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.
74 * Return: source string if it is in .rodata section otherwise
75 * fallback to kstrdup.
77 const char *kstrdup_const(const char *s, gfp_t gfp)
79 if (is_kernel_rodata((unsigned long)s))
82 return kstrdup(s, gfp);
84 EXPORT_SYMBOL(kstrdup_const);
87 * kstrndup - allocate space for and copy an existing string
88 * @s: the string to duplicate
89 * @max: read at most @max chars from @s
90 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
92 * Note: Use kmemdup_nul() instead if the size is known exactly.
94 * Return: newly allocated copy of @s or %NULL in case of error
96 char *kstrndup(const char *s, size_t max, gfp_t gfp)
104 len = strnlen(s, max);
105 buf = kmalloc_track_caller(len+1, gfp);
112 EXPORT_SYMBOL(kstrndup);
115 * kmemdup - duplicate region of memory
117 * @src: memory region to duplicate
118 * @len: memory region length
119 * @gfp: GFP mask to use
121 * Return: newly allocated copy of @src or %NULL in case of error
123 void *kmemdup(const void *src, size_t len, gfp_t gfp)
127 p = kmalloc_track_caller(len, gfp);
132 EXPORT_SYMBOL(kmemdup);
135 * kmemdup_nul - Create a NUL-terminated string from unterminated data
136 * @s: The data to stringify
137 * @len: The size of the data
138 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
140 * Return: newly allocated copy of @s with NUL-termination or %NULL in
143 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
150 buf = kmalloc_track_caller(len + 1, gfp);
157 EXPORT_SYMBOL(kmemdup_nul);
160 * memdup_user - duplicate memory region from user space
162 * @src: source address in user space
163 * @len: number of bytes to copy
165 * Return: an ERR_PTR() on failure. Result is physically
166 * contiguous, to be freed by kfree().
168 void *memdup_user(const void __user *src, size_t len)
172 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
174 return ERR_PTR(-ENOMEM);
176 if (copy_from_user(p, src, len)) {
178 return ERR_PTR(-EFAULT);
183 EXPORT_SYMBOL(memdup_user);
186 * vmemdup_user - duplicate memory region from user space
188 * @src: source address in user space
189 * @len: number of bytes to copy
191 * Return: an ERR_PTR() on failure. Result may be not
192 * physically contiguous. Use kvfree() to free.
194 void *vmemdup_user(const void __user *src, size_t len)
198 p = kvmalloc(len, GFP_USER);
200 return ERR_PTR(-ENOMEM);
202 if (copy_from_user(p, src, len)) {
204 return ERR_PTR(-EFAULT);
209 EXPORT_SYMBOL(vmemdup_user);
212 * strndup_user - duplicate an existing string from user space
213 * @s: The string to duplicate
214 * @n: Maximum number of bytes to copy, including the trailing NUL.
216 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
218 char *strndup_user(const char __user *s, long n)
223 length = strnlen_user(s, n);
226 return ERR_PTR(-EFAULT);
229 return ERR_PTR(-EINVAL);
231 p = memdup_user(s, length);
236 p[length - 1] = '\0';
240 EXPORT_SYMBOL(strndup_user);
243 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
245 * @src: source address in user space
246 * @len: number of bytes to copy
248 * Return: an ERR_PTR() on failure.
