210 *	@sg: The scatter-gather list to map into
5211 *	@offset: The offset into the buffer's contents to start mapping
5212 *	@len: Length of buffer space to be mapped
5213 *
5214 *	Fill the specified scatter-gather list with mappings/pointers into a
5215 *	region of the buffer space attached to a socket buffer. Returns either
5216 *	the number of scatterlist items used, or -EMSGSIZE if the contents
5217 *	could not fit.
5218 */
5219int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
5220{
5221	int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
5222
5223	if (nsg <= 0)
5224		return nsg;
5225
5226	sg_mark_end(&sg[nsg - 1]);
5227
5228	return nsg;
5229}
5230EXPORT_SYMBOL_GPL(skb_to_sgvec);
5231
5232/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
5233 * sglist without mark the sg which contain last skb data as the end.
5234 * So the caller can mannipulate sg list as will when padding new data after
5235 * the first call without calling sg_unmark_end to expend sg list.
5236 *
5237 * Scenario to use skb_to_sgvec_nomark:
5238 * 1. sg_init_table
5239 * 2. skb_to_sgvec_nomark(payload1)
5240 * 3. skb_to_sgvec_nomark(payload2)
5241 *
5242 * This is equivalent to:
5243 * 1. sg_init_table
5244 * 2. skb_to_sgvec(payload1)
5245 * 3. sg_unmark_end
5246 * 4. skb_to_sgvec(payload2)
5247 *
5248 * When mapping multiple payload conditionally, skb_to_sgvec_nomark
5249 * is more preferable.
5250 */
5251int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
5252			int offset, int len)
5253{
5254	return __skb_to_sgvec(skb, sg, offset, len, 0);
5255}
5256EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
5257
5258
5259
5260/**
5261 *	skb_cow_data - Check that a socket buffer's data buffers are writable
5262 *	@skb: The socket buffer to check.
5263 *	@tailbits: Amount of trailing space to be added
5264 *	@trailer: Returned pointer to the skb where the @tailbits space begins
5265 *
5266 *	Make sure that the data buffers attached to a socket buffer are
5267 *	writable. If they are not, private copies are made of the data buffers
5268 *	and the socket buffer is set to use these instead.
5269 *
5270 *	If @tailbits is given, make sure that there is space to write @tailbits
5271 *	bytes of data beyond current end of socket buffer.  @trailer will be
5272 *	set to point to the skb in which this space begins.
5273 *
5274 *	The number of scatterlist elements required to completely map the
5275 *	COW'd and extended socket buffer will be returned.
5276 */
5277int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
5278{
5279	int copyflag;
5280	int elt;
5281	struct sk_buff *skb1, **skb_p;
5282
5283	/* If skb is cloned or its head is paged, reallocate
5284	 * head pulling out all the pages (pages are considered not writable
5285	 * at the moment even if they are anonymous).
5286	 */
5287	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
5288	    !__pskb_pull_tail(skb, __skb_pagelen(skb)))
5289		return -ENOMEM;
5290
5291	/* Easy case. Most of packets will go this way. */
5292	if (!skb_has_frag_list(skb)) {
5293		/* A little of trouble, not enough of space for trailer.
5294		 * This should not happen, when stack is tuned to generate
5295		 * good frames. OK, on miss we reallocate and reserve even more
5296		 * space, 128 bytes is fair. */
5297
5298		if (skb_tailroom(skb) < tailbits &&
5299		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
5300			return -ENOMEM;
5301
5302		/* Voila! */
5303		*trailer = skb;
5304		return 1;
5305	}
5306
5307	/* Misery. We are in troubles, going to mincer fragments... */
5308
5309	elt = 1;
5310	skb_p = &skb_shinfo(skb)->frag_list;
5311	copyflag = 0;
5312
5313	while ((skb1 = *skb_p) != NULL) {
5314		int ntail = 0;
5315
5316		/* The fragment is partially pulled by someone,
5317		 * this can happen on input. Copy it and everything
5318		 * after it. */
5319
5320		if (skb_shared(skb1))
5321			copyflag = 1;
5322
5323		/* If the skb is the last, worry about trailer. */
5324
5325		if (skb1->next == NULL && tailbits) {
5326			if (skb_shinfo(skb1)->nr_frags ||
5327			    skb_has_frag_list(skb1) ||
5328			    skb_tailroom(skb1) < tailbits)
5329				ntail = tailbits + 128;
5330		}
5331
5332		if (copyflag ||
5333		    skb_cloned(skb1) ||
5334		    ntail ||
5335		    skb_shinfo(skb1)->nr_frags ||
5336		    skb_has_frag_list(skb1)) {
5337			struct sk_buff *skb2;
5338
5339			/* Fuck, we are miserable poor guys... */
5340			if (ntail == 0)
5341				skb2 = skb_copy(skb1, GFP_ATOMIC);
5342			else
5343				skb2 = skb_copy_expand(skb1,
5344						       skb_headroom(skb1),
5345						       ntail,
5346						       GFP_ATOMIC);
5347			if (unlikely(skb2 == NULL))
5348				return -ENOMEM;
5349
5350			if (skb1->sk)
5351				skb_set_owner_w(skb2, skb1->sk);
5352
5353			/* Looking around. Are we still alive?
5354			 * OK, link new skb, drop old one */
5355
5356			skb2->next = skb1->next;
5357			*skb_p = skb2;
5358			kfree_skb(skb1);
5359			skb1 = skb2;
5360		}
5361		elt++;
5362		*trailer = skb1;
5363		skb_p = &skb1->next;
5364	}
5365
5366	return elt;
5367}
5368EXPORT_SYMBOL_GPL(skb_cow_data);
5369
5370static void sock_rmem_free(struct sk_buff *skb)
5371{
5372	struct sock *sk = skb->sk;
5373
5374	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
5375}
5376
5377static void skb_set_err_queue(struct sk_buff *skb)
5378{
5379	/* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
5380	 * So, it is safe to (mis)use it to mark skbs on the error queue.
5381	 */
5382	skb->pkt_type = PACKET_OUTGOING;
5383	BUILD_BUG_ON(PACKET_OUTGOING == 0);
5384}
5385
5386/*
5387 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
5388 */
5389int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
5390{
 
 
5391	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
5392	    (unsigned int)READ_ONCE(sk->sk_rcvbuf))
5393		return -ENOMEM;
5394
5395	skb_orphan(skb);
5396	skb->sk = sk;
5397	skb->destructor = sock_rmem_free;
5398	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
5399	skb_set_err_queue(skb);
5400
5401	/* before exiting rcu section, make sure dst is refcounted */
5402	skb_dst_force(skb);
5403
5404	skb_queue_tail(&sk->sk_error_queue, skb);
5405	if (!sock_flag(sk, SOCK_DEAD))
5406		sk_error_report(sk);
5407	return 0;
5408}
5409EXPORT_SYMBOL(sock_queue_err_skb);
5410
5411static bool is_icmp_err_skb(const struct sk_buff *skb)
5412{
5413	return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
5414		       SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
5415}
5416
5417struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
5418{
5419	struct sk_buff_head *q = &sk->sk_error_queue;
5420	struct sk_buff *skb, *skb_next = NULL;
5421	bool icmp_next = false;
5422	unsigned long flags;
5423
5424	if (skb_queue_empty_lockless(q))
5425		return NULL;
5426
5427	spin_lock_irqsave(&q->lock, flags);
5428	skb = __skb_dequeue(q);
5429	if (skb && (skb_next = skb_peek(q))) {
5430		icmp_next = is_icmp_err_skb(skb_next);
5431		if (icmp_next)
5432			sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
5433	}
5434	spin_unlock_irqrestore(&q->lock, flags);
5435
5436	if (is_icmp_err_skb(skb) && !icmp_next)
5437		sk->sk_err = 0;
5438
5439	if (skb_next)
5440		sk_error_report(sk);
5441
5442	return skb;
5443}
5444EXPORT_SYMBOL(sock_dequeue_err_skb);
5445
5446/**
5447 * skb_clone_sk - create clone of skb, and take reference to socket
5448 * @skb: the skb to clone
5449 *
5450 * This function creates a clone of a buffer that holds a reference on
5451 * sk_refcnt.  Buffers created via this function are meant to be
5452 * returned using sock_queue_err_skb, or free via kfree_skb.
