Linux Audio

Check our new training course

Loading...
   1// SPDX-License-Identifier: GPL-2.0-only
   2#include <linux/mm.h>
   3#include <linux/slab.h>
   4#include <linux/string.h>
   5#include <linux/compiler.h>
   6#include <linux/export.h>
   7#include <linux/err.h>
   8#include <linux/sched.h>
   9#include <linux/sched/mm.h>
  10#include <linux/sched/signal.h>
  11#include <linux/sched/task_stack.h>
  12#include <linux/security.h>
  13#include <linux/swap.h>
  14#include <linux/swapops.h>
  15#include <linux/mman.h>
  16#include <linux/hugetlb.h>
  17#include <linux/vmalloc.h>
  18#include <linux/userfaultfd_k.h>
  19#include <linux/elf.h>
  20#include <linux/elf-randomize.h>
  21#include <linux/personality.h>
  22#include <linux/random.h>
  23#include <linux/processor.h>
  24#include <linux/sizes.h>
  25#include <linux/compat.h>
  26
  27#include <linux/uaccess.h>
  28
  29#include <kunit/visibility.h>
  30
  31#include "internal.h"
  32#include "swap.h"
  33
  34/**
  35 * kfree_const - conditionally free memory
  36 * @x: pointer to the memory
  37 *
  38 * Function calls kfree only if @x is not in .rodata section.
  39 */
  40void kfree_const(const void *x)
  41{
  42	if (!is_kernel_rodata((unsigned long)x))
  43		kfree(x);
  44}
  45EXPORT_SYMBOL(kfree_const);
  46
  47/**
  48 * __kmemdup_nul - Create a NUL-terminated string from @s, which might be unterminated.
  49 * @s: The data to copy
  50 * @len: The size of the data, not including the NUL terminator
  51 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
  52 *
  53 * Return: newly allocated copy of @s with NUL-termination or %NULL in
  54 * case of error
  55 */
  56static __always_inline char *__kmemdup_nul(const char *s, size_t len, gfp_t gfp)
  57{
  58	char *buf;
  59
  60	/* '+1' for the NUL terminator */
  61	buf = kmalloc_track_caller(len + 1, gfp);
  62	if (!buf)
  63		return NULL;
  64
  65	memcpy(buf, s, len);
  66	/* Ensure the buf is always NUL-terminated, regardless of @s. */
  67	buf[len] = '\0';
  68	return buf;
  69}
  70
  71/**
  72 * kstrdup - allocate space for and copy an existing string
  73 * @s: the string to duplicate
  74 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
  75 *
  76 * Return: newly allocated copy of @s or %NULL in case of error
  77 */
  78noinline
  79char *kstrdup(const char *s, gfp_t gfp)
  80{
  81	return s ? __kmemdup_nul(s, strlen(s), gfp) : NULL;
  82}
  83EXPORT_SYMBOL(kstrdup);
  84
  85/**
  86 * kstrdup_const - conditionally duplicate an existing const string
  87 * @s: the string to duplicate
  88 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
  89 *
  90 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
  91 * must not be passed to krealloc().
  92 *
  93 * Return: source string if it is in .rodata section otherwise
  94 * fallback to kstrdup.
  95 */
  96const char *kstrdup_const(const char *s, gfp_t gfp)
  97{
  98	if (is_kernel_rodata((unsigned long)s))
  99		return s;
 100
 101	return kstrdup(s, gfp);
 102}
 103EXPORT_SYMBOL(kstrdup_const);
 104
 105/**
 106 * kstrndup - allocate space for and copy an existing string
 107 * @s: the string to duplicate
 108 * @max: read at most @max chars from @s
 109 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 110 *
 111 * Note: Use kmemdup_nul() instead if the size is known exactly.
 112 *
 113 * Return: newly allocated copy of @s or %NULL in case of error
 114 */
 115char *kstrndup(const char *s, size_t max, gfp_t gfp)
 116{
 117	return s ? __kmemdup_nul(s, strnlen(s, max), gfp) : NULL;
 118}
 119EXPORT_SYMBOL(kstrndup);
 120
 121/**
 122 * kmemdup - duplicate region of memory
 123 *
 124 * @src: memory region to duplicate
 125 * @len: memory region length
 126 * @gfp: GFP mask to use
 127 *
 128 * Return: newly allocated copy of @src or %NULL in case of error,
 129 * result is physically contiguous. Use kfree() to free.
 130 */
 131void *kmemdup_noprof(const void *src, size_t len, gfp_t gfp)
 132{
 133	void *p;
 134
 135	p = kmalloc_node_track_caller_noprof(len, gfp, NUMA_NO_NODE, _RET_IP_);
 136	if (p)
 137		memcpy(p, src, len);
 138	return p;
 139}
 140EXPORT_SYMBOL(kmemdup_noprof);
 141
 142/**
 143 * kmemdup_array - duplicate a given array.
 144 *
 145 * @src: array to duplicate.
 146 * @count: number of elements to duplicate from array.
 147 * @element_size: size of each element of array.
 148 * @gfp: GFP mask to use.
