1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/slab.h>
4 #include <linux/string.h>
5 #include <linux/compiler.h>
6 #include <linux/export.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>
27 #include <linux/uaccess.h>
33 * kfree_const - conditionally free memory
34 * @x: pointer to the memory
36 * Function calls kfree only if @x is not in .rodata section.
38 void kfree_const(const void *x)
40 if (!is_kernel_rodata((unsigned long)x))
43 EXPORT_SYMBOL(kfree_const);
46 * kstrdup - allocate space for and copy an existing string
47 * @s: the string to duplicate
48 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
50 * Return: newly allocated copy of @s or %NULL in case of error
53 char *kstrdup(const char *s, gfp_t gfp)
62 buf = kmalloc_track_caller(len, gfp);
67 EXPORT_SYMBOL(kstrdup);
70 * kstrdup_const - conditionally duplicate an existing const string
71 * @s: the string to duplicate
72 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
74 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
75 * must not be passed to krealloc().
77 * Return: source string if it is in .rodata section otherwise
78 * fallback to kstrdup.
80 const char *kstrdup_const(const char *s, gfp_t gfp)
82 if (is_kernel_rodata((unsigned long)s))
85 return kstrdup(s, gfp);
87 EXPORT_SYMBOL(kstrdup_const);
90 * kstrndup - allocate space for and copy an existing string
91 * @s: the string to duplicate
92 * @max: read at most @max chars from @s
93 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
95 * Note: Use kmemdup_nul() instead if the size is known exactly.
97 * Return: newly allocated copy of @s or %NULL in case of error
99 char *kstrndup(const char *s, size_t max, gfp_t gfp)
107 len = strnlen(s, max);
108 buf = kmalloc_track_caller(len+1, gfp);
115 EXPORT_SYMBOL(kstrndup);
118 * kmemdup - duplicate region of memory
120 * @src: memory region to duplicate
121 * @len: memory region length
122 * @gfp: GFP mask to use
124 * Return: newly allocated copy of @src or %NULL in case of error,
125 * result is physically contiguous. Use kfree() to free.
127 void *kmemdup(const void *src, size_t len, gfp_t gfp)
131 p = kmalloc_track_caller(len, gfp);
136 EXPORT_SYMBOL(kmemdup);
139 * kvmemdup - duplicate region of memory
141 * @src: memory region to duplicate
142 * @len: memory region length
143 * @gfp: GFP mask to use
145 * Return: newly allocated copy of @src or %NULL in case of error,
146 * result may be not physically contiguous. Use kvfree() to free.
148 void *kvmemdup(const void *src, size_t len, gfp_t gfp)
152 p = kvmalloc(len, gfp);
157 EXPORT_SYMBOL(kvmemdup);
160 * kmemdup_nul - Create a NUL-terminated string from unterminated data
161 * @s: The data to stringify
162 * @len: The size of the data
163 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
165 * Return: newly allocated copy of @s with NUL-termination or %NULL in
168 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
175 buf = kmalloc_track_caller(len + 1, gfp);
182 EXPORT_SYMBOL(kmemdup_nul);
185 * memdup_user - duplicate memory region from user space
187 * @src: source address in user space
188 * @len: number of bytes to copy
190 * Return: an ERR_PTR() on failure. Result is physically
191 * contiguous, to be freed by kfree().
193 void *memdup_user(const void __user *src, size_t len)
197 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
199 return ERR_PTR(-ENOMEM);
201 if (copy_from_user(p, src, len)) {
203 return ERR_PTR(-EFAULT);
208 EXPORT_SYMBOL(memdup_user);
211 * vmemdup_user - duplicate memory region from user space
213 * @src: source address in user space
214 * @len: number of bytes to copy
216 * Return: an ERR_PTR() on failure. Result may be not
217 * physically contiguous. Use kvfree() to free.
219 void *vmemdup_user(const void __user *src, size_t len)
223 p = kvmalloc(len, GFP_USER);
225 return ERR_PTR(-ENOMEM);
227 if (copy_from_user(p, src, len)) {
229 return ERR_PTR(-EFAULT);
234 EXPORT_SYMBOL(vmemdup_user);
237 * strndup_user - duplicate an existing string from user space
238 * @s: The string to duplicate
239 * @n: Maximum number of bytes to copy, including the trailing NUL.
