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
;
346 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
347 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
349 /* Is the current task 32bit ? */
350 if (!IS_ENABLED(CONFIG_64BIT
) || is_compat_task())
351 return randomize_page(mm
->brk
, SZ_32M
);
353 return randomize_page(mm
->brk
, SZ_1G
);
356 unsigned long arch_mmap_rnd(void)
360 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
361 if (is_compat_task())
362 rnd
= get_random_long() & ((1UL << mmap_rnd_compat_bits
) - 1);
364 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
365 rnd
= get_random_long() & ((1UL << mmap_rnd_bits
) - 1);
367 return rnd
<< PAGE_SHIFT
;
370 static int mmap_is_legacy(struct rlimit
*rlim_stack
)
372 if (current
->personality
& ADDR_COMPAT_LAYOUT
)
375 if (rlim_stack
->rlim_cur
== RLIM_INFINITY
)
378 return sysctl_legacy_va_layout
;
382 * Leave enough space between the mmap area and the stack to honour ulimit in
383 * the face of randomisation.
385 #define MIN_GAP (SZ_128M)
386 #define MAX_GAP (STACK_TOP / 6 * 5)
388 static unsigned long mmap_base(unsigned long rnd
, struct rlimit
*rlim_stack
)
390 unsigned long gap
= rlim_stack
->rlim_cur
;
391 unsigned long pad
= stack_guard_gap
;
393 /* Account for stack randomization if necessary */
394 if (current
->flags
& PF_RANDOMIZE
)
395 pad
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
397 /* Values close to RLIM_INFINITY can overflow. */
403 else if (gap
> MAX_GAP
)
406 return PAGE_ALIGN(STACK_TOP
- gap
- rnd
);
409 void arch_pick_mmap_layout(struct mm_struct
*mm
, struct rlimit
*rlim_stack
)
411 unsigned long random_factor
= 0UL;
413 if (current
->flags
& PF_RANDOMIZE
)
414 random_factor
= arch_mmap_rnd();
416 if (mmap_is_legacy(rlim_stack
)) {
417 mm
->mmap_base
= TASK_UNMAPPED_BASE
+ random_factor
;
418 mm
->get_unmapped_area
= arch_get_unmapped_area
;
420 mm
->mmap_base
= mmap_base(random_factor
, rlim_stack
);
421 mm
->get_unmapped_area
= arch_get_unmapped_area_topdown
;
424 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
425 void arch_pick_mmap_layout(struct mm_struct
*mm
, struct rlimit
*rlim_stack
)
427 mm
->mmap_base
= TASK_UNMAPPED_BASE
;
428 mm
->get_unmapped_area
= arch_get_unmapped_area
;
433 * __account_locked_vm - account locked pages to an mm's locked_vm
434 * @mm: mm to account against
435 * @pages: number of pages to account
436 * @inc: %true if @pages should be considered positive, %false if not
437 * @task: task used to check RLIMIT_MEMLOCK
438 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
440 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
441 * that mmap_lock is held as writer.
