1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/mm.h>
3 #include <linux/slab.h>
4 #include <linux/string.h>
5 #include <linux/compiler.h>
6 #include <linux/export.h>
7 #include <linux/err.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>
26
27 #include <linux/uaccess.h>
28
29 #include "internal.h"
30
31 /**
32 * kfree_const - conditionally free memory
33 * @x: pointer to the memory
34 *
35 * Function calls kfree only if @x is not in .rodata section.
36 */
37 void kfree_const(const void *x)
38 {
39 if (!is_kernel_rodata((unsigned long)x))
40 kfree(x);
41 }
42 EXPORT_SYMBOL(kfree_const);
43
44 /**
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
48 *
49 * Return: newly allocated copy of @s or %NULL in case of error
50 */
51 char *kstrdup(const char *s, gfp_t gfp)
52 {
53 size_t len;
54 char *buf;
55
56 if (!s)
57 return NULL;
58
59 len = strlen(s) + 1;
60 buf = kmalloc_track_caller(len, gfp);
61 if (buf)
62 memcpy(buf, s, len);
63 return buf;
64 }
65 EXPORT_SYMBOL(kstrdup);
66
67 /**
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
71 *
72 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
73 * must not be passed to krealloc().
74 *
75 * Return: source string if it is in .rodata section otherwise
76 * fallback to kstrdup.
77 */
78 const char *kstrdup_const(const char *s, gfp_t gfp)
79 {
80 if (is_kernel_rodata((unsigned long)s))
81 return s;
82
83 return kstrdup(s, gfp);
84 }
85 EXPORT_SYMBOL(kstrdup_const);
86
87 /**
88 * kstrndup - allocate space for and copy an existing string
89 * @s: the string to duplicate
90 * @max: read at most @max chars from @s
91 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
92 *
93 * Note: Use kmemdup_nul() instead if the size is known exactly.
94 *
95 * Return: newly allocated copy of @s or %NULL in case of error
96 */
97 char *kstrndup(const char *s, size_t max, gfp_t gfp)
98 {
99 size_t len;
100 char *buf;
101
102 if (!s)
103 return NULL;
104
105 len = strnlen(s, max);
106 buf = kmalloc_track_caller(len+1, gfp);
107 if (buf) {
108 memcpy(buf, s, len);
109 buf[len] = '\0';
110 }
111 return buf;
112 }
113 EXPORT_SYMBOL(kstrndup);
114
115 /**
116 * kmemdup - duplicate region of memory
117 *
118 * @src: memory region to duplicate
119 * @len: memory region length
120 * @gfp: GFP mask to use
121 *
122 * Return: newly allocated copy of @src or %NULL in case of error
123 */
124 void *kmemdup(const void *src, size_t len, gfp_t gfp)
125 {
126 void *p;
127
128 p = kmalloc_track_caller(len, gfp);
129 if (p)
130 memcpy(p, src, len);
131 return p;
132 }
133 EXPORT_SYMBOL(kmemdup);
134
135 /**
136 * kmemdup_nul - Create a NUL-terminated string from unterminated data
137 * @s: The data to stringify
138 * @len: The size of the data
139 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
140 *
141 * Return: newly allocated copy of @s with NUL-termination or %NULL in
142 * case of error
143 */
144 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
145 {
146 char *buf;
147
148 if (!s)
149 return NULL;
150
151 buf = kmalloc_track_caller(len + 1, gfp);
152 if (buf) {
153 memcpy(buf, s, len);
154 buf[len] = '\0';
155 }
156 return buf;
157 }
158 EXPORT_SYMBOL(kmemdup_nul);
159
160 /**
161 * memdup_user - duplicate memory region from user space
162 *
163 * @src: source address in user space
164 * @len: number of bytes to copy
165 *
166 * Return: an ERR_PTR() on failure. Result is physically
167 * contiguous, to be freed by kfree().