5453 *
5454 * When passing buffers allocated with this function to sock_queue_err_skb
5455 * it is necessary to wrap the call with sock_hold/sock_put in order to
5456 * prevent the socket from being released prior to being enqueued on
5457 * the sk_error_queue.
5458 */
5459struct sk_buff *skb_clone_sk(struct sk_buff *skb)
5460{
5461	struct sock *sk = skb->sk;
5462	struct sk_buff *clone;
5463
5464	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
5465		return NULL;
 
5466
5467	clone = skb_clone(skb, GFP_ATOMIC);
5468	if (!clone) {
5469		sock_put(sk);
5470		return NULL;
 
 
 
 
 
 
5471	}
5472
5473	clone->sk = sk;
5474	clone->destructor = sock_efree;
5475
5476	return clone;
5477}
5478EXPORT_SYMBOL(skb_clone_sk);
5479
5480static void __skb_complete_tx_timestamp(struct sk_buff *skb,
5481					struct sock *sk,
5482					int tstype,
5483					bool opt_stats)
5484{
5485	struct sock_exterr_skb *serr;
5486	int err;
5487
5488	BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
5489
5490	serr = SKB_EXT_ERR(skb);
5491	memset(serr, 0, sizeof(*serr));
5492	serr->ee.ee_errno = ENOMSG;
5493	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
5494	serr->ee.ee_info = tstype;
5495	serr->opt_stats = opt_stats;
5496	serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
5497	if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
5498		serr->ee.ee_data = skb_shinfo(skb)->tskey;
5499		if (sk_is_tcp(sk))
5500			serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
5501	}
5502
5503	err = sock_queue_err_skb(sk, skb);
5504
5505	if (err)
5506		kfree_skb(skb);
5507}
5508
5509static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
5510{
5511	bool ret;
5512
5513	if (likely(tsonly || READ_ONCE(sock_net(sk)->core.sysctl_tstamp_allow_data)))
5514		return true;
5515
5516	read_lock_bh(&sk->sk_callback_lock);
5517	ret = sk->sk_socket && sk->sk_socket->file &&
5518	      file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
5519	read_unlock_bh(&sk->sk_callback_lock);
5520	return ret;
5521}
5522
5523void skb_complete_tx_timestamp(struct sk_buff *skb,
5524			       struct skb_shared_hwtstamps *hwtstamps)
5525{
5526	struct sock *sk = skb->sk;
5527
5528	if (!skb_may_tx_timestamp(sk, false))
5529		goto err;
5530
5531	/* Take a reference to prevent skb_orphan() from freeing the socket,
5532	 * but only if the socket refcount is not zero.
5533	 */
5534	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5535		*skb_hwtstamps(skb) = *hwtstamps;
5536		__skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5537		sock_put(sk);
5538		return;
5539	}
5540
5541err:
5542	kfree_skb(skb);
5543}
5544EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5545
5546void __skb_tstamp_tx(struct sk_buff *orig_skb,
5547		     const struct sk_buff *ack_skb,
5548		     struct skb_shared_hwtstamps *hwtstamps,
5549		     struct sock *sk, int tstype)
5550{
5551	struct sk_buff *skb;
5552	bool tsonly, opt_stats = false;
5553	u32 tsflags;
5554
5555	if (!sk)
5556		return;
5557
5558	tsflags = READ_ONCE(sk->sk_tsflags);
5559	if (!hwtstamps && !(tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5560	    skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5561		return;
5562
5563	tsonly = tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5564	if (!skb_may_tx_timestamp(sk, tsonly))
5565		return;
5566
5567	if (tsonly) {
5568#ifdef CONFIG_INET
5569		if ((tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5570		    sk_is_tcp(sk)) {
5571			skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5572							     ack_skb);
5573			opt_stats = true;
5574		} else
5575#endif
5576			skb = alloc_skb(0, GFP_ATOMIC);
5577	} else {
5578		skb = skb_clone(orig_skb, GFP_ATOMIC);
5579
5580		if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
5581			kfree_skb(skb);
5582			return;
5583		}
5584	}
5585	if (!skb)
5586		return;
5587
5588	if (tsonly) {
5589		skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5590					     SKBTX_ANY_TSTAMP;
5591		skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5592	}
5593
5594	if (hwtstamps)
5595		*skb_hwtstamps(skb) = *hwtstamps;
5596	else
5597		__net_timestamp(skb);
5598
5599	__skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5600}
5601EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5602
5603void skb_tstamp_tx(struct sk_buff *orig_skb,
5604		   struct skb_shared_hwtstamps *hwtstamps)
5605{
5606	return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5607			       SCM_TSTAMP_SND);
5608}
5609EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5610
5611#ifdef CONFIG_WIRELESS
5612void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5613{
5614	struct sock *sk = skb->sk;
5615	struct sock_exterr_skb *serr;
5616	int err = 1;
5617
5618	skb->wifi_acked_valid = 1;
5619	skb->wifi_acked = acked;
5620
5621	serr = SKB_EXT_ERR(skb);
5622	memset(serr, 0, sizeof(*serr));
5623	serr->ee.ee_errno = ENOMSG;
5624	serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5625
5626	/* Take a reference to prevent skb_orphan() from freeing the socket,
5627	 * but only if the socket refcount is not zero.
5628	 */
5629	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5630		err = sock_queue_err_skb(sk, skb);
5631		sock_put(sk);
5632	}
5633	if (err)
5634		kfree_skb(skb);
5635}
5636EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5637#endif /* CONFIG_WIRELESS */
5638
5639/**
5640 * skb_partial_csum_set - set up and verify partial csum values for packet
5641 * @skb: the skb to set
5642 * @start: the number of bytes after skb->data to start checksumming.
5643 * @off: the offset from start to place the checksum.
5644 *
5645 * For untrusted partially-checksummed packets, we need to make sure the values
5646 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5647 *
5648 * This function checks and sets those values and skb->ip_summed: if this
5649 * returns false you should drop the packet.
5650 */
5651bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5652{
5653	u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5654	u32 csum_start = skb_headroom(skb) + (u32)start;
5655
5656	if (unlikely(csum_start >= U16_MAX || csum_end > skb_headlen(skb))) {
5657		net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5658				     start, off, skb_headroom(skb), skb_headlen(skb));
5659		return false;
5660	}
5661	skb->ip_summed = CHECKSUM_PARTIAL;
5662	skb->csum_start = csum_start;
5663	skb->csum_offset = off;
5664	skb->transport_header = csum_start;
5665	return true;
5666}
5667EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5668
5669static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5670			       unsigned int max)
5671{
5672	if (skb_headlen(skb) >= len)
5673		return 0;
5674
5675	/* If we need to pullup then pullup to the max, so we
5676	 * won't need to do it again.
5677	 */
5678	if (max > skb->len)
5679		max = skb->len;
5680
5681	if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5682		return -ENOMEM;
5683
5684	if (skb_headlen(skb) < len)
5685		return -EPROTO;
5686
5687	return 0;
5688}
5689
5690#define MAX_TCP_HDR_LEN (15 * 4)
5691
5692static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5693				      typeof(IPPROTO_IP) proto,
5694				      unsigned int off)
5695{
5696	int err;
5697
5698	switch (proto) {
5699	case IPPROTO_TCP:
5700		err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5701					  off + MAX_TCP_HDR_LEN);
5702		if (!err && !skb_partial_csum_set(skb, off,
5703						  offsetof(struct tcphdr,
5704							   check)))
5705			err = -EPROTO;
5706		return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5707
5708	case IPPROTO_UDP:
5709		err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5710					  off + sizeof(struct udphdr));
5711		if (!err && !skb_partial_csum_set(skb, off,
5712						  offsetof(struct udphdr,
5713							   check)))
5714			err = -EPROTO;
5715		return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5716	}
5717
5718	return ERR_PTR(-EPROTO);
5719}
5720
5721/* This value should be large enough to cover a tagged ethernet header plus
5722 * maximally sized IP and TCP or UDP headers.
