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>
32 * kfree_const - conditionally free memory
33 * @x: pointer to the memory
35 * Function calls kfree only if @x is not in .rodata section.
37 void kfree_const(const void *x)
39 if (!is_kernel_rodata((unsigned long)x))
42 EXPORT_SYMBOL(kfree_const);
45 * kstrdup - allocate space for and copy an existing string
46 * @s: the string to duplicate
47 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
49 * Return: newly allocated copy of @s or %NULL in case of error
51 char *kstrdup(const char *s, gfp_t gfp)
60 buf = kmalloc_track_caller(len, gfp);
65 EXPORT_SYMBOL(kstrdup);
68 * kstrdup_const - conditionally duplicate an existing const string
69 * @s: the string to duplicate
70 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
72 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
73 * must not be passed to krealloc().
75 * Return: source string if it is in .rodata section otherwise
76 * fallback to kstrdup.
78 const char *kstrdup_const(const char *s, gfp_t gfp)
80 if (is_kernel_rodata((unsigned long)s))
83 return kstrdup(s, gfp);
85 EXPORT_SYMBOL(kstrdup_const);
88 * kstrndup - allocate space for and copy an existing string
89 * @s: the string to duplicate
90 * @max: read at most @max chars from @s
91 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
93 * Note: Use kmemdup_nul() instead if the size is known exactly.
95 * Return: newly allocated copy of @s or %NULL in case of error
97 char *kstrndup(const char *s, size_t max, gfp_t gfp)
105 len = strnlen(s, max);
106 buf = kmalloc_track_caller(len+1, gfp);
113 EXPORT_SYMBOL(kstrndup);
116 * kmemdup - duplicate region of memory
118 * @src: memory region to duplicate
119 * @len: memory region length
120 * @gfp: GFP mask to use
122 * Return: newly allocated copy of @src or %NULL in case of error
124 void *kmemdup(const void *src, size_t len, gfp_t gfp)
128 p = kmalloc_track_caller(len, gfp);
133 EXPORT_SYMBOL(kmemdup);
136 * kmemdup_nul - Create a NUL-terminated string from unterminated data
137 * @s: The data to stringify
138 * @len: The size of the data
139 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
141 * Return: newly allocated copy of @s with NUL-termination or %NULL in
144 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
151 buf = kmalloc_track_caller(len + 1, gfp);
158 EXPORT_SYMBOL(kmemdup_nul);
161 * memdup_user - duplicate memory region from user space
163 * @src: source address in user space
164 * @len: number of bytes to copy
166 * Return: an ERR_PTR() on failure. Result is physically
167 * contiguous, to be freed by kfree().
169 void *memdup_user(const void __user *src, size_t len)
173 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
175 return ERR_PTR(-ENOMEM);
177 if (copy_from_user(p, src, len)) {
179 return ERR_PTR(-EFAULT);
184 EXPORT_SYMBOL(memdup_user);
187 * vmemdup_user - duplicate memory region from user space
189 * @src: source address in user space
190 * @len: number of bytes to copy
192 * Return: an ERR_PTR() on failure. Result may be not
193 * physically contiguous. Use kvfree() to free.
195 void *vmemdup_user(const void __user *src, size_t len)
199 p = kvmalloc(len, GFP_USER);
201 return ERR_PTR(-ENOMEM);
203 if (copy_from_user(p, src, len)) {
205 return ERR_PTR(-EFAULT);
210 EXPORT_SYMBOL(vmemdup_user);
213 * strndup_user - duplicate an existing string from user space
214 * @s: The string to duplicate
215 * @n: Maximum number of bytes to copy, including the trailing NUL.
217 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
219 char *strndup_user(const char __user *s, long n)
224 length = strnlen_user(s, n);
227 return ERR_PTR(-EFAULT);
230 return ERR_PTR(-EINVAL);
232 p = memdup_user(s, length);
237 p[length - 1] = '\0';
241 EXPORT_SYMBOL(strndup_user);
244 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
246 * @src: source address in user space
247 * @len: number of bytes to copy
249 * Return: an ERR_PTR() on failure.
