Loading...
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 */
37void kfree_const(const void *x)
38{
39 if (!is_kernel_rodata((unsigned long)x))
40 kfree(x);
41}
42EXPORT_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 */
51char *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}
65EXPORT_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.
73 *
74 * Return: source string if it is in .rodata section otherwise
75 * fallback to kstrdup.
76 */
77const char *kstrdup_const(const char *s, gfp_t gfp)
78{
79 if (is_kernel_rodata((unsigned long)s))
80 return s;
81
82 return kstrdup(s, gfp);
83}
84EXPORT_SYMBOL(kstrdup_const);
85
86/**
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
91 *
92 * Note: Use kmemdup_nul() instead if the size is known exactly.
93 *
94 * Return: newly allocated copy of @s or %NULL in case of error
95 */
96char *kstrndup(const char *s, size_t max, gfp_t gfp)
97{
98 size_t len;
99 char *buf;
100
101 if (!s)
102 return NULL;
103
104 len = strnlen(s, max);
105 buf = kmalloc_track_caller(len+1, gfp);
106 if (buf) {
107 memcpy(buf, s, len);
108 buf[len] = '\0';
109 }
110 return buf;
111}
112EXPORT_SYMBOL(kstrndup);
113
114/**
115 * kmemdup - duplicate region of memory
116 *
117 * @src: memory region to duplicate
118 * @len: memory region length
119 * @gfp: GFP mask to use
120 *
121 * Return: newly allocated copy of @src or %NULL in case of error
122 */
123void *kmemdup(const void *src, size_t len, gfp_t gfp)
124{
125 void *p;
126
127 p = kmalloc_track_caller(len, gfp);
128 if (p)
129 memcpy(p, src, len);
130 return p;
131}
132EXPORT_SYMBOL(kmemdup);
133
134/**
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
139 *
140 * Return: newly allocated copy of @s with NUL-termination or %NULL in
141 * case of error
142 */
143char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
144{
145 char *buf;
146
147 if (!s)
148 return NULL;
149
150 buf = kmalloc_track_caller(len + 1, gfp);
151 if (buf) {
152 memcpy(buf, s, len);
153 buf[len] = '\0';
154 }
155 return buf;
156}
157EXPORT_SYMBOL(kmemdup_nul);
158
159/**
160 * memdup_user - duplicate memory region from user space
161 *
162 * @src: source address in user space
163 * @len: number of bytes to copy
164 *
165 * Return: an ERR_PTR() on failure. Result is physically
166 * contiguous, to be freed by kfree().
167 */
168void *memdup_user(const void __user *src, size_t len)
169{
170 void *p;
171
172 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
173 if (!p)
174 return ERR_PTR(-ENOMEM);
175
176 if (copy_from_user(p, src, len)) {
177 kfree(p);
178 return ERR_PTR(-EFAULT);
179 }
180
181 return p;
182}
183EXPORT_SYMBOL(memdup_user);
184
185/**
186 * vmemdup_user - duplicate memory region from user space
187 *
188 * @src: source address in user space
189 * @len: number of bytes to copy
190 *
191 * Return: an ERR_PTR() on failure. Result may be not
192 * physically contiguous. Use kvfree() to free.
193 */
194void *vmemdup_user(const void __user *src, size_t len)
195{
196 void *p;
197
198 p = kvmalloc(len, GFP_USER);
199 if (!p)
200 return ERR_PTR(-ENOMEM);
201
202 if (copy_from_user(p, src, len)) {
203 kvfree(p);
204 return ERR_PTR(-EFAULT);
205 }
206
207 return p;
208}
209EXPORT_SYMBOL(vmemdup_user);
210
211/**
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.
215 *
216 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
217 */
218char *strndup_user(const char __user *s, long n)
219{
220 char *p;
221 long length;
222
223 length = strnlen_user(s, n);
224
225 if (!length)
226 return ERR_PTR(-EFAULT);
227
228 if (length > n)
229 return ERR_PTR(-EINVAL);
230
231 p = memdup_user(s, length);
232
233 if (IS_ERR(p))
234 return p;
235
236 p[length - 1] = '\0';
237
238 return p;
239}
240EXPORT_SYMBOL(strndup_user);
241
242/**
243 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
244 *
245 * @src: source address in user space
246 * @len: number of bytes to copy
247 *
248 * Return: an ERR_PTR() on failure.
249 */
250void *memdup_user_nul(const void __user *src, size_t len)
251{
252 char *p;
253
254 /*
255 * Always use GFP_KERNEL, since copy_from_user() can sleep and
256 * cause pagefault, which makes it pointless to use GFP_NOFS
257 * or GFP_ATOMIC.