5723 */
5724#define MAX_IP_HDR_LEN 128
5725
5726static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5727{
5728	unsigned int off;
5729	bool fragment;
5730	__sum16 *csum;
5731	int err;
5732
5733	fragment = false;
5734
5735	err = skb_maybe_pull_tail(skb,
5736				  sizeof(struct iphdr),
5737				  MAX_IP_HDR_LEN);
5738	if (err < 0)
5739		goto out;
5740
5741	if (ip_is_fragment(ip_hdr(skb)))
5742		fragment = true;
5743
5744	off = ip_hdrlen(skb);
5745
5746	err = -EPROTO;
5747
5748	if (fragment)
5749		goto out;
5750
5751	csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5752	if (IS_ERR(csum))
5753		return PTR_ERR(csum);
5754
5755	if (recalculate)
5756		*csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5757					   ip_hdr(skb)->daddr,
5758					   skb->len - off,
5759					   ip_hdr(skb)->protocol, 0);
5760	err = 0;
5761
5762out:
5763	return err;
5764}
5765
5766/* This value should be large enough to cover a tagged ethernet header plus
5767 * an IPv6 header, all options, and a maximal TCP or UDP header.
5768 */
5769#define MAX_IPV6_HDR_LEN 256
5770
5771#define OPT_HDR(type, skb, off) \
5772	(type *)(skb_network_header(skb) + (off))
5773
5774static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5775{
5776	int err;
5777	u8 nexthdr;
5778	unsigned int off;
5779	unsigned int len;
5780	bool fragment;
5781	bool done;
5782	__sum16 *csum;
5783
5784	fragment = false;
5785	done = false;
5786
5787	off = sizeof(struct ipv6hdr);
5788
5789	err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5790	if (err < 0)
5791		goto out;
5792
5793	nexthdr = ipv6_hdr(skb)->nexthdr;
5794
5795	len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5796	while (off <= len && !done) {
5797		switch (nexthdr) {
5798		case IPPROTO_DSTOPTS:
5799		case IPPROTO_HOPOPTS:
5800		case IPPROTO_ROUTING: {
5801			struct ipv6_opt_hdr *hp;
5802
5803			err = skb_maybe_pull_tail(skb,
5804						  off +
5805						  sizeof(struct ipv6_opt_hdr),
5806						  MAX_IPV6_HDR_LEN);
5807			if (err < 0)
5808				goto out;
5809
5810			hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5811			nexthdr = hp->nexthdr;
5812			off += ipv6_optlen(hp);
5813			break;
5814		}
5815		case IPPROTO_AH: {
5816			struct ip_auth_hdr *hp;
5817
5818			err = skb_maybe_pull_tail(skb,
5819						  off +
5820						  sizeof(struct ip_auth_hdr),
5821						  MAX_IPV6_HDR_LEN);
5822			if (err < 0)
5823				goto out;
5824
5825			hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5826			nexthdr = hp->nexthdr;
5827			off += ipv6_authlen(hp);
5828			break;
5829		}
5830		case IPPROTO_FRAGMENT: {
5831			struct frag_hdr *hp;
5832
5833			err = skb_maybe_pull_tail(skb,
5834						  off +
5835						  sizeof(struct frag_hdr),
5836						  MAX_IPV6_HDR_LEN);
5837			if (err < 0)
5838				goto out;
5839
5840			hp = OPT_HDR(struct frag_hdr, skb, off);
5841
5842			if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5843				fragment = true;
5844
5845			nexthdr = hp->nexthdr;
5846			off += sizeof(struct frag_hdr);
5847			break;
5848		}
5849		default:
5850			done = true;
5851			break;
5852		}
5853	}
5854
5855	err = -EPROTO;
5856
5857	if (!done || fragment)
5858		goto out;
5859
5860	csum = skb_checksum_setup_ip(skb, nexthdr, off);
5861	if (IS_ERR(csum))
5862		return PTR_ERR(csum);
5863
5864	if (recalculate)
5865		*csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5866					 &ipv6_hdr(skb)->daddr,
5867					 skb->len - off, nexthdr, 0);
5868	err = 0;
5869
5870out:
5871	return err;
5872}
5873
5874/**
5875 * skb_checksum_setup - set up partial checksum offset
5876 * @skb: the skb to set up
5877 * @recalculate: if true the pseudo-header checksum will be recalculated
5878 */
5879int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5880{
5881	int err;
5882
5883	switch (skb->protocol) {
5884	case htons(ETH_P_IP):
5885		err = skb_checksum_setup_ipv4(skb, recalculate);
5886		break;
5887
5888	case htons(ETH_P_IPV6):
5889		err = skb_checksum_setup_ipv6(skb, recalculate);
5890		break;
5891
5892	default:
5893		err = -EPROTO;
5894		break;
5895	}
5896
5897	return err;
5898}
5899EXPORT_SYMBOL(skb_checksum_setup);
5900
5901/**
5902 * skb_checksum_maybe_trim - maybe trims the given skb
5903 * @skb: the skb to check
5904 * @transport_len: the data length beyond the network header
5905 *
5906 * Checks whether the given skb has data beyond the given transport length.
5907 * If so, returns a cloned skb trimmed to this transport length.
5908 * Otherwise returns the provided skb. Returns NULL in error cases
5909 * (e.g. transport_len exceeds skb length or out-of-memory).
5910 *
5911 * Caller needs to set the skb transport header and free any returned skb if it
5912 * differs from the provided skb.
5913 */
5914static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5915					       unsigned int transport_len)
5916{
5917	struct sk_buff *skb_chk;
5918	unsigned int len = skb_transport_offset(skb) + transport_len;
5919	int ret;
5920
5921	if (skb->len < len)
5922		return NULL;
5923	else if (skb->len == len)
5924		return skb;
5925
5926	skb_chk = skb_clone(skb, GFP_ATOMIC);
5927	if (!skb_chk)
5928		return NULL;
5929
5930	ret = pskb_trim_rcsum(skb_chk, len);
5931	if (ret) {
5932		kfree_skb(skb_chk);
5933		return NULL;
5934	}
5935
5936	return skb_chk;
5937}
5938
5939/**
5940 * skb_checksum_trimmed - validate checksum of an skb
5941 * @skb: the skb to check
5942 * @transport_len: the data length beyond the network header
5943 * @skb_chkf: checksum function to use
5944 *
5945 * Applies the given checksum function skb_chkf to the provided skb.
5946 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5947 *
5948 * If the skb has data beyond the given transport length, then a
5949 * trimmed & cloned skb is checked and returned.
5950 *
5951 * Caller needs to set the skb transport header and free any returned skb if it
5952 * differs from the provided skb.
5953 */
5954struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5955				     unsigned int transport_len,
5956				     __sum16(*skb_chkf)(struct sk_buff *skb))
5957{
5958	struct sk_buff *skb_chk;
5959	unsigned int offset = skb_transport_offset(skb);
5960	__sum16 ret;
5961
5962	skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5963	if (!skb_chk)
5964		goto err;
5965
5966	if (!pskb_may_pull(skb_chk, offset))
5967		goto err;
5968
5969	skb_pull_rcsum(skb_chk, offset);
5970	ret = skb_chkf(skb_chk);
5971	skb_push_rcsum(skb_chk, offset);
5972
5973	if (ret)
5974		goto err;
5975
5976	return skb_chk;
5977
5978err:
5979	if (skb_chk && skb_chk != skb)
5980		kfree_skb(skb_chk);
5981
5982	return NULL;
5983
5984}
5985EXPORT_SYMBOL(skb_checksum_trimmed);
5986
5987void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5988{
5989	net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5990			     skb->dev->name);
5991}
5992EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5993
5994void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5995{
5996	if (head_stolen) {
5997		skb_release_head_state(skb);
5998		kmem_cache_free(net_hotdata.skbuff_cache, skb);
5999	} else {
6000		__kfree_skb(skb);
6001	}
6002}
6003EXPORT_SYMBOL(kfree_skb_partial);
6004
6005/**
6006 * skb_try_coalesce - try to merge skb to prior one
6007 * @to: prior buffer
6008 * @from: buffer to add
6009 * @fragstolen: pointer to boolean
6010 * @delta_truesize: how much more was allocated than was requested
6011 */
6012bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
6013		      bool *fragstolen, int *delta_truesize)
6014{
6015	struct skb_shared_info *to_shinfo, *from_shinfo;
6016	int i, delta, len = from->len;
6017
6018	*fragstolen = false;
6019
6020	if (skb_cloned(to))
6021		return false;
6022
6023	/* In general, avoid mixing page_pool and non-page_pool allocated
6024	 * pages within the same SKB. In theory we could take full
6025	 * references if @from is cloned and !@to->pp_recycle but its
6026	 * tricky (due to potential race with the clone disappearing) and
6027	 * rare, so not worth dealing with.