251 void *memdup_user_nul(const void __user *src, size_t len)
256 * Always use GFP_KERNEL, since copy_from_user() can sleep and
257 * cause pagefault, which makes it pointless to use GFP_NOFS
260 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
262 return ERR_PTR(-ENOMEM);
264 if (copy_from_user(p, src, len)) {
266 return ERR_PTR(-EFAULT);
272 EXPORT_SYMBOL(memdup_user_nul);
274 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
275 struct vm_area_struct *prev)
277 struct vm_area_struct *next;
281 next = prev->vm_next;
292 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma)
294 struct vm_area_struct *prev, *next;
299 prev->vm_next = next;
303 next->vm_prev = prev;
306 /* Check if the vma is being used as a stack by this task */
307 int vma_is_stack_for_current(struct vm_area_struct *vma)
309 struct task_struct * __maybe_unused t = current;
311 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
315 * Change backing file, only valid to use during initial VMA setup.
317 void vma_set_file(struct vm_area_struct *vma, struct file *file)
319 /* Changing an anonymous vma with this is illegal */
321 swap(vma->vm_file, file);
324 EXPORT_SYMBOL(vma_set_file);
326 #ifndef STACK_RND_MASK
327 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
330 unsigned long randomize_stack_top(unsigned long stack_top)
332 unsigned long random_variable = 0;
334 if (current->flags & PF_RANDOMIZE) {
335 random_variable = get_random_long();
336 random_variable &= STACK_RND_MASK;
337 random_variable <<= PAGE_SHIFT;
339 #ifdef CONFIG_STACK_GROWSUP
340 return PAGE_ALIGN(stack_top) + random_variable;
342 return PAGE_ALIGN(stack_top) - random_variable;
347 * randomize_page - Generate a random, page aligned address
348 * @start: The smallest acceptable address the caller will take.
349 * @range: The size of the area, starting at @start, within which the
350 * random address must fall.
352 * If @start + @range would overflow, @range is capped.
354 * NOTE: Historical use of randomize_range, which this replaces, presumed that
355 * @start was already page aligned. We now align it regardless.
357 * Return: A page aligned address within [start, start + range). On error,
358 * @start is returned.
360 unsigned long randomize_page(unsigned long start, unsigned long range)
362 if (!PAGE_ALIGNED(start)) {
363 range -= PAGE_ALIGN(start) - start;
364 start = PAGE_ALIGN(start);
367 if (start > ULONG_MAX - range)
368 range = ULONG_MAX - start;
370 range >>= PAGE_SHIFT;
375 return start + (get_random_long() % range << PAGE_SHIFT);
378 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
379 unsigned long arch_randomize_brk(struct mm_struct *mm)
381 /* Is the current task 32bit ? */
382 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
383 return randomize_page(mm->brk, SZ_32M);
385 return randomize_page(mm->brk, SZ_1G);
388 unsigned long arch_mmap_rnd(void)
392 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
393 if (is_compat_task())
394 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
396 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
397 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
399 return rnd << PAGE_SHIFT;
402 static int mmap_is_legacy(struct rlimit *rlim_stack)
404 if (current->personality & ADDR_COMPAT_LAYOUT)
407 if (rlim_stack->rlim_cur == RLIM_INFINITY)
410 return sysctl_legacy_va_layout;
414 * Leave enough space between the mmap area and the stack to honour ulimit in
415 * the face of randomisation.
417 #define MIN_GAP (SZ_128M)
418 #define MAX_GAP (STACK_TOP / 6 * 5)
420 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
422 unsigned long gap = rlim_stack->rlim_cur;
423 unsigned long pad = stack_guard_gap;
425 /* Account for stack randomization if necessary */
426 if (current->flags & PF_RANDOMIZE)
427 pad += (STACK_RND_MASK << PAGE_SHIFT);
429 /* Values close to RLIM_INFINITY can overflow. */
435 else if (gap > MAX_GAP)
438 return PAGE_ALIGN(STACK_TOP - gap - rnd);
441 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
443 unsigned long random_factor = 0UL;
445 if (current->flags & PF_RANDOMIZE)
446 random_factor = arch_mmap_rnd();
448 if (mmap_is_legacy(rlim_stack)) {
449 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
450 mm->get_unmapped_area = arch_get_unmapped_area;
452 mm->mmap_base = mmap_base(random_factor, rlim_stack);
453 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
456 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
457 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
459 mm->mmap_base = TASK_UNMAPPED_BASE;
460 mm->get_unmapped_area = arch_get_unmapped_area;
465 * __account_locked_vm - account locked pages to an mm's locked_vm
466 * @mm: mm to account against
467 * @pages: number of pages to account
468 * @inc: %true if @pages should be considered positive, %false if not
469 * @task: task used to check RLIMIT_MEMLOCK
470 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
472 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
473 * that mmap_lock is held as writer.
