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>
18 #include <asm/sections.h>
19 #include <linux/uaccess.h>
23 static inline int is_kernel_rodata(unsigned long addr)
25 return addr >= (unsigned long)__start_rodata &&
26 addr < (unsigned long)__end_rodata;
30 * kfree_const - conditionally free memory
31 * @x: pointer to the memory
33 * Function calls kfree only if @x is not in .rodata section.
35 void kfree_const(const void *x)
37 if (!is_kernel_rodata((unsigned long)x))
40 EXPORT_SYMBOL(kfree_const);
43 * kstrdup - allocate space for and copy an existing string
44 * @s: the string to duplicate
45 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
47 char *kstrdup(const char *s, gfp_t gfp)
56 buf = kmalloc_track_caller(len, gfp);
61 EXPORT_SYMBOL(kstrdup);
64 * kstrdup_const - conditionally duplicate an existing const string
65 * @s: the string to duplicate
66 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
68 * Function returns source string if it is in .rodata section otherwise it
69 * fallbacks to kstrdup.
70 * Strings allocated by kstrdup_const should be freed by kfree_const.
72 const char *kstrdup_const(const char *s, gfp_t gfp)
74 if (is_kernel_rodata((unsigned long)s))
77 return kstrdup(s, gfp);
79 EXPORT_SYMBOL(kstrdup_const);
82 * kstrndup - allocate space for and copy an existing string
83 * @s: the string to duplicate
84 * @max: read at most @max chars from @s
85 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
87 * Note: Use kmemdup_nul() instead if the size is known exactly.
89 char *kstrndup(const char *s, size_t max, gfp_t gfp)
97 len = strnlen(s, max);
98 buf = kmalloc_track_caller(len+1, gfp);
105 EXPORT_SYMBOL(kstrndup);
108 * kmemdup - duplicate region of memory
110 * @src: memory region to duplicate
111 * @len: memory region length
112 * @gfp: GFP mask to use
114 void *kmemdup(const void *src, size_t len, gfp_t gfp)
118 p = kmalloc_track_caller(len, gfp);
123 EXPORT_SYMBOL(kmemdup);
126 * kmemdup_nul - Create a NUL-terminated string from unterminated data
127 * @s: The data to stringify
128 * @len: The size of the data
129 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
131 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
138 buf = kmalloc_track_caller(len + 1, gfp);
145 EXPORT_SYMBOL(kmemdup_nul);
148 * memdup_user - duplicate memory region from user space
150 * @src: source address in user space
151 * @len: number of bytes to copy
153 * Returns an ERR_PTR() on failure. Result is physically
154 * contiguous, to be freed by kfree().
156 void *memdup_user(const void __user *src, size_t len)
160 p = kmalloc_track_caller(len, GFP_USER);
162 return ERR_PTR(-ENOMEM);
164 if (copy_from_user(p, src, len)) {
166 return ERR_PTR(-EFAULT);
171 EXPORT_SYMBOL(memdup_user);
174 * vmemdup_user - duplicate memory region from user space
176 * @src: source address in user space
177 * @len: number of bytes to copy
179 * Returns an ERR_PTR() on failure. Result may be not
180 * physically contiguous. Use kvfree() to free.
182 void *vmemdup_user(const void __user *src, size_t len)
186 p = kvmalloc(len, GFP_USER);
188 return ERR_PTR(-ENOMEM);
190 if (copy_from_user(p, src, len)) {
192 return ERR_PTR(-EFAULT);
197 EXPORT_SYMBOL(vmemdup_user);
200 * strndup_user - duplicate an existing string from user space
201 * @s: The string to duplicate
202 * @n: Maximum number of bytes to copy, including the trailing NUL.
204 char *strndup_user(const char __user *s, long n)
209 length = strnlen_user(s, n);
212 return ERR_PTR(-EFAULT);
215 return ERR_PTR(-EINVAL);
217 p = memdup_user(s, length);
222 p[length - 1] = '\0';
226 EXPORT_SYMBOL(strndup_user);
229 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
231 * @src: source address in user space
232 * @len: number of bytes to copy
234 * Returns an ERR_PTR() on failure.