 149 *
 150 * Return: duplicated array of @src or %NULL in case of error,
 151 * result is physically contiguous. Use kfree() to free.
 152 */
 153void *kmemdup_array(const void *src, size_t count, size_t element_size, gfp_t gfp)
 154{
 155	return kmemdup(src, size_mul(element_size, count), gfp);
 156}
 157EXPORT_SYMBOL(kmemdup_array);
 158
 159/**
 160 * kvmemdup - duplicate region of memory
 161 *
 162 * @src: memory region to duplicate
 163 * @len: memory region length
 164 * @gfp: GFP mask to use
 165 *
 166 * Return: newly allocated copy of @src or %NULL in case of error,
 167 * result may be not physically contiguous. Use kvfree() to free.
 168 */
 169void *kvmemdup(const void *src, size_t len, gfp_t gfp)
 170{
 171	void *p;
 172
 173	p = kvmalloc(len, gfp);
 174	if (p)
 175		memcpy(p, src, len);
 176	return p;
 177}
 178EXPORT_SYMBOL(kvmemdup);
 179
 180/**
 181 * kmemdup_nul - Create a NUL-terminated string from unterminated data
 182 * @s: The data to stringify
 183 * @len: The size of the data
 184 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
 185 *
 186 * Return: newly allocated copy of @s with NUL-termination or %NULL in
 187 * case of error
 188 */
 189char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
 190{
 191	return s ? __kmemdup_nul(s, len, gfp) : NULL;
 192}
 193EXPORT_SYMBOL(kmemdup_nul);
 194
 195static kmem_buckets *user_buckets __ro_after_init;
 196
 197static int __init init_user_buckets(void)
 198{
 199	user_buckets = kmem_buckets_create("memdup_user", 0, 0, INT_MAX, NULL);
 200
 201	return 0;
 202}
 203subsys_initcall(init_user_buckets);
 204
 205/**
 206 * memdup_user - duplicate memory region from user space
 207 *
 208 * @src: source address in user space
 209 * @len: number of bytes to copy
 210 *
 211 * Return: an ERR_PTR() on failure.  Result is physically
 212 * contiguous, to be freed by kfree().
 213 */
 214void *memdup_user(const void __user *src, size_t len)
 215{
 216	void *p;
 217
 218	p = kmem_buckets_alloc_track_caller(user_buckets, len, GFP_USER | __GFP_NOWARN);
 219	if (!p)
 220		return ERR_PTR(-ENOMEM);
 221
 222	if (copy_from_user(p, src, len)) {
 223		kfree(p);
 224		return ERR_PTR(-EFAULT);
 225	}
 226
 227	return p;
 228}
 229EXPORT_SYMBOL(memdup_user);
 230
 231/**
 232 * vmemdup_user - duplicate memory region from user space
 233 *
 234 * @src: source address in user space
 235 * @len: number of bytes to copy
 236 *
 237 * Return: an ERR_PTR() on failure.  Result may be not
 238 * physically contiguous.  Use kvfree() to free.
 239 */
 240void *vmemdup_user(const void __user *src, size_t len)
 241{
 242	void *p;
 243
 244	p = kmem_buckets_valloc(user_buckets, len, GFP_USER);
 245	if (!p)
 246		return ERR_PTR(-ENOMEM);
 247
 248	if (copy_from_user(p, src, len)) {
 249		kvfree(p);
 250		return ERR_PTR(-EFAULT);
 251	}
 252
 253	return p;
 254}
 255EXPORT_SYMBOL(vmemdup_user);
 256
 257/**
 258 * strndup_user - duplicate an existing string from user space
 259 * @s: The string to duplicate
 260 * @n: Maximum number of bytes to copy, including the trailing NUL.
 261 *
 262 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
 263 */
 264char *strndup_user(const char __user *s, long n)
 265{
 266	char *p;
 267	long length;
 268
 269	length = strnlen_user(s, n);
 270
 271	if (!length)
 272		return ERR_PTR(-EFAULT);
 273
 274	if (length > n)
 275		return ERR_PTR(-EINVAL);
 276
 277	p = memdup_user(s, length);
 278
 279	if (IS_ERR(p))
 280		return p;
 281
 282	p[length - 1] = '\0';
 283
 284	return p;
 285}
 286EXPORT_SYMBOL(strndup_user);
 287
 288/**
 289 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
 290 *
 291 * @src: source address in user space
 292 * @len: number of bytes to copy
 293 *
 294 * Return: an ERR_PTR() on failure.
 295 */
 296void *memdup_user_nul(const void __user *src, size_t len)
 297{
 298	char *p;
 299
 300	p = kmem_buckets_alloc_track_caller(user_buckets, len + 1, GFP_USER | __GFP_NOWARN);
 301	if (!p)
 302		return ERR_PTR(-ENOMEM);
 303
 304	if (copy_from_user(p, src, len)) {
 305		kfree(p);
 306		return ERR_PTR(-EFAULT);
 307	}
 308	p[len] = '\0';
 309
 310	return p;
 311}
 312EXPORT_SYMBOL(memdup_user_nul);
 313
 314/* Check if the vma is being used as a stack by this task */
 315int vma_is_stack_for_current(struct vm_area_struct *vma)
 316{
 317	struct task_struct * __maybe_unused t = current;
 318
 319	return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
 320}
 321
 322/*
 323 * Change backing file, only valid to use during initial VMA setup.
