2 #include <linux/slab.h>
3 #include <linux/string.h>
4 #include <linux/compiler.h>
5 #include <linux/export.h>
7 #include <linux/sched.h>
8 #include <linux/sched/mm.h>
9 #include <linux/sched/task_stack.h>
10 #include <linux/security.h>
11 #include <linux/swap.h>
12 #include <linux/swapops.h>
13 #include <linux/mman.h>
14 #include <linux/hugetlb.h>
15 #include <linux/vmalloc.h>
16 #include <linux/userfaultfd_k.h>
17 #include <linux/random.h>
19 #include <asm/sections.h>
20 #include <linux/uaccess.h>
24 static inline int is_kernel_rodata(unsigned long addr)
26 return addr >= (unsigned long)__start_rodata &&
27 addr < (unsigned long)__end_rodata;
31 * kfree_const - conditionally free memory
32 * @x: pointer to the memory
34 * Function calls kfree only if @x is not in .rodata section.
36 void kfree_const(const void *x)
38 if (!is_kernel_rodata((unsigned long)x))
41 EXPORT_SYMBOL(kfree_const);
44 * kstrdup - allocate space for and copy an existing string
45 * @s: the string to duplicate
46 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
48 char *kstrdup(const char *s, gfp_t gfp)
57 buf = kmalloc_track_caller(len, gfp);
62 EXPORT_SYMBOL(kstrdup);
65 * kstrdup_const - conditionally duplicate an existing const string
66 * @s: the string to duplicate
67 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
69 * Function returns source string if it is in .rodata section otherwise it
70 * fallbacks to kstrdup.
71 * Strings allocated by kstrdup_const should be freed by kfree_const.
73 const char *kstrdup_const(const char *s, gfp_t gfp)
75 if (is_kernel_rodata((unsigned long)s))
78 return kstrdup(s, gfp);
80 EXPORT_SYMBOL(kstrdup_const);
83 * kstrndup - allocate space for and copy an existing string
84 * @s: the string to duplicate
85 * @max: read at most @max chars from @s
86 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
88 * Note: Use kmemdup_nul() instead if the size is known exactly.
90 char *kstrndup(const char *s, size_t max, gfp_t gfp)
98 len = strnlen(s, max);
99 buf = kmalloc_track_caller(len+1, gfp);
106 EXPORT_SYMBOL(kstrndup);
109 * kmemdup - duplicate region of memory
111 * @src: memory region to duplicate
112 * @len: memory region length
113 * @gfp: GFP mask to use
115 void *kmemdup(const void *src, size_t len, gfp_t gfp)
119 p = kmalloc_track_caller(len, gfp);
124 EXPORT_SYMBOL(kmemdup);
127 * kmemdup_nul - Create a NUL-terminated string from unterminated data
128 * @s: The data to stringify
129 * @len: The size of the data
130 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
132 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
139 buf = kmalloc_track_caller(len + 1, gfp);
146 EXPORT_SYMBOL(kmemdup_nul);
149 * memdup_user - duplicate memory region from user space
151 * @src: source address in user space
152 * @len: number of bytes to copy
154 * Returns an ERR_PTR() on failure.
156 void *memdup_user(const void __user *src, size_t len)
161 * Always use GFP_KERNEL, since copy_from_user() can sleep and
162 * cause pagefault, which makes it pointless to use GFP_NOFS
165 p = kmalloc_track_caller(len, GFP_KERNEL);
167 return ERR_PTR(-ENOMEM);
169 if (copy_from_user(p, src, len)) {
171 return ERR_PTR(-EFAULT);
176 EXPORT_SYMBOL(memdup_user);
179 * strndup_user - duplicate an existing string from user space
180 * @s: The string to duplicate
181 * @n: Maximum number of bytes to copy, including the trailing NUL.
183 char *strndup_user(const char __user *s, long n)
188 length = strnlen_user(s, n);
191 return ERR_PTR(-EFAULT);
194 return ERR_PTR(-EINVAL);
196 p = memdup_user(s, length);
201 p[length - 1] = '\0';
205 EXPORT_SYMBOL(strndup_user);
208 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
210 * @src: source address in user space
211 * @len: number of bytes to copy
213 * Returns an ERR_PTR() on failure.