236 void *memdup_user_nul(const void __user *src, size_t len)
241 * Always use GFP_KERNEL, since copy_from_user() can sleep and
242 * cause pagefault, which makes it pointless to use GFP_NOFS
245 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
247 return ERR_PTR(-ENOMEM);
249 if (copy_from_user(p, src, len)) {
251 return ERR_PTR(-EFAULT);
257 EXPORT_SYMBOL(memdup_user_nul);
259 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
260 struct vm_area_struct *prev, struct rb_node *rb_parent)
262 struct vm_area_struct *next;
266 next = prev->vm_next;
271 next = rb_entry(rb_parent,
272 struct vm_area_struct, vm_rb);
281 /* Check if the vma is being used as a stack by this task */
282 int vma_is_stack_for_current(struct vm_area_struct *vma)
284 struct task_struct * __maybe_unused t = current;
286 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
289 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
290 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
292 mm->mmap_base = TASK_UNMAPPED_BASE;
293 mm->get_unmapped_area = arch_get_unmapped_area;
298 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
299 * back to the regular GUP.
300 * Note a difference with get_user_pages_fast: this always returns the
301 * number of pages pinned, 0 if no pages were pinned.
302 * If the architecture does not support this function, simply return with no
305 int __weak __get_user_pages_fast(unsigned long start,
306 int nr_pages, int write, struct page **pages)
310 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
313 * get_user_pages_fast() - pin user pages in memory
314 * @start: starting user address
315 * @nr_pages: number of pages from start to pin
316 * @write: whether pages will be written to
317 * @pages: array that receives pointers to the pages pinned.
318 * Should be at least nr_pages long.
320 * Returns number of pages pinned. This may be fewer than the number
321 * requested. If nr_pages is 0 or negative, returns 0. If no pages
322 * were pinned, returns -errno.
324 * get_user_pages_fast provides equivalent functionality to get_user_pages,
325 * operating on current and current->mm, with force=0 and vma=NULL. However
326 * unlike get_user_pages, it must be called without mmap_sem held.
328 * get_user_pages_fast may take mmap_sem and page table locks, so no
329 * assumptions can be made about lack of locking. get_user_pages_fast is to be
330 * implemented in a way that is advantageous (vs get_user_pages()) when the
331 * user memory area is already faulted in and present in ptes. However if the
332 * pages have to be faulted in, it may turn out to be slightly slower so
333 * callers need to carefully consider what to use. On many architectures,
334 * get_user_pages_fast simply falls back to get_user_pages.
336 int __weak get_user_pages_fast(unsigned long start,
337 int nr_pages, int write, struct page **pages)
339 return get_user_pages_unlocked(start, nr_pages, pages,
340 write ? FOLL_WRITE : 0);
342 EXPORT_SYMBOL_GPL(get_user_pages_fast);
344 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
345 unsigned long len, unsigned long prot,
346 unsigned long flag, unsigned long pgoff)
349 struct mm_struct *mm = current->mm;
350 unsigned long populate;
353 ret = security_mmap_file(file, prot, flag);
355 if (down_write_killable(&mm->mmap_sem))
357 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
359 up_write(&mm->mmap_sem);
360 userfaultfd_unmap_complete(mm, &uf);
362 mm_populate(ret, populate);
367 unsigned long vm_mmap(struct file *file, unsigned long addr,
368 unsigned long len, unsigned long prot,
369 unsigned long flag, unsigned long offset)
371 if (unlikely(offset + PAGE_ALIGN(len) < offset))
373 if (unlikely(offset_in_page(offset)))
376 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
378 EXPORT_SYMBOL(vm_mmap);
381 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
382 * failure, fall back to non-contiguous (vmalloc) allocation.
383 * @size: size of the request.
384 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
385 * @node: numa node to allocate from
387 * Uses kmalloc to get the memory but if the allocation fails then falls back
388 * to the vmalloc allocator. Use kvfree for freeing the memory.
390 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
391 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
392 * preferable to the vmalloc fallback, due to visible performance drawbacks.
394 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
395 * fall back to vmalloc.