258 */
259 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
260 if (!p)
261 return ERR_PTR(-ENOMEM);
262
263 if (copy_from_user(p, src, len)) {
264 kfree(p);
265 return ERR_PTR(-EFAULT);
266 }
267 p[len] = '\0';
268
269 return p;
270}
271EXPORT_SYMBOL(memdup_user_nul);
272
273void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
274 struct vm_area_struct *prev, struct rb_node *rb_parent)
275{
276 struct vm_area_struct *next;
277
278 vma->vm_prev = prev;
279 if (prev) {
280 next = prev->vm_next;
281 prev->vm_next = vma;
282 } else {
283 mm->mmap = vma;
284 if (rb_parent)
285 next = rb_entry(rb_parent,
286 struct vm_area_struct, vm_rb);
287 else
288 next = NULL;
289 }
290 vma->vm_next = next;
291 if (next)
292 next->vm_prev = vma;
293}
294
295/* Check if the vma is being used as a stack by this task */
296int vma_is_stack_for_current(struct vm_area_struct *vma)
297{
298 struct task_struct * __maybe_unused t = current;
299
300 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
301}
302
303#ifndef STACK_RND_MASK
304#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
305#endif
306
307unsigned long randomize_stack_top(unsigned long stack_top)
308{
309 unsigned long random_variable = 0;
310
311 if (current->flags & PF_RANDOMIZE) {
312 random_variable = get_random_long();
313 random_variable &= STACK_RND_MASK;
314 random_variable <<= PAGE_SHIFT;
315 }
316#ifdef CONFIG_STACK_GROWSUP
317 return PAGE_ALIGN(stack_top) + random_variable;
318#else
319 return PAGE_ALIGN(stack_top) - random_variable;
320#endif
321}
322
323#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
324unsigned long arch_randomize_brk(struct mm_struct *mm)
325{
326 /* Is the current task 32bit ? */
327 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
328 return randomize_page(mm->brk, SZ_32M);
329
330 return randomize_page(mm->brk, SZ_1G);
331}
332
333unsigned long arch_mmap_rnd(void)
334{
335 unsigned long rnd;
336
337#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
338 if (is_compat_task())
339 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
340 else
341#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
342 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
343
344 return rnd << PAGE_SHIFT;
345}
346
347static int mmap_is_legacy(struct rlimit *rlim_stack)
348{
349 if (current->personality & ADDR_COMPAT_LAYOUT)
350 return 1;
351
352 if (rlim_stack->rlim_cur == RLIM_INFINITY)
353 return 1;
354
355 return sysctl_legacy_va_layout;
356}
357
358/*
359 * Leave enough space between the mmap area and the stack to honour ulimit in
360 * the face of randomisation.
361 */
362#define MIN_GAP (SZ_128M)
363#define MAX_GAP (STACK_TOP / 6 * 5)
364
365static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
366{
367 unsigned long gap = rlim_stack->rlim_cur;
368 unsigned long pad = stack_guard_gap;
369
370 /* Account for stack randomization if necessary */
371 if (current->flags & PF_RANDOMIZE)
372 pad += (STACK_RND_MASK << PAGE_SHIFT);
373
374 /* Values close to RLIM_INFINITY can overflow. */
375 if (gap + pad > gap)
376 gap += pad;
377
378 if (gap < MIN_GAP)
379 gap = MIN_GAP;
380 else if (gap > MAX_GAP)
381 gap = MAX_GAP;
382
383 return PAGE_ALIGN(STACK_TOP - gap - rnd);
384}
385
386void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
387{
388 unsigned long random_factor = 0UL;
389
390 if (current->flags & PF_RANDOMIZE)
391 random_factor = arch_mmap_rnd();
392
393 if (mmap_is_legacy(rlim_stack)) {
394 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
395 mm->get_unmapped_area = arch_get_unmapped_area;
396 } else {
397 mm->mmap_base = mmap_base(random_factor, rlim_stack);
398 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
399 }
400}
401#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
402void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
403{
404 mm->mmap_base = TASK_UNMAPPED_BASE;
405 mm->get_unmapped_area = arch_get_unmapped_area;
406}
407#endif
408
409/**
410 * __account_locked_vm - account locked pages to an mm's locked_vm
411 * @mm: mm to account against
412 * @pages: number of pages to account
413 * @inc: %true if @pages should be considered positive, %false if not
414 * @task: task used to check RLIMIT_MEMLOCK
415 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
416 *
417 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
418 * that mmap_sem is held as writer.