250 void *memdup_user_nul(const void __user *src, size_t len)
255 * Always use GFP_KERNEL, since copy_from_user() can sleep and
256 * cause pagefault, which makes it pointless to use GFP_NOFS
259 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
261 return ERR_PTR(-ENOMEM);
263 if (copy_from_user(p, src, len)) {
265 return ERR_PTR(-EFAULT);
271 EXPORT_SYMBOL(memdup_user_nul);
273 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
274 struct vm_area_struct *prev)
276 struct vm_area_struct *next;
280 next = prev->vm_next;
291 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma)
293 struct vm_area_struct *prev, *next;
298 prev->vm_next = next;
302 next->vm_prev = prev;
305 /* Check if the vma is being used as a stack by this task */
306 int vma_is_stack_for_current(struct vm_area_struct *vma)
308 struct task_struct * __maybe_unused t = current;
310 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
313 #ifndef STACK_RND_MASK
314 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
317 unsigned long randomize_stack_top(unsigned long stack_top)
319 unsigned long random_variable = 0;
321 if (current->flags & PF_RANDOMIZE) {
322 random_variable = get_random_long();
323 random_variable &= STACK_RND_MASK;
324 random_variable <<= PAGE_SHIFT;
326 #ifdef CONFIG_STACK_GROWSUP
327 return PAGE_ALIGN(stack_top) + random_variable;
329 return PAGE_ALIGN(stack_top) - random_variable;
333 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
334 unsigned long arch_randomize_brk(struct mm_struct *mm)
336 /* Is the current task 32bit ? */
337 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
338 return randomize_page(mm->brk, SZ_32M);
340 return randomize_page(mm->brk, SZ_1G);
343 unsigned long arch_mmap_rnd(void)
347 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
348 if (is_compat_task())
349 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
351 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
352 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
354 return rnd << PAGE_SHIFT;
357 static int mmap_is_legacy(struct rlimit *rlim_stack)
359 if (current->personality & ADDR_COMPAT_LAYOUT)
362 if (rlim_stack->rlim_cur == RLIM_INFINITY)
365 return sysctl_legacy_va_layout;
369 * Leave enough space between the mmap area and the stack to honour ulimit in
370 * the face of randomisation.
372 #define MIN_GAP (SZ_128M)
373 #define MAX_GAP (STACK_TOP / 6 * 5)
375 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
377 unsigned long gap = rlim_stack->rlim_cur;
378 unsigned long pad = stack_guard_gap;
380 /* Account for stack randomization if necessary */
381 if (current->flags & PF_RANDOMIZE)
382 pad += (STACK_RND_MASK << PAGE_SHIFT);
384 /* Values close to RLIM_INFINITY can overflow. */
390 else if (gap > MAX_GAP)
393 return PAGE_ALIGN(STACK_TOP - gap - rnd);
396 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
398 unsigned long random_factor = 0UL;
400 if (current->flags & PF_RANDOMIZE)
401 random_factor = arch_mmap_rnd();
403 if (mmap_is_legacy(rlim_stack)) {
404 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
405 mm->get_unmapped_area = arch_get_unmapped_area;
407 mm->mmap_base = mmap_base(random_factor, rlim_stack);
408 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
411 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
412 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
414 mm->mmap_base = TASK_UNMAPPED_BASE;
415 mm->get_unmapped_area = arch_get_unmapped_area;
420 * __account_locked_vm - account locked pages to an mm's locked_vm
421 * @mm: mm to account against
422 * @pages: number of pages to account
423 * @inc: %true if @pages should be considered positive, %false if not
424 * @task: task used to check RLIMIT_MEMLOCK
425 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
427 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
428 * that mmap_sem is held as writer.
432 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
434 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
435 struct task_struct *task, bool bypass_rlim)
437 unsigned long locked_vm, limit;
440 lockdep_assert_held_write(&mm->mmap_sem);
442 locked_vm = mm->locked_vm;
445 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
446 if (locked_vm + pages > limit)
450 mm->locked_vm = locked_vm + pages;
452 WARN_ON_ONCE(pages > locked_vm);
453 mm->locked_vm = locked_vm - pages;
456 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
457 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
458 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
459 ret ? " - exceeded" : "");
463 EXPORT_SYMBOL_GPL(__account_locked_vm);
466 * account_locked_vm - account locked pages to an mm's locked_vm
467 * @mm: mm to account against, may be NULL
468 * @pages: number of pages to account
469 * @inc: %true if @pages should be considered positive, %false if not
471 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
474 * * 0 on success, or if mm is NULL
475 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
477 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
481 if (pages == 0 || !mm)
484 down_write(&mm->mmap_sem);
485 ret = __account_locked_vm(mm, pages, inc, current,
486 capable(CAP_IPC_LOCK));
487 up_write(&mm->mmap_sem);
491 EXPORT_SYMBOL_GPL(account_locked_vm);
493 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
494 unsigned long len, unsigned long prot,
495 unsigned long flag, unsigned long pgoff)
498 struct mm_struct *mm = current->mm;
499 unsigned long populate;
502 ret = security_mmap_file(file, prot, flag);
504 if (down_write_killable(&mm->mmap_sem))
506 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
508 up_write(&mm->mmap_sem);
509 userfaultfd_unmap_complete(mm, &uf);
511 mm_populate(ret, populate);
516 unsigned long vm_mmap(struct file *file, unsigned long addr,
517 unsigned long len, unsigned long prot,
518 unsigned long flag, unsigned long offset)
520 if (unlikely(offset + PAGE_ALIGN(len) < offset))
522 if (unlikely(offset_in_page(offset)))
525 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
527 EXPORT_SYMBOL(vm_mmap);
530 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
531 * failure, fall back to non-contiguous (vmalloc) allocation.