168 */
169 void *memdup_user(const void __user *src, size_t len)
170 {
171 void *p;
172
173 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
174 if (!p)
175 return ERR_PTR(-ENOMEM);
176
177 if (copy_from_user(p, src, len)) {
178 kfree(p);
179 return ERR_PTR(-EFAULT);
180 }
181
182 return p;
183 }
184 EXPORT_SYMBOL(memdup_user);
185
186 /**
187 * vmemdup_user - duplicate memory region from user space
188 *
189 * @src: source address in user space
190 * @len: number of bytes to copy
191 *
192 * Return: an ERR_PTR() on failure. Result may be not
193 * physically contiguous. Use kvfree() to free.
194 */
195 void *vmemdup_user(const void __user *src, size_t len)
196 {
197 void *p;
198
199 p = kvmalloc(len, GFP_USER);
200 if (!p)
201 return ERR_PTR(-ENOMEM);
202
203 if (copy_from_user(p, src, len)) {
204 kvfree(p);
205 return ERR_PTR(-EFAULT);
206 }
207
208 return p;
209 }
210 EXPORT_SYMBOL(vmemdup_user);
211
212 /**
213 * strndup_user - duplicate an existing string from user space
214 * @s: The string to duplicate
215 * @n: Maximum number of bytes to copy, including the trailing NUL.
216 *
217 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
218 */
219 char *strndup_user(const char __user *s, long n)
220 {
221 char *p;
222 long length;
223
224 length = strnlen_user(s, n);
225
226 if (!length)
227 return ERR_PTR(-EFAULT);
228
229 if (length > n)
230 return ERR_PTR(-EINVAL);
231
232 p = memdup_user(s, length);
233
234 if (IS_ERR(p))
235 return p;
236
237 p[length - 1] = '\0';
238
239 return p;
240 }
241 EXPORT_SYMBOL(strndup_user);
242
243 /**
244 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
245 *
246 * @src: source address in user space
247 * @len: number of bytes to copy
248 *
249 * Return: an ERR_PTR() on failure.
250 */
251 void *memdup_user_nul(const void __user *src, size_t len)
252 {
253 char *p;
254
255 /*
256 * Always use GFP_KERNEL, since copy_from_user() can sleep and
257 * cause pagefault, which makes it pointless to use GFP_NOFS
258 * or GFP_ATOMIC.
259 */
260 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
261 if (!p)
262 return ERR_PTR(-ENOMEM);
263
264 if (copy_from_user(p, src, len)) {
265 kfree(p);
266 return ERR_PTR(-EFAULT);
267 }
268 p[len] = '\0';
269
270 return p;
271 }
272 EXPORT_SYMBOL(memdup_user_nul);
273
274 void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
275 struct vm_area_struct *prev)
276 {
277 struct vm_area_struct *next;
278
279 vma->vm_prev = prev;
280 if (prev) {
281 next = prev->vm_next;
282 prev->vm_next = vma;
283 } else {
284 next = mm->mmap;
285 mm->mmap = vma;
286 }
287 vma->vm_next = next;
288 if (next)
289 next->vm_prev = vma;
290 }
291
292 void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma)
293 {
294 struct vm_area_struct *prev, *next;
295
296 next = vma->vm_next;
297 prev = vma->vm_prev;
298 if (prev)
299 prev->vm_next = next;
300 else
301 mm->mmap = next;
302 if (next)
303 next->vm_prev = prev;
304 }
305
306 /* Check if the vma is being used as a stack by this task */
307 int vma_is_stack_for_current(struct vm_area_struct *vma)
308 {
309 struct task_struct * __maybe_unused t = current;
310
311 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
312 }
313
314 /*
315 * Change backing file, only valid to use during initial VMA setup.