419 *
420 * Return:
421 * * 0 on success
422 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
423 */
424int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
425 struct task_struct *task, bool bypass_rlim)
426{
427 unsigned long locked_vm, limit;
428 int ret = 0;
429
430 lockdep_assert_held_write(&mm->mmap_sem);
431
432 locked_vm = mm->locked_vm;
433 if (inc) {
434 if (!bypass_rlim) {
435 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
436 if (locked_vm + pages > limit)
437 ret = -ENOMEM;
438 }
439 if (!ret)
440 mm->locked_vm = locked_vm + pages;
441 } else {
442 WARN_ON_ONCE(pages > locked_vm);
443 mm->locked_vm = locked_vm - pages;
444 }
445
446 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
447 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
448 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
449 ret ? " - exceeded" : "");
450
451 return ret;
452}
453EXPORT_SYMBOL_GPL(__account_locked_vm);
454
455/**
456 * account_locked_vm - account locked pages to an mm's locked_vm
457 * @mm: mm to account against, may be NULL
458 * @pages: number of pages to account
459 * @inc: %true if @pages should be considered positive, %false if not
460 *
461 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
462 *
463 * Return:
464 * * 0 on success, or if mm is NULL
465 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
466 */
467int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
468{
469 int ret;
470
471 if (pages == 0 || !mm)
472 return 0;
473
474 down_write(&mm->mmap_sem);
475 ret = __account_locked_vm(mm, pages, inc, current,
476 capable(CAP_IPC_LOCK));
477 up_write(&mm->mmap_sem);
478
479 return ret;
480}
481EXPORT_SYMBOL_GPL(account_locked_vm);
482
483unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
484 unsigned long len, unsigned long prot,
485 unsigned long flag, unsigned long pgoff)
486{
487 unsigned long ret;
488 struct mm_struct *mm = current->mm;
489 unsigned long populate;
490 LIST_HEAD(uf);
491
492 ret = security_mmap_file(file, prot, flag);
493 if (!ret) {
494 if (down_write_killable(&mm->mmap_sem))
495 return -EINTR;
496 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
497 &populate, &uf);
498 up_write(&mm->mmap_sem);
499 userfaultfd_unmap_complete(mm, &uf);
500 if (populate)
501 mm_populate(ret, populate);
502 }
503 return ret;
504}
505
506unsigned long vm_mmap(struct file *file, unsigned long addr,
507 unsigned long len, unsigned long prot,
508 unsigned long flag, unsigned long offset)
509{
510 if (unlikely(offset + PAGE_ALIGN(len) < offset))
511 return -EINVAL;
512 if (unlikely(offset_in_page(offset)))
513 return -EINVAL;
514
515 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
516}
517EXPORT_SYMBOL(vm_mmap);
518
519/**
520 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
521 * failure, fall back to non-contiguous (vmalloc) allocation.
522 * @size: size of the request.
523 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
524 * @node: numa node to allocate from
525 *
526 * Uses kmalloc to get the memory but if the allocation fails then falls back
527 * to the vmalloc allocator. Use kvfree for freeing the memory.
528 *
529 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
530 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
531 * preferable to the vmalloc fallback, due to visible performance drawbacks.
532 *
533 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
534 * fall back to vmalloc.
535 *
536 * Return: pointer to the allocated memory of %NULL in case of failure
537 */
538void *kvmalloc_node(size_t size, gfp_t flags, int node)
539{
540 gfp_t kmalloc_flags = flags;
541 void *ret;
542
543 /*
544 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
545 * so the given set of flags has to be compatible.
546 */
547 if ((flags & GFP_KERNEL) != GFP_KERNEL)
548 return kmalloc_node(size, flags, node);
549
550 /*
551 * We want to attempt a large physically contiguous block first because
552 * it is less likely to fragment multiple larger blocks and therefore
553 * contribute to a long term fragmentation less than vmalloc fallback.
554 * However make sure that larger requests are not too disruptive - no
555 * OOM killer and no allocation failure warnings as we have a fallback.
556 */
557 if (size > PAGE_SIZE) {
558 kmalloc_flags |= __GFP_NOWARN;
559
560 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
561 kmalloc_flags |= __GFP_NORETRY;
562 }
563
564 ret = kmalloc_node(size, kmalloc_flags, node);
565
566 /*
567 * It doesn't really make sense to fallback to vmalloc for sub page
568 * requests
569 */
570 if (ret || size <= PAGE_SIZE)
571 return ret;
572
573 return __vmalloc_node_flags_caller(size, node, flags,
574 __builtin_return_address(0));
575}
576EXPORT_SYMBOL(kvmalloc_node);
577
578/**
579 * kvfree() - Free memory.
580 * @addr: Pointer to allocated memory.
581 *
582 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
583 * It is slightly more efficient to use kfree() or vfree() if you are certain
584 * that you know which one to use.
585 *
586 * Context: Either preemptible task context or not-NMI interrupt.
587 */
588void kvfree(const void *addr)
589{
590 if (is_vmalloc_addr(addr))
591 vfree(addr);
592 else
593 kfree(addr);
594}
595EXPORT_SYMBOL(kvfree);
596
597static inline void *__page_rmapping(struct page *page)
598{
599 unsigned long mapping;
600
601 mapping = (unsigned long)page->mapping;
602 mapping &= ~PAGE_MAPPING_FLAGS;
603
604 return (void *)mapping;
605}
606
607/* Neutral page->mapping pointer to address_space or anon_vma or other */
608void *page_rmapping(struct page *page)
609{
610 page = compound_head(page);
611 return __page_rmapping(page);
612}
613
614/*
615 * Return true if this page is mapped into pagetables.