241 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
243 char *strndup_user(const char __user *s, long n)
248 length = strnlen_user(s, n);
251 return ERR_PTR(-EFAULT);
254 return ERR_PTR(-EINVAL);
256 p = memdup_user(s, length);
261 p[length - 1] = '\0';
265 EXPORT_SYMBOL(strndup_user);
268 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
270 * @src: source address in user space
271 * @len: number of bytes to copy
273 * Return: an ERR_PTR() on failure.
275 void *memdup_user_nul(const void __user *src, size_t len)
280 * Always use GFP_KERNEL, since copy_from_user() can sleep and
281 * cause pagefault, which makes it pointless to use GFP_NOFS
284 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
286 return ERR_PTR(-ENOMEM);
288 if (copy_from_user(p, src, len)) {
290 return ERR_PTR(-EFAULT);
296 EXPORT_SYMBOL(memdup_user_nul);
298 /* Check if the vma is being used as a stack by this task */
299 int vma_is_stack_for_current(struct vm_area_struct *vma)
301 struct task_struct * __maybe_unused t = current;
303 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
307 * Change backing file, only valid to use during initial VMA setup.
309 void vma_set_file(struct vm_area_struct *vma, struct file *file)
311 /* Changing an anonymous vma with this is illegal */
313 swap(vma->vm_file, file);
316 EXPORT_SYMBOL(vma_set_file);
318 #ifndef STACK_RND_MASK
319 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
322 unsigned long randomize_stack_top(unsigned long stack_top)
324 unsigned long random_variable = 0;
326 if (current->flags & PF_RANDOMIZE) {
327 random_variable = get_random_long();
328 random_variable &= STACK_RND_MASK;
329 random_variable <<= PAGE_SHIFT;
331 #ifdef CONFIG_STACK_GROWSUP
332 return PAGE_ALIGN(stack_top) + random_variable;
334 return PAGE_ALIGN(stack_top) - random_variable;
339 * randomize_page - Generate a random, page aligned address
340 * @start: The smallest acceptable address the caller will take.
341 * @range: The size of the area, starting at @start, within which the
342 * random address must fall.
344 * If @start + @range would overflow, @range is capped.
346 * NOTE: Historical use of randomize_range, which this replaces, presumed that
347 * @start was already page aligned. We now align it regardless.
349 * Return: A page aligned address within [start, start + range). On error,
350 * @start is returned.
352 unsigned long randomize_page(unsigned long start, unsigned long range)
354 if (!PAGE_ALIGNED(start)) {
355 range -= PAGE_ALIGN(start) - start;
356 start = PAGE_ALIGN(start);
359 if (start > ULONG_MAX - range)
360 range = ULONG_MAX - start;
362 range >>= PAGE_SHIFT;
367 return start + (get_random_long() % range << PAGE_SHIFT);
370 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
371 unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
373 /* Is the current task 32bit ? */
374 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
375 return randomize_page(mm->brk, SZ_32M);
377 return randomize_page(mm->brk, SZ_1G);
380 unsigned long arch_mmap_rnd(void)
384 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
385 if (is_compat_task())
386 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
388 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
389 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
391 return rnd << PAGE_SHIFT;
394 static int mmap_is_legacy(struct rlimit *rlim_stack)
396 if (current->personality & ADDR_COMPAT_LAYOUT)
399 if (rlim_stack->rlim_cur == RLIM_INFINITY)
402 return sysctl_legacy_va_layout;
406 * Leave enough space between the mmap area and the stack to honour ulimit in
407 * the face of randomisation.