215 void *memdup_user_nul(const void __user *src, size_t len)
220 * Always use GFP_KERNEL, since copy_from_user() can sleep and
221 * cause pagefault, which makes it pointless to use GFP_NOFS
224 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
226 return ERR_PTR(-ENOMEM);
228 if (copy_from_user(p, src, len)) {
230 return ERR_PTR(-EFAULT);
236 EXPORT_SYMBOL(memdup_user_nul);
238 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
239 struct vm_area_struct *prev, struct rb_node *rb_parent)
241 struct vm_area_struct *next;
245 next = prev->vm_next;
250 next = rb_entry(rb_parent,
251 struct vm_area_struct, vm_rb);
260 /* Check if the vma is being used as a stack by this task */
261 int vma_is_stack_for_current(struct vm_area_struct *vma)
263 struct task_struct * __maybe_unused t = current;
265 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
269 * randomize_page - Generate a random, page aligned address
270 * @start: The smallest acceptable address the caller will take.
271 * @range: The size of the area, starting at @start, within which the
272 * random address must fall.
274 * If @start + @range would overflow, @range is capped.
276 * NOTE: Historical use of randomize_range, which this replaces, presumed that
277 * @start was already page aligned. We now align it regardless.
279 * Return: A page aligned address within [start, start + range). On error,
280 * @start is returned.
282 unsigned long randomize_page(unsigned long start, unsigned long range)
284 if (!PAGE_ALIGNED(start)) {
285 range -= PAGE_ALIGN(start) - start;
286 start = PAGE_ALIGN(start);
289 if (start > ULONG_MAX - range)
290 range = ULONG_MAX - start;
292 range >>= PAGE_SHIFT;
297 return start + (get_random_long() % range << PAGE_SHIFT);
300 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
301 void arch_pick_mmap_layout(struct mm_struct *mm)
303 mm->mmap_base = TASK_UNMAPPED_BASE;
304 mm->get_unmapped_area = arch_get_unmapped_area;
309 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
310 * back to the regular GUP.
311 * If the architecture not support this function, simply return with no
314 int __weak __get_user_pages_fast(unsigned long start,
315 int nr_pages, int write, struct page **pages)
319 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
322 * get_user_pages_fast() - pin user pages in memory
323 * @start: starting user address
324 * @nr_pages: number of pages from start to pin
325 * @write: whether pages will be written to
326 * @pages: array that receives pointers to the pages pinned.
327 * Should be at least nr_pages long.
329 * Returns number of pages pinned. This may be fewer than the number
330 * requested. If nr_pages is 0 or negative, returns 0. If no pages
331 * were pinned, returns -errno.
333 * get_user_pages_fast provides equivalent functionality to get_user_pages,
334 * operating on current and current->mm, with force=0 and vma=NULL. However
335 * unlike get_user_pages, it must be called without mmap_sem held.
337 * get_user_pages_fast may take mmap_sem and page table locks, so no
338 * assumptions can be made about lack of locking. get_user_pages_fast is to be
339 * implemented in a way that is advantageous (vs get_user_pages()) when the
340 * user memory area is already faulted in and present in ptes. However if the
341 * pages have to be faulted in, it may turn out to be slightly slower so
342 * callers need to carefully consider what to use. On many architectures,
343 * get_user_pages_fast simply falls back to get_user_pages.
345 int __weak get_user_pages_fast(unsigned long start,
346 int nr_pages, int write, struct page **pages)
348 return get_user_pages_unlocked(start, nr_pages, pages,
349 write ? FOLL_WRITE : 0);
351 EXPORT_SYMBOL_GPL(get_user_pages_fast);
353 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
354 unsigned long len, unsigned long prot,
355 unsigned long flag, unsigned long pgoff)
358 struct mm_struct *mm = current->mm;
359 unsigned long populate;
362 ret = security_mmap_file(file, prot, flag);
364 if (down_write_killable(&mm->mmap_sem))
366 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
368 up_write(&mm->mmap_sem);
369 userfaultfd_unmap_complete(mm, &uf);
371 mm_populate(ret, populate);
376 unsigned long vm_mmap(struct file *file, unsigned long addr,
377 unsigned long len, unsigned long prot,
378 unsigned long flag, unsigned long offset)
380 if (unlikely(offset + PAGE_ALIGN(len) < offset))
382 if (unlikely(offset_in_page(offset)))
385 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
387 EXPORT_SYMBOL(vm_mmap);
390 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
391 * failure, fall back to non-contiguous (vmalloc) allocation.
392 * @size: size of the request.
