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, struct rb_node *rb_parent)
276 struct vm_area_struct *next;
280 next = prev->vm_next;
285 next = rb_entry(rb_parent,
286 struct vm_area_struct, vm_rb);
295 /* Check if the vma is being used as a stack by this task */
296 int vma_is_stack_for_current(struct vm_area_struct *vma)
298 struct task_struct * __maybe_unused t = current;
300 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
303 #ifndef STACK_RND_MASK
304 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
307 unsigned long randomize_stack_top(unsigned long stack_top)
309 unsigned long random_variable = 0;
311 if (current->flags & PF_RANDOMIZE) {
312 random_variable = get_random_long();
313 random_variable &= STACK_RND_MASK;
314 random_variable <<= PAGE_SHIFT;
316 #ifdef CONFIG_STACK_GROWSUP
317 return PAGE_ALIGN(stack_top) + random_variable;
319 return PAGE_ALIGN(stack_top) - random_variable;
323 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
324 #ifdef CONFIG_ARCH_HAS_ELF_RANDOMIZE
325 unsigned long arch_mmap_rnd(void)
329 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
330 if (is_compat_task())
331 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
333 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
334 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
336 return rnd << PAGE_SHIFT;
338 #endif /* CONFIG_ARCH_HAS_ELF_RANDOMIZE */
340 static int mmap_is_legacy(struct rlimit *rlim_stack)
342 if (current->personality & ADDR_COMPAT_LAYOUT)
345 if (rlim_stack->rlim_cur == RLIM_INFINITY)
348 return sysctl_legacy_va_layout;
352 * Leave enough space between the mmap area and the stack to honour ulimit in
353 * the face of randomisation.
355 #define MIN_GAP (SZ_128M)
356 #define MAX_GAP (STACK_TOP / 6 * 5)
358 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
360 unsigned long gap = rlim_stack->rlim_cur;
361 unsigned long pad = stack_guard_gap;
363 /* Account for stack randomization if necessary */
364 if (current->flags & PF_RANDOMIZE)
365 pad += (STACK_RND_MASK << PAGE_SHIFT);
367 /* Values close to RLIM_INFINITY can overflow. */
373 else if (gap > MAX_GAP)
376 return PAGE_ALIGN(STACK_TOP - gap - rnd);
379 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
381 unsigned long random_factor = 0UL;
383 if (current->flags & PF_RANDOMIZE)
384 random_factor = arch_mmap_rnd();
386 if (mmap_is_legacy(rlim_stack)) {
387 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
388 mm->get_unmapped_area = arch_get_unmapped_area;
390 mm->mmap_base = mmap_base(random_factor, rlim_stack);
391 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
394 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
395 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
397 mm->mmap_base = TASK_UNMAPPED_BASE;
398 mm->get_unmapped_area = arch_get_unmapped_area;
403 * __account_locked_vm - account locked pages to an mm's locked_vm
404 * @mm: mm to account against
405 * @pages: number of pages to account
406 * @inc: %true if @pages should be considered positive, %false if not
407 * @task: task used to check RLIMIT_MEMLOCK
408 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
410 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
411 * that mmap_sem is held as writer.
415 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
417 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
418 struct task_struct *task, bool bypass_rlim)
420 unsigned long locked_vm, limit;
423 lockdep_assert_held_write(&mm->mmap_sem);
425 locked_vm = mm->locked_vm;
428 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
429 if (locked_vm + pages > limit)
433 mm->locked_vm = locked_vm + pages;
435 WARN_ON_ONCE(pages > locked_vm);
436 mm->locked_vm = locked_vm - pages;
439 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
440 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
441 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
442 ret ? " - exceeded" : "");
446 EXPORT_SYMBOL_GPL(__account_locked_vm);
449 * account_locked_vm - account locked pages to an mm's locked_vm
450 * @mm: mm to account against, may be NULL
451 * @pages: number of pages to account
452 * @inc: %true if @pages should be considered positive, %false if not
454 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
457 * * 0 on success, or if mm is NULL
458 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
460 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
464 if (pages == 0 || !mm)
467 down_write(&mm->mmap_sem);
468 ret = __account_locked_vm(mm, pages, inc, current,
469 capable(CAP_IPC_LOCK));
470 up_write(&mm->mmap_sem);
474 EXPORT_SYMBOL_GPL(account_locked_vm);
476 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
477 unsigned long len, unsigned long prot,
478 unsigned long flag, unsigned long pgoff)
481 struct mm_struct *mm = current->mm;
482 unsigned long populate;
485 ret = security_mmap_file(file, prot, flag);
487 if (down_write_killable(&mm->mmap_sem))
489 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
491 up_write(&mm->mmap_sem);
492 userfaultfd_unmap_complete(mm, &uf);
494 mm_populate(ret, populate);
499 unsigned long vm_mmap(struct file *file, unsigned long addr,
500 unsigned long len, unsigned long prot,
501 unsigned long flag, unsigned long offset)
503 if (unlikely(offset + PAGE_ALIGN(len) < offset))
505 if (unlikely(offset_in_page(offset)))
508 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
510 EXPORT_SYMBOL(vm_mmap);
513 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
514 * failure, fall back to non-contiguous (vmalloc) allocation.