409 #define MIN_GAP (SZ_128M)
410 #define MAX_GAP (STACK_TOP / 6 * 5)
412 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
414 unsigned long gap = rlim_stack->rlim_cur;
415 unsigned long pad = stack_guard_gap;
417 /* Account for stack randomization if necessary */
418 if (current->flags & PF_RANDOMIZE)
419 pad += (STACK_RND_MASK << PAGE_SHIFT);
421 /* Values close to RLIM_INFINITY can overflow. */
427 else if (gap > MAX_GAP)
430 return PAGE_ALIGN(STACK_TOP - gap - rnd);
433 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
435 unsigned long random_factor = 0UL;
437 if (current->flags & PF_RANDOMIZE)
438 random_factor = arch_mmap_rnd();
440 if (mmap_is_legacy(rlim_stack)) {
441 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
442 mm->get_unmapped_area = arch_get_unmapped_area;
444 mm->mmap_base = mmap_base(random_factor, rlim_stack);
445 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
448 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
449 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
451 mm->mmap_base = TASK_UNMAPPED_BASE;
452 mm->get_unmapped_area = arch_get_unmapped_area;
457 * __account_locked_vm - account locked pages to an mm's locked_vm
458 * @mm: mm to account against
459 * @pages: number of pages to account
460 * @inc: %true if @pages should be considered positive, %false if not
461 * @task: task used to check RLIMIT_MEMLOCK
462 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
464 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
465 * that mmap_lock is held as writer.
469 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
471 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
472 struct task_struct *task, bool bypass_rlim)
474 unsigned long locked_vm, limit;
477 mmap_assert_write_locked(mm);
479 locked_vm = mm->locked_vm;
482 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
483 if (locked_vm + pages > limit)
487 mm->locked_vm = locked_vm + pages;
489 WARN_ON_ONCE(pages > locked_vm);
490 mm->locked_vm = locked_vm - pages;
493 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
494 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
495 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
496 ret ? " - exceeded" : "");
500 EXPORT_SYMBOL_GPL(__account_locked_vm);
503 * account_locked_vm - account locked pages to an mm's locked_vm
504 * @mm: mm to account against, may be NULL
505 * @pages: number of pages to account
506 * @inc: %true if @pages should be considered positive, %false if not
508 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
511 * * 0 on success, or if mm is NULL
512 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
514 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
518 if (pages == 0 || !mm)
522 ret = __account_locked_vm(mm, pages, inc, current,
523 capable(CAP_IPC_LOCK));
524 mmap_write_unlock(mm);
528 EXPORT_SYMBOL_GPL(account_locked_vm);
530 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
531 unsigned long len, unsigned long prot,
532 unsigned long flag, unsigned long pgoff)
535 struct mm_struct *mm = current->mm;
536 unsigned long populate;
539 ret = security_mmap_file(file, prot, flag);
541 if (mmap_write_lock_killable(mm))
543 ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
545 mmap_write_unlock(mm);
546 userfaultfd_unmap_complete(mm, &uf);
548 mm_populate(ret, populate);
553 unsigned long vm_mmap(struct file *file, unsigned long addr,
554 unsigned long len, unsigned long prot,
555 unsigned long flag, unsigned long offset)
557 if (unlikely(offset + PAGE_ALIGN(len) < offset))
559 if (unlikely(offset_in_page(offset)))
562 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
564 EXPORT_SYMBOL(vm_mmap);
567 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
568 * failure, fall back to non-contiguous (vmalloc) allocation.
569 * @size: size of the request.
570 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
571 * @node: numa node to allocate from
573 * Uses kmalloc to get the memory but if the allocation fails then falls back
574 * to the vmalloc allocator. Use kvfree for freeing the memory.
576 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
577 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
578 * preferable to the vmalloc fallback, due to visible performance drawbacks.
580 * Return: pointer to the allocated memory of %NULL in case of failure
582 void *kvmalloc_node(size_t size, gfp_t flags, int node)
584 gfp_t kmalloc_flags = flags;
588 * We want to attempt a large physically contiguous block first because
589 * it is less likely to fragment multiple larger blocks and therefore
590 * contribute to a long term fragmentation less than vmalloc fallback.
591 * However make sure that larger requests are not too disruptive - no
592 * OOM killer and no allocation failure warnings as we have a fallback.