 324 */
 325void vma_set_file(struct vm_area_struct *vma, struct file *file)
 326{
 327	/* Changing an anonymous vma with this is illegal */
 328	get_file(file);
 329	swap(vma->vm_file, file);
 330	fput(file);
 331}
 332EXPORT_SYMBOL(vma_set_file);
 333
 334#ifndef STACK_RND_MASK
 335#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 336#endif
 337
 338unsigned long randomize_stack_top(unsigned long stack_top)
 339{
 340	unsigned long random_variable = 0;
 341
 342	if (current->flags & PF_RANDOMIZE) {
 343		random_variable = get_random_long();
 344		random_variable &= STACK_RND_MASK;
 345		random_variable <<= PAGE_SHIFT;
 346	}
 347#ifdef CONFIG_STACK_GROWSUP
 348	return PAGE_ALIGN(stack_top) + random_variable;
 349#else
 350	return PAGE_ALIGN(stack_top) - random_variable;
 351#endif
 352}
 353
 354/**
 355 * randomize_page - Generate a random, page aligned address
 356 * @start:	The smallest acceptable address the caller will take.
 357 * @range:	The size of the area, starting at @start, within which the
 358 *		random address must fall.
 359 *
 360 * If @start + @range would overflow, @range is capped.
 361 *
 362 * NOTE: Historical use of randomize_range, which this replaces, presumed that
 363 * @start was already page aligned.  We now align it regardless.
 364 *
 365 * Return: A page aligned address within [start, start + range).  On error,
 366 * @start is returned.
 367 */
 368unsigned long randomize_page(unsigned long start, unsigned long range)
 369{
 370	if (!PAGE_ALIGNED(start)) {
 371		range -= PAGE_ALIGN(start) - start;
 372		start = PAGE_ALIGN(start);
 373	}
 374
 375	if (start > ULONG_MAX - range)
 376		range = ULONG_MAX - start;
 377
 378	range >>= PAGE_SHIFT;
 379
 380	if (range == 0)
 381		return start;
 382
 383	return start + (get_random_long() % range << PAGE_SHIFT);
 384}
 385
 386#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
 387unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
 388{
 389	/* Is the current task 32bit ? */
 390	if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
 391		return randomize_page(mm->brk, SZ_32M);
 392
 393	return randomize_page(mm->brk, SZ_1G);
 394}
 395
 396unsigned long arch_mmap_rnd(void)
 397{
 398	unsigned long rnd;
 399
 400#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
 401	if (is_compat_task())
 402		rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
 403	else
 404#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
 405		rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
 406
 407	return rnd << PAGE_SHIFT;
 408}
 409
 410static int mmap_is_legacy(struct rlimit *rlim_stack)
 411{
 412	if (current->personality & ADDR_COMPAT_LAYOUT)
 413		return 1;
 414
 415	/* On parisc the stack always grows up - so a unlimited stack should
 416	 * not be an indicator to use the legacy memory layout. */
 417	if (rlim_stack->rlim_cur == RLIM_INFINITY &&
 418		!IS_ENABLED(CONFIG_STACK_GROWSUP))
 419		return 1;
 420
 421	return sysctl_legacy_va_layout;
 422}
 423
 424/*
 425 * Leave enough space between the mmap area and the stack to honour ulimit in
 426 * the face of randomisation.
 427 */
 428#define MIN_GAP		(SZ_128M)
 429#define MAX_GAP		(STACK_TOP / 6 * 5)
 430
 431static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
 432{
 433#ifdef CONFIG_STACK_GROWSUP
 434	/*
 435	 * For an upwards growing stack the calculation is much simpler.
 436	 * Memory for the maximum stack size is reserved at the top of the
 437	 * task. mmap_base starts directly below the stack and grows
 438	 * downwards.