397 void *kvmalloc_node(size_t size, gfp_t flags, int node)
399 gfp_t kmalloc_flags = flags;
403 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
404 * so the given set of flags has to be compatible.
406 if ((flags & GFP_KERNEL) != GFP_KERNEL)
407 return kmalloc_node(size, flags, node);
410 * We want to attempt a large physically contiguous block first because
411 * it is less likely to fragment multiple larger blocks and therefore
412 * contribute to a long term fragmentation less than vmalloc fallback.
413 * However make sure that larger requests are not too disruptive - no
414 * OOM killer and no allocation failure warnings as we have a fallback.
416 if (size > PAGE_SIZE) {
417 kmalloc_flags |= __GFP_NOWARN;
419 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
420 kmalloc_flags |= __GFP_NORETRY;
423 ret = kmalloc_node(size, kmalloc_flags, node);
426 * It doesn't really make sense to fallback to vmalloc for sub page
429 if (ret || size <= PAGE_SIZE)
432 return __vmalloc_node_flags_caller(size, node, flags,
433 __builtin_return_address(0));
435 EXPORT_SYMBOL(kvmalloc_node);
438 * kvfree - free memory allocated with kvmalloc
439 * @addr: pointer returned by kvmalloc
441 * If the memory is allocated from vmalloc area it is freed with vfree().
442 * Otherwise kfree() is used.
444 void kvfree(const void *addr)
446 if (is_vmalloc_addr(addr))
451 EXPORT_SYMBOL(kvfree);
453 static inline void *__page_rmapping(struct page *page)
455 unsigned long mapping;
457 mapping = (unsigned long)page->mapping;
458 mapping &= ~PAGE_MAPPING_FLAGS;
460 return (void *)mapping;
463 /* Neutral page->mapping pointer to address_space or anon_vma or other */
464 void *page_rmapping(struct page *page)
466 page = compound_head(page);
467 return __page_rmapping(page);
471 * Return true if this page is mapped into pagetables.
472 * For compound page it returns true if any subpage of compound page is mapped.
474 bool page_mapped(struct page *page)
478 if (likely(!PageCompound(page)))
479 return atomic_read(&page->_mapcount) >= 0;
480 page = compound_head(page);
481 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
485 for (i = 0; i < hpage_nr_pages(page); i++) {
486 if (atomic_read(&page[i]._mapcount) >= 0)
491 EXPORT_SYMBOL(page_mapped);
493 struct anon_vma *page_anon_vma(struct page *page)
495 unsigned long mapping;
497 page = compound_head(page);
498 mapping = (unsigned long)page->mapping;
499 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
501 return __page_rmapping(page);
504 struct address_space *page_mapping(struct page *page)
506 struct address_space *mapping;
508 page = compound_head(page);
510 /* This happens if someone calls flush_dcache_page on slab page */
511 if (unlikely(PageSlab(page)))
514 if (unlikely(PageSwapCache(page))) {
517 entry.val = page_private(page);
518 return swap_address_space(entry);
521 mapping = page->mapping;
522 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
525 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
527 EXPORT_SYMBOL(page_mapping);
530 * For file cache pages, return the address_space, otherwise return NULL
532 struct address_space *page_mapping_file(struct page *page)
534 if (unlikely(PageSwapCache(page)))
536 return page_mapping(page);
539 /* Slow path of page_mapcount() for compound pages */
540 int __page_mapcount(struct page *page)
544 ret = atomic_read(&page->_mapcount) + 1;
546 * For file THP page->_mapcount contains total number of mapping
547 * of the page: no need to look into compound_mapcount.
