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 <linux/uaccess.h>
23 * kfree_const - conditionally free memory
24 * @x: pointer to the memory
26 * Function calls kfree only if @x is not in .rodata section.
28 void kfree_const(const void *x)
30 if (!is_kernel_rodata((unsigned long)x))
33 EXPORT_SYMBOL(kfree_const);
36 * kstrdup - allocate space for and copy an existing string
37 * @s: the string to duplicate
38 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
40 char *kstrdup(const char *s, gfp_t gfp)
49 buf = kmalloc_track_caller(len, gfp);
54 EXPORT_SYMBOL(kstrdup);
57 * kstrdup_const - conditionally duplicate an existing const string
58 * @s: the string to duplicate
59 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
61 * Function returns source string if it is in .rodata section otherwise it
62 * fallbacks to kstrdup.
63 * Strings allocated by kstrdup_const should be freed by kfree_const.
65 const char *kstrdup_const(const char *s, gfp_t gfp)
67 if (is_kernel_rodata((unsigned long)s))
70 return kstrdup(s, gfp);
72 EXPORT_SYMBOL(kstrdup_const);
75 * kstrndup - allocate space for and copy an existing string
76 * @s: the string to duplicate
77 * @max: read at most @max chars from @s
78 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
80 * Note: Use kmemdup_nul() instead if the size is known exactly.
82 char *kstrndup(const char *s, size_t max, gfp_t gfp)
90 len = strnlen(s, max);
91 buf = kmalloc_track_caller(len+1, gfp);
98 EXPORT_SYMBOL(kstrndup);
101 * kmemdup - duplicate region of memory
103 * @src: memory region to duplicate
104 * @len: memory region length
105 * @gfp: GFP mask to use
107 void *kmemdup(const void *src, size_t len, gfp_t gfp)
111 p = kmalloc_track_caller(len, gfp);
116 EXPORT_SYMBOL(kmemdup);
119 * kmemdup_nul - Create a NUL-terminated string from unterminated data
120 * @s: The data to stringify
121 * @len: The size of the data
122 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
124 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
131 buf = kmalloc_track_caller(len + 1, gfp);
138 EXPORT_SYMBOL(kmemdup_nul);
141 * memdup_user - duplicate memory region from user space
143 * @src: source address in user space
144 * @len: number of bytes to copy
146 * Returns an ERR_PTR() on failure. Result is physically
147 * contiguous, to be freed by kfree().
149 void *memdup_user(const void __user *src, size_t len)
153 p = kmalloc_track_caller(len, GFP_USER);
155 return ERR_PTR(-ENOMEM);
157 if (copy_from_user(p, src, len)) {
159 return ERR_PTR(-EFAULT);
164 EXPORT_SYMBOL(memdup_user);
167 * vmemdup_user - duplicate memory region from user space
169 * @src: source address in user space
170 * @len: number of bytes to copy
172 * Returns an ERR_PTR() on failure. Result may be not
173 * physically contiguous. Use kvfree() to free.
175 void *vmemdup_user(const void __user *src, size_t len)
179 p = kvmalloc(len, GFP_USER);
181 return ERR_PTR(-ENOMEM);
183 if (copy_from_user(p, src, len)) {
185 return ERR_PTR(-EFAULT);
190 EXPORT_SYMBOL(vmemdup_user);
193 * strndup_user - duplicate an existing string from user space
194 * @s: The string to duplicate
195 * @n: Maximum number of bytes to copy, including the trailing NUL.
197 char *strndup_user(const char __user *s, long n)
202 length = strnlen_user(s, n);
205 return ERR_PTR(-EFAULT);
208 return ERR_PTR(-EINVAL);
210 p = memdup_user(s, length);
215 p[length - 1] = '\0';
219 EXPORT_SYMBOL(strndup_user);
222 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
224 * @src: source address in user space
225 * @len: number of bytes to copy
227 * Returns an ERR_PTR() on failure.