594 if (size > PAGE_SIZE) {
595 kmalloc_flags |= __GFP_NOWARN;
597 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
598 kmalloc_flags |= __GFP_NORETRY;
600 /* nofail semantic is implemented by the vmalloc fallback */
601 kmalloc_flags &= ~__GFP_NOFAIL;
604 ret = kmalloc_node(size, kmalloc_flags, node);
607 * It doesn't really make sense to fallback to vmalloc for sub page
610 if (ret || size <= PAGE_SIZE)
613 /* non-sleeping allocations are not supported by vmalloc */
614 if (!gfpflags_allow_blocking(flags))
617 /* Don't even allow crazy sizes */
618 if (unlikely(size > INT_MAX)) {
619 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
624 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
625 * since the callers already cannot assume anything
626 * about the resulting pointer, and cannot play
629 return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
630 flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
631 node, __builtin_return_address(0));
633 EXPORT_SYMBOL(kvmalloc_node);
636 * kvfree() - Free memory.
637 * @addr: Pointer to allocated memory.
639 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
640 * It is slightly more efficient to use kfree() or vfree() if you are certain
641 * that you know which one to use.
643 * Context: Either preemptible task context or not-NMI interrupt.
645 void kvfree(const void *addr)
647 if (is_vmalloc_addr(addr))
652 EXPORT_SYMBOL(kvfree);
655 * kvfree_sensitive - Free a data object containing sensitive information.
656 * @addr: address of the data object to be freed.
657 * @len: length of the data object.
659 * Use the special memzero_explicit() function to clear the content of a
660 * kvmalloc'ed object containing sensitive data to make sure that the
661 * compiler won't optimize out the data clearing.
663 void kvfree_sensitive(const void *addr, size_t len)
665 if (likely(!ZERO_OR_NULL_PTR(addr))) {
666 memzero_explicit((void *)addr, len);
670 EXPORT_SYMBOL(kvfree_sensitive);
672 void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
676 if (oldsize >= newsize)
678 newp = kvmalloc(newsize, flags);
681 memcpy(newp, p, oldsize);
685 EXPORT_SYMBOL(kvrealloc);
688 * __vmalloc_array - allocate memory for a virtually contiguous array.
689 * @n: number of elements.
690 * @size: element size.
691 * @flags: the type of memory to allocate (see kmalloc).
693 void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
697 if (unlikely(check_mul_overflow(n, size, &bytes)))
699 return __vmalloc(bytes, flags);
701 EXPORT_SYMBOL(__vmalloc_array);
704 * vmalloc_array - allocate memory for a virtually contiguous array.
705 * @n: number of elements.
706 * @size: element size.
708 void *vmalloc_array(size_t n, size_t size)
710 return __vmalloc_array(n, size, GFP_KERNEL);
712 EXPORT_SYMBOL(vmalloc_array);
715 * __vcalloc - allocate and zero memory for a virtually contiguous array.
716 * @n: number of elements.
717 * @size: element size.
718 * @flags: the type of memory to allocate (see kmalloc).
720 void *__vcalloc(size_t n, size_t size, gfp_t flags)
722 return __vmalloc_array(n, size, flags | __GFP_ZERO);
724 EXPORT_SYMBOL(__vcalloc);
727 * vcalloc - allocate and zero memory for a virtually contiguous array.
728 * @n: number of elements.
729 * @size: element size.
731 void *vcalloc(size_t n, size_t size)
733 return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
735 EXPORT_SYMBOL(vcalloc);
737 /* Neutral page->mapping pointer to address_space or anon_vma or other */
738 void *page_rmapping(struct page *page)
740 return folio_raw_mapping(page_folio(page));
743 struct anon_vma *folio_anon_vma(struct folio *folio)
745 unsigned long mapping = (unsigned long)folio->mapping;
747 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
749 return (void *)(mapping - PAGE_MAPPING_ANON);
753 * folio_mapping - Find the mapping where this folio is stored.
756 * For folios which are in the page cache, return the mapping that this
757 * page belongs to. Folios in the swap cache return the swap mapping
758 * this page is stored in (which is different from the mapping for the
759 * swap file or swap device where the data is stored).
761 * You can call this for folios which aren't in the swap cache or page
762 * cache and it will return NULL.