515 * @size: size of the request.
516 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
517 * @node: numa node to allocate from
519 * Uses kmalloc to get the memory but if the allocation fails then falls back
520 * to the vmalloc allocator. Use kvfree for freeing the memory.
522 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
523 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
524 * preferable to the vmalloc fallback, due to visible performance drawbacks.
526 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
527 * fall back to vmalloc.
529 * Return: pointer to the allocated memory of %NULL in case of failure
531 void *kvmalloc_node(size_t size, gfp_t flags, int node)
533 gfp_t kmalloc_flags = flags;
537 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
538 * so the given set of flags has to be compatible.
540 if ((flags & GFP_KERNEL) != GFP_KERNEL)
541 return kmalloc_node(size, flags, node);
544 * We want to attempt a large physically contiguous block first because
545 * it is less likely to fragment multiple larger blocks and therefore
546 * contribute to a long term fragmentation less than vmalloc fallback.
547 * However make sure that larger requests are not too disruptive - no
548 * OOM killer and no allocation failure warnings as we have a fallback.
550 if (size > PAGE_SIZE) {
551 kmalloc_flags |= __GFP_NOWARN;
553 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
554 kmalloc_flags |= __GFP_NORETRY;
557 ret = kmalloc_node(size, kmalloc_flags, node);
560 * It doesn't really make sense to fallback to vmalloc for sub page
563 if (ret || size <= PAGE_SIZE)
566 return __vmalloc_node_flags_caller(size, node, flags,
567 __builtin_return_address(0));
569 EXPORT_SYMBOL(kvmalloc_node);
572 * kvfree() - Free memory.
573 * @addr: Pointer to allocated memory.
575 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
576 * It is slightly more efficient to use kfree() or vfree() if you are certain
577 * that you know which one to use.
579 * Context: Either preemptible task context or not-NMI interrupt.
581 void kvfree(const void *addr)
583 if (is_vmalloc_addr(addr))
588 EXPORT_SYMBOL(kvfree);
590 static inline void *__page_rmapping(struct page *page)
592 unsigned long mapping;
594 mapping = (unsigned long)page->mapping;
595 mapping &= ~PAGE_MAPPING_FLAGS;
597 return (void *)mapping;
600 /* Neutral page->mapping pointer to address_space or anon_vma or other */
601 void *page_rmapping(struct page *page)
603 page = compound_head(page);
604 return __page_rmapping(page);
608 * Return true if this page is mapped into pagetables.
609 * For compound page it returns true if any subpage of compound page is mapped.
611 bool page_mapped(struct page *page)
615 if (likely(!PageCompound(page)))
616 return atomic_read(&page->_mapcount) >= 0;
617 page = compound_head(page);
618 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
622 for (i = 0; i < compound_nr(page); i++) {
623 if (atomic_read(&page[i]._mapcount) >= 0)
628 EXPORT_SYMBOL(page_mapped);
630 struct anon_vma *page_anon_vma(struct page *page)
632 unsigned long mapping;
634 page = compound_head(page);
635 mapping = (unsigned long)page->mapping;
636 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
638 return __page_rmapping(page);
641 struct address_space *page_mapping(struct page *page)
643 struct address_space *mapping;
645 page = compound_head(page);
647 /* This happens if someone calls flush_dcache_page on slab page */
648 if (unlikely(PageSlab(page)))
651 if (unlikely(PageSwapCache(page))) {
654 entry.val = page_private(page);
655 return swap_address_space(entry);
658 mapping = page->mapping;
659 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
662 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
664 EXPORT_SYMBOL(page_mapping);
667 * For file cache pages, return the address_space, otherwise return NULL
669 struct address_space *page_mapping_file(struct page *page)
671 if (unlikely(PageSwapCache(page)))
673 return page_mapping(page);
676 /* Slow path of page_mapcount() for compound pages */
677 int __page_mapcount(struct page *page)
681 ret = atomic_read(&page->_mapcount) + 1;
683 * For file THP page->_mapcount contains total number of mapping
684 * of the page: no need to look into compound_mapcount.