477 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
479 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
480 struct task_struct *task, bool bypass_rlim)
482 unsigned long locked_vm, limit;
485 mmap_assert_write_locked(mm);
487 locked_vm = mm->locked_vm;
490 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
491 if (locked_vm + pages > limit)
495 mm->locked_vm = locked_vm + pages;
497 WARN_ON_ONCE(pages > locked_vm);
498 mm->locked_vm = locked_vm - pages;
501 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
502 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
503 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
504 ret ? " - exceeded" : "");
508 EXPORT_SYMBOL_GPL(__account_locked_vm);
511 * account_locked_vm - account locked pages to an mm's locked_vm
512 * @mm: mm to account against, may be NULL
513 * @pages: number of pages to account
514 * @inc: %true if @pages should be considered positive, %false if not
516 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
519 * * 0 on success, or if mm is NULL
520 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
522 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
526 if (pages == 0 || !mm)
530 ret = __account_locked_vm(mm, pages, inc, current,
531 capable(CAP_IPC_LOCK));
532 mmap_write_unlock(mm);
536 EXPORT_SYMBOL_GPL(account_locked_vm);
538 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
539 unsigned long len, unsigned long prot,
540 unsigned long flag, unsigned long pgoff)
543 struct mm_struct *mm = current->mm;
544 unsigned long populate;
547 ret = security_mmap_file(file, prot, flag);
549 if (mmap_write_lock_killable(mm))
551 ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
553 mmap_write_unlock(mm);
554 userfaultfd_unmap_complete(mm, &uf);
556 mm_populate(ret, populate);
561 unsigned long vm_mmap(struct file *file, unsigned long addr,
562 unsigned long len, unsigned long prot,
563 unsigned long flag, unsigned long offset)
565 if (unlikely(offset + PAGE_ALIGN(len) < offset))
567 if (unlikely(offset_in_page(offset)))
570 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
572 EXPORT_SYMBOL(vm_mmap);
575 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
576 * failure, fall back to non-contiguous (vmalloc) allocation.
577 * @size: size of the request.
578 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
579 * @node: numa node to allocate from
581 * Uses kmalloc to get the memory but if the allocation fails then falls back
582 * to the vmalloc allocator. Use kvfree for freeing the memory.
584 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
585 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
586 * preferable to the vmalloc fallback, due to visible performance drawbacks.
588 * Return: pointer to the allocated memory of %NULL in case of failure
590 void *kvmalloc_node(size_t size, gfp_t flags, int node)
592 gfp_t kmalloc_flags = flags;
596 * We want to attempt a large physically contiguous block first because
597 * it is less likely to fragment multiple larger blocks and therefore
598 * contribute to a long term fragmentation less than vmalloc fallback.
599 * However make sure that larger requests are not too disruptive - no
600 * OOM killer and no allocation failure warnings as we have a fallback.
602 if (size > PAGE_SIZE) {
603 kmalloc_flags |= __GFP_NOWARN;
605 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
606 kmalloc_flags |= __GFP_NORETRY;
608 /* nofail semantic is implemented by the vmalloc fallback */
609 kmalloc_flags &= ~__GFP_NOFAIL;
612 ret = kmalloc_node(size, kmalloc_flags, node);
615 * It doesn't really make sense to fallback to vmalloc for sub page
618 if (ret || size <= PAGE_SIZE)
621 /* Don't even allow crazy sizes */
622 if (unlikely(size > INT_MAX)) {
623 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
628 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
629 * since the callers already cannot assume anything
630 * about the resulting pointer, and cannot play
633 return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
634 flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
635 node, __builtin_return_address(0));
637 EXPORT_SYMBOL(kvmalloc_node);
640 * kvfree() - Free memory.
641 * @addr: Pointer to allocated memory.
643 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
644 * It is slightly more efficient to use kfree() or vfree() if you are certain
645 * that you know which one to use.
647 * Context: Either preemptible task context or not-NMI interrupt.
649 void kvfree(const void *addr)
651 if (is_vmalloc_addr(addr))
656 EXPORT_SYMBOL(kvfree);
659 * kvfree_sensitive - Free a data object containing sensitive information.
660 * @addr: address of the data object to be freed.
661 * @len: length of the data object.