 439	 */
 440	return PAGE_ALIGN_DOWN(mmap_upper_limit(rlim_stack) - rnd);
 441#else
 442	unsigned long gap = rlim_stack->rlim_cur;
 443	unsigned long pad = stack_guard_gap;
 444
 445	/* Account for stack randomization if necessary */
 446	if (current->flags & PF_RANDOMIZE)
 447		pad += (STACK_RND_MASK << PAGE_SHIFT);
 448
 449	/* Values close to RLIM_INFINITY can overflow. */
 450	if (gap + pad > gap)
 451		gap += pad;
 452
 453	if (gap < MIN_GAP && MIN_GAP < MAX_GAP)
 454		gap = MIN_GAP;
 455	else if (gap > MAX_GAP)
 456		gap = MAX_GAP;
 457
 458	return PAGE_ALIGN(STACK_TOP - gap - rnd);
 459#endif
 460}
 461
 462void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
 463{
 464	unsigned long random_factor = 0UL;
 465
 466	if (current->flags & PF_RANDOMIZE)
 467		random_factor = arch_mmap_rnd();
 468
 469	if (mmap_is_legacy(rlim_stack)) {
 470		mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
 471		clear_bit(MMF_TOPDOWN, &mm->flags);
 472	} else {
 473		mm->mmap_base = mmap_base(random_factor, rlim_stack);
 474		set_bit(MMF_TOPDOWN, &mm->flags);
 475	}
 476}
 477#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
 478void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
 479{
 480	mm->mmap_base = TASK_UNMAPPED_BASE;
 481	clear_bit(MMF_TOPDOWN, &mm->flags);
 482}
 483#endif
 484#ifdef CONFIG_MMU
 485EXPORT_SYMBOL_IF_KUNIT(arch_pick_mmap_layout);
 486#endif
 487
 488/**
 489 * __account_locked_vm - account locked pages to an mm's locked_vm
 490 * @mm:          mm to account against
 491 * @pages:       number of pages to account
 492 * @inc:         %true if @pages should be considered positive, %false if not
 493 * @task:        task used to check RLIMIT_MEMLOCK
 494 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
 495 *
 496 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
 497 * that mmap_lock is held as writer.
 498 *
 499 * Return:
 500 * * 0       on success
 501 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
 502 */
 503int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
 504			struct task_struct *task, bool bypass_rlim)
 505{
 506	unsigned long locked_vm, limit;
 507	int ret = 0;
 508
 509	mmap_assert_write_locked(mm);
 510
 511	locked_vm = mm->locked_vm;
 512	if (inc) {
 513		if (!bypass_rlim) {
 514			limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 515			if (locked_vm + pages > limit)
 516				ret = -ENOMEM;
 517		}
 518		if (!ret)
 519			mm->locked_vm = locked_vm + pages;
 520	} else {
 521		WARN_ON_ONCE(pages > locked_vm);
 522		mm->locked_vm = locked_vm - pages;
 523	}
 524
 525	pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
 526		 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
 527		 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
 528		 ret ? " - exceeded" : "");
 529
 530	return ret;
 531}
 532EXPORT_SYMBOL_GPL(__account_locked_vm);
 533
 534/**
 535 * account_locked_vm - account locked pages to an mm's locked_vm
 536 * @mm:          mm to account against, may be NULL
 537 * @pages:       number of pages to account
 538 * @inc:         %true if @pages should be considered positive, %false if not
 539 *
 540 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
 541 *
 542 * Return:
 543 * * 0       on success, or if mm is NULL
 544 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
 545 */
 546int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
 547{
 548	int ret;
 549
 550	if (pages == 0 || !mm)
 551		return 0;
 552
 553	mmap_write_lock(mm);
 554	ret = __account_locked_vm(mm, pages, inc, current,
 555				  capable(CAP_IPC_LOCK));
 556	mmap_write_unlock(mm);
 557
 558	return ret;
 559}
 560EXPORT_SYMBOL_GPL(account_locked_vm);
 561
 562unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
 563	unsigned long len, unsigned long prot,
 564	unsigned long flag, unsigned long pgoff)
 565{
 566	unsigned long ret;
 567	struct mm_struct *mm = current->mm;
 568	unsigned long populate;
 569	LIST_HEAD(uf);
 570
 571	ret = security_mmap_file(file, prot, flag);
 572	if (!ret) {
 573		if (mmap_write_lock_killable(mm))
 574			return -EINTR;
 575		ret = do_mmap(file, addr, len, prot, flag, 0, pgoff, &populate,
 576			      &uf);
 577		mmap_write_unlock(mm);
 578		userfaultfd_unmap_complete(mm, &uf);
 579		if (populate)
 580			mm_populate(ret, populate);
 581	}
 582	return ret;
 583}
 584
 585unsigned long vm_mmap(struct file *file, unsigned long addr,
 586	unsigned long len, unsigned long prot,
 587	unsigned long flag, unsigned long offset)
 588{
 589	if (unlikely(offset + PAGE_ALIGN(len) < offset))
 590		return -EINVAL;
 591	if (unlikely(offset_in_page(offset)))
 592		return -EINVAL;
 593
 594	return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
 595}
 596EXPORT_SYMBOL(vm_mmap);
 597
 598static gfp_t kmalloc_gfp_adjust(gfp_t flags, size_t size)
 599{
 600	/*
 601	 * We want to attempt a large physically contiguous block first because
 602	 * it is less likely to fragment multiple larger blocks and therefore
 603	 * contribute to a long term fragmentation less than vmalloc fallback.
 604	 * However make sure that larger requests are not too disruptive - no
 605	 * OOM killer and no allocation failure warnings as we have a fallback.
 606	 */
 607	if (size > PAGE_SIZE) {
 608		flags |= __GFP_NOWARN;
 609
 610		if (!(flags & __GFP_RETRY_MAYFAIL))
 611			flags |= __GFP_NORETRY;
 612
 613		/* nofail semantic is implemented by the vmalloc fallback */
 614		flags &= ~__GFP_NOFAIL;
 615	}
 616
 617	return flags;
 618}
 619
 620/**
 621 * __kvmalloc_node - attempt to allocate physically contiguous memory, but upon
 622 * failure, fall back to non-contiguous (vmalloc) allocation.