316 */
317 void vma_set_file(struct vm_area_struct *vma, struct file *file)
318 {
319 /* Changing an anonymous vma with this is illegal */
320 get_file(file);
321 swap(vma->vm_file, file);
322 fput(file);
323 }
324 EXPORT_SYMBOL(vma_set_file);
325
326 #ifndef STACK_RND_MASK
327 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
328 #endif
329
330 unsigned long randomize_stack_top(unsigned long stack_top)
331 {
332 unsigned long random_variable = 0;
333
334 if (current->flags & PF_RANDOMIZE) {
335 random_variable = get_random_long();
336 random_variable &= STACK_RND_MASK;
337 random_variable <<= PAGE_SHIFT;
338 }
339 #ifdef CONFIG_STACK_GROWSUP
340 return PAGE_ALIGN(stack_top) + random_variable;
341 #else
342 return PAGE_ALIGN(stack_top) - random_variable;
343 #endif
344 }
345
346 /**
347 * randomize_page - Generate a random, page aligned address
348 * @start: The smallest acceptable address the caller will take.
349 * @range: The size of the area, starting at @start, within which the
350 * random address must fall.
351 *
352 * If @start + @range would overflow, @range is capped.
353 *
354 * NOTE: Historical use of randomize_range, which this replaces, presumed that
355 * @start was already page aligned. We now align it regardless.
356 *
357 * Return: A page aligned address within [start, start + range). On error,
358 * @start is returned.
359 */
360 unsigned long randomize_page(unsigned long start, unsigned long range)
361 {
362 if (!PAGE_ALIGNED(start)) {
363 range -= PAGE_ALIGN(start) - start;
364 start = PAGE_ALIGN(start);
365 }
366
367 if (start > ULONG_MAX - range)
368 range = ULONG_MAX - start;
369
370 range >>= PAGE_SHIFT;
371
372 if (range == 0)
373 return start;
374
375 return start + (get_random_long() % range << PAGE_SHIFT);
376 }
377
378 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
379 unsigned long arch_randomize_brk(struct mm_struct *mm)
380 {
381 /* Is the current task 32bit ? */
382 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
383 return randomize_page(mm->brk, SZ_32M);
384
385 return randomize_page(mm->brk, SZ_1G);
386 }
387
388 unsigned long arch_mmap_rnd(void)
389 {
390 unsigned long rnd;
391
392 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
393 if (is_compat_task())
394 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
395 else
396 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
397 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
398
399 return rnd << PAGE_SHIFT;
400 }
401
402 static int mmap_is_legacy(struct rlimit *rlim_stack)
403 {
404 if (current->personality & ADDR_COMPAT_LAYOUT)
405 return 1;
406
407 if (rlim_stack->rlim_cur == RLIM_INFINITY)
408 return 1;
409
410 return sysctl_legacy_va_layout;
411 }
412
413 /*
414 * Leave enough space between the mmap area and the stack to honour ulimit in
415 * the face of randomisation.
416 */
417 #define MIN_GAP (SZ_128M)
418 #define MAX_GAP (STACK_TOP / 6 * 5)
419
420 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
421 {
422 unsigned long gap = rlim_stack->rlim_cur;
423 unsigned long pad = stack_guard_gap;
424
425 /* Account for stack randomization if necessary */
426 if (current->flags & PF_RANDOMIZE)
427 pad += (STACK_RND_MASK << PAGE_SHIFT);
428
429 /* Values close to RLIM_INFINITY can overflow. */
430 if (gap + pad > gap)
431 gap += pad;
432
433 if (gap < MIN_GAP)
434 gap = MIN_GAP;
435 else if (gap > MAX_GAP)
436 gap = MAX_GAP;
437
438 return PAGE_ALIGN(STACK_TOP - gap - rnd);
439 }
440
441 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
442 {
443 unsigned long random_factor = 0UL;
444
445 if (current->flags & PF_RANDOMIZE)
446 random_factor = arch_mmap_rnd();
447
448 if (mmap_is_legacy(rlim_stack)) {
449 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
450 mm->get_unmapped_area = arch_get_unmapped_area;
451 } else {
452 mm->mmap_base = mmap_base(random_factor, rlim_stack);
453 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
454 }
455 }
456 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
457 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
458 {
459 mm->mmap_base = TASK_UNMAPPED_BASE;
460 mm->get_unmapped_area = arch_get_unmapped_area;
461 }
462 #endif
463
464 /**
465 * __account_locked_vm - account locked pages to an mm's locked_vm
466 * @mm: mm to account against
467 * @pages: number of pages to account
468 * @inc: %true if @pages should be considered positive, %false if not
469 * @task: task used to check RLIMIT_MEMLOCK
470 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
471 *
472 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
473 * that mmap_lock is held as writer.