616 * For compound page it returns true if any subpage of compound page is mapped.
617 */
618bool page_mapped(struct page *page)
619{
620 int i;
621
622 if (likely(!PageCompound(page)))
623 return atomic_read(&page->_mapcount) >= 0;
624 page = compound_head(page);
625 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
626 return true;
627 if (PageHuge(page))
628 return false;
629 for (i = 0; i < compound_nr(page); i++) {
630 if (atomic_read(&page[i]._mapcount) >= 0)
631 return true;
632 }
633 return false;
634}
635EXPORT_SYMBOL(page_mapped);
636
637struct anon_vma *page_anon_vma(struct page *page)
638{
639 unsigned long mapping;
640
641 page = compound_head(page);
642 mapping = (unsigned long)page->mapping;
643 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
644 return NULL;
645 return __page_rmapping(page);
646}
647
648struct address_space *page_mapping(struct page *page)
649{
650 struct address_space *mapping;
651
652 page = compound_head(page);
653
654 /* This happens if someone calls flush_dcache_page on slab page */
655 if (unlikely(PageSlab(page)))
656 return NULL;
657
658 if (unlikely(PageSwapCache(page))) {
659 swp_entry_t entry;
660
661 entry.val = page_private(page);
662 return swap_address_space(entry);
663 }
664
665 mapping = page->mapping;
666 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
667 return NULL;
668
669 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
670}
671EXPORT_SYMBOL(page_mapping);
672
673/*
674 * For file cache pages, return the address_space, otherwise return NULL
675 */
676struct address_space *page_mapping_file(struct page *page)
677{
678 if (unlikely(PageSwapCache(page)))
679 return NULL;
680 return page_mapping(page);
681}
682
683/* Slow path of page_mapcount() for compound pages */
684int __page_mapcount(struct page *page)
685{
686 int ret;
687
688 ret = atomic_read(&page->_mapcount) + 1;
689 /*
690 * For file THP page->_mapcount contains total number of mapping
691 * of the page: no need to look into compound_mapcount.
692 */
693 if (!PageAnon(page) && !PageHuge(page))
694 return ret;
695 page = compound_head(page);
696 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
697 if (PageDoubleMap(page))
698 ret--;
699 return ret;
700}
701EXPORT_SYMBOL_GPL(__page_mapcount);
702
703int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
704int sysctl_overcommit_ratio __read_mostly = 50;
705unsigned long sysctl_overcommit_kbytes __read_mostly;
706int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
707unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
708unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
709
710int overcommit_ratio_handler(struct ctl_table *table, int write,
711 void __user *buffer, size_t *lenp,
712 loff_t *ppos)
713{
714 int ret;
715
716 ret = proc_dointvec(table, write, buffer, lenp, ppos);
717 if (ret == 0 && write)
718 sysctl_overcommit_kbytes = 0;
719 return ret;
720}
721
722int overcommit_kbytes_handler(struct ctl_table *table, int write,
723 void __user *buffer, size_t *lenp,
724 loff_t *ppos)
725{
726 int ret;
727
728 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
729 if (ret == 0 && write)
730 sysctl_overcommit_ratio = 0;
731 return ret;
732}
733
734/*
735 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
736 */
737unsigned long vm_commit_limit(void)
738{
739 unsigned long allowed;
740
741 if (sysctl_overcommit_kbytes)
742 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
743 else
744 allowed = ((totalram_pages() - hugetlb_total_pages())
745 * sysctl_overcommit_ratio / 100);
746 allowed += total_swap_pages;
747
748 return allowed;
749}
750
751/*
752 * Make sure vm_committed_as in one cacheline and not cacheline shared with
753 * other variables. It can be updated by several CPUs frequently.
754 */
755struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
756
757/*
758 * The global memory commitment made in the system can be a metric
759 * that can be used to drive ballooning decisions when Linux is hosted
760 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
761 * balancing memory across competing virtual machines that are hosted.
762 * Several metrics drive this policy engine including the guest reported
763 * memory commitment.
764 */
765unsigned long vm_memory_committed(void)
766{
767 return percpu_counter_read_positive(&vm_committed_as);
768}
769EXPORT_SYMBOL_GPL(vm_memory_committed);
770
771/*
772 * Check that a process has enough memory to allocate a new virtual
773 * mapping. 0 means there is enough memory for the allocation to
774 * succeed and -ENOMEM implies there is not.
775 *
776 * We currently support three overcommit policies, which are set via the
777 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
778 *
779 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
780 * Additional code 2002 Jul 20 by Robert Love.