549 if (!PageAnon(page) && !PageHuge(page))
551 page = compound_head(page);
552 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
553 if (PageDoubleMap(page))
557 EXPORT_SYMBOL_GPL(__page_mapcount);
559 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
560 int sysctl_overcommit_ratio __read_mostly = 50;
561 unsigned long sysctl_overcommit_kbytes __read_mostly;
562 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
563 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
564 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
566 int overcommit_ratio_handler(struct ctl_table *table, int write,
567 void __user *buffer, size_t *lenp,
572 ret = proc_dointvec(table, write, buffer, lenp, ppos);
573 if (ret == 0 && write)
574 sysctl_overcommit_kbytes = 0;
578 int overcommit_kbytes_handler(struct ctl_table *table, int write,
579 void __user *buffer, size_t *lenp,
584 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
585 if (ret == 0 && write)
586 sysctl_overcommit_ratio = 0;
591 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
593 unsigned long vm_commit_limit(void)
595 unsigned long allowed;
597 if (sysctl_overcommit_kbytes)
598 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
600 allowed = ((totalram_pages - hugetlb_total_pages())
601 * sysctl_overcommit_ratio / 100);
602 allowed += total_swap_pages;
608 * Make sure vm_committed_as in one cacheline and not cacheline shared with
609 * other variables. It can be updated by several CPUs frequently.
611 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
614 * The global memory commitment made in the system can be a metric
615 * that can be used to drive ballooning decisions when Linux is hosted
616 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
617 * balancing memory across competing virtual machines that are hosted.
618 * Several metrics drive this policy engine including the guest reported
621 unsigned long vm_memory_committed(void)
623 return percpu_counter_read_positive(&vm_committed_as);
625 EXPORT_SYMBOL_GPL(vm_memory_committed);
628 * Check that a process has enough memory to allocate a new virtual
629 * mapping. 0 means there is enough memory for the allocation to
630 * succeed and -ENOMEM implies there is not.
632 * We currently support three overcommit policies, which are set via the
633 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
635 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
636 * Additional code 2002 Jul 20 by Robert Love.
638 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
640 * Note this is a helper function intended to be used by LSMs which
641 * wish to use this logic.
643 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
645 long free, allowed, reserve;
647 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
648 -(s64)vm_committed_as_batch * num_online_cpus(),
649 "memory commitment underflow");
651 vm_acct_memory(pages);
654 * Sometimes we want to use more memory than we have
656 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
659 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
660 free = global_zone_page_state(NR_FREE_PAGES);
661 free += global_node_page_state(NR_FILE_PAGES);
664 * shmem pages shouldn't be counted as free in this
665 * case, they can't be purged, only swapped out, and
666 * that won't affect the overall amount of available
667 * memory in the system.
669 free -= global_node_page_state(NR_SHMEM);
671 free += get_nr_swap_pages();
674 * Any slabs which are created with the
675 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
676 * which are reclaimable, under pressure. The dentry
677 * cache and most inode caches should fall into this
679 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
682 * Part of the kernel memory, which can be released
683 * under memory pressure.
685 free += global_node_page_state(
686 NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
689 * Leave reserved pages. The pages are not for anonymous pages.
691 if (free <= totalreserve_pages)
694 free -= totalreserve_pages;
697 * Reserve some for root
700 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
708 allowed = vm_commit_limit();
710 * Reserve some for root
713 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
716 * Don't let a single process grow so big a user can't recover
719 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
720 allowed -= min_t(long, mm->total_vm / 32, reserve);
723 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
726 vm_unacct_memory(pages);
732 * get_cmdline() - copy the cmdline value to a buffer.
733 * @task: the task whose cmdline value to copy.
734 * @buffer: the buffer to copy to.
735 * @buflen: the length of the buffer. Larger cmdline values are truncated
737 * Returns the size of the cmdline field copied. Note that the copy does
738 * not guarantee an ending NULL byte.
740 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
744 struct mm_struct *mm = get_task_mm(task);
745 unsigned long arg_start, arg_end, env_start, env_end;
749 goto out_mm; /* Shh! No looking before we're done */
751 down_read(&mm->mmap_sem);
752 arg_start = mm->arg_start;
753 arg_end = mm->arg_end;
754 env_start = mm->env_start;
755 env_end = mm->env_end;
756 up_read(&mm->mmap_sem);
758 len = arg_end - arg_start;
763 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
766 * If the nul at the end of args has been overwritten, then
767 * assume application is using setproctitle(3).
769 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
770 len = strnlen(buffer, res);
774 len = env_end - env_start;
775 if (len > buflen - res)
777 res += access_process_vm(task, env_start,
780 res = strnlen(buffer, res);