764 struct address_space *folio_mapping(struct folio *folio)
766 struct address_space *mapping;
768 /* This happens if someone calls flush_dcache_page on slab page */
769 if (unlikely(folio_test_slab(folio)))
772 if (unlikely(folio_test_swapcache(folio)))
773 return swap_address_space(folio_swap_entry(folio));
775 mapping = folio->mapping;
776 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
781 EXPORT_SYMBOL(folio_mapping);
784 * folio_copy - Copy the contents of one folio to another.
785 * @dst: Folio to copy to.
786 * @src: Folio to copy from.
788 * The bytes in the folio represented by @src are copied to @dst.
789 * Assumes the caller has validated that @dst is at least as large as @src.
790 * Can be called in atomic context for order-0 folios, but if the folio is
791 * larger, it may sleep.
793 void folio_copy(struct folio *dst, struct folio *src)
796 long nr = folio_nr_pages(src);
799 copy_highpage(folio_page(dst, i), folio_page(src, i));
806 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
807 int sysctl_overcommit_ratio __read_mostly = 50;
808 unsigned long sysctl_overcommit_kbytes __read_mostly;
809 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
810 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
811 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
813 int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
814 size_t *lenp, loff_t *ppos)
818 ret = proc_dointvec(table, write, buffer, lenp, ppos);
819 if (ret == 0 && write)
820 sysctl_overcommit_kbytes = 0;
824 static void sync_overcommit_as(struct work_struct *dummy)
826 percpu_counter_sync(&vm_committed_as);
829 int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
830 size_t *lenp, loff_t *ppos)
837 * The deviation of sync_overcommit_as could be big with loose policy
838 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
839 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
840 * with the strict "NEVER", and to avoid possible race condition (even
841 * though user usually won't too frequently do the switching to policy
842 * OVERCOMMIT_NEVER), the switch is done in the following order:
843 * 1. changing the batch
844 * 2. sync percpu count on each CPU
845 * 3. switch the policy
849 t.data = &new_policy;
850 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
851 if (ret || new_policy == -1)
854 mm_compute_batch(new_policy);
855 if (new_policy == OVERCOMMIT_NEVER)
856 schedule_on_each_cpu(sync_overcommit_as);
857 sysctl_overcommit_memory = new_policy;
859 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
865 int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
866 size_t *lenp, loff_t *ppos)
870 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
871 if (ret == 0 && write)
872 sysctl_overcommit_ratio = 0;
877 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
879 unsigned long vm_commit_limit(void)
881 unsigned long allowed;
883 if (sysctl_overcommit_kbytes)
884 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
886 allowed = ((totalram_pages() - hugetlb_total_pages())
887 * sysctl_overcommit_ratio / 100);
888 allowed += total_swap_pages;
894 * Make sure vm_committed_as in one cacheline and not cacheline shared with
895 * other variables. It can be updated by several CPUs frequently.
897 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
900 * The global memory commitment made in the system can be a metric
901 * that can be used to drive ballooning decisions when Linux is hosted
902 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
903 * balancing memory across competing virtual machines that are hosted.
904 * Several metrics drive this policy engine including the guest reported
907 * The time cost of this is very low for small platforms, and for big
908 * platform like a 2S/36C/72T Skylake server, in worst case where
909 * vm_committed_as's spinlock is under severe contention, the time cost
910 * could be about 30~40 microseconds.
912 unsigned long vm_memory_committed(void)
914 return percpu_counter_sum_positive(&vm_committed_as);
916 EXPORT_SYMBOL_GPL(vm_memory_committed);
919 * Check that a process has enough memory to allocate a new virtual
920 * mapping. 0 means there is enough memory for the allocation to
921 * succeed and -ENOMEM implies there is not.
923 * We currently support three overcommit policies, which are set via the
924 * vm.overcommit_memory sysctl. See Documentation/mm/overcommit-accounting.rst
926 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
927 * Additional code 2002 Jul 20 by Robert Love.
929 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
931 * Note this is a helper function intended to be used by LSMs which
932 * wish to use this logic.
934 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
938 vm_acct_memory(pages);
941 * Sometimes we want to use more memory than we have
943 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
946 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
947 if (pages > totalram_pages() + total_swap_pages)
952 allowed = vm_commit_limit();
954 * Reserve some for root
957 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
960 * Don't let a single process grow so big a user can't recover
963 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
965 allowed -= min_t(long, mm->total_vm / 32, reserve);
968 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
971 pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
972 __func__, current->pid, current->comm);
973 vm_unacct_memory(pages);
979 * get_cmdline() - copy the cmdline value to a buffer.