 623 * @size: size of the request.
 624 * @b: which set of kmalloc buckets to allocate from.
 625 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
 626 * @node: numa node to allocate from
 627 *
 628 * Uses kmalloc to get the memory but if the allocation fails then falls back
 629 * to the vmalloc allocator. Use kvfree for freeing the memory.
 630 *
 631 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
 632 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
 633 * preferable to the vmalloc fallback, due to visible performance drawbacks.
 634 *
 635 * Return: pointer to the allocated memory of %NULL in case of failure
 636 */
 637void *__kvmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node)
 638{
 639	void *ret;
 640
 641	/*
 642	 * It doesn't really make sense to fallback to vmalloc for sub page
 643	 * requests
 644	 */
 645	ret = __kmalloc_node_noprof(PASS_BUCKET_PARAMS(size, b),
 646				    kmalloc_gfp_adjust(flags, size),
 647				    node);
 648	if (ret || size <= PAGE_SIZE)
 649		return ret;
 650
 651	/* non-sleeping allocations are not supported by vmalloc */
 652	if (!gfpflags_allow_blocking(flags))
 653		return NULL;
 654
 655	/* Don't even allow crazy sizes */
 656	if (unlikely(size > INT_MAX)) {
 657		WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 658		return NULL;
 659	}
 660
 661	/*
 662	 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
 663	 * since the callers already cannot assume anything
 664	 * about the resulting pointer, and cannot play
 665	 * protection games.
 666	 */
 667	return __vmalloc_node_range_noprof(size, 1, VMALLOC_START, VMALLOC_END,
 668			flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
 669			node, __builtin_return_address(0));
 670}
 671EXPORT_SYMBOL(__kvmalloc_node_noprof);
 672
 673/**
 674 * kvfree() - Free memory.
 675 * @addr: Pointer to allocated memory.
 676 *
 677 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
 678 * It is slightly more efficient to use kfree() or vfree() if you are certain
 679 * that you know which one to use.
 680 *
 681 * Context: Either preemptible task context or not-NMI interrupt.
 682 */
 683void kvfree(const void *addr)
 684{
 685	if (is_vmalloc_addr(addr))
 686		vfree(addr);
 687	else
 688		kfree(addr);
 689}
 690EXPORT_SYMBOL(kvfree);
 691
 692/**
 693 * kvfree_sensitive - Free a data object containing sensitive information.
 694 * @addr: address of the data object to be freed.
 695 * @len: length of the data object.
 696 *
 697 * Use the special memzero_explicit() function to clear the content of a
 698 * kvmalloc'ed object containing sensitive data to make sure that the
 699 * compiler won't optimize out the data clearing.
 700 */
 701void kvfree_sensitive(const void *addr, size_t len)
 702{
 703	if (likely(!ZERO_OR_NULL_PTR(addr))) {
 704		memzero_explicit((void *)addr, len);
 705		kvfree(addr);
 706	}
 707}
 708EXPORT_SYMBOL(kvfree_sensitive);
 709
 710/**
 711 * kvrealloc - reallocate memory; contents remain unchanged
 712 * @p: object to reallocate memory for
 713 * @size: the size to reallocate
 714 * @flags: the flags for the page level allocator
 715 *
 716 * If @p is %NULL, kvrealloc() behaves exactly like kvmalloc(). If @size is 0
 717 * and @p is not a %NULL pointer, the object pointed to is freed.
 718 *
 719 * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
 720 * initial memory allocation, every subsequent call to this API for the same
 721 * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
 722 * __GFP_ZERO is not fully honored by this API.
 723 *
 724 * In any case, the contents of the object pointed to are preserved up to the
 725 * lesser of the new and old sizes.
 726 *
 727 * This function must not be called concurrently with itself or kvfree() for the
 728 * same memory allocation.
 729 *
 730 * Return: pointer to the allocated memory or %NULL in case of error
 731 */
 732void *kvrealloc_noprof(const void *p, size_t size, gfp_t flags)
 733{
 734	void *n;
 735
 736	if (is_vmalloc_addr(p))
 737		return vrealloc_noprof(p, size, flags);
 738
 739	n = krealloc_noprof(p, size, kmalloc_gfp_adjust(flags, size));
 740	if (!n) {
 741		/* We failed to krealloc(), fall back to kvmalloc(). */
 742		n = kvmalloc_noprof(size, flags);
 743		if (!n)
 744			return NULL;
 745
 746		if (p) {
 747			/* We already know that `p` is not a vmalloc address. */
 748			kasan_disable_current();
 749			memcpy(n, kasan_reset_tag(p), ksize(p));
 750			kasan_enable_current();
 751
 752			kfree(p);
 753		}
 754	}
 755
 756	return n;
 757}
 758EXPORT_SYMBOL(kvrealloc_noprof);
 759
 760/**
 761 * __vmalloc_array - allocate memory for a virtually contiguous array.