6028	 */
6029	if (to->pp_recycle != from->pp_recycle)
6030		return false;
6031
6032	if (skb_frags_readable(from) != skb_frags_readable(to))
6033		return false;
6034
6035	if (len <= skb_tailroom(to) && skb_frags_readable(from)) {
6036		if (len)
6037			BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
6038		*delta_truesize = 0;
6039		return true;
6040	}
6041
6042	to_shinfo = skb_shinfo(to);
6043	from_shinfo = skb_shinfo(from);
6044	if (to_shinfo->frag_list || from_shinfo->frag_list)
6045		return false;
6046	if (skb_zcopy(to) || skb_zcopy(from))
6047		return false;
6048
6049	if (skb_headlen(from) != 0) {
6050		struct page *page;
6051		unsigned int offset;
6052
6053		if (to_shinfo->nr_frags +
6054		    from_shinfo->nr_frags >= MAX_SKB_FRAGS)
6055			return false;
6056
6057		if (skb_head_is_locked(from))
6058			return false;
6059
6060		delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
6061
6062		page = virt_to_head_page(from->head);
6063		offset = from->data - (unsigned char *)page_address(page);
6064
6065		skb_fill_page_desc(to, to_shinfo->nr_frags,
6066				   page, offset, skb_headlen(from));
6067		*fragstolen = true;
6068	} else {
6069		if (to_shinfo->nr_frags +
6070		    from_shinfo->nr_frags > MAX_SKB_FRAGS)
6071			return false;
6072
6073		delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
 
6074	}
6075
6076	WARN_ON_ONCE(delta < len);
6077
6078	memcpy(to_shinfo->frags + to_shinfo->nr_frags,
6079	       from_shinfo->frags,
6080	       from_shinfo->nr_frags * sizeof(skb_frag_t));
6081	to_shinfo->nr_frags += from_shinfo->nr_frags;
6082
6083	if (!skb_cloned(from))
6084		from_shinfo->nr_frags = 0;
6085
6086	/* if the skb is not cloned this does nothing
6087	 * since we set nr_frags to 0.
6088	 */
6089	if (skb_pp_frag_ref(from)) {
6090		for (i = 0; i < from_shinfo->nr_frags; i++)
6091			__skb_frag_ref(&from_shinfo->frags[i]);
6092	}
6093
6094	to->truesize += delta;
6095	to->len += len;
6096	to->data_len += len;
6097
6098	*delta_truesize = delta;
6099	return true;
6100}
6101EXPORT_SYMBOL(skb_try_coalesce);
6102
6103/**
6104 * skb_scrub_packet - scrub an skb
6105 *
6106 * @skb: buffer to clean
6107 * @xnet: packet is crossing netns
6108 *
6109 * skb_scrub_packet can be used after encapsulating or decapsulating a packet
6110 * into/from a tunnel. Some information have to be cleared during these
6111 * operations.
6112 * skb_scrub_packet can also be used to clean a skb before injecting it in
6113 * another namespace (@xnet == true). We have to clear all information in the
6114 * skb that could impact namespace isolation.
6115 */
6116void skb_scrub_packet(struct sk_buff *skb, bool xnet)
6117{
6118	skb->pkt_type = PACKET_HOST;
6119	skb->skb_iif = 0;
6120	skb->ignore_df = 0;
6121	skb_dst_drop(skb);
6122	skb_ext_reset(skb);
6123	nf_reset_ct(skb);
6124	nf_reset_trace(skb);
6125
6126#ifdef CONFIG_NET_SWITCHDEV
6127	skb->offload_fwd_mark = 0;
6128	skb->offload_l3_fwd_mark = 0;
6129#endif
6130	ipvs_reset(skb);
6131
6132	if (!xnet)
6133		return;
6134
6135	skb->mark = 0;
6136	skb_clear_tstamp(skb);
6137}
6138EXPORT_SYMBOL_GPL(skb_scrub_packet);
6139
6140static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
6141{
6142	int mac_len, meta_len;
6143	void *meta;
6144
6145	if (skb_cow(skb, skb_headroom(skb)) < 0) {
6146		kfree_skb(skb);
6147		return NULL;
6148	}
6149
6150	mac_len = skb->data - skb_mac_header(skb);
6151	if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
6152		memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
6153			mac_len - VLAN_HLEN - ETH_TLEN);
6154	}
6155
6156	meta_len = skb_metadata_len(skb);
6157	if (meta_len) {
6158		meta = skb_metadata_end(skb) - meta_len;
6159		memmove(meta + VLAN_HLEN, meta, meta_len);
6160	}
6161
6162	skb->mac_header += VLAN_HLEN;
6163	return skb;
6164}
6165
6166struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
6167{
6168	struct vlan_hdr *vhdr;
6169	u16 vlan_tci;
6170
6171	if (unlikely(skb_vlan_tag_present(skb))) {
6172		/* vlan_tci is already set-up so leave this for another time */
6173		return skb;
6174	}
6175
6176	skb = skb_share_check(skb, GFP_ATOMIC);
6177	if (unlikely(!skb))
6178		goto err_free;
6179	/* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
6180	if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
6181		goto err_free;
6182
6183	vhdr = (struct vlan_hdr *)skb->data;
6184	vlan_tci = ntohs(vhdr->h_vlan_TCI);
6185	__vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
6186
6187	skb_pull_rcsum(skb, VLAN_HLEN);
6188	vlan_set_encap_proto(skb, vhdr);
6189
6190	skb = skb_reorder_vlan_header(skb);
6191	if (unlikely(!skb))
6192		goto err_free;
6193
6194	skb_reset_network_header(skb);
6195	if (!skb_transport_header_was_set(skb))
6196		skb_reset_transport_header(skb);
6197	skb_reset_mac_len(skb);
6198
6199	return skb;
6200
6201err_free:
6202	kfree_skb(skb);
6203	return NULL;
6204}
6205EXPORT_SYMBOL(skb_vlan_untag);
6206
6207int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
6208{
6209	if (!pskb_may_pull(skb, write_len))
6210		return -ENOMEM;
6211
6212	if (!skb_frags_readable(skb))
6213		return -EFAULT;
6214
6215	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
6216		return 0;
6217
6218	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6219}
6220EXPORT_SYMBOL(skb_ensure_writable);
6221
6222int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev)
6223{
6224	int needed_headroom = dev->needed_headroom;
6225	int needed_tailroom = dev->needed_tailroom;
6226
6227	/* For tail taggers, we need to pad short frames ourselves, to ensure
6228	 * that the tail tag does not fail at its role of being at the end of
6229	 * the packet, once the conduit interface pads the frame. Account for
6230	 * that pad length here, and pad later.
6231	 */
6232	if (unlikely(needed_tailroom && skb->len < ETH_ZLEN))
6233		needed_tailroom += ETH_ZLEN - skb->len;
6234	/* skb_headroom() returns unsigned int... */
6235	needed_headroom = max_t(int, needed_headroom - skb_headroom(skb), 0);
6236	needed_tailroom = max_t(int, needed_tailroom - skb_tailroom(skb), 0);
6237
6238	if (likely(!needed_headroom && !needed_tailroom && !skb_cloned(skb)))
6239		/* No reallocation needed, yay! */
6240		return 0;
6241
6242	return pskb_expand_head(skb, needed_headroom, needed_tailroom,
6243				GFP_ATOMIC);
6244}
6245EXPORT_SYMBOL(skb_ensure_writable_head_tail);
6246
6247/* remove VLAN header from packet and update csum accordingly.