229 void *memdup_user_nul(const void __user *src, size_t len)
234 * Always use GFP_KERNEL, since copy_from_user() can sleep and
235 * cause pagefault, which makes it pointless to use GFP_NOFS
238 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
240 return ERR_PTR(-ENOMEM);
242 if (copy_from_user(p, src, len)) {
244 return ERR_PTR(-EFAULT);
250 EXPORT_SYMBOL(memdup_user_nul);
252 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
253 struct vm_area_struct *prev, struct rb_node *rb_parent)
255 struct vm_area_struct *next;
259 next = prev->vm_next;
264 next = rb_entry(rb_parent,
265 struct vm_area_struct, vm_rb);
274 /* Check if the vma is being used as a stack by this task */
275 int vma_is_stack_for_current(struct vm_area_struct *vma)
277 struct task_struct * __maybe_unused t = current;
279 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
282 #if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
283 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
285 mm->mmap_base = TASK_UNMAPPED_BASE;
286 mm->get_unmapped_area = arch_get_unmapped_area;
291 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
292 * back to the regular GUP.
293 * Note a difference with get_user_pages_fast: this always returns the
294 * number of pages pinned, 0 if no pages were pinned.
295 * If the architecture does not support this function, simply return with no
298 int __weak __get_user_pages_fast(unsigned long start,
299 int nr_pages, int write, struct page **pages)
303 EXPORT_SYMBOL_GPL(__get_user_pages_fast);
306 * get_user_pages_fast() - pin user pages in memory
307 * @start: starting user address
308 * @nr_pages: number of pages from start to pin
309 * @write: whether pages will be written to
310 * @pages: array that receives pointers to the pages pinned.
311 * Should be at least nr_pages long.
313 * Returns number of pages pinned. This may be fewer than the number
314 * requested. If nr_pages is 0 or negative, returns 0. If no pages
315 * were pinned, returns -errno.
317 * get_user_pages_fast provides equivalent functionality to get_user_pages,
318 * operating on current and current->mm, with force=0 and vma=NULL. However
319 * unlike get_user_pages, it must be called without mmap_sem held.
321 * get_user_pages_fast may take mmap_sem and page table locks, so no
322 * assumptions can be made about lack of locking. get_user_pages_fast is to be
323 * implemented in a way that is advantageous (vs get_user_pages()) when the
324 * user memory area is already faulted in and present in ptes. However if the
325 * pages have to be faulted in, it may turn out to be slightly slower so
326 * callers need to carefully consider what to use. On many architectures,
327 * get_user_pages_fast simply falls back to get_user_pages.
329 int __weak get_user_pages_fast(unsigned long start,
330 int nr_pages, int write, struct page **pages)
332 return get_user_pages_unlocked(start, nr_pages, pages,
333 write ? FOLL_WRITE : 0);
335 EXPORT_SYMBOL_GPL(get_user_pages_fast);
337 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
338 unsigned long len, unsigned long prot,
339 unsigned long flag, unsigned long pgoff)
342 struct mm_struct *mm = current->mm;
343 unsigned long populate;
346 ret = security_mmap_file(file, prot, flag);
348 if (down_write_killable(&mm->mmap_sem))
350 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
352 up_write(&mm->mmap_sem);
353 userfaultfd_unmap_complete(mm, &uf);
355 mm_populate(ret, populate);
360 unsigned long vm_mmap(struct file *file, unsigned long addr,
361 unsigned long len, unsigned long prot,
362 unsigned long flag, unsigned long offset)
364 if (unlikely(offset + PAGE_ALIGN(len) < offset))
366 if (unlikely(offset_in_page(offset)))
369 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
371 EXPORT_SYMBOL(vm_mmap);
374 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
375 * failure, fall back to non-contiguous (vmalloc) allocation.
376 * @size: size of the request.
377 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
378 * @node: numa node to allocate from
380 * Uses kmalloc to get the memory but if the allocation fails then falls back
381 * to the vmalloc allocator. Use kvfree for freeing the memory.
383 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
384 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
385 * preferable to the vmalloc fallback, due to visible performance drawbacks.
387 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
388 * fall back to vmalloc.
390 void *kvmalloc_node(size_t size, gfp_t flags, int node)
392 gfp_t kmalloc_flags = flags;
396 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
397 * so the given set of flags has to be compatible.