 762 * @n: number of elements.
 763 * @size: element size.
 764 * @flags: the type of memory to allocate (see kmalloc).
 765 */
 766void *__vmalloc_array_noprof(size_t n, size_t size, gfp_t flags)
 767{
 768	size_t bytes;
 769
 770	if (unlikely(check_mul_overflow(n, size, &bytes)))
 771		return NULL;
 772	return __vmalloc_noprof(bytes, flags);
 773}
 774EXPORT_SYMBOL(__vmalloc_array_noprof);
 775
 776/**
 777 * vmalloc_array - allocate memory for a virtually contiguous array.
 778 * @n: number of elements.
 779 * @size: element size.
 780 */
 781void *vmalloc_array_noprof(size_t n, size_t size)
 782{
 783	return __vmalloc_array_noprof(n, size, GFP_KERNEL);
 784}
 785EXPORT_SYMBOL(vmalloc_array_noprof);
 786
 787/**
 788 * __vcalloc - allocate and zero memory for a virtually contiguous array.
 789 * @n: number of elements.
 790 * @size: element size.
 791 * @flags: the type of memory to allocate (see kmalloc).
 792 */
 793void *__vcalloc_noprof(size_t n, size_t size, gfp_t flags)
 794{
 795	return __vmalloc_array_noprof(n, size, flags | __GFP_ZERO);
 796}
 797EXPORT_SYMBOL(__vcalloc_noprof);
 798
 799/**
 800 * vcalloc - allocate and zero memory for a virtually contiguous array.
 801 * @n: number of elements.
 802 * @size: element size.
 803 */
 804void *vcalloc_noprof(size_t n, size_t size)
 805{
 806	return __vmalloc_array_noprof(n, size, GFP_KERNEL | __GFP_ZERO);
 807}
 808EXPORT_SYMBOL(vcalloc_noprof);
 809
 810struct anon_vma *folio_anon_vma(const struct folio *folio)
 811{
 812	unsigned long mapping = (unsigned long)folio->mapping;
 813
 814	if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 815		return NULL;
 816	return (void *)(mapping - PAGE_MAPPING_ANON);
 817}
 818
 819/**
 820 * folio_mapping - Find the mapping where this folio is stored.
 821 * @folio: The folio.
 822 *
 823 * For folios which are in the page cache, return the mapping that this
 824 * page belongs to.  Folios in the swap cache return the swap mapping
 825 * this page is stored in (which is different from the mapping for the
 826 * swap file or swap device where the data is stored).
 827 *
 828 * You can call this for folios which aren't in the swap cache or page
 829 * cache and it will return NULL.
 830 */
 831struct address_space *folio_mapping(struct folio *folio)
 832{
 833	struct address_space *mapping;
 834
 835	/* This happens if someone calls flush_dcache_page on slab page */
 836	if (unlikely(folio_test_slab(folio)))
 837		return NULL;
 838
 839	if (unlikely(folio_test_swapcache(folio)))
 840		return swap_address_space(folio->swap);
 841
 842	mapping = folio->mapping;
 843	if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
 844		return NULL;
 845
 846	return mapping;
 847}
 848EXPORT_SYMBOL(folio_mapping);
 849
 850/**
 851 * folio_copy - Copy the contents of one folio to another.
 852 * @dst: Folio to copy to.
 853 * @src: Folio to copy from.
 854 *
 855 * The bytes in the folio represented by @src are copied to @dst.
 856 * Assumes the caller has validated that @dst is at least as large as @src.
 857 * Can be called in atomic context for order-0 folios, but if the folio is
 858 * larger, it may sleep.