6248 * expects a non skb_vlan_tag_present skb with a vlan tag payload
6249 */
6250int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
6251{
6252	int offset = skb->data - skb_mac_header(skb);
6253	int err;
6254
6255	if (WARN_ONCE(offset,
6256		      "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
6257		      offset)) {
6258		return -EINVAL;
6259	}
6260
6261	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
6262	if (unlikely(err))
6263		return err;
6264
6265	skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6266
6267	vlan_remove_tag(skb, vlan_tci);
6268
6269	skb->mac_header += VLAN_HLEN;
6270
6271	if (skb_network_offset(skb) < ETH_HLEN)
6272		skb_set_network_header(skb, ETH_HLEN);
6273
6274	skb_reset_mac_len(skb);
6275
6276	return err;
6277}
6278EXPORT_SYMBOL(__skb_vlan_pop);
6279
6280/* Pop a vlan tag either from hwaccel or from payload.
6281 * Expects skb->data at mac header.
6282 */
6283int skb_vlan_pop(struct sk_buff *skb)
6284{
6285	u16 vlan_tci;
6286	__be16 vlan_proto;
6287	int err;
6288
6289	if (likely(skb_vlan_tag_present(skb))) {
6290		__vlan_hwaccel_clear_tag(skb);
6291	} else {
6292		if (unlikely(!eth_type_vlan(skb->protocol)))
6293			return 0;
6294
6295		err = __skb_vlan_pop(skb, &vlan_tci);
6296		if (err)
6297			return err;
6298	}
6299	/* move next vlan tag to hw accel tag */
6300	if (likely(!eth_type_vlan(skb->protocol)))
6301		return 0;
6302
6303	vlan_proto = skb->protocol;
6304	err = __skb_vlan_pop(skb, &vlan_tci);
6305	if (unlikely(err))
6306		return err;
6307
6308	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6309	return 0;
6310}
6311EXPORT_SYMBOL(skb_vlan_pop);
6312
6313/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
6314 * Expects skb->data at mac header.
6315 */
6316int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
6317{
6318	if (skb_vlan_tag_present(skb)) {
6319		int offset = skb->data - skb_mac_header(skb);
6320		int err;
6321
6322		if (WARN_ONCE(offset,
6323			      "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
6324			      offset)) {
6325			return -EINVAL;
6326		}
6327
6328		err = __vlan_insert_tag(skb, skb->vlan_proto,
6329					skb_vlan_tag_get(skb));
6330		if (err)
6331			return err;
6332
6333		skb->protocol = skb->vlan_proto;
6334		skb->network_header -= VLAN_HLEN;
6335
6336		skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6337	}
6338	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6339	return 0;
6340}
6341EXPORT_SYMBOL(skb_vlan_push);
6342
6343/**
6344 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
6345 *
6346 * @skb: Socket buffer to modify
6347 *
6348 * Drop the Ethernet header of @skb.
6349 *
6350 * Expects that skb->data points to the mac header and that no VLAN tags are
6351 * present.
6352 *
6353 * Returns 0 on success, -errno otherwise.
6354 */
6355int skb_eth_pop(struct sk_buff *skb)
6356{
6357	if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
6358	    skb_network_offset(skb) < ETH_HLEN)
6359		return -EPROTO;
6360
6361	skb_pull_rcsum(skb, ETH_HLEN);
6362	skb_reset_mac_header(skb);
6363	skb_reset_mac_len(skb);
6364
6365	return 0;
6366}
6367EXPORT_SYMBOL(skb_eth_pop);
6368
6369/**
6370 * skb_eth_push() - Add a new Ethernet header at the head of a packet
6371 *
6372 * @skb: Socket buffer to modify
6373 * @dst: Destination MAC address of the new header
6374 * @src: Source MAC address of the new header
6375 *
6376 * Prepend @skb with a new Ethernet header.
6377 *
6378 * Expects that skb->data points to the mac header, which must be empty.
6379 *
6380 * Returns 0 on success, -errno otherwise.
6381 */
6382int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
6383		 const unsigned char *src)
6384{
6385	struct ethhdr *eth;
6386	int err;
6387
6388	if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
6389		return -EPROTO;
6390
6391	err = skb_cow_head(skb, sizeof(*eth));
6392	if (err < 0)
6393		return err;
6394
6395	skb_push(skb, sizeof(*eth));
6396	skb_reset_mac_header(skb);
6397	skb_reset_mac_len(skb);
6398
6399	eth = eth_hdr(skb);
6400	ether_addr_copy(eth->h_dest, dst);
6401	ether_addr_copy(eth->h_source, src);
6402	eth->h_proto = skb->protocol;
6403
6404	skb_postpush_rcsum(skb, eth, sizeof(*eth));
6405
6406	return 0;
6407}
6408EXPORT_SYMBOL(skb_eth_push);
6409
6410/* Update the ethertype of hdr and the skb csum value if required. */
6411static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
6412			     __be16 ethertype)
6413{
6414	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6415		__be16 diff[] = { ~hdr->h_proto, ethertype };
6416
6417		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6418	}
6419
6420	hdr->h_proto = ethertype;
6421}
6422
6423/**
6424 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6425 *                   the packet
6426 *
6427 * @skb: buffer
6428 * @mpls_lse: MPLS label stack entry to push
6429 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6430 * @mac_len: length of the MAC header
6431 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6432 *            ethernet
6433 *
6434 * Expects skb->data at mac header.
6435 *
6436 * Returns 0 on success, -errno otherwise.
6437 */
6438int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6439		  int mac_len, bool ethernet)
6440{
6441	struct mpls_shim_hdr *lse;
6442	int err;
6443
6444	if (unlikely(!eth_p_mpls(mpls_proto)))
6445		return -EINVAL;
6446
6447	/* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6448	if (skb->encapsulation)
6449		return -EINVAL;
6450
6451	err = skb_cow_head(skb, MPLS_HLEN);
6452	if (unlikely(err))
6453		return err;
6454
6455	if (!skb->inner_protocol) {
6456		skb_set_inner_network_header(skb, skb_network_offset(skb));
6457		skb_set_inner_protocol(skb, skb->protocol);
6458	}
6459
6460	skb_push(skb, MPLS_HLEN);
6461	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6462		mac_len);
6463	skb_reset_mac_header(skb);
6464	skb_set_network_header(skb, mac_len);
6465	skb_reset_mac_len(skb);
6466
6467	lse = mpls_hdr(skb);
6468	lse->label_stack_entry = mpls_lse;
6469	skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6470
6471	if (ethernet && mac_len >= ETH_HLEN)
6472		skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6473	skb->protocol = mpls_proto;
6474
6475	return 0;
6476}
6477EXPORT_SYMBOL_GPL(skb_mpls_push);
6478
6479/**
6480 * skb_mpls_pop() - pop the outermost MPLS header
6481 *
6482 * @skb: buffer
6483 * @next_proto: ethertype of header after popped MPLS header
6484 * @mac_len: length of the MAC header
6485 * @ethernet: flag to indicate if the packet is ethernet
6486 *
6487 * Expects skb->data at mac header.
6488 *
6489 * Returns 0 on success, -errno otherwise.
6490 */
6491int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6492		 bool ethernet)
6493{
6494	int err;
6495
6496	if (unlikely(!eth_p_mpls(skb->protocol)))
6497		return 0;
6498
6499	err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6500	if (unlikely(err))
6501		return err;
6502
6503	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6504	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6505		mac_len);
6506
6507	__skb_pull(skb, MPLS_HLEN);
6508	skb_reset_mac_header(skb);
6509	skb_set_network_header(skb, mac_len);
6510
6511	if (ethernet && mac_len >= ETH_HLEN) {
6512		struct ethhdr *hdr;
6513
6514		/* use mpls_hdr() to get ethertype to account for VLANs. */
6515		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6516		skb_mod_eth_type(skb, hdr, next_proto);
6517	}
6518	skb->protocol = next_proto;
6519
6520	return 0;
6521}
6522EXPORT_SYMBOL_GPL(skb_mpls_pop);
6523
6524/**
6525 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6526 *
6527 * @skb: buffer
6528 * @mpls_lse: new MPLS label stack entry to update to
6529 *
6530 * Expects skb->data at mac header.
6531 *
6532 * Returns 0 on success, -errno otherwise.