474 *
475 * Return:
476 * * 0 on success
477 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
478 */
479 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
480 struct task_struct *task, bool bypass_rlim)
481 {
482 unsigned long locked_vm, limit;
483 int ret = 0;
484
485 mmap_assert_write_locked(mm);
486
487 locked_vm = mm->locked_vm;
488 if (inc) {
489 if (!bypass_rlim) {
490 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
491 if (locked_vm + pages > limit)
492 ret = -ENOMEM;
493 }
494 if (!ret)
495 mm->locked_vm = locked_vm + pages;
496 } else {
497 WARN_ON_ONCE(pages > locked_vm);
498 mm->locked_vm = locked_vm - pages;
499 }
500
501 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
502 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
503 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
504 ret ? " - exceeded" : "");
505
506 return ret;
507 }
508 EXPORT_SYMBOL_GPL(__account_locked_vm);
509
510 /**
511 * account_locked_vm - account locked pages to an mm's locked_vm
512 * @mm: mm to account against, may be NULL
513 * @pages: number of pages to account
514 * @inc: %true if @pages should be considered positive, %false if not
515 *
516 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
517 *
518 * Return:
519 * * 0 on success, or if mm is NULL
520 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
521 */
522 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
523 {
524 int ret;
525
526 if (pages == 0 || !mm)
527 return 0;
528
529 mmap_write_lock(mm);
530 ret = __account_locked_vm(mm, pages, inc, current,
531 capable(CAP_IPC_LOCK));
532 mmap_write_unlock(mm);
533
534 return ret;
535 }
536 EXPORT_SYMBOL_GPL(account_locked_vm);
537
538 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
539 unsigned long len, unsigned long prot,
540 unsigned long flag, unsigned long pgoff)
541 {
542 unsigned long ret;
543 struct mm_struct *mm = current->mm;
544 unsigned long populate;
545 LIST_HEAD(uf);
546
547 ret = security_mmap_file(file, prot, flag);
548 if (!ret) {
549 if (mmap_write_lock_killable(mm))
550 return -EINTR;
551 ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
552 &uf);
553 mmap_write_unlock(mm);
554 userfaultfd_unmap_complete(mm, &uf);
555 if (populate)
556 mm_populate(ret, populate);
557 }
558 return ret;
559 }
560
561 unsigned long vm_mmap(struct file *file, unsigned long addr,
562 unsigned long len, unsigned long prot,
563 unsigned long flag, unsigned long offset)
564 {
565 if (unlikely(offset + PAGE_ALIGN(len) < offset))
566 return -EINVAL;
567 if (unlikely(offset_in_page(offset)))
568 return -EINVAL;
569
570 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
571 }
572 EXPORT_SYMBOL(vm_mmap);
573
574 /**
575 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
576 * failure, fall back to non-contiguous (vmalloc) allocation.
577 * @size: size of the request.
578 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
579 * @node: numa node to allocate from
580 *
581 * Uses kmalloc to get the memory but if the allocation fails then falls back
582 * to the vmalloc allocator. Use kvfree for freeing the memory.
583 *
584 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
585 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
586 * preferable to the vmalloc fallback, due to visible performance drawbacks.
587 *
588 * Return: pointer to the allocated memory of %NULL in case of failure
589 */
590 void *kvmalloc_node(size_t size, gfp_t flags, int node)
591 {
592 gfp_t kmalloc_flags = flags;
593 void *ret;
594
595 /*
596 * We want to attempt a large physically contiguous block first because
597 * it is less likely to fragment multiple larger blocks and therefore
598 * contribute to a long term fragmentation less than vmalloc fallback.