 859 */
 860void folio_copy(struct folio *dst, struct folio *src)
 861{
 862	long i = 0;
 863	long nr = folio_nr_pages(src);
 864
 865	for (;;) {
 866		copy_highpage(folio_page(dst, i), folio_page(src, i));
 867		if (++i == nr)
 868			break;
 869		cond_resched();
 870	}
 871}
 872EXPORT_SYMBOL(folio_copy);
 873
 874int folio_mc_copy(struct folio *dst, struct folio *src)
 875{
 876	long nr = folio_nr_pages(src);
 877	long i = 0;
 878
 879	for (;;) {
 880		if (copy_mc_highpage(folio_page(dst, i), folio_page(src, i)))
 881			return -EHWPOISON;
 882		if (++i == nr)
 883			break;
 884		cond_resched();
 885	}
 886
 887	return 0;
 888}
 889EXPORT_SYMBOL(folio_mc_copy);
 890
 891int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
 892int sysctl_overcommit_ratio __read_mostly = 50;
 893unsigned long sysctl_overcommit_kbytes __read_mostly;
 894int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
 895unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
 896unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
 897
 898int overcommit_ratio_handler(const struct ctl_table *table, int write, void *buffer,
 899		size_t *lenp, loff_t *ppos)
 900{
 901	int ret;
 902
 903	ret = proc_dointvec(table, write, buffer, lenp, ppos);
 904	if (ret == 0 && write)
 905		sysctl_overcommit_kbytes = 0;
 906	return ret;
 907}
 908
 909static void sync_overcommit_as(struct work_struct *dummy)
 910{
 911	percpu_counter_sync(&vm_committed_as);
 912}
 913
 914int overcommit_policy_handler(const struct ctl_table *table, int write, void *buffer,
 915		size_t *lenp, loff_t *ppos)
 916{
 917	struct ctl_table t;
 918	int new_policy = -1;
 919	int ret;
 920
 921	/*
 922	 * The deviation of sync_overcommit_as could be big with loose policy
 923	 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
 924	 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
 925	 * with the strict "NEVER", and to avoid possible race condition (even
 926	 * though user usually won't too frequently do the switching to policy
 927	 * OVERCOMMIT_NEVER), the switch is done in the following order:
 928	 *	1. changing the batch
 929	 *	2. sync percpu count on each CPU
 930	 *	3. switch the policy
 931	 */
 932	if (write) {
 933		t = *table;
 934		t.data = &new_policy;
 935		ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
 936		if (ret || new_policy == -1)
 937			return ret;
 938
 939		mm_compute_batch(new_policy);
 940		if (new_policy == OVERCOMMIT_NEVER)
 941			schedule_on_each_cpu(sync_overcommit_as);
 942		sysctl_overcommit_memory = new_policy;
 943	} else {
 944		ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 945	}
 946
 947	return ret;
 948}
 949
 950int overcommit_kbytes_handler(const struct ctl_table *table, int write, void *buffer,
 951		size_t *lenp, loff_t *ppos)
 952{
 953	int ret;
 954
 955	ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 956	if (ret == 0 && write)
 957		sysctl_overcommit_ratio = 0;
 958	return ret;
 959}
 960
 961/*
 962 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
 963 */
 964unsigned long vm_commit_limit(void)
 965{
 966	unsigned long allowed;
 967
 968	if (sysctl_overcommit_kbytes)
 969		allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
 970	else
 971		allowed = ((totalram_pages() - hugetlb_total_pages())
 972			   * sysctl_overcommit_ratio / 100);
 973	allowed += total_swap_pages;
 974
 975	return allowed;
 976}
 977
 978/*
 979 * Make sure vm_committed_as in one cacheline and not cacheline shared with
 980 * other variables. It can be updated by several CPUs frequently.
 981 */
 982struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
 983
 984/*
 985 * The global memory commitment made in the system can be a metric
 986 * that can be used to drive ballooning decisions when Linux is hosted
 987 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
 988 * balancing memory across competing virtual machines that are hosted.
 989 * Several metrics drive this policy engine including the guest reported
 990 * memory commitment.
 991 *
 992 * The time cost of this is very low for small platforms, and for big
 993 * platform like a 2S/36C/72T Skylake server, in worst case where
 994 * vm_committed_as's spinlock is under severe contention, the time cost
 995 * could be about 30~40 microseconds.
 996 */
 997unsigned long vm_memory_committed(void)
 998{
 999	return percpu_counter_sum_positive(&vm_committed_as);
1000}
1001EXPORT_SYMBOL_GPL(vm_memory_committed);
1002
1003/*
1004 * Check that a process has enough memory to allocate a new virtual
1005 * mapping. 0 means there is enough memory for the allocation to
1006 * succeed and -ENOMEM implies there is not.
1007 *
1008 * We currently support three overcommit policies, which are set via the
1009 * vm.overcommit_memory sysctl.  See Documentation/mm/overcommit-accounting.rst
1010 *
1011 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1012 * Additional code 2002 Jul 20 by Robert Love.
1013 *
1014 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1015 *
1016 * Note this is a helper function intended to be used by LSMs which
1017 * wish to use this logic.
1018 */
1019int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1020{
1021	long allowed;
1022	unsigned long bytes_failed;
1023
1024	vm_acct_memory(pages);
1025
1026	/*
1027	 * Sometimes we want to use more memory than we have
1028	 */
1029	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1030		return 0;
1031
1032	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1033		if (pages > totalram_pages() + total_swap_pages)
1034			goto error;
1035		return 0;
1036	}
1037
1038	allowed = vm_commit_limit();
1039	/*
1040	 * Reserve some for root
1041	 */
1042	if (!cap_sys_admin)
1043		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1044
1045	/*
1046	 * Don't let a single process grow so big a user can't recover
1047	 */
1048	if (mm) {
1049		long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1050
1051		allowed -= min_t(long, mm->total_vm / 32, reserve);
1052	}
1053
1054	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1055		return 0;
1056error:
1057	bytes_failed = pages << PAGE_SHIFT;
1058	pr_warn_ratelimited("%s: pid: %d, comm: %s, bytes: %lu not enough memory for the allocation\n",
1059			    __func__, current->pid, current->comm, bytes_failed);
1060	vm_unacct_memory(pages);
1061
1062	return -ENOMEM;
1063}
1064
1065/**
1066 * get_cmdline() - copy the cmdline value to a buffer.
1067 * @task:     the task whose cmdline value to copy.