6533 */
6534int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6535{
6536	int err;
6537
6538	if (unlikely(!eth_p_mpls(skb->protocol)))
6539		return -EINVAL;
6540
6541	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6542	if (unlikely(err))
6543		return err;
6544
6545	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6546		__be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6547
6548		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6549	}
6550
6551	mpls_hdr(skb)->label_stack_entry = mpls_lse;
6552
6553	return 0;
6554}
6555EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6556
6557/**
6558 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6559 *
6560 * @skb: buffer
6561 *
6562 * Expects skb->data at mac header.
6563 *
6564 * Returns 0 on success, -errno otherwise.
6565 */
6566int skb_mpls_dec_ttl(struct sk_buff *skb)
6567{
6568	u32 lse;
6569	u8 ttl;
6570
6571	if (unlikely(!eth_p_mpls(skb->protocol)))
6572		return -EINVAL;
6573
6574	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6575		return -ENOMEM;
6576
6577	lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6578	ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6579	if (!--ttl)
6580		return -EINVAL;
6581
6582	lse &= ~MPLS_LS_TTL_MASK;
6583	lse |= ttl << MPLS_LS_TTL_SHIFT;
6584
6585	return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6586}
6587EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6588
6589/**
6590 * alloc_skb_with_frags - allocate skb with page frags
6591 *
6592 * @header_len: size of linear part
6593 * @data_len: needed length in frags
6594 * @order: max page order desired.
6595 * @errcode: pointer to error code if any
6596 * @gfp_mask: allocation mask
6597 *
6598 * This can be used to allocate a paged skb, given a maximal order for frags.
6599 */
6600struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6601				     unsigned long data_len,
6602				     int order,
6603				     int *errcode,
6604				     gfp_t gfp_mask)
6605{
6606	unsigned long chunk;
6607	struct sk_buff *skb;
6608	struct page *page;
6609	int nr_frags = 0;
6610
6611	*errcode = -EMSGSIZE;
6612	if (unlikely(data_len > MAX_SKB_FRAGS * (PAGE_SIZE << order)))
6613		return NULL;
6614
6615	*errcode = -ENOBUFS;
6616	skb = alloc_skb(header_len, gfp_mask);
6617	if (!skb)
6618		return NULL;
6619
6620	while (data_len) {
6621		if (nr_frags == MAX_SKB_FRAGS - 1)
6622			goto failure;
6623		while (order && PAGE_ALIGN(data_len) < (PAGE_SIZE << order))
6624			order--;
6625
6626		if (order) {
6627			page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6628					   __GFP_COMP |
6629					   __GFP_NOWARN,
6630					   order);
6631			if (!page) {
6632				order--;
6633				continue;
6634			}
6635		} else {
6636			page = alloc_page(gfp_mask);
6637			if (!page)
6638				goto failure;
6639		}
6640		chunk = min_t(unsigned long, data_len,
6641			      PAGE_SIZE << order);
6642		skb_fill_page_desc(skb, nr_frags, page, 0, chunk);
6643		nr_frags++;
6644		skb->truesize += (PAGE_SIZE << order);
6645		data_len -= chunk;
6646	}
6647	return skb;
6648
6649failure:
6650	kfree_skb(skb);
6651	return NULL;
6652}
6653EXPORT_SYMBOL(alloc_skb_with_frags);
6654
6655/* carve out the first off bytes from skb when off < headlen */
6656static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6657				    const int headlen, gfp_t gfp_mask)
6658{
6659	int i;
6660	unsigned int size = skb_end_offset(skb);
6661	int new_hlen = headlen - off;
6662	u8 *data;
6663
6664	if (skb_pfmemalloc(skb))
6665		gfp_mask |= __GFP_MEMALLOC;
6666
6667	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6668	if (!data)
6669		return -ENOMEM;
6670	size = SKB_WITH_OVERHEAD(size);
6671
6672	/* Copy real data, and all frags */
6673	skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6674	skb->len -= off;
6675
6676	memcpy((struct skb_shared_info *)(data + size),
6677	       skb_shinfo(skb),
6678	       offsetof(struct skb_shared_info,
6679			frags[skb_shinfo(skb)->nr_frags]));
6680	if (skb_cloned(skb)) {
6681		/* drop the old head gracefully */
6682		if (skb_orphan_frags(skb, gfp_mask)) {
6683			skb_kfree_head(data, size);
6684			return -ENOMEM;
6685		}
6686		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6687			skb_frag_ref(skb, i);
6688		if (skb_has_frag_list(skb))
6689			skb_clone_fraglist(skb);
6690		skb_release_data(skb, SKB_CONSUMED);
6691	} else {
6692		/* we can reuse existing recount- all we did was
6693		 * relocate values
6694		 */
6695		skb_free_head(skb);
6696	}
6697
6698	skb->head = data;
6699	skb->data = data;
6700	skb->head_frag = 0;
6701	skb_set_end_offset(skb, size);
6702	skb_set_tail_pointer(skb, skb_headlen(skb));
6703	skb_headers_offset_update(skb, 0);
6704	skb->cloned = 0;
6705	skb->hdr_len = 0;
6706	skb->nohdr = 0;
6707	atomic_set(&skb_shinfo(skb)->dataref, 1);
6708
6709	return 0;
6710}
6711
6712static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6713
6714/* carve out the first eat bytes from skb's frag_list. May recurse into
6715 * pskb_carve()
6716 */
6717static int pskb_carve_frag_list(struct sk_buff *skb,
6718				struct skb_shared_info *shinfo, int eat,
6719				gfp_t gfp_mask)
6720{
6721	struct sk_buff *list = shinfo->frag_list;
6722	struct sk_buff *clone = NULL;
6723	struct sk_buff *insp = NULL;
6724
6725	do {
6726		if (!list) {
6727			pr_err("Not enough bytes to eat. Want %d\n", eat);
6728			return -EFAULT;
6729		}
6730		if (list->len <= eat) {
6731			/* Eaten as whole. */
6732			eat -= list->len;
6733			list = list->next;
6734			insp = list;
6735		} else {
6736			/* Eaten partially. */
6737			if (skb_shared(list)) {
6738				clone = skb_clone(list, gfp_mask);
6739				if (!clone)
6740					return -ENOMEM;
6741				insp = list->next;
6742				list = clone;
6743			} else {
6744				/* This may be pulled without problems. */
6745				insp = list;
6746			}
6747			if (pskb_carve(list, eat, gfp_mask) < 0) {
6748				kfree_skb(clone);
6749				return -ENOMEM;
6750			}
6751			break;
6752		}
6753	} while (eat);
6754
6755	/* Free pulled out fragments. */
6756	while ((list = shinfo->frag_list) != insp) {
6757		shinfo->frag_list = list->next;
6758		consume_skb(list);
6759	}
6760	/* And insert new clone at head. */
6761	if (clone) {
6762		clone->next = list;
6763		shinfo->frag_list = clone;
6764	}
6765	return 0;
6766}
6767
6768/* carve off first len bytes from skb. Split line (off) is in the
6769 * non-linear part of skb
6770 */
6771static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6772				       int pos, gfp_t gfp_mask)
6773{
6774	int i, k = 0;
6775	unsigned int size = skb_end_offset(skb);
6776	u8 *data;
6777	const int nfrags = skb_shinfo(skb)->nr_frags;
6778	struct skb_shared_info *shinfo;
6779
6780	if (skb_pfmemalloc(skb))
6781		gfp_mask |= __GFP_MEMALLOC;
6782
6783	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6784	if (!data)
6785		return -ENOMEM;
6786	size = SKB_WITH_OVERHEAD(size);
6787
6788	memcpy((struct skb_shared_info *)(data + size),
6789	       skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6790	if (skb_orphan_frags(skb, gfp_mask)) {
6791		skb_kfree_head(data, size);
6792		return -ENOMEM;
6793	}
6794	shinfo = (struct skb_shared_info *)(data + size);
6795	for (i = 0; i < nfrags; i++) {
6796		int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6797
6798		if (pos + fsize > off) {
6799			shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6800
6801			if (pos < off) {
6802				/* Split frag.
6803				 * We have two variants in this case:
6804				 * 1. Move all the frag to the second
6805				 *    part, if it is possible. F.e.
6806				 *    this approach is mandatory for TUX,
6807				 *    where splitting is expensive.