599 * However make sure that larger requests are not too disruptive - no
600 * OOM killer and no allocation failure warnings as we have a fallback.
601 */
602 if (size > PAGE_SIZE) {
603 kmalloc_flags |= __GFP_NOWARN;
604
605 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
606 kmalloc_flags |= __GFP_NORETRY;
607
608 /* nofail semantic is implemented by the vmalloc fallback */
609 kmalloc_flags &= ~__GFP_NOFAIL;
610 }
611
612 ret = kmalloc_node(size, kmalloc_flags, node);
613
614 /*
615 * It doesn't really make sense to fallback to vmalloc for sub page
616 * requests
617 */
618 if (ret || size <= PAGE_SIZE)
619 return ret;
620
621 /* Don't even allow crazy sizes */
622 if (unlikely(size > INT_MAX)) {
623 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
624 return NULL;
625 }
626
627 return __vmalloc_node(size, 1, flags, node,
628 __builtin_return_address(0));
629 }
630 EXPORT_SYMBOL(kvmalloc_node);
631
632 /**
633 * kvfree() - Free memory.
634 * @addr: Pointer to allocated memory.
635 *
636 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
637 * It is slightly more efficient to use kfree() or vfree() if you are certain
638 * that you know which one to use.
639 *
640 * Context: Either preemptible task context or not-NMI interrupt.
641 */
642 void kvfree(const void *addr)
643 {
644 if (is_vmalloc_addr(addr))
645 vfree(addr);
646 else
647 kfree(addr);
648 }
649 EXPORT_SYMBOL(kvfree);
650
651 /**
652 * kvfree_sensitive - Free a data object containing sensitive information.
653 * @addr: address of the data object to be freed.
654 * @len: length of the data object.
655 *
656 * Use the special memzero_explicit() function to clear the content of a
657 * kvmalloc'ed object containing sensitive data to make sure that the
658 * compiler won't optimize out the data clearing.
659 */
660 void kvfree_sensitive(const void *addr, size_t len)
661 {
662 if (likely(!ZERO_OR_NULL_PTR(addr))) {
663 memzero_explicit((void *)addr, len);
664 kvfree(addr);
665 }
666 }
667 EXPORT_SYMBOL(kvfree_sensitive);
668
669 void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
670 {
671 void *newp;
672
673 if (oldsize >= newsize)
674 return (void *)p;
675 newp = kvmalloc(newsize, flags);
676 if (!newp)
677 return NULL;
678 memcpy(newp, p, oldsize);
679 kvfree(p);
680 return newp;
681 }
682 EXPORT_SYMBOL(kvrealloc);
683
684 /* Neutral page->mapping pointer to address_space or anon_vma or other */
685 void *page_rmapping(struct page *page)
686 {
687 return folio_raw_mapping(page_folio(page));
688 }
689
690 /**
691 * folio_mapped - Is this folio mapped into userspace?
692 * @folio: The folio.
693 *
694 * Return: True if any page in this folio is referenced by user page tables.
695 */
696 bool folio_mapped(struct folio *folio)
697 {
698 long i, nr;
699
700 if (!folio_test_large(folio))
701 return atomic_read(&folio->_mapcount) >= 0;
702 if (atomic_read(folio_mapcount_ptr(folio)) >= 0)
703 return true;
704 if (folio_test_hugetlb(folio))
705 return false;
706
707 nr = folio_nr_pages(folio);
708 for (i = 0; i < nr; i++) {
709 if (atomic_read(&folio_page(folio, i)->_mapcount) >= 0)
710 return true;
711 }
712 return false;
713 }
714 EXPORT_SYMBOL(folio_mapped);
715
716 struct anon_vma *page_anon_vma(struct page *page)
717 {
718 struct folio *folio = page_folio(page);
719 unsigned long mapping = (unsigned long)folio->mapping;
720
721 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
722 return NULL;
723 return (void *)(mapping - PAGE_MAPPING_ANON);
724 }
725
726 /**
727 * folio_mapping - Find the mapping where this folio is stored.