663 * Use the special memzero_explicit() function to clear the content of a
664 * kvmalloc'ed object containing sensitive data to make sure that the
665 * compiler won't optimize out the data clearing.
667 void kvfree_sensitive(const void *addr, size_t len)
669 if (likely(!ZERO_OR_NULL_PTR(addr))) {
670 memzero_explicit((void *)addr, len);
674 EXPORT_SYMBOL(kvfree_sensitive);
676 void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
680 if (oldsize >= newsize)
682 newp = kvmalloc(newsize, flags);
685 memcpy(newp, p, oldsize);
689 EXPORT_SYMBOL(kvrealloc);
692 * __vmalloc_array - allocate memory for a virtually contiguous array.
693 * @n: number of elements.
694 * @size: element size.
695 * @flags: the type of memory to allocate (see kmalloc).
697 void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
701 if (unlikely(check_mul_overflow(n, size, &bytes)))
703 return __vmalloc(bytes, flags);
705 EXPORT_SYMBOL(__vmalloc_array);
708 * vmalloc_array - allocate memory for a virtually contiguous array.
709 * @n: number of elements.
710 * @size: element size.
712 void *vmalloc_array(size_t n, size_t size)
714 return __vmalloc_array(n, size, GFP_KERNEL);
716 EXPORT_SYMBOL(vmalloc_array);
719 * __vcalloc - allocate and zero memory for a virtually contiguous array.
720 * @n: number of elements.
721 * @size: element size.
722 * @flags: the type of memory to allocate (see kmalloc).
724 void *__vcalloc(size_t n, size_t size, gfp_t flags)
726 return __vmalloc_array(n, size, flags | __GFP_ZERO);
728 EXPORT_SYMBOL(__vcalloc);
731 * vcalloc - allocate and zero memory for a virtually contiguous array.
732 * @n: number of elements.
733 * @size: element size.
735 void *vcalloc(size_t n, size_t size)
737 return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
739 EXPORT_SYMBOL(vcalloc);
741 /* Neutral page->mapping pointer to address_space or anon_vma or other */
742 void *page_rmapping(struct page *page)
744 return folio_raw_mapping(page_folio(page));
748 * folio_mapped - Is this folio mapped into userspace?
751 * Return: True if any page in this folio is referenced by user page tables.
753 bool folio_mapped(struct folio *folio)
757 if (!folio_test_large(folio))
758 return atomic_read(&folio->_mapcount) >= 0;
759 if (atomic_read(folio_mapcount_ptr(folio)) >= 0)
761 if (folio_test_hugetlb(folio))
764 nr = folio_nr_pages(folio);
765 for (i = 0; i < nr; i++) {
766 if (atomic_read(&folio_page(folio, i)->_mapcount) >= 0)
771 EXPORT_SYMBOL(folio_mapped);
773 struct anon_vma *folio_anon_vma(struct folio *folio)
775 unsigned long mapping = (unsigned long)folio->mapping;
777 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
779 return (void *)(mapping - PAGE_MAPPING_ANON);
783 * folio_mapping - Find the mapping where this folio is stored.
786 * For folios which are in the page cache, return the mapping that this
787 * page belongs to. Folios in the swap cache return the swap mapping
788 * this page is stored in (which is different from the mapping for the
789 * swap file or swap device where the data is stored).
791 * You can call this for folios which aren't in the swap cache or page
792 * cache and it will return NULL.
794 struct address_space *folio_mapping(struct folio *folio)
796 struct address_space *mapping;
798 /* This happens if someone calls flush_dcache_page on slab page */
799 if (unlikely(folio_test_slab(folio)))
802 if (unlikely(folio_test_swapcache(folio)))
803 return swap_address_space(folio_swap_entry(folio));
805 mapping = folio->mapping;
806 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
809 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
811 EXPORT_SYMBOL(folio_mapping);
813 /* Slow path of page_mapcount() for compound pages */
814 int __page_mapcount(struct page *page)
818 ret = atomic_read(&page->_mapcount) + 1;
820 * For file THP page->_mapcount contains total number of mapping
821 * of the page: no need to look into compound_mapcount.
823 if (!PageAnon(page) && !PageHuge(page))
825 page = compound_head(page);
826 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
827 if (PageDoubleMap(page))
831 EXPORT_SYMBOL_GPL(__page_mapcount);
834 * folio_mapcount() - Calculate the number of mappings of this folio.
837 * A large folio tracks both how many times the entire folio is mapped,
838 * and how many times each individual page in the folio is mapped.