1068 * @buffer:   the buffer to copy to.
1069 * @buflen:   the length of the buffer. Larger cmdline values are truncated
1070 *            to this length.
1071 *
1072 * Return: the size of the cmdline field copied. Note that the copy does
1073 * not guarantee an ending NULL byte.
1074 */
1075int get_cmdline(struct task_struct *task, char *buffer, int buflen)
1076{
1077	int res = 0;
1078	unsigned int len;
1079	struct mm_struct *mm = get_task_mm(task);
1080	unsigned long arg_start, arg_end, env_start, env_end;
1081	if (!mm)
1082		goto out;
1083	if (!mm->arg_end)
1084		goto out_mm;	/* Shh! No looking before we're done */
1085
1086	spin_lock(&mm->arg_lock);
1087	arg_start = mm->arg_start;
1088	arg_end = mm->arg_end;
1089	env_start = mm->env_start;
1090	env_end = mm->env_end;
1091	spin_unlock(&mm->arg_lock);
1092
1093	len = arg_end - arg_start;
1094
1095	if (len > buflen)
1096		len = buflen;
1097
1098	res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1099
1100	/*
1101	 * If the nul at the end of args has been overwritten, then
1102	 * assume application is using setproctitle(3).
1103	 */
1104	if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1105		len = strnlen(buffer, res);
1106		if (len < res) {
1107			res = len;
1108		} else {
1109			len = env_end - env_start;
1110			if (len > buflen - res)
1111				len = buflen - res;
1112			res += access_process_vm(task, env_start,
1113						 buffer+res, len,
1114						 FOLL_FORCE);
1115			res = strnlen(buffer, res);
1116		}
1117	}
1118out_mm:
1119	mmput(mm);
1120out:
1121	return res;
1122}
1123
1124int __weak memcmp_pages(struct page *page1, struct page *page2)
1125{
1126	char *addr1, *addr2;
1127	int ret;
1128
1129	addr1 = kmap_local_page(page1);
1130	addr2 = kmap_local_page(page2);
1131	ret = memcmp(addr1, addr2, PAGE_SIZE);
1132	kunmap_local(addr2);
1133	kunmap_local(addr1);
1134	return ret;
1135}
1136
1137#ifdef CONFIG_PRINTK
1138/**
1139 * mem_dump_obj - Print available provenance information
1140 * @object: object for which to find provenance information.
1141 *
1142 * This function uses pr_cont(), so that the caller is expected to have
1143 * printed out whatever preamble is appropriate.  The provenance information
1144 * depends on the type of object and on how much debugging is enabled.
1145 * For example, for a slab-cache object, the slab name is printed, and,
1146 * if available, the return address and stack trace from the allocation
1147 * and last free path of that object.
1148 */
1149void mem_dump_obj(void *object)
1150{
1151	const char *type;
1152
1153	if (kmem_dump_obj(object))
1154		return;
1155
1156	if (vmalloc_dump_obj(object))
1157		return;
1158
1159	if (is_vmalloc_addr(object))
1160		type = "vmalloc memory";
1161	else if (virt_addr_valid(object))
1162		type = "non-slab/vmalloc memory";
1163	else if (object == NULL)
1164		type = "NULL pointer";
1165	else if (object == ZERO_SIZE_PTR)
1166		type = "zero-size pointer";
1167	else
1168		type = "non-paged memory";
1169
1170	pr_cont(" %s\n", type);
1171}
1172EXPORT_SYMBOL_GPL(mem_dump_obj);
1173#endif
1174
1175/*
1176 * A driver might set a page logically offline -- PageOffline() -- and
1177 * turn the page inaccessible in the hypervisor; after that, access to page
1178 * content can be fatal.
1179 *
1180 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1181 * pages after checking PageOffline(); however, these PFN walkers can race
1182 * with drivers that set PageOffline().
1183 *
1184 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1185 * synchronize with such drivers, achieving that a page cannot be set
1186 * PageOffline() while frozen.
1187 *
1188 * page_offline_begin()/page_offline_end() is used by drivers that care about
1189 * such races when setting a page PageOffline().
1190 */
1191static DECLARE_RWSEM(page_offline_rwsem);
1192
1193void page_offline_freeze(void)
1194{
1195	down_read(&page_offline_rwsem);
1196}
1197
1198void page_offline_thaw(void)
1199{
1200	up_read(&page_offline_rwsem);
1201}
1202
1203void page_offline_begin(void)
1204{
1205	down_write(&page_offline_rwsem);
1206}
1207EXPORT_SYMBOL(page_offline_begin);
1208
1209void page_offline_end(void)
1210{
1211	up_write(&page_offline_rwsem);
1212}
1213EXPORT_SYMBOL(page_offline_end);
1214
1215#ifndef flush_dcache_folio
1216void flush_dcache_folio(struct folio *folio)
1217{
1218	long i, nr = folio_nr_pages(folio);
1219
1220	for (i = 0; i < nr; i++)
1221		flush_dcache_page(folio_page(folio, i));
1222}
1223EXPORT_SYMBOL(flush_dcache_folio);
1224#endif