393 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
394 * @node: numa node to allocate from
396 * Uses kmalloc to get the memory but if the allocation fails then falls back
397 * to the vmalloc allocator. Use kvfree for freeing the memory.
399 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
400 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
401 * preferable to the vmalloc fallback, due to visible performance drawbacks.
403 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
404 * fall back to vmalloc.
406 void *kvmalloc_node(size_t size, gfp_t flags, int node)
408 gfp_t kmalloc_flags = flags;
412 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
413 * so the given set of flags has to be compatible.
415 if ((flags & GFP_KERNEL) != GFP_KERNEL)
416 return kmalloc_node(size, flags, node);
419 * We want to attempt a large physically contiguous block first because
420 * it is less likely to fragment multiple larger blocks and therefore
421 * contribute to a long term fragmentation less than vmalloc fallback.
422 * However make sure that larger requests are not too disruptive - no
423 * OOM killer and no allocation failure warnings as we have a fallback.
425 if (size > PAGE_SIZE) {
426 kmalloc_flags |= __GFP_NOWARN;
428 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
429 kmalloc_flags |= __GFP_NORETRY;
432 ret = kmalloc_node(size, kmalloc_flags, node);
435 * It doesn't really make sense to fallback to vmalloc for sub page
438 if (ret || size <= PAGE_SIZE)
441 return __vmalloc_node_flags_caller(size, node, flags,
442 __builtin_return_address(0));
444 EXPORT_SYMBOL(kvmalloc_node);
446 void kvfree(const void *addr)
448 if (is_vmalloc_addr(addr))
453 EXPORT_SYMBOL(kvfree);
456 * kvfree_sensitive - Free a data object containing sensitive information.
457 * @addr: address of the data object to be freed.
458 * @len: length of the data object.
460 * Use the special memzero_explicit() function to clear the content of a
461 * kvmalloc'ed object containing sensitive data to make sure that the
462 * compiler won't optimize out the data clearing.
464 void kvfree_sensitive(const void *addr, size_t len)
466 if (likely(!ZERO_OR_NULL_PTR(addr))) {
467 memzero_explicit((void *)addr, len);
471 EXPORT_SYMBOL(kvfree_sensitive);
473 static inline void *__page_rmapping(struct page *page)
475 unsigned long mapping;
477 mapping = (unsigned long)page->mapping;
478 mapping &= ~PAGE_MAPPING_FLAGS;
480 return (void *)mapping;
483 /* Neutral page->mapping pointer to address_space or anon_vma or other */
484 void *page_rmapping(struct page *page)
486 page = compound_head(page);
487 return __page_rmapping(page);
491 * Return true if this page is mapped into pagetables.
492 * For compound page it returns true if any subpage of compound page is mapped.
494 bool page_mapped(struct page *page)
498 if (likely(!PageCompound(page)))
499 return atomic_read(&page->_mapcount) >= 0;
500 page = compound_head(page);
501 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
505 for (i = 0; i < (1 << compound_order(page)); i++) {
506 if (atomic_read(&page[i]._mapcount) >= 0)
511 EXPORT_SYMBOL(page_mapped);
513 struct anon_vma *page_anon_vma(struct page *page)
515 unsigned long mapping;
517 page = compound_head(page);
518 mapping = (unsigned long)page->mapping;
519 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
521 return __page_rmapping(page);
524 struct address_space *page_mapping(struct page *page)
526 struct address_space *mapping;
528 page = compound_head(page);
530 /* This happens if someone calls flush_dcache_page on slab page */
531 if (unlikely(PageSlab(page)))
534 if (unlikely(PageSwapCache(page))) {
537 entry.val = page_private(page);
538 return swap_address_space(entry);
541 mapping = page->mapping;
542 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
545 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
547 EXPORT_SYMBOL(page_mapping);
549 /* Slow path of page_mapcount() for compound pages */
550 int __page_mapcount(struct page *page)
554 ret = atomic_read(&page->_mapcount) + 1;
556 * For file THP page->_mapcount contains total number of mapping
557 * of the page: no need to look into compound_mapcount.