781 *
782 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
783 *
784 * Note this is a helper function intended to be used by LSMs which
785 * wish to use this logic.
786 */
787int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
788{
789 long allowed;
790
791 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
792 -(s64)vm_committed_as_batch * num_online_cpus(),
793 "memory commitment underflow");
794
795 vm_acct_memory(pages);
796
797 /*
798 * Sometimes we want to use more memory than we have
799 */
800 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
801 return 0;
802
803 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
804 if (pages > totalram_pages() + total_swap_pages)
805 goto error;
806 return 0;
807 }
808
809 allowed = vm_commit_limit();
810 /*
811 * Reserve some for root
812 */
813 if (!cap_sys_admin)
814 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
815
816 /*
817 * Don't let a single process grow so big a user can't recover
818 */
819 if (mm) {
820 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
821
822 allowed -= min_t(long, mm->total_vm / 32, reserve);
823 }
824
825 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
826 return 0;
827error:
828 vm_unacct_memory(pages);
829
830 return -ENOMEM;
831}
832
833/**
834 * get_cmdline() - copy the cmdline value to a buffer.
835 * @task: the task whose cmdline value to copy.
836 * @buffer: the buffer to copy to.
837 * @buflen: the length of the buffer. Larger cmdline values are truncated
838 * to this length.
839 *
840 * Return: the size of the cmdline field copied. Note that the copy does
841 * not guarantee an ending NULL byte.
842 */
843int get_cmdline(struct task_struct *task, char *buffer, int buflen)
844{
845 int res = 0;
846 unsigned int len;
847 struct mm_struct *mm = get_task_mm(task);
848 unsigned long arg_start, arg_end, env_start, env_end;
849 if (!mm)
850 goto out;
851 if (!mm->arg_end)
852 goto out_mm; /* Shh! No looking before we're done */
853
854 spin_lock(&mm->arg_lock);
855 arg_start = mm->arg_start;
856 arg_end = mm->arg_end;
857 env_start = mm->env_start;
858 env_end = mm->env_end;
859 spin_unlock(&mm->arg_lock);
860
861 len = arg_end - arg_start;
862
863 if (len > buflen)
864 len = buflen;
865
866 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
867
868 /*
869 * If the nul at the end of args has been overwritten, then
870 * assume application is using setproctitle(3).
871 */
872 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
873 len = strnlen(buffer, res);
874 if (len < res) {
875 res = len;
876 } else {
877 len = env_end - env_start;
878 if (len > buflen - res)
879 len = buflen - res;
880 res += access_process_vm(task, env_start,
881 buffer+res, len,
882 FOLL_FORCE);
883 res = strnlen(buffer, res);
884 }
885 }
886out_mm:
887 mmput(mm);
888out:
889 return res;
890}
891
892int memcmp_pages(struct page *page1, struct page *page2)
893{
894 char *addr1, *addr2;
895 int ret;
896
897 addr1 = kmap_atomic(page1);
898 addr2 = kmap_atomic(page2);
899 ret = memcmp(addr1, addr2, PAGE_SIZE);
900 kunmap_atomic(addr2);
901 kunmap_atomic(addr1);
902 return ret;
903}
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#include "swap.h"
31
32/**
33 * kfree_const - conditionally free memory
34 * @x: pointer to the memory
35 *
36 * Function calls kfree only if @x is not in .rodata section.
37 */
38void kfree_const(const void *x)
39{
40 if (!is_kernel_rodata((unsigned long)x))
41 kfree(x);
42}
43EXPORT_SYMBOL(kfree_const);
44
45/**
46 * kstrdup - allocate space for and copy an existing string
47 * @s: the string to duplicate
48 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
49 *
50 * Return: newly allocated copy of @s or %NULL in case of error
51 */
52noinline
53char *kstrdup(const char *s, gfp_t gfp)
54{
55 size_t len;
56 char *buf;
57
58 if (!s)
59 return NULL;
60
61 len = strlen(s) + 1;
62 buf = kmalloc_track_caller(len, gfp);
63 if (buf)
64 memcpy(buf, s, len);
65 return buf;
66}
67EXPORT_SYMBOL(kstrdup);
68
69/**
70 * kstrdup_const - conditionally duplicate an existing const string
71 * @s: the string to duplicate
72 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
73 *
74 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
75 * must not be passed to krealloc().
76 *
77 * Return: source string if it is in .rodata section otherwise
78 * fallback to kstrdup.
79 */
80const char *kstrdup_const(const char *s, gfp_t gfp)
81{
82 if (is_kernel_rodata((unsigned long)s))
83 return s;
84
85 return kstrdup(s, gfp);
86}
87EXPORT_SYMBOL(kstrdup_const);
88
89/**
90 * kstrndup - allocate space for and copy an existing string
91 * @s: the string to duplicate
92 * @max: read at most @max chars from @s
93 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
94 *
95 * Note: Use kmemdup_nul() instead if the size is known exactly.