445 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
447 int __account_locked_vm(struct mm_struct
*mm
, unsigned long pages
, bool inc
,
448 struct task_struct
*task
, bool bypass_rlim
)
450 unsigned long locked_vm
, limit
;
453 mmap_assert_write_locked(mm
);
455 locked_vm
= mm
->locked_vm
;
458 limit
= task_rlimit(task
, RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
459 if (locked_vm
+ pages
> limit
)
463 mm
->locked_vm
= locked_vm
+ pages
;
465 WARN_ON_ONCE(pages
> locked_vm
);
466 mm
->locked_vm
= locked_vm
- pages
;
469 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__
, task
->pid
,
470 (void *)_RET_IP_
, (inc
) ? '+' : '-', pages
<< PAGE_SHIFT
,
471 locked_vm
<< PAGE_SHIFT
, task_rlimit(task
, RLIMIT_MEMLOCK
),
472 ret
? " - exceeded" : "");
476 EXPORT_SYMBOL_GPL(__account_locked_vm
);
479 * account_locked_vm - account locked pages to an mm's locked_vm
480 * @mm: mm to account against, may be NULL
481 * @pages: number of pages to account
482 * @inc: %true if @pages should be considered positive, %false if not
484 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
487 * * 0 on success, or if mm is NULL
488 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
490 int account_locked_vm(struct mm_struct
*mm
, unsigned long pages
, bool inc
)
494 if (pages
== 0 || !mm
)
498 ret
= __account_locked_vm(mm
, pages
, inc
, current
,
499 capable(CAP_IPC_LOCK
));
500 mmap_write_unlock(mm
);
504 EXPORT_SYMBOL_GPL(account_locked_vm
);
506 unsigned long vm_mmap_pgoff(struct file
*file
, unsigned long addr
,
507 unsigned long len
, unsigned long prot
,
508 unsigned long flag
, unsigned long pgoff
)
511 struct mm_struct
*mm
= current
->mm
;
512 unsigned long populate
;
515 ret
= security_mmap_file(file
, prot
, flag
);
517 if (mmap_write_lock_killable(mm
))
519 ret
= do_mmap(file
, addr
, len
, prot
, flag
, pgoff
, &populate
,
521 mmap_write_unlock(mm
);
522 userfaultfd_unmap_complete(mm
, &uf
);
524 mm_populate(ret
, populate
);
529 unsigned long vm_mmap(struct file
*file
, unsigned long addr
,
530 unsigned long len
, unsigned long prot
,
531 unsigned long flag
, unsigned long offset
)
533 if (unlikely(offset
+ PAGE_ALIGN(len
) < offset
))
535 if (unlikely(offset_in_page(offset
)))
538 return vm_mmap_pgoff(file
, addr
, len
, prot
, flag
, offset
>> PAGE_SHIFT
);
540 EXPORT_SYMBOL(vm_mmap
);
543 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
544 * failure, fall back to non-contiguous (vmalloc) allocation.
545 * @size: size of the request.
546 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
547 * @node: numa node to allocate from
549 * Uses kmalloc to get the memory but if the allocation fails then falls back
550 * to the vmalloc allocator. Use kvfree for freeing the memory.
552 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
553 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
554 * preferable to the vmalloc fallback, due to visible performance drawbacks.
556 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
557 * fall back to vmalloc.
559 * Return: pointer to the allocated memory of %NULL in case of failure
561 void *kvmalloc_node(size_t size
, gfp_t flags
, int node
)
563 gfp_t kmalloc_flags
= flags
;
567 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
568 * so the given set of flags has to be compatible.
570 if ((flags
& GFP_KERNEL
) != GFP_KERNEL
)
571 return kmalloc_node(size
, flags
, node
);
574 * We want to attempt a large physically contiguous block first because
575 * it is less likely to fragment multiple larger blocks and therefore
576 * contribute to a long term fragmentation less than vmalloc fallback.
577 * However make sure that larger requests are not too disruptive - no
578 * OOM killer and no allocation failure warnings as we have a fallback.
580 if (size
> PAGE_SIZE
) {
581 kmalloc_flags
|= __GFP_NOWARN
;
583 if (!(kmalloc_flags
& __GFP_RETRY_MAYFAIL
))
584 kmalloc_flags
|= __GFP_NORETRY
;
587 ret
= kmalloc_node(size
, kmalloc_flags
, node
);
590 * It doesn't really make sense to fallback to vmalloc for sub page
593 if (ret
|| size
<= PAGE_SIZE
)
596 /* Don't even allow crazy sizes */
597 if (unlikely(size
> INT_MAX
)) {
598 WARN_ON_ONCE(!(flags
& __GFP_NOWARN
));
602 return __vmalloc_node(size
, 1, flags
, node
,
603 __builtin_return_address(0));
605 EXPORT_SYMBOL(kvmalloc_node
);
608 * kvfree() - Free memory.
609 * @addr: Pointer to allocated memory.