728 * @folio: The folio.
729 *
730 * For folios which are in the page cache, return the mapping that this
731 * page belongs to. Folios in the swap cache return the swap mapping
732 * this page is stored in (which is different from the mapping for the
733 * swap file or swap device where the data is stored).
734 *
735 * You can call this for folios which aren't in the swap cache or page
736 * cache and it will return NULL.
737 */
738 struct address_space *folio_mapping(struct folio *folio)
739 {
740 struct address_space *mapping;
741
742 /* This happens if someone calls flush_dcache_page on slab page */
743 if (unlikely(folio_test_slab(folio)))
744 return NULL;
745
746 if (unlikely(folio_test_swapcache(folio)))
747 return swap_address_space(folio_swap_entry(folio));
748
749 mapping = folio->mapping;
750 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
751 return NULL;
752
753 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
754 }
755 EXPORT_SYMBOL(folio_mapping);
756
757 /* Slow path of page_mapcount() for compound pages */
758 int __page_mapcount(struct page *page)
759 {
760 int ret;
761
762 ret = atomic_read(&page->_mapcount) + 1;
763 /*
764 * For file THP page->_mapcount contains total number of mapping
765 * of the page: no need to look into compound_mapcount.
766 */
767 if (!PageAnon(page) && !PageHuge(page))
768 return ret;
769 page = compound_head(page);
770 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
771 if (PageDoubleMap(page))
772 ret--;
773 return ret;
774 }
775 EXPORT_SYMBOL_GPL(__page_mapcount);
776
777 /**
778 * folio_copy - Copy the contents of one folio to another.
779 * @dst: Folio to copy to.
780 * @src: Folio to copy from.
781 *
782 * The bytes in the folio represented by @src are copied to @dst.
783 * Assumes the caller has validated that @dst is at least as large as @src.
784 * Can be called in atomic context for order-0 folios, but if the folio is
785 * larger, it may sleep.
786 */
787 void folio_copy(struct folio *dst, struct folio *src)
788 {
789 long i = 0;
790 long nr = folio_nr_pages(src);
791
792 for (;;) {
793 copy_highpage(folio_page(dst, i), folio_page(src, i));
794 if (++i == nr)
795 break;
796 cond_resched();
797 }
798 }
799
800 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
801 int sysctl_overcommit_ratio __read_mostly = 50;
802 unsigned long sysctl_overcommit_kbytes __read_mostly;
803 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
804 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
805 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
806
807 int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
808 size_t *lenp, loff_t *ppos)
809 {
810 int ret;
811
812 ret = proc_dointvec(table, write, buffer, lenp, ppos);
813 if (ret == 0 && write)
814 sysctl_overcommit_kbytes = 0;
815 return ret;
816 }
817
818 static void sync_overcommit_as(struct work_struct *dummy)
819 {
820 percpu_counter_sync(&vm_committed_as);
821 }
822
823 int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
824 size_t *lenp, loff_t *ppos)
825 {
826 struct ctl_table t;
827 int new_policy = -1;
828 int ret;
829
830 /*
831 * The deviation of sync_overcommit_as could be big with loose policy
832 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
833 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
834 * with the strict "NEVER", and to avoid possible race condition (even
835 * though user usually won't too frequently do the switching to policy
836 * OVERCOMMIT_NEVER), the switch is done in the following order:
837 * 1. changing the batch
838 * 2. sync percpu count on each CPU
839 * 3. switch the policy
840 */
841 if (write) {
842 t = *table;
843 t.data = &new_policy;
844 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
845 if (ret || new_policy == -1)
846 return ret;
847
848 mm_compute_batch(new_policy);
849 if (new_policy == OVERCOMMIT_NEVER)
850 schedule_on_each_cpu(sync_overcommit_as);
851 sysctl_overcommit_memory = new_policy;
852 } else {
853 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
854 }
855
856 return ret;
857 }
858
859 int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
860 size_t *lenp, loff_t *ppos)
861 {
862 int ret;
863
864 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
865 if (ret == 0 && write)
866 sysctl_overcommit_ratio = 0;
867 return ret;
868 }
869
870 /*
871 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
872 */
873 unsigned long vm_commit_limit(void)
874 {
875 unsigned long allowed;
876
877 if (sysctl_overcommit_kbytes)
878 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
879 else
880 allowed = ((totalram_pages() - hugetlb_total_pages())
881 * sysctl_overcommit_ratio / 100);
882 allowed += total_swap_pages;
883
884 return allowed;
885 }
886
887 /*
888 * Make sure vm_committed_as in one cacheline and not cacheline shared with
889 * other variables. It can be updated by several CPUs frequently.