559 if (!PageAnon(page) && !PageHuge(page))
561 page = compound_head(page);
562 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
563 if (PageDoubleMap(page))
567 EXPORT_SYMBOL_GPL(__page_mapcount);
569 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
570 int sysctl_overcommit_ratio __read_mostly = 50;
571 unsigned long sysctl_overcommit_kbytes __read_mostly;
572 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
573 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
574 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
576 int overcommit_ratio_handler(struct ctl_table *table, int write,
577 void __user *buffer, size_t *lenp,
582 ret = proc_dointvec(table, write, buffer, lenp, ppos);
583 if (ret == 0 && write)
584 sysctl_overcommit_kbytes = 0;
588 int overcommit_kbytes_handler(struct ctl_table *table, int write,
589 void __user *buffer, size_t *lenp,
594 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
595 if (ret == 0 && write)
596 sysctl_overcommit_ratio = 0;
601 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
603 unsigned long vm_commit_limit(void)
605 unsigned long allowed;
607 if (sysctl_overcommit_kbytes)
608 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
610 allowed = ((totalram_pages - hugetlb_total_pages())
611 * sysctl_overcommit_ratio / 100);
612 allowed += total_swap_pages;
618 * Make sure vm_committed_as in one cacheline and not cacheline shared with
619 * other variables. It can be updated by several CPUs frequently.
621 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
624 * The global memory commitment made in the system can be a metric
625 * that can be used to drive ballooning decisions when Linux is hosted
626 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
627 * balancing memory across competing virtual machines that are hosted.
628 * Several metrics drive this policy engine including the guest reported
631 unsigned long vm_memory_committed(void)
633 return percpu_counter_read_positive(&vm_committed_as);
635 EXPORT_SYMBOL_GPL(vm_memory_committed);
638 * Check that a process has enough memory to allocate a new virtual
639 * mapping. 0 means there is enough memory for the allocation to
640 * succeed and -ENOMEM implies there is not.
642 * We currently support three overcommit policies, which are set via the
643 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
645 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
646 * Additional code 2002 Jul 20 by Robert Love.
648 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
650 * Note this is a helper function intended to be used by LSMs which
651 * wish to use this logic.
653 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
655 long free, allowed, reserve;
657 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
658 -(s64)vm_committed_as_batch * num_online_cpus(),
659 "memory commitment underflow");
661 vm_acct_memory(pages);
664 * Sometimes we want to use more memory than we have
666 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
669 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
670 free = global_zone_page_state(NR_FREE_PAGES);
671 free += global_node_page_state(NR_FILE_PAGES);
674 * shmem pages shouldn't be counted as free in this
675 * case, they can't be purged, only swapped out, and
676 * that won't affect the overall amount of available
677 * memory in the system.
679 free -= global_node_page_state(NR_SHMEM);
681 free += get_nr_swap_pages();
684 * Any slabs which are created with the
685 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
686 * which are reclaimable, under pressure. The dentry
687 * cache and most inode caches should fall into this
689 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
692 * Part of the kernel memory, which can be released
693 * under memory pressure.
695 free += global_node_page_state(
696 NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
699 * Leave reserved pages. The pages are not for anonymous pages.
701 if (free <= totalreserve_pages)
704 free -= totalreserve_pages;
707 * Reserve some for root
710 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
718 allowed = vm_commit_limit();
720 * Reserve some for root
723 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
726 * Don't let a single process grow so big a user can't recover
729 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
730 allowed -= min_t(long, mm->total_vm / 32, reserve);
733 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
736 vm_unacct_memory(pages);
742 * get_cmdline() - copy the cmdline value to a buffer.
743 * @task: the task whose cmdline value to copy.
744 * @buffer: the buffer to copy to.
745 * @buflen: the length of the buffer. Larger cmdline values are truncated
747 * Returns the size of the cmdline field copied. Note that the copy does
748 * not guarantee an ending NULL byte.
750 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
754 struct mm_struct *mm = get_task_mm(task);
755 unsigned long arg_start, arg_end, env_start, env_end;
759 goto out_mm; /* Shh! No looking before we're done */
761 down_read(&mm->mmap_sem);
762 arg_start = mm->arg_start;
763 arg_end = mm->arg_end;
764 env_start = mm->env_start;
765 env_end = mm->env_end;
766 up_read(&mm->mmap_sem);
768 len = arg_end - arg_start;
773 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
776 * If the nul at the end of args has been overwritten, then
777 * assume application is using setproctitle(3).
779 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
780 len = strnlen(buffer, res);
784 len = env_end - env_start;
785 if (len > buflen - res)
787 res += access_process_vm(task, env_start,
790 res = strnlen(buffer, res);