611 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
612 * It is slightly more efficient to use kfree() or vfree() if you are certain
613 * that you know which one to use.
615 * Context: Either preemptible task context or not-NMI interrupt.
617 void kvfree(const void *addr
)
619 if (is_vmalloc_addr(addr
))
624 EXPORT_SYMBOL(kvfree
);
627 * kvfree_sensitive - Free a data object containing sensitive information.
628 * @addr: address of the data object to be freed.
629 * @len: length of the data object.
631 * Use the special memzero_explicit() function to clear the content of a
632 * kvmalloc'ed object containing sensitive data to make sure that the
633 * compiler won't optimize out the data clearing.
635 void kvfree_sensitive(const void *addr
, size_t len
)
637 if (likely(!ZERO_OR_NULL_PTR(addr
))) {
638 memzero_explicit((void *)addr
, len
);
642 EXPORT_SYMBOL(kvfree_sensitive
);
644 void *kvrealloc(const void *p
, size_t oldsize
, size_t newsize
, gfp_t flags
)
648 if (oldsize
>= newsize
)
650 newp
= kvmalloc(newsize
, flags
);
653 memcpy(newp
, p
, oldsize
);
657 EXPORT_SYMBOL(kvrealloc
);
659 static inline void *__page_rmapping(struct page
*page
)
661 unsigned long mapping
;
663 mapping
= (unsigned long)page
->mapping
;
664 mapping
&= ~PAGE_MAPPING_FLAGS
;
666 return (void *)mapping
;
669 /* Neutral page->mapping pointer to address_space or anon_vma or other */
670 void *page_rmapping(struct page
*page
)
672 page
= compound_head(page
);
673 return __page_rmapping(page
);
677 * Return true if this page is mapped into pagetables.
678 * For compound page it returns true if any subpage of compound page is mapped.
680 bool page_mapped(struct page
*page
)
684 if (likely(!PageCompound(page
)))
685 return atomic_read(&page
->_mapcount
) >= 0;
686 page
= compound_head(page
);
687 if (atomic_read(compound_mapcount_ptr(page
)) >= 0)
691 for (i
= 0; i
< compound_nr(page
); i
++) {
692 if (atomic_read(&page
[i
]._mapcount
) >= 0)
697 EXPORT_SYMBOL(page_mapped
);
699 struct anon_vma
*page_anon_vma(struct page
*page
)
701 unsigned long mapping
;
703 page
= compound_head(page
);
704 mapping
= (unsigned long)page
->mapping
;
705 if ((mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
707 return __page_rmapping(page
);
710 struct address_space
*page_mapping(struct page
*page
)
712 struct address_space
*mapping
;
714 page
= compound_head(page
);
716 /* This happens if someone calls flush_dcache_page on slab page */
717 if (unlikely(PageSlab(page
)))
720 if (unlikely(PageSwapCache(page
))) {
723 entry
.val
= page_private(page
);
724 return swap_address_space(entry
);
727 mapping
= page
->mapping
;
728 if ((unsigned long)mapping
& PAGE_MAPPING_ANON
)
731 return (void *)((unsigned long)mapping
& ~PAGE_MAPPING_FLAGS
);
733 EXPORT_SYMBOL(page_mapping
);
735 /* Slow path of page_mapcount() for compound pages */
736 int __page_mapcount(struct page
*page
)
740 ret
= atomic_read(&page
->_mapcount
) + 1;
742 * For file THP page->_mapcount contains total number of mapping
743 * of the page: no need to look into compound_mapcount.