6808				 * 2. Split is accurately. We make this.
6809				 */
6810				skb_frag_off_add(&shinfo->frags[0], off - pos);
6811				skb_frag_size_sub(&shinfo->frags[0], off - pos);
6812			}
6813			skb_frag_ref(skb, i);
6814			k++;
6815		}
6816		pos += fsize;
6817	}
6818	shinfo->nr_frags = k;
6819	if (skb_has_frag_list(skb))
6820		skb_clone_fraglist(skb);
6821
6822	/* split line is in frag list */
6823	if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6824		/* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6825		if (skb_has_frag_list(skb))
6826			kfree_skb_list(skb_shinfo(skb)->frag_list);
6827		skb_kfree_head(data, size);
6828		return -ENOMEM;
6829	}
6830	skb_release_data(skb, SKB_CONSUMED);
6831
6832	skb->head = data;
6833	skb->head_frag = 0;
6834	skb->data = data;
6835	skb_set_end_offset(skb, size);
6836	skb_reset_tail_pointer(skb);
6837	skb_headers_offset_update(skb, 0);
6838	skb->cloned   = 0;
6839	skb->hdr_len  = 0;
6840	skb->nohdr    = 0;
6841	skb->len -= off;
6842	skb->data_len = skb->len;
6843	atomic_set(&skb_shinfo(skb)->dataref, 1);
6844	return 0;
6845}
6846
6847/* remove len bytes from the beginning of the skb */
6848static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6849{
6850	int headlen = skb_headlen(skb);
6851
6852	if (len < headlen)
6853		return pskb_carve_inside_header(skb, len, headlen, gfp);
6854	else
6855		return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6856}
6857
6858/* Extract to_copy bytes starting at off from skb, and return this in
6859 * a new skb
6860 */
6861struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6862			     int to_copy, gfp_t gfp)
6863{
6864	struct sk_buff  *clone = skb_clone(skb, gfp);
6865
6866	if (!clone)
6867		return NULL;
6868
6869	if (pskb_carve(clone, off, gfp) < 0 ||
6870	    pskb_trim(clone, to_copy)) {
6871		kfree_skb(clone);
6872		return NULL;
6873	}
6874	return clone;
6875}
6876EXPORT_SYMBOL(pskb_extract);
6877
6878/**
6879 * skb_condense - try to get rid of fragments/frag_list if possible
6880 * @skb: buffer
6881 *
6882 * Can be used to save memory before skb is added to a busy queue.
6883 * If packet has bytes in frags and enough tail room in skb->head,
6884 * pull all of them, so that we can free the frags right now and adjust
6885 * truesize.
6886 * Notes:
6887 *	We do not reallocate skb->head thus can not fail.
6888 *	Caller must re-evaluate skb->truesize if needed.
6889 */
6890void skb_condense(struct sk_buff *skb)
6891{
6892	if (skb->data_len) {
6893		if (skb->data_len > skb->end - skb->tail ||
6894		    skb_cloned(skb) || !skb_frags_readable(skb))
6895			return;
6896
6897		/* Nice, we can free page frag(s) right now */
6898		__pskb_pull_tail(skb, skb->data_len);
6899	}
6900	/* At this point, skb->truesize might be over estimated,
6901	 * because skb had a fragment, and fragments do not tell
6902	 * their truesize.
6903	 * When we pulled its content into skb->head, fragment
6904	 * was freed, but __pskb_pull_tail() could not possibly
6905	 * adjust skb->truesize, not knowing the frag truesize.
6906	 */
6907	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6908}
6909EXPORT_SYMBOL(skb_condense);
6910
6911#ifdef CONFIG_SKB_EXTENSIONS
6912static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6913{
6914	return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6915}
6916
6917/**
6918 * __skb_ext_alloc - allocate a new skb extensions storage
6919 *
6920 * @flags: See kmalloc().
6921 *
6922 * Returns the newly allocated pointer. The pointer can later attached to a
6923 * skb via __skb_ext_set().
6924 * Note: caller must handle the skb_ext as an opaque data.
6925 */
6926struct skb_ext *__skb_ext_alloc(gfp_t flags)
6927{
6928	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6929
6930	if (new) {
6931		memset(new->offset, 0, sizeof(new->offset));
6932		refcount_set(&new->refcnt, 1);
6933	}
6934
6935	return new;
6936}
6937
6938static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6939					 unsigned int old_active)
6940{
6941	struct skb_ext *new;
6942
6943	if (refcount_read(&old->refcnt) == 1)
6944		return old;
6945
6946	new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6947	if (!new)
6948		return NULL;
6949
6950	memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6951	refcount_set(&new->refcnt, 1);
6952
6953#ifdef CONFIG_XFRM
6954	if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6955		struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6956		unsigned int i;
6957
6958		for (i = 0; i < sp->len; i++)
6959			xfrm_state_hold(sp->xvec[i]);
6960	}
6961#endif
6962#ifdef CONFIG_MCTP_FLOWS
6963	if (old_active & (1 << SKB_EXT_MCTP)) {
6964		struct mctp_flow *flow = skb_ext_get_ptr(old, SKB_EXT_MCTP);
6965
6966		if (flow->key)
6967			refcount_inc(&flow->key->refs);
6968	}
6969#endif
6970	__skb_ext_put(old);
6971	return new;
6972}
6973
6974/**
6975 * __skb_ext_set - attach the specified extension storage to this skb
6976 * @skb: buffer
6977 * @id: extension id
6978 * @ext: extension storage previously allocated via __skb_ext_alloc()
6979 *
6980 * Existing extensions, if any, are cleared.
6981 *
6982 * Returns the pointer to the extension.
6983 */
6984void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6985		    struct skb_ext *ext)
6986{
6987	unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6988
6989	skb_ext_put(skb);
6990	newlen = newoff + skb_ext_type_len[id];
6991	ext->chunks = newlen;
6992	ext->offset[id] = newoff;
6993	skb->extensions = ext;
6994	skb->active_extensions = 1 << id;
6995	return skb_ext_get_ptr(ext, id);
6996}
6997
6998/**
6999 * skb_ext_add - allocate space for given extension, COW if needed
7000 * @skb: buffer
7001 * @id: extension to allocate space for
7002 *
7003 * Allocates enough space for the given extension.
7004 * If the extension is already present, a pointer to that extension
7005 * is returned.
7006 *
7007 * If the skb was cloned, COW applies and the returned memory can be
7008 * modified without changing the extension space of clones buffers.
7009 *
7010 * Returns pointer to the extension or NULL on allocation failure.
7011 */
7012void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
7013{
7014	struct skb_ext *new, *old = NULL;
7015	unsigned int newlen, newoff;
7016
7017	if (skb->active_extensions) {
7018		old = skb->extensions;
7019
7020		new = skb_ext_maybe_cow(old, skb->active_extensions);
7021		if (!new)
7022			return NULL;
7023
7024		if (__skb_ext_exist(new, id))
7025			goto set_active;
7026
7027		newoff = new->chunks;
7028	} else {
7029		newoff = SKB_EXT_CHUNKSIZEOF(*new);
7030
7031		new = __skb_ext_alloc(GFP_ATOMIC);
7032		if (!new)
7033			return NULL;
7034	}
7035
7036	newlen = newoff + skb_ext_type_len[id];
7037	new->chunks = newlen;
7038	new->offset[id] = newoff;
7039set_active:
7040	skb->slow_gro = 1;
7041	skb->extensions = new;
7042	skb->active_extensions |= 1 << id;
7043	return skb_ext_get_ptr(new, id);
7044}
7045EXPORT_SYMBOL(skb_ext_add);
7046
7047#ifdef CONFIG_XFRM
7048static void skb_ext_put_sp(struct sec_path *sp)
7049{
7050	unsigned int i;
7051
7052	for (i = 0; i < sp->len; i++)
7053		xfrm_state_put(sp->xvec[i]);
7054}
7055#endif
7056
7057#ifdef CONFIG_MCTP_FLOWS
7058static void skb_ext_put_mctp(struct mctp_flow *flow)
7059{
7060	if (flow->key)
7061		mctp_key_unref(flow->key);
7062}
7063#endif
7064
7065void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
7066{
7067	struct skb_ext *ext = skb->extensions;
7068
7069	skb->active_extensions &= ~(1 << id);
7070	if (skb->active_extensions == 0) {
7071		skb->extensions = NULL;
7072		__skb_ext_put(ext);
7073#ifdef CONFIG_XFRM
7074	} else if (id == SKB_EXT_SEC_PATH &&
7075		   refcount_read(&ext->refcnt) == 1) {
7076		struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
7077
7078		skb_ext_put_sp(sp);
7079		sp->len = 0;
7080#endif
7081	}
7082}
7083EXPORT_SYMBOL(__skb_ext_del);
7084
7085void __skb_ext_put(struct skb_ext *ext)
7086{
7087	/* If this is last clone, nothing can increment
7088	 * it after check passes.  Avoids one atomic op.