399 if ((flags & GFP_KERNEL) != GFP_KERNEL)
400 return kmalloc_node(size, flags, node);
403 * We want to attempt a large physically contiguous block first because
404 * it is less likely to fragment multiple larger blocks and therefore
405 * contribute to a long term fragmentation less than vmalloc fallback.
406 * However make sure that larger requests are not too disruptive - no
407 * OOM killer and no allocation failure warnings as we have a fallback.
409 if (size > PAGE_SIZE) {
410 kmalloc_flags |= __GFP_NOWARN;
412 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
413 kmalloc_flags |= __GFP_NORETRY;
416 ret = kmalloc_node(size, kmalloc_flags, node);
419 * It doesn't really make sense to fallback to vmalloc for sub page
422 if (ret || size <= PAGE_SIZE)
425 return __vmalloc_node_flags_caller(size, node, flags,
426 __builtin_return_address(0));
428 EXPORT_SYMBOL(kvmalloc_node);
431 * kvfree() - Free memory.
432 * @addr: Pointer to allocated memory.
434 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
435 * It is slightly more efficient to use kfree() or vfree() if you are certain
436 * that you know which one to use.
438 * Context: Either preemptible task context or not-NMI interrupt.
440 void kvfree(const void *addr)
442 if (is_vmalloc_addr(addr))
447 EXPORT_SYMBOL(kvfree);
449 static inline void *__page_rmapping(struct page *page)
451 unsigned long mapping;
453 mapping = (unsigned long)page->mapping;
454 mapping &= ~PAGE_MAPPING_FLAGS;
456 return (void *)mapping;
459 /* Neutral page->mapping pointer to address_space or anon_vma or other */
460 void *page_rmapping(struct page *page)
462 page = compound_head(page);
463 return __page_rmapping(page);
467 * Return true if this page is mapped into pagetables.
468 * For compound page it returns true if any subpage of compound page is mapped.
470 bool page_mapped(struct page *page)
474 if (likely(!PageCompound(page)))
475 return atomic_read(&page->_mapcount) >= 0;
476 page = compound_head(page);
477 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
481 for (i = 0; i < (1 << compound_order(page)); i++) {
482 if (atomic_read(&page[i]._mapcount) >= 0)
487 EXPORT_SYMBOL(page_mapped);
489 struct anon_vma *page_anon_vma(struct page *page)
491 unsigned long mapping;
493 page = compound_head(page);
494 mapping = (unsigned long)page->mapping;
495 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
497 return __page_rmapping(page);
500 struct address_space *page_mapping(struct page *page)
502 struct address_space *mapping;
504 page = compound_head(page);
506 /* This happens if someone calls flush_dcache_page on slab page */
507 if (unlikely(PageSlab(page)))
510 if (unlikely(PageSwapCache(page))) {
513 entry.val = page_private(page);
514 return swap_address_space(entry);
517 mapping = page->mapping;
518 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
521 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
523 EXPORT_SYMBOL(page_mapping);
526 * For file cache pages, return the address_space, otherwise return NULL
528 struct address_space *page_mapping_file(struct page *page)
530 if (unlikely(PageSwapCache(page)))
532 return page_mapping(page);
535 /* Slow path of page_mapcount() for compound pages */
536 int __page_mapcount(struct page *page)
540 ret = atomic_read(&page->_mapcount) + 1;
542 * For file THP page->_mapcount contains total number of mapping
543 * of the page: no need to look into compound_mapcount.