745 if (!PageAnon(page
) && !PageHuge(page
))
747 page
= compound_head(page
);
748 ret
+= atomic_read(compound_mapcount_ptr(page
)) + 1;
749 if (PageDoubleMap(page
))
753 EXPORT_SYMBOL_GPL(__page_mapcount
);
755 void copy_huge_page(struct page
*dst
, struct page
*src
)
757 unsigned i
, nr
= compound_nr(src
);
759 for (i
= 0; i
< nr
; i
++) {
761 copy_highpage(nth_page(dst
, i
), nth_page(src
, i
));
765 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
;
766 int sysctl_overcommit_ratio __read_mostly
= 50;
767 unsigned long sysctl_overcommit_kbytes __read_mostly
;
768 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
769 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
770 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
772 int overcommit_ratio_handler(struct ctl_table
*table
, int write
, void *buffer
,
773 size_t *lenp
, loff_t
*ppos
)
777 ret
= proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
778 if (ret
== 0 && write
)
779 sysctl_overcommit_kbytes
= 0;
783 static void sync_overcommit_as(struct work_struct
*dummy
)
785 percpu_counter_sync(&vm_committed_as
);
788 int overcommit_policy_handler(struct ctl_table
*table
, int write
, void *buffer
,
789 size_t *lenp
, loff_t
*ppos
)
796 * The deviation of sync_overcommit_as could be big with loose policy
797 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
798 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
799 * with the strict "NEVER", and to avoid possible race condition (even
800 * though user usually won't too frequently do the switching to policy
801 * OVERCOMMIT_NEVER), the switch is done in the following order:
802 * 1. changing the batch
803 * 2. sync percpu count on each CPU
804 * 3. switch the policy
808 t
.data
= &new_policy
;
809 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
810 if (ret
|| new_policy
== -1)
813 mm_compute_batch(new_policy
);
814 if (new_policy
== OVERCOMMIT_NEVER
)
815 schedule_on_each_cpu(sync_overcommit_as
);
816 sysctl_overcommit_memory
= new_policy
;
818 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
824 int overcommit_kbytes_handler(struct ctl_table
*table
, int write
, void *buffer
,
825 size_t *lenp
, loff_t
*ppos
)
829 ret
= proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
830 if (ret
== 0 && write
)
831 sysctl_overcommit_ratio
= 0;
836 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
838 unsigned long vm_commit_limit(void)
840 unsigned long allowed
;
842 if (sysctl_overcommit_kbytes
)
843 allowed
= sysctl_overcommit_kbytes
>> (PAGE_SHIFT
- 10);
845 allowed
= ((totalram_pages() - hugetlb_total_pages())
846 * sysctl_overcommit_ratio
/ 100);
847 allowed
+= total_swap_pages
;
853 * Make sure vm_committed_as in one cacheline and not cacheline shared with
854 * other variables. It can be updated by several CPUs frequently.
856 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
859 * The global memory commitment made in the system can be a metric
860 * that can be used to drive ballooning decisions when Linux is hosted
861 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
862 * balancing memory across competing virtual machines that are hosted.
863 * Several metrics drive this policy engine including the guest reported
866 * The time cost of this is very low for small platforms, and for big
867 * platform like a 2S/36C/72T Skylake server, in worst case where
868 * vm_committed_as's spinlock is under severe contention, the time cost
869 * could be about 30~40 microseconds.
871 unsigned long vm_memory_committed(void)
873 return percpu_counter_sum_positive(&vm_committed_as
);
875 EXPORT_SYMBOL_GPL(vm_memory_committed
);
878 * Check that a process has enough memory to allocate a new virtual
879 * mapping. 0 means there is enough memory for the allocation to
880 * succeed and -ENOMEM implies there is not.
882 * We currently support three overcommit policies, which are set via the
883 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
885 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
886 * Additional code 2002 Jul 20 by Robert Love.
888 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
890 * Note this is a helper function intended to be used by LSMs which
891 * wish to use this logic.
893 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
897 vm_acct_memory(pages
);
900 * Sometimes we want to use more memory than we have
902 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
905 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
906 if (pages
> totalram_pages() + total_swap_pages
)
911 allowed
= vm_commit_limit();
913 * Reserve some for root
916 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
919 * Don't let a single process grow so big a user can't recover
922 long reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
924 allowed
-= min_t(long, mm
->total_vm
/ 32, reserve
);
927 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
930 vm_unacct_memory(pages
);
936 * get_cmdline() - copy the cmdline value to a buffer.