96 *
97 * Return: newly allocated copy of @s or %NULL in case of error
98 */
99char *kstrndup(const char *s, size_t max, gfp_t gfp)
100{
101 size_t len;
102 char *buf;
103
104 if (!s)
105 return NULL;
106
107 len = strnlen(s, max);
108 buf = kmalloc_track_caller(len+1, gfp);
109 if (buf) {
110 memcpy(buf, s, len);
111 buf[len] = '\0';
112 }
113 return buf;
114}
115EXPORT_SYMBOL(kstrndup);
116
117/**
118 * kmemdup - duplicate region of memory
119 *
120 * @src: memory region to duplicate
121 * @len: memory region length
122 * @gfp: GFP mask to use
123 *
124 * Return: newly allocated copy of @src or %NULL in case of error,
125 * result is physically contiguous. Use kfree() to free.
126 */
127void *kmemdup(const void *src, size_t len, gfp_t gfp)
128{
129 void *p;
130
131 p = kmalloc_track_caller(len, gfp);
132 if (p)
133 memcpy(p, src, len);
134 return p;
135}
136EXPORT_SYMBOL(kmemdup);
137
138/**
139 * kmemdup_array - duplicate a given array.
140 *
141 * @src: array to duplicate.
142 * @element_size: size of each element of array.
143 * @count: number of elements to duplicate from array.
144 * @gfp: GFP mask to use.
145 *
146 * Return: duplicated array of @src or %NULL in case of error,
147 * result is physically contiguous. Use kfree() to free.
148 */
149void *kmemdup_array(const void *src, size_t element_size, size_t count, gfp_t gfp)
150{
151 return kmemdup(src, size_mul(element_size, count), gfp);
152}
153EXPORT_SYMBOL(kmemdup_array);
154
155/**
156 * kvmemdup - duplicate region of memory
157 *
158 * @src: memory region to duplicate
159 * @len: memory region length
160 * @gfp: GFP mask to use
161 *
162 * Return: newly allocated copy of @src or %NULL in case of error,
163 * result may be not physically contiguous. Use kvfree() to free.
164 */
165void *kvmemdup(const void *src, size_t len, gfp_t gfp)
166{
167 void *p;
168
169 p = kvmalloc(len, gfp);
170 if (p)
171 memcpy(p, src, len);
172 return p;
173}
174EXPORT_SYMBOL(kvmemdup);
175
176/**
177 * kmemdup_nul - Create a NUL-terminated string from unterminated data
178 * @s: The data to stringify
179 * @len: The size of the data
180 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
181 *
182 * Return: newly allocated copy of @s with NUL-termination or %NULL in
183 * case of error
184 */
185char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
186{
187 char *buf;
188
189 if (!s)
190 return NULL;
191
192 buf = kmalloc_track_caller(len + 1, gfp);
193 if (buf) {
194 memcpy(buf, s, len);
195 buf[len] = '\0';
196 }
197 return buf;
198}
199EXPORT_SYMBOL(kmemdup_nul);
200
201/**
202 * memdup_user - duplicate memory region from user space
203 *
204 * @src: source address in user space
205 * @len: number of bytes to copy
206 *
207 * Return: an ERR_PTR() on failure. Result is physically
208 * contiguous, to be freed by kfree().
209 */
210void *memdup_user(const void __user *src, size_t len)
211{
212 void *p;
213
214 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
215 if (!p)
216 return ERR_PTR(-ENOMEM);
217
218 if (copy_from_user(p, src, len)) {
219 kfree(p);
220 return ERR_PTR(-EFAULT);
221 }
222
223 return p;
224}
225EXPORT_SYMBOL(memdup_user);
226
227/**
228 * vmemdup_user - duplicate memory region from user space
229 *
230 * @src: source address in user space
231 * @len: number of bytes to copy
232 *
233 * Return: an ERR_PTR() on failure. Result may be not
234 * physically contiguous. Use kvfree() to free.
235 */
236void *vmemdup_user(const void __user *src, size_t len)
237{
238 void *p;
239
240 p = kvmalloc(len, GFP_USER);
241 if (!p)
242 return ERR_PTR(-ENOMEM);
243
244 if (copy_from_user(p, src, len)) {
245 kvfree(p);
246 return ERR_PTR(-EFAULT);
247 }
248
249 return p;
250}
251EXPORT_SYMBOL(vmemdup_user);
252
253/**
254 * strndup_user - duplicate an existing string from user space
255 * @s: The string to duplicate
256 * @n: Maximum number of bytes to copy, including the trailing NUL.