980 * @task: the task whose cmdline value to copy.
981 * @buffer: the buffer to copy to.
982 * @buflen: the length of the buffer. Larger cmdline values are truncated
985 * Return: the size of the cmdline field copied. Note that the copy does
986 * not guarantee an ending NULL byte.
988 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
992 struct mm_struct *mm = get_task_mm(task);
993 unsigned long arg_start, arg_end, env_start, env_end;
997 goto out_mm; /* Shh! No looking before we're done */
999 spin_lock(&mm->arg_lock);
1000 arg_start = mm->arg_start;
1001 arg_end = mm->arg_end;
1002 env_start = mm->env_start;
1003 env_end = mm->env_end;
1004 spin_unlock(&mm->arg_lock);
1006 len = arg_end - arg_start;
1011 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1014 * If the nul at the end of args has been overwritten, then
1015 * assume application is using setproctitle(3).
1017 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1018 len = strnlen(buffer, res);
1022 len = env_end - env_start;
1023 if (len > buflen - res)
1025 res += access_process_vm(task, env_start,
1028 res = strnlen(buffer, res);
1037 int __weak memcmp_pages(struct page *page1, struct page *page2)
1039 char *addr1, *addr2;
1042 addr1 = kmap_atomic(page1);
1043 addr2 = kmap_atomic(page2);
1044 ret = memcmp(addr1, addr2, PAGE_SIZE);
1045 kunmap_atomic(addr2);
1046 kunmap_atomic(addr1);
1050 #ifdef CONFIG_PRINTK
1052 * mem_dump_obj - Print available provenance information
1053 * @object: object for which to find provenance information.
1055 * This function uses pr_cont(), so that the caller is expected to have
1056 * printed out whatever preamble is appropriate. The provenance information
1057 * depends on the type of object and on how much debugging is enabled.
1058 * For example, for a slab-cache object, the slab name is printed, and,
1059 * if available, the return address and stack trace from the allocation
1060 * and last free path of that object.
1062 void mem_dump_obj(void *object)
1066 if (kmem_valid_obj(object)) {
1067 kmem_dump_obj(object);
1071 if (vmalloc_dump_obj(object))
1074 if (virt_addr_valid(object))
1075 type = "non-slab/vmalloc memory";
1076 else if (object == NULL)
1077 type = "NULL pointer";
1078 else if (object == ZERO_SIZE_PTR)
1079 type = "zero-size pointer";
1081 type = "non-paged memory";
1083 pr_cont(" %s\n", type);
1085 EXPORT_SYMBOL_GPL(mem_dump_obj);
1089 * A driver might set a page logically offline -- PageOffline() -- and
1090 * turn the page inaccessible in the hypervisor; after that, access to page
1091 * content can be fatal.
1093 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1094 * pages after checking PageOffline(); however, these PFN walkers can race
1095 * with drivers that set PageOffline().
1097 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1098 * synchronize with such drivers, achieving that a page cannot be set
1099 * PageOffline() while frozen.
1101 * page_offline_begin()/page_offline_end() is used by drivers that care about
1102 * such races when setting a page PageOffline().
1104 static DECLARE_RWSEM(page_offline_rwsem);
1106 void page_offline_freeze(void)
1108 down_read(&page_offline_rwsem);
1111 void page_offline_thaw(void)
1113 up_read(&page_offline_rwsem);
1116 void page_offline_begin(void)
1118 down_write(&page_offline_rwsem);
1120 EXPORT_SYMBOL(page_offline_begin);
1122 void page_offline_end(void)
1124 up_write(&page_offline_rwsem);
1126 EXPORT_SYMBOL(page_offline_end);
1128 #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_FOLIO
1129 void flush_dcache_folio(struct folio *folio)
1131 long i, nr = folio_nr_pages(folio);
1133 for (i = 0; i < nr; i++)
1134 flush_dcache_page(folio_page(folio, i));
1136 EXPORT_SYMBOL(flush_dcache_folio);