890 */
891 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
892
893 /*
894 * The global memory commitment made in the system can be a metric
895 * that can be used to drive ballooning decisions when Linux is hosted
896 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
897 * balancing memory across competing virtual machines that are hosted.
898 * Several metrics drive this policy engine including the guest reported
899 * memory commitment.
900 *
901 * The time cost of this is very low for small platforms, and for big
902 * platform like a 2S/36C/72T Skylake server, in worst case where
903 * vm_committed_as's spinlock is under severe contention, the time cost
904 * could be about 30~40 microseconds.
905 */
906 unsigned long vm_memory_committed(void)
907 {
908 return percpu_counter_sum_positive(&vm_committed_as);
909 }
910 EXPORT_SYMBOL_GPL(vm_memory_committed);
911
912 /*
913 * Check that a process has enough memory to allocate a new virtual
914 * mapping. 0 means there is enough memory for the allocation to
915 * succeed and -ENOMEM implies there is not.
916 *
917 * We currently support three overcommit policies, which are set via the
918 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
919 *
920 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
921 * Additional code 2002 Jul 20 by Robert Love.
922 *
923 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
924 *
925 * Note this is a helper function intended to be used by LSMs which
926 * wish to use this logic.
927 */
928 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
929 {
930 long allowed;
931
932 vm_acct_memory(pages);
933
934 /*
935 * Sometimes we want to use more memory than we have
936 */
937 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
938 return 0;
939
940 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
941 if (pages > totalram_pages() + total_swap_pages)
942 goto error;
943 return 0;
944 }
945
946 allowed = vm_commit_limit();
947 /*
948 * Reserve some for root
949 */
950 if (!cap_sys_admin)
951 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
952
953 /*
954 * Don't let a single process grow so big a user can't recover
955 */
956 if (mm) {
957 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
958
959 allowed -= min_t(long, mm->total_vm / 32, reserve);
960 }
961
962 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
963 return 0;
964 error:
965 vm_unacct_memory(pages);
966
967 return -ENOMEM;
968 }
969
970 /**
971 * get_cmdline() - copy the cmdline value to a buffer.
972 * @task: the task whose cmdline value to copy.
973 * @buffer: the buffer to copy to.
974 * @buflen: the length of the buffer. Larger cmdline values are truncated
975 * to this length.
976 *
977 * Return: the size of the cmdline field copied. Note that the copy does
978 * not guarantee an ending NULL byte.
979 */
980 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
981 {
982 int res = 0;
983 unsigned int len;
984 struct mm_struct *mm = get_task_mm(task);
985 unsigned long arg_start, arg_end, env_start, env_end;
986 if (!mm)
987 goto out;
988 if (!mm->arg_end)
989 goto out_mm; /* Shh! No looking before we're done */
990
991 spin_lock(&mm->arg_lock);
992 arg_start = mm->arg_start;
993 arg_end = mm->arg_end;
994 env_start = mm->env_start;
995 env_end = mm->env_end;
996 spin_unlock(&mm->arg_lock);
997
998 len = arg_end - arg_start;
999
1000 if (len > buflen)
1001 len = buflen;
1002
1003 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1004
1005 /*
1006 * If the nul at the end of args has been overwritten, then
1007 * assume application is using setproctitle(3).