839 * This function calculates the total number of times the folio is
842 * Return: The number of times this folio is mapped.
844 int folio_mapcount(struct folio *folio)
846 int i, compound, nr, ret;
848 if (likely(!folio_test_large(folio)))
849 return atomic_read(&folio->_mapcount) + 1;
851 compound = folio_entire_mapcount(folio);
852 nr = folio_nr_pages(folio);
853 if (folio_test_hugetlb(folio))
856 for (i = 0; i < nr; i++)
857 ret += atomic_read(&folio_page(folio, i)->_mapcount) + 1;
858 /* File pages has compound_mapcount included in _mapcount */
859 if (!folio_test_anon(folio))
860 return ret - compound * nr;
861 if (folio_test_double_map(folio))
867 * folio_copy - Copy the contents of one folio to another.
868 * @dst: Folio to copy to.
869 * @src: Folio to copy from.
871 * The bytes in the folio represented by @src are copied to @dst.
872 * Assumes the caller has validated that @dst is at least as large as @src.
873 * Can be called in atomic context for order-0 folios, but if the folio is
874 * larger, it may sleep.
876 void folio_copy(struct folio *dst, struct folio *src)
879 long nr = folio_nr_pages(src);
882 copy_highpage(folio_page(dst, i), folio_page(src, i));
889 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
890 int sysctl_overcommit_ratio __read_mostly = 50;
891 unsigned long sysctl_overcommit_kbytes __read_mostly;
892 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
893 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
894 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
896 int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
897 size_t *lenp, loff_t *ppos)
901 ret = proc_dointvec(table, write, buffer, lenp, ppos);
902 if (ret == 0 && write)
903 sysctl_overcommit_kbytes = 0;
907 static void sync_overcommit_as(struct work_struct *dummy)
909 percpu_counter_sync(&vm_committed_as);
912 int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
913 size_t *lenp, loff_t *ppos)
920 * The deviation of sync_overcommit_as could be big with loose policy
921 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
922 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
923 * with the strict "NEVER", and to avoid possible race condition (even
924 * though user usually won't too frequently do the switching to policy
925 * OVERCOMMIT_NEVER), the switch is done in the following order:
926 * 1. changing the batch
927 * 2. sync percpu count on each CPU
928 * 3. switch the policy
932 t.data = &new_policy;
933 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
934 if (ret || new_policy == -1)
937 mm_compute_batch(new_policy);
938 if (new_policy == OVERCOMMIT_NEVER)
939 schedule_on_each_cpu(sync_overcommit_as);
940 sysctl_overcommit_memory = new_policy;
942 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
948 int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
949 size_t *lenp, loff_t *ppos)
953 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
954 if (ret == 0 && write)
955 sysctl_overcommit_ratio = 0;
960 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
962 unsigned long vm_commit_limit(void)
964 unsigned long allowed;
966 if (sysctl_overcommit_kbytes)
967 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
969 allowed = ((totalram_pages() - hugetlb_total_pages())
970 * sysctl_overcommit_ratio / 100);
971 allowed += total_swap_pages;
977 * Make sure vm_committed_as in one cacheline and not cacheline shared with
978 * other variables. It can be updated by several CPUs frequently.
980 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
983 * The global memory commitment made in the system can be a metric
984 * that can be used to drive ballooning decisions when Linux is hosted
985 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
986 * balancing memory across competing virtual machines that are hosted.
987 * Several metrics drive this policy engine including the guest reported
990 * The time cost of this is very low for small platforms, and for big
991 * platform like a 2S/36C/72T Skylake server, in worst case where
992 * vm_committed_as's spinlock is under severe contention, the time cost
993 * could be about 30~40 microseconds.
995 unsigned long vm_memory_committed(void)
997 return percpu_counter_sum_positive(&vm_committed_as);
999 EXPORT_SYMBOL_GPL(vm_memory_committed);
1002 * Check that a process has enough memory to allocate a new virtual
1003 * mapping. 0 means there is enough memory for the allocation to
1004 * succeed and -ENOMEM implies there is not.
1006 * We currently support three overcommit policies, which are set via the
1007 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
1009 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1010 * Additional code 2002 Jul 20 by Robert Love.
1012 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1014 * Note this is a helper function intended to be used by LSMs which
1015 * wish to use this logic.