257 *
258 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
259 */
260char *strndup_user(const char __user *s, long n)
261{
262 char *p;
263 long length;
264
265 length = strnlen_user(s, n);
266
267 if (!length)
268 return ERR_PTR(-EFAULT);
269
270 if (length > n)
271 return ERR_PTR(-EINVAL);
272
273 p = memdup_user(s, length);
274
275 if (IS_ERR(p))
276 return p;
277
278 p[length - 1] = '\0';
279
280 return p;
281}
282EXPORT_SYMBOL(strndup_user);
283
284/**
285 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
286 *
287 * @src: source address in user space
288 * @len: number of bytes to copy
289 *
290 * Return: an ERR_PTR() on failure.
291 */
292void *memdup_user_nul(const void __user *src, size_t len)
293{
294 char *p;
295
296 /*
297 * Always use GFP_KERNEL, since copy_from_user() can sleep and
298 * cause pagefault, which makes it pointless to use GFP_NOFS
299 * or GFP_ATOMIC.
300 */
301 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
302 if (!p)
303 return ERR_PTR(-ENOMEM);
304
305 if (copy_from_user(p, src, len)) {
306 kfree(p);
307 return ERR_PTR(-EFAULT);
308 }
309 p[len] = '\0';
310
311 return p;
312}
313EXPORT_SYMBOL(memdup_user_nul);
314
315/* Check if the vma is being used as a stack by this task */
316int vma_is_stack_for_current(struct vm_area_struct *vma)
317{
318 struct task_struct * __maybe_unused t = current;
319
320 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
321}
322
323/*
324 * Change backing file, only valid to use during initial VMA setup.
325 */
326void vma_set_file(struct vm_area_struct *vma, struct file *file)
327{
328 /* Changing an anonymous vma with this is illegal */
329 get_file(file);
330 swap(vma->vm_file, file);
331 fput(file);
332}
333EXPORT_SYMBOL(vma_set_file);
334
335#ifndef STACK_RND_MASK
336#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
337#endif
338
339unsigned long randomize_stack_top(unsigned long stack_top)
340{
341 unsigned long random_variable = 0;
342
343 if (current->flags & PF_RANDOMIZE) {
344 random_variable = get_random_long();
345 random_variable &= STACK_RND_MASK;
346 random_variable <<= PAGE_SHIFT;
347 }
348#ifdef CONFIG_STACK_GROWSUP
349 return PAGE_ALIGN(stack_top) + random_variable;
350#else
351 return PAGE_ALIGN(stack_top) - random_variable;
352#endif
353}
354
355/**
356 * randomize_page - Generate a random, page aligned address
357 * @start: The smallest acceptable address the caller will take.
358 * @range: The size of the area, starting at @start, within which the
359 * random address must fall.
360 *
361 * If @start + @range would overflow, @range is capped.
362 *
363 * NOTE: Historical use of randomize_range, which this replaces, presumed that
364 * @start was already page aligned. We now align it regardless.
365 *
366 * Return: A page aligned address within [start, start + range). On error,
367 * @start is returned.
368 */
369unsigned long randomize_page(unsigned long start, unsigned long range)
370{
371 if (!PAGE_ALIGNED(start)) {
372 range -= PAGE_ALIGN(start) - start;
373 start = PAGE_ALIGN(start);
374 }
375
376 if (start > ULONG_MAX - range)
377 range = ULONG_MAX - start;
378
379 range >>= PAGE_SHIFT;
380
381 if (range == 0)
382 return start;
383
384 return start + (get_random_long() % range << PAGE_SHIFT);
385}
386
387#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
388unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
389{
390 /* Is the current task 32bit ? */
391 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
392 return randomize_page(mm->brk, SZ_32M);
393
394 return randomize_page(mm->brk, SZ_1G);
395}
396
397unsigned long arch_mmap_rnd(void)
398{
399 unsigned long rnd;
400
401#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
402 if (is_compat_task())
403 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
404 else
405#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
406 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
407
408 return rnd << PAGE_SHIFT;
409}
410
411static int mmap_is_legacy(struct rlimit *rlim_stack)
412{
413 if (current->personality & ADDR_COMPAT_LAYOUT)
414 return 1;
415
416 /* On parisc the stack always grows up - so a unlimited stack should
417 * not be an indicator to use the legacy memory layout. */
418 if (rlim_stack->rlim_cur == RLIM_INFINITY &&
419 !IS_ENABLED(CONFIG_STACK_GROWSUP))
420 return 1;
421
422 return sysctl_legacy_va_layout;
423}
424
425/*
426 * Leave enough space between the mmap area and the stack to honour ulimit in
427 * the face of randomisation.
428 */
429#define MIN_GAP (SZ_128M)
430#define MAX_GAP (STACK_TOP / 6 * 5)
431
432static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
433{
434#ifdef CONFIG_STACK_GROWSUP
435 /*
436 * For an upwards growing stack the calculation is much simpler.
437 * Memory for the maximum stack size is reserved at the top of the
438 * task. mmap_base starts directly below the stack and grows
439 * downwards.