686 if (!PageAnon(page) && !PageHuge(page))
688 page = compound_head(page);
689 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
690 if (PageDoubleMap(page))
694 EXPORT_SYMBOL_GPL(__page_mapcount);
696 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
697 int sysctl_overcommit_ratio __read_mostly = 50;
698 unsigned long sysctl_overcommit_kbytes __read_mostly;
699 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
700 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
701 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
703 int overcommit_ratio_handler(struct ctl_table *table, int write,
704 void __user *buffer, size_t *lenp,
709 ret = proc_dointvec(table, write, buffer, lenp, ppos);
710 if (ret == 0 && write)
711 sysctl_overcommit_kbytes = 0;
715 int overcommit_kbytes_handler(struct ctl_table *table, int write,
716 void __user *buffer, size_t *lenp,
721 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
722 if (ret == 0 && write)
723 sysctl_overcommit_ratio = 0;
728 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
730 unsigned long vm_commit_limit(void)
732 unsigned long allowed;
734 if (sysctl_overcommit_kbytes)
735 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
737 allowed = ((totalram_pages() - hugetlb_total_pages())
738 * sysctl_overcommit_ratio / 100);
739 allowed += total_swap_pages;
745 * Make sure vm_committed_as in one cacheline and not cacheline shared with
746 * other variables. It can be updated by several CPUs frequently.
748 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
751 * The global memory commitment made in the system can be a metric
752 * that can be used to drive ballooning decisions when Linux is hosted
753 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
754 * balancing memory across competing virtual machines that are hosted.
755 * Several metrics drive this policy engine including the guest reported
758 unsigned long vm_memory_committed(void)
760 return percpu_counter_read_positive(&vm_committed_as);
762 EXPORT_SYMBOL_GPL(vm_memory_committed);
765 * Check that a process has enough memory to allocate a new virtual
766 * mapping. 0 means there is enough memory for the allocation to
767 * succeed and -ENOMEM implies there is not.
769 * We currently support three overcommit policies, which are set via the
770 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
772 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
773 * Additional code 2002 Jul 20 by Robert Love.
775 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
777 * Note this is a helper function intended to be used by LSMs which
778 * wish to use this logic.
780 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
784 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
785 -(s64)vm_committed_as_batch * num_online_cpus(),
786 "memory commitment underflow");
788 vm_acct_memory(pages);
791 * Sometimes we want to use more memory than we have
793 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
796 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
797 if (pages > totalram_pages() + total_swap_pages)
802 allowed = vm_commit_limit();
804 * Reserve some for root
807 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
810 * Don't let a single process grow so big a user can't recover
813 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
815 allowed -= min_t(long, mm->total_vm / 32, reserve);
818 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
821 vm_unacct_memory(pages);
827 * get_cmdline() - copy the cmdline value to a buffer.
828 * @task: the task whose cmdline value to copy.
829 * @buffer: the buffer to copy to.
830 * @buflen: the length of the buffer. Larger cmdline values are truncated
833 * Return: the size of the cmdline field copied. Note that the copy does
834 * not guarantee an ending NULL byte.
836 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
840 struct mm_struct *mm = get_task_mm(task);
841 unsigned long arg_start, arg_end, env_start, env_end;
845 goto out_mm; /* Shh! No looking before we're done */
847 spin_lock(&mm->arg_lock);
848 arg_start = mm->arg_start;
849 arg_end = mm->arg_end;
850 env_start = mm->env_start;
851 env_end = mm->env_end;
852 spin_unlock(&mm->arg_lock);
854 len = arg_end - arg_start;
859 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
862 * If the nul at the end of args has been overwritten, then
863 * assume application is using setproctitle(3).
865 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
866 len = strnlen(buffer, res);
870 len = env_end - env_start;
871 if (len > buflen - res)
873 res += access_process_vm(task, env_start,
876 res = strnlen(buffer, res);
885 int memcmp_pages(struct page *page1, struct page *page2)
890 addr1 = kmap_atomic(page1);
891 addr2 = kmap_atomic(page2);
892 ret = memcmp(addr1, addr2, PAGE_SIZE);
893 kunmap_atomic(addr2);
894 kunmap_atomic(addr1);