937 * @task: the task whose cmdline value to copy.
938 * @buffer: the buffer to copy to.
939 * @buflen: the length of the buffer. Larger cmdline values are truncated
942 * Return: the size of the cmdline field copied. Note that the copy does
943 * not guarantee an ending NULL byte.
945 int get_cmdline(struct task_struct
*task
, char *buffer
, int buflen
)
949 struct mm_struct
*mm
= get_task_mm(task
);
950 unsigned long arg_start
, arg_end
, env_start
, env_end
;
954 goto out_mm
; /* Shh! No looking before we're done */
956 spin_lock(&mm
->arg_lock
);
957 arg_start
= mm
->arg_start
;
958 arg_end
= mm
->arg_end
;
959 env_start
= mm
->env_start
;
960 env_end
= mm
->env_end
;
961 spin_unlock(&mm
->arg_lock
);
963 len
= arg_end
- arg_start
;
968 res
= access_process_vm(task
, arg_start
, buffer
, len
, FOLL_FORCE
);
971 * If the nul at the end of args has been overwritten, then
972 * assume application is using setproctitle(3).
974 if (res
> 0 && buffer
[res
-1] != '\0' && len
< buflen
) {
975 len
= strnlen(buffer
, res
);
979 len
= env_end
- env_start
;
980 if (len
> buflen
- res
)
982 res
+= access_process_vm(task
, env_start
,
985 res
= strnlen(buffer
, res
);
994 int __weak
memcmp_pages(struct page
*page1
, struct page
*page2
)
999 addr1
= kmap_atomic(page1
);
1000 addr2
= kmap_atomic(page2
);
1001 ret
= memcmp(addr1
, addr2
, PAGE_SIZE
);
1002 kunmap_atomic(addr2
);
1003 kunmap_atomic(addr1
);
1007 #ifdef CONFIG_PRINTK
1009 * mem_dump_obj - Print available provenance information
1010 * @object: object for which to find provenance information.
1012 * This function uses pr_cont(), so that the caller is expected to have
1013 * printed out whatever preamble is appropriate. The provenance information
1014 * depends on the type of object and on how much debugging is enabled.
1015 * For example, for a slab-cache object, the slab name is printed, and,
1016 * if available, the return address and stack trace from the allocation
1017 * and last free path of that object.
1019 void mem_dump_obj(void *object
)
1023 if (kmem_valid_obj(object
)) {
1024 kmem_dump_obj(object
);
1028 if (vmalloc_dump_obj(object
))
1031 if (virt_addr_valid(object
))
1032 type
= "non-slab/vmalloc memory";
1033 else if (object
== NULL
)
1034 type
= "NULL pointer";
1035 else if (object
== ZERO_SIZE_PTR
)
1036 type
= "zero-size pointer";
1038 type
= "non-paged memory";
1040 pr_cont(" %s\n", type
);
1042 EXPORT_SYMBOL_GPL(mem_dump_obj
);
1046 * A driver might set a page logically offline -- PageOffline() -- and
1047 * turn the page inaccessible in the hypervisor; after that, access to page
1048 * content can be fatal.
1050 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1051 * pages after checking PageOffline(); however, these PFN walkers can race
1052 * with drivers that set PageOffline().
1054 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1055 * synchronize with such drivers, achieving that a page cannot be set
1056 * PageOffline() while frozen.
1058 * page_offline_begin()/page_offline_end() is used by drivers that care about
1059 * such races when setting a page PageOffline().
1061 static DECLARE_RWSEM(page_offline_rwsem
);
1063 void page_offline_freeze(void)
1065 down_read(&page_offline_rwsem
);
1068 void page_offline_thaw(void)
1070 up_read(&page_offline_rwsem
);
1073 void page_offline_begin(void)
1075 down_write(&page_offline_rwsem
);
1077 EXPORT_SYMBOL(page_offline_begin
);
1079 void page_offline_end(void)
1081 up_write(&page_offline_rwsem
);
1083 EXPORT_SYMBOL(page_offline_end
);