1017 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1021 vm_acct_memory(pages);
1024 * Sometimes we want to use more memory than we have
1026 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1029 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1030 if (pages > totalram_pages() + total_swap_pages)
1035 allowed = vm_commit_limit();
1037 * Reserve some for root
1040 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1043 * Don't let a single process grow so big a user can't recover
1046 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1048 allowed -= min_t(long, mm->total_vm / 32, reserve);
1051 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1054 vm_unacct_memory(pages);
1060 * get_cmdline() - copy the cmdline value to a buffer.
1061 * @task: the task whose cmdline value to copy.
1062 * @buffer: the buffer to copy to.
1063 * @buflen: the length of the buffer. Larger cmdline values are truncated
1066 * Return: the size of the cmdline field copied. Note that the copy does
1067 * not guarantee an ending NULL byte.
1069 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
1073 struct mm_struct *mm = get_task_mm(task);
1074 unsigned long arg_start, arg_end, env_start, env_end;
1078 goto out_mm; /* Shh! No looking before we're done */
1080 spin_lock(&mm->arg_lock);
1081 arg_start = mm->arg_start;
1082 arg_end = mm->arg_end;
1083 env_start = mm->env_start;
1084 env_end = mm->env_end;
1085 spin_unlock(&mm->arg_lock);
1087 len = arg_end - arg_start;
1092 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1095 * If the nul at the end of args has been overwritten, then
1096 * assume application is using setproctitle(3).
1098 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1099 len = strnlen(buffer, res);
1103 len = env_end - env_start;
1104 if (len > buflen - res)
1106 res += access_process_vm(task, env_start,
1109 res = strnlen(buffer, res);
1118 int __weak memcmp_pages(struct page *page1, struct page *page2)
1120 char *addr1, *addr2;
1123 addr1 = kmap_atomic(page1);
1124 addr2 = kmap_atomic(page2);
1125 ret = memcmp(addr1, addr2, PAGE_SIZE);
1126 kunmap_atomic(addr2);
1127 kunmap_atomic(addr1);
1131 #ifdef CONFIG_PRINTK
1133 * mem_dump_obj - Print available provenance information
1134 * @object: object for which to find provenance information.
1136 * This function uses pr_cont(), so that the caller is expected to have
1137 * printed out whatever preamble is appropriate. The provenance information
1138 * depends on the type of object and on how much debugging is enabled.
1139 * For example, for a slab-cache object, the slab name is printed, and,
1140 * if available, the return address and stack trace from the allocation
1141 * and last free path of that object.
1143 void mem_dump_obj(void *object)
1147 if (kmem_valid_obj(object)) {
1148 kmem_dump_obj(object);
1152 if (vmalloc_dump_obj(object))
1155 if (virt_addr_valid(object))
1156 type = "non-slab/vmalloc memory";
1157 else if (object == NULL)
1158 type = "NULL pointer";
1159 else if (object == ZERO_SIZE_PTR)
1160 type = "zero-size pointer";
1162 type = "non-paged memory";
1164 pr_cont(" %s\n", type);
1166 EXPORT_SYMBOL_GPL(mem_dump_obj);
1170 * A driver might set a page logically offline -- PageOffline() -- and
1171 * turn the page inaccessible in the hypervisor; after that, access to page
1172 * content can be fatal.
1174 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1175 * pages after checking PageOffline(); however, these PFN walkers can race
1176 * with drivers that set PageOffline().
1178 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1179 * synchronize with such drivers, achieving that a page cannot be set
1180 * PageOffline() while frozen.
1182 * page_offline_begin()/page_offline_end() is used by drivers that care about
1183 * such races when setting a page PageOffline().
1185 static DECLARE_RWSEM(page_offline_rwsem);
1187 void page_offline_freeze(void)
1189 down_read(&page_offline_rwsem);
1192 void page_offline_thaw(void)
1194 up_read(&page_offline_rwsem);
1197 void page_offline_begin(void)
1199 down_write(&page_offline_rwsem);
1201 EXPORT_SYMBOL(page_offline_begin);
1203 void page_offline_end(void)
1205 up_write(&page_offline_rwsem);
1207 EXPORT_SYMBOL(page_offline_end);
1209 #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_FOLIO
1210 void flush_dcache_folio(struct folio *folio)
1212 long i, nr = folio_nr_pages(folio);
1214 for (i = 0; i < nr; i++)
1215 flush_dcache_page(folio_page(folio, i));
1217 EXPORT_SYMBOL(flush_dcache_folio);