532 * @size: size of the request.
533 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
534 * @node: numa node to allocate from
536 * Uses kmalloc to get the memory but if the allocation fails then falls back
537 * to the vmalloc allocator. Use kvfree for freeing the memory.
539 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
540 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
541 * preferable to the vmalloc fallback, due to visible performance drawbacks.
543 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
544 * fall back to vmalloc.
546 * Return: pointer to the allocated memory of %NULL in case of failure
548 void *kvmalloc_node(size_t size, gfp_t flags, int node)
550 gfp_t kmalloc_flags = flags;
554 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
555 * so the given set of flags has to be compatible.
557 if ((flags & GFP_KERNEL) != GFP_KERNEL)
558 return kmalloc_node(size, flags, node);
561 * We want to attempt a large physically contiguous block first because
562 * it is less likely to fragment multiple larger blocks and therefore
563 * contribute to a long term fragmentation less than vmalloc fallback.
564 * However make sure that larger requests are not too disruptive - no
565 * OOM killer and no allocation failure warnings as we have a fallback.
567 if (size > PAGE_SIZE) {
568 kmalloc_flags |= __GFP_NOWARN;
570 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
571 kmalloc_flags |= __GFP_NORETRY;
574 ret = kmalloc_node(size, kmalloc_flags, node);
577 * It doesn't really make sense to fallback to vmalloc for sub page
580 if (ret || size <= PAGE_SIZE)
583 return __vmalloc_node_flags_caller(size, node, flags,
584 __builtin_return_address(0));
586 EXPORT_SYMBOL(kvmalloc_node);
589 * kvfree() - Free memory.
590 * @addr: Pointer to allocated memory.
592 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
593 * It is slightly more efficient to use kfree() or vfree() if you are certain
594 * that you know which one to use.
596 * Context: Either preemptible task context or not-NMI interrupt.
598 void kvfree(const void *addr)
600 if (is_vmalloc_addr(addr))
605 EXPORT_SYMBOL(kvfree);
607 static inline void *__page_rmapping(struct page *page)
609 unsigned long mapping;
611 mapping = (unsigned long)page->mapping;
612 mapping &= ~PAGE_MAPPING_FLAGS;
614 return (void *)mapping;
617 /* Neutral page->mapping pointer to address_space or anon_vma or other */
618 void *page_rmapping(struct page *page)
620 page = compound_head(page);
621 return __page_rmapping(page);
625 * Return true if this page is mapped into pagetables.
626 * For compound page it returns true if any subpage of compound page is mapped.
628 bool page_mapped(struct page *page)
632 if (likely(!PageCompound(page)))
633 return atomic_read(&page->_mapcount) >= 0;
634 page = compound_head(page);
635 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
639 for (i = 0; i < compound_nr(page); i++) {
640 if (atomic_read(&page[i]._mapcount) >= 0)
645 EXPORT_SYMBOL(page_mapped);
647 struct anon_vma *page_anon_vma(struct page *page)
649 unsigned long mapping;
651 page = compound_head(page);
652 mapping = (unsigned long)page->mapping;
653 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
655 return __page_rmapping(page);
658 struct address_space *page_mapping(struct page *page)
660 struct address_space *mapping;
662 page = compound_head(page);
664 /* This happens if someone calls flush_dcache_page on slab page */
665 if (unlikely(PageSlab(page)))
668 if (unlikely(PageSwapCache(page))) {
671 entry.val = page_private(page);
672 return swap_address_space(entry);
675 mapping = page->mapping;
676 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
679 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
681 EXPORT_SYMBOL(page_mapping);
684 * For file cache pages, return the address_space, otherwise return NULL
686 struct address_space *page_mapping_file(struct page *page)
688 if (unlikely(PageSwapCache(page)))
690 return page_mapping(page);
693 /* Slow path of page_mapcount() for compound pages */
694 int __page_mapcount(struct page *page)
698 ret = atomic_read(&page->_mapcount) + 1;
700 * For file THP page->_mapcount contains total number of mapping
701 * of the page: no need to look into compound_mapcount.