545 if (!PageAnon(page) && !PageHuge(page))
547 page = compound_head(page);
548 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
549 if (PageDoubleMap(page))
553 EXPORT_SYMBOL_GPL(__page_mapcount);
555 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
556 int sysctl_overcommit_ratio __read_mostly = 50;
557 unsigned long sysctl_overcommit_kbytes __read_mostly;
558 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
559 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
560 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
562 int overcommit_ratio_handler(struct ctl_table *table, int write,
563 void __user *buffer, size_t *lenp,
568 ret = proc_dointvec(table, write, buffer, lenp, ppos);
569 if (ret == 0 && write)
570 sysctl_overcommit_kbytes = 0;
574 int overcommit_kbytes_handler(struct ctl_table *table, int write,
575 void __user *buffer, size_t *lenp,
580 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
581 if (ret == 0 && write)
582 sysctl_overcommit_ratio = 0;
587 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
589 unsigned long vm_commit_limit(void)
591 unsigned long allowed;
593 if (sysctl_overcommit_kbytes)
594 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
596 allowed = ((totalram_pages() - hugetlb_total_pages())
597 * sysctl_overcommit_ratio / 100);
598 allowed += total_swap_pages;
604 * Make sure vm_committed_as in one cacheline and not cacheline shared with
605 * other variables. It can be updated by several CPUs frequently.
607 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
610 * The global memory commitment made in the system can be a metric
611 * that can be used to drive ballooning decisions when Linux is hosted
612 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
613 * balancing memory across competing virtual machines that are hosted.
614 * Several metrics drive this policy engine including the guest reported
617 unsigned long vm_memory_committed(void)
619 return percpu_counter_read_positive(&vm_committed_as);
621 EXPORT_SYMBOL_GPL(vm_memory_committed);
624 * Check that a process has enough memory to allocate a new virtual
625 * mapping. 0 means there is enough memory for the allocation to
626 * succeed and -ENOMEM implies there is not.
628 * We currently support three overcommit policies, which are set via the
629 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
631 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
632 * Additional code 2002 Jul 20 by Robert Love.
634 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
636 * Note this is a helper function intended to be used by LSMs which
637 * wish to use this logic.
639 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
641 long free, allowed, reserve;
643 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
644 -(s64)vm_committed_as_batch * num_online_cpus(),
645 "memory commitment underflow");
647 vm_acct_memory(pages);
650 * Sometimes we want to use more memory than we have
652 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
655 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
656 free = global_zone_page_state(NR_FREE_PAGES);
657 free += global_node_page_state(NR_FILE_PAGES);
660 * shmem pages shouldn't be counted as free in this
661 * case, they can't be purged, only swapped out, and
662 * that won't affect the overall amount of available
663 * memory in the system.
665 free -= global_node_page_state(NR_SHMEM);
667 free += get_nr_swap_pages();
670 * Any slabs which are created with the
671 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
672 * which are reclaimable, under pressure. The dentry
673 * cache and most inode caches should fall into this
675 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
678 * Part of the kernel memory, which can be released
679 * under memory pressure.
681 free += global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
684 * Leave reserved pages. The pages are not for anonymous pages.
686 if (free <= totalreserve_pages)
689 free -= totalreserve_pages;
692 * Reserve some for root
695 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
703 allowed = vm_commit_limit();
705 * Reserve some for root
708 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
711 * Don't let a single process grow so big a user can't recover
714 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
715 allowed -= min_t(long, mm->total_vm / 32, reserve);
718 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
721 vm_unacct_memory(pages);
727 * get_cmdline() - copy the cmdline value to a buffer.
728 * @task: the task whose cmdline value to copy.
729 * @buffer: the buffer to copy to.
730 * @buflen: the length of the buffer. Larger cmdline values are truncated
732 * Returns the size of the cmdline field copied. Note that the copy does
733 * not guarantee an ending NULL byte.
735 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
739 struct mm_struct *mm = get_task_mm(task);
740 unsigned long arg_start, arg_end, env_start, env_end;
744 goto out_mm; /* Shh! No looking before we're done */
746 down_read(&mm->mmap_sem);
747 arg_start = mm->arg_start;
748 arg_end = mm->arg_end;
749 env_start = mm->env_start;
750 env_end = mm->env_end;
751 up_read(&mm->mmap_sem);
753 len = arg_end - arg_start;
758 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
761 * If the nul at the end of args has been overwritten, then
762 * assume application is using setproctitle(3).
764 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
765 len = strnlen(buffer, res);
769 len = env_end - env_start;
770 if (len > buflen - res)
772 res += access_process_vm(task, env_start,
775 res = strnlen(buffer, res);