1008 */
1009 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1010 len = strnlen(buffer, res);
1011 if (len < res) {
1012 res = len;
1013 } else {
1014 len = env_end - env_start;
1015 if (len > buflen - res)
1016 len = buflen - res;
1017 res += access_process_vm(task, env_start,
1018 buffer+res, len,
1019 FOLL_FORCE);
1020 res = strnlen(buffer, res);
1021 }
1022 }
1023 out_mm:
1024 mmput(mm);
1025 out:
1026 return res;
1027 }
1028
1029 int __weak memcmp_pages(struct page *page1, struct page *page2)
1030 {
1031 char *addr1, *addr2;
1032 int ret;
1033
1034 addr1 = kmap_atomic(page1);
1035 addr2 = kmap_atomic(page2);
1036 ret = memcmp(addr1, addr2, PAGE_SIZE);
1037 kunmap_atomic(addr2);
1038 kunmap_atomic(addr1);
1039 return ret;
1040 }
1041
1042 #ifdef CONFIG_PRINTK
1043 /**
1044 * mem_dump_obj - Print available provenance information
1045 * @object: object for which to find provenance information.
1046 *
1047 * This function uses pr_cont(), so that the caller is expected to have
1048 * printed out whatever preamble is appropriate. The provenance information
1049 * depends on the type of object and on how much debugging is enabled.
1050 * For example, for a slab-cache object, the slab name is printed, and,
1051 * if available, the return address and stack trace from the allocation
1052 * and last free path of that object.
1053 */
1054 void mem_dump_obj(void *object)
1055 {
1056 const char *type;
1057
1058 if (kmem_valid_obj(object)) {
1059 kmem_dump_obj(object);
1060 return;
1061 }
1062
1063 if (vmalloc_dump_obj(object))
1064 return;
1065
1066 if (virt_addr_valid(object))
1067 type = "non-slab/vmalloc memory";
1068 else if (object == NULL)
1069 type = "NULL pointer";
1070 else if (object == ZERO_SIZE_PTR)
1071 type = "zero-size pointer";
1072 else
1073 type = "non-paged memory";
1074
1075 pr_cont(" %s\n", type);
1076 }
1077 EXPORT_SYMBOL_GPL(mem_dump_obj);
1078 #endif
1079
1080 /*
1081 * A driver might set a page logically offline -- PageOffline() -- and
1082 * turn the page inaccessible in the hypervisor; after that, access to page
1083 * content can be fatal.
1084 *
1085 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1086 * pages after checking PageOffline(); however, these PFN walkers can race
1087 * with drivers that set PageOffline().
1088 *
1089 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1090 * synchronize with such drivers, achieving that a page cannot be set
1091 * PageOffline() while frozen.
1092 *
1093 * page_offline_begin()/page_offline_end() is used by drivers that care about
1094 * such races when setting a page PageOffline().
1095 */
1096 static DECLARE_RWSEM(page_offline_rwsem);
1097
1098 void page_offline_freeze(void)
1099 {
1100 down_read(&page_offline_rwsem);
1101 }
1102
1103 void page_offline_thaw(void)
1104 {
1105 up_read(&page_offline_rwsem);
1106 }
1107
1108 void page_offline_begin(void)
1109 {
1110 down_write(&page_offline_rwsem);
1111 }
1112 EXPORT_SYMBOL(page_offline_begin);
1113
1114 void page_offline_end(void)
1115 {
1116 up_write(&page_offline_rwsem);
1117 }
1118 EXPORT_SYMBOL(page_offline_end);
1119
1120 #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_FOLIO
1121 void flush_dcache_folio(struct folio *folio)
1122 {
1123 long i, nr = folio_nr_pages(folio);
1124
1125 for (i = 0; i < nr; i++)
1126 flush_dcache_page(folio_page(folio, i));
1127 }
1128 EXPORT_SYMBOL(flush_dcache_folio);
1129 #endif
1130
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.