7089	 */
7090	if (refcount_read(&ext->refcnt) == 1)
7091		goto free_now;
7092
7093	if (!refcount_dec_and_test(&ext->refcnt))
7094		return;
7095free_now:
7096#ifdef CONFIG_XFRM
7097	if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
7098		skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
7099#endif
7100#ifdef CONFIG_MCTP_FLOWS
7101	if (__skb_ext_exist(ext, SKB_EXT_MCTP))
7102		skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
7103#endif
7104
7105	kmem_cache_free(skbuff_ext_cache, ext);
7106}
7107EXPORT_SYMBOL(__skb_ext_put);
7108#endif /* CONFIG_SKB_EXTENSIONS */
7109
7110static void kfree_skb_napi_cache(struct sk_buff *skb)
7111{
7112	/* if SKB is a clone, don't handle this case */
7113	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
7114		__kfree_skb(skb);
7115		return;
7116	}
7117
7118	local_bh_disable();
7119	__napi_kfree_skb(skb, SKB_CONSUMED);
7120	local_bh_enable();
7121}
7122
7123/**
7124 * skb_attempt_defer_free - queue skb for remote freeing
7125 * @skb: buffer
7126 *
7127 * Put @skb in a per-cpu list, using the cpu which
7128 * allocated the skb/pages to reduce false sharing
7129 * and memory zone spinlock contention.
7130 */
7131void skb_attempt_defer_free(struct sk_buff *skb)
7132{
7133	int cpu = skb->alloc_cpu;
7134	struct softnet_data *sd;
7135	unsigned int defer_max;
7136	bool kick;
7137
7138	if (cpu == raw_smp_processor_id() ||
7139	    WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
7140	    !cpu_online(cpu)) {
7141nodefer:	kfree_skb_napi_cache(skb);
7142		return;
7143	}
7144
7145	DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
7146	DEBUG_NET_WARN_ON_ONCE(skb->destructor);
7147
7148	sd = &per_cpu(softnet_data, cpu);
7149	defer_max = READ_ONCE(net_hotdata.sysctl_skb_defer_max);
7150	if (READ_ONCE(sd->defer_count) >= defer_max)
7151		goto nodefer;
7152
7153	spin_lock_bh(&sd->defer_lock);
7154	/* Send an IPI every time queue reaches half capacity. */
7155	kick = sd->defer_count == (defer_max >> 1);
7156	/* Paired with the READ_ONCE() few lines above */
7157	WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
7158
7159	skb->next = sd->defer_list;
7160	/* Paired with READ_ONCE() in skb_defer_free_flush() */
7161	WRITE_ONCE(sd->defer_list, skb);
7162	spin_unlock_bh(&sd->defer_lock);
7163
7164	/* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
7165	 * if we are unlucky enough (this seems very unlikely).
7166	 */
7167	if (unlikely(kick))
7168		kick_defer_list_purge(sd, cpu);
7169}
7170
7171static void skb_splice_csum_page(struct sk_buff *skb, struct page *page,
7172				 size_t offset, size_t len)
7173{
7174	const char *kaddr;
7175	__wsum csum;
7176
7177	kaddr = kmap_local_page(page);
7178	csum = csum_partial(kaddr + offset, len, 0);
7179	kunmap_local(kaddr);
7180	skb->csum = csum_block_add(skb->csum, csum, skb->len);
7181}
7182
7183/**
7184 * skb_splice_from_iter - Splice (or copy) pages to skbuff
7185 * @skb: The buffer to add pages to
7186 * @iter: Iterator representing the pages to be added
7187 * @maxsize: Maximum amount of pages to be added
7188 * @gfp: Allocation flags
7189 *
7190 * This is a common helper function for supporting MSG_SPLICE_PAGES.  It
7191 * extracts pages from an iterator and adds them to the socket buffer if
7192 * possible, copying them to fragments if not possible (such as if they're slab
7193 * pages).
7194 *
7195 * Returns the amount of data spliced/copied or -EMSGSIZE if there's
7196 * insufficient space in the buffer to transfer anything.
7197 */
7198ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
7199			     ssize_t maxsize, gfp_t gfp)
7200{
7201	size_t frag_limit = READ_ONCE(net_hotdata.sysctl_max_skb_frags);
7202	struct page *pages[8], **ppages = pages;
7203	ssize_t spliced = 0, ret = 0;
7204	unsigned int i;
7205
7206	while (iter->count > 0) {
7207		ssize_t space, nr, len;
7208		size_t off;
7209
7210		ret = -EMSGSIZE;
7211		space = frag_limit - skb_shinfo(skb)->nr_frags;
7212		if (space < 0)
7213			break;
7214
7215		/* We might be able to coalesce without increasing nr_frags */
7216		nr = clamp_t(size_t, space, 1, ARRAY_SIZE(pages));
7217
7218		len = iov_iter_extract_pages(iter, &ppages, maxsize, nr, 0, &off);
7219		if (len <= 0) {
7220			ret = len ?: -EIO;
7221			break;
7222		}
7223
7224		i = 0;
7225		do {
7226			struct page *page = pages[i++];
7227			size_t part = min_t(size_t, PAGE_SIZE - off, len);
7228
7229			ret = -EIO;
7230			if (WARN_ON_ONCE(!sendpage_ok(page)))
7231				goto out;
7232
7233			ret = skb_append_pagefrags(skb, page, off, part,
7234						   frag_limit);
7235			if (ret < 0) {
7236				iov_iter_revert(iter, len);
7237				goto out;
7238			}
7239
7240			if (skb->ip_summed == CHECKSUM_NONE)
7241				skb_splice_csum_page(skb, page, off, part);
7242
7243			off = 0;
7244			spliced += part;
7245			maxsize -= part;
7246			len -= part;
7247		} while (len > 0);
7248
7249		if (maxsize <= 0)
7250			break;
7251	}
7252
7253out:
7254	skb_len_add(skb, spliced);
7255	return spliced ?: ret;
7256}
7257EXPORT_SYMBOL(skb_splice_from_iter);
7258
7259static __always_inline
7260size_t memcpy_from_iter_csum(void *iter_from, size_t progress,
7261			     size_t len, void *to, void *priv2)
7262{
7263	__wsum *csum = priv2;
7264	__wsum next = csum_partial_copy_nocheck(iter_from, to + progress, len);
7265
7266	*csum = csum_block_add(*csum, next, progress);
7267	return 0;
7268}
7269
7270static __always_inline
7271size_t copy_from_user_iter_csum(void __user *iter_from, size_t progress,
7272				size_t len, void *to, void *priv2)
7273{
7274	__wsum next, *csum = priv2;
7275
7276	next = csum_and_copy_from_user(iter_from, to + progress, len);
7277	*csum = csum_block_add(*csum, next, progress);
7278	return next ? 0 : len;
7279}
7280
7281bool csum_and_copy_from_iter_full(void *addr, size_t bytes,
7282				  __wsum *csum, struct iov_iter *i)
7283{
7284	size_t copied;
7285
7286	if (WARN_ON_ONCE(!i->data_source))
7287		return false;
7288	copied = iterate_and_advance2(i, bytes, addr, csum,
7289				      copy_from_user_iter_csum,
7290				      memcpy_from_iter_csum);
7291	if (likely(copied == bytes))
7292		return true;
7293	iov_iter_revert(i, copied);
7294	return false;
7295}
7296EXPORT_SYMBOL(csum_and_copy_from_iter_full);