703 if (!PageAnon(page) && !PageHuge(page))
705 page = compound_head(page);
706 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
707 if (PageDoubleMap(page))
711 EXPORT_SYMBOL_GPL(__page_mapcount);
713 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
714 int sysctl_overcommit_ratio __read_mostly = 50;
715 unsigned long sysctl_overcommit_kbytes __read_mostly;
716 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
717 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
718 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
720 int overcommit_ratio_handler(struct ctl_table *table, int write,
721 void __user *buffer, size_t *lenp,
726 ret = proc_dointvec(table, write, buffer, lenp, ppos);
727 if (ret == 0 && write)
728 sysctl_overcommit_kbytes = 0;
732 int overcommit_kbytes_handler(struct ctl_table *table, int write,
733 void __user *buffer, size_t *lenp,
738 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
739 if (ret == 0 && write)
740 sysctl_overcommit_ratio = 0;
745 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
747 unsigned long vm_commit_limit(void)
749 unsigned long allowed;
751 if (sysctl_overcommit_kbytes)
752 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
754 allowed = ((totalram_pages() - hugetlb_total_pages())
755 * sysctl_overcommit_ratio / 100);
756 allowed += total_swap_pages;
762 * Make sure vm_committed_as in one cacheline and not cacheline shared with
763 * other variables. It can be updated by several CPUs frequently.
765 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
768 * The global memory commitment made in the system can be a metric
769 * that can be used to drive ballooning decisions when Linux is hosted
770 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
771 * balancing memory across competing virtual machines that are hosted.
772 * Several metrics drive this policy engine including the guest reported
775 unsigned long vm_memory_committed(void)
777 return percpu_counter_read_positive(&vm_committed_as);
779 EXPORT_SYMBOL_GPL(vm_memory_committed);
782 * Check that a process has enough memory to allocate a new virtual
783 * mapping. 0 means there is enough memory for the allocation to
784 * succeed and -ENOMEM implies there is not.
786 * We currently support three overcommit policies, which are set via the
787 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
789 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
790 * Additional code 2002 Jul 20 by Robert Love.
792 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
794 * Note this is a helper function intended to be used by LSMs which
795 * wish to use this logic.
797 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
801 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
802 -(s64)vm_committed_as_batch * num_online_cpus(),
803 "memory commitment underflow");
805 vm_acct_memory(pages);
808 * Sometimes we want to use more memory than we have
810 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
813 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
814 if (pages > totalram_pages() + total_swap_pages)
819 allowed = vm_commit_limit();
821 * Reserve some for root
824 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
827 * Don't let a single process grow so big a user can't recover
830 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
832 allowed -= min_t(long, mm->total_vm / 32, reserve);
835 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
838 vm_unacct_memory(pages);
844 * get_cmdline() - copy the cmdline value to a buffer.
845 * @task: the task whose cmdline value to copy.
846 * @buffer: the buffer to copy to.
847 * @buflen: the length of the buffer. Larger cmdline values are truncated
850 * Return: the size of the cmdline field copied. Note that the copy does
851 * not guarantee an ending NULL byte.
853 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
857 struct mm_struct *mm = get_task_mm(task);
858 unsigned long arg_start, arg_end, env_start, env_end;
862 goto out_mm; /* Shh! No looking before we're done */
864 spin_lock(&mm->arg_lock);
865 arg_start = mm->arg_start;
866 arg_end = mm->arg_end;
867 env_start = mm->env_start;
868 env_end = mm->env_end;
869 spin_unlock(&mm->arg_lock);
871 len = arg_end - arg_start;
876 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
879 * If the nul at the end of args has been overwritten, then
880 * assume application is using setproctitle(3).
882 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
883 len = strnlen(buffer, res);
887 len = env_end - env_start;
888 if (len > buflen - res)
890 res += access_process_vm(task, env_start,
893 res = strnlen(buffer, res);
902 int memcmp_pages(struct page *page1, struct page *page2)
907 addr1 = kmap_atomic(page1);
908 addr2 = kmap_atomic(page2);
909 ret = memcmp(addr1, addr2, PAGE_SIZE);
910 kunmap_atomic(addr2);
911 kunmap_atomic(addr1);