440 */
441 return PAGE_ALIGN_DOWN(mmap_upper_limit(rlim_stack) - rnd);
442#else
443 unsigned long gap = rlim_stack->rlim_cur;
444 unsigned long pad = stack_guard_gap;
445
446 /* Account for stack randomization if necessary */
447 if (current->flags & PF_RANDOMIZE)
448 pad += (STACK_RND_MASK << PAGE_SHIFT);
449
450 /* Values close to RLIM_INFINITY can overflow. */
451 if (gap + pad > gap)
452 gap += pad;
453
454 if (gap < MIN_GAP)
455 gap = MIN_GAP;
456 else if (gap > MAX_GAP)
457 gap = MAX_GAP;
458
459 return PAGE_ALIGN(STACK_TOP - gap - rnd);
460#endif
461}
462
463void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
464{
465 unsigned long random_factor = 0UL;
466
467 if (current->flags & PF_RANDOMIZE)
468 random_factor = arch_mmap_rnd();
469
470 if (mmap_is_legacy(rlim_stack)) {
471 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
472 mm->get_unmapped_area = arch_get_unmapped_area;
473 } else {
474 mm->mmap_base = mmap_base(random_factor, rlim_stack);
475 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
476 }
477}
478#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
479void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
480{
481 mm->mmap_base = TASK_UNMAPPED_BASE;
482 mm->get_unmapped_area = arch_get_unmapped_area;
483}
484#endif
485
486/**
487 * __account_locked_vm - account locked pages to an mm's locked_vm
488 * @mm: mm to account against
489 * @pages: number of pages to account
490 * @inc: %true if @pages should be considered positive, %false if not
491 * @task: task used to check RLIMIT_MEMLOCK
492 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
493 *
494 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
495 * that mmap_lock is held as writer.
496 *
497 * Return:
498 * * 0 on success
499 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
500 */
501int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
502 struct task_struct *task, bool bypass_rlim)
503{
504 unsigned long locked_vm, limit;
505 int ret = 0;
506
507 mmap_assert_write_locked(mm);
508
509 locked_vm = mm->locked_vm;
510 if (inc) {
511 if (!bypass_rlim) {
512 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
513 if (locked_vm + pages > limit)
514 ret = -ENOMEM;
515 }
516 if (!ret)
517 mm->locked_vm = locked_vm + pages;
518 } else {
519 WARN_ON_ONCE(pages > locked_vm);
520 mm->locked_vm = locked_vm - pages;
521 }
522
523 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
524 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
525 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
526 ret ? " - exceeded" : "");
527
528 return ret;
529}
530EXPORT_SYMBOL_GPL(__account_locked_vm);
531
532/**
533 * account_locked_vm - account locked pages to an mm's locked_vm
534 * @mm: mm to account against, may be NULL
535 * @pages: number of pages to account
536 * @inc: %true if @pages should be considered positive, %false if not
537 *
538 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
539 *
540 * Return:
541 * * 0 on success, or if mm is NULL
542 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
543 */
544int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
545{
546 int ret;
547
548 if (pages == 0 || !mm)
549 return 0;
550
551 mmap_write_lock(mm);
552 ret = __account_locked_vm(mm, pages, inc, current,
553 capable(CAP_IPC_LOCK));
554 mmap_write_unlock(mm);
555
556 return ret;
557}
558EXPORT_SYMBOL_GPL(account_locked_vm);
559
560unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
561 unsigned long len, unsigned long prot,
562 unsigned long flag, unsigned long pgoff)
563{
564 unsigned long ret;
565 struct mm_struct *mm = current->mm;
566 unsigned long populate;
567 LIST_HEAD(uf);
568
569 ret = security_mmap_file(file, prot, flag);
570 if (!ret) {
571 if (mmap_write_lock_killable(mm))
572 return -EINTR;
573 ret = do_mmap(file, addr, len, prot, flag, 0, pgoff, &populate,
574 &uf);
575 mmap_write_unlock(mm);
576 userfaultfd_unmap_complete(mm, &uf);
577 if (populate)
578 mm_populate(ret, populate);
579 }
580 return ret;
581}
582
583unsigned long vm_mmap(struct file *file, unsigned long addr,
584 unsigned long len, unsigned long prot,
585 unsigned long flag, unsigned long offset)
586{
587 if (unlikely(offset + PAGE_ALIGN(len) < offset))
588 return -EINVAL;
589 if (unlikely(offset_in_page(offset)))
590 return -EINVAL;
591
592 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
593}
594EXPORT_SYMBOL(vm_mmap);
595
596/**
597 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
598 * failure, fall back to non-contiguous (vmalloc) allocation.
599 * @size: size of the request.
600 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
601 * @node: numa node to allocate from
602 *
603 * Uses kmalloc to get the memory but if the allocation fails then falls back
604 * to the vmalloc allocator. Use kvfree for freeing the memory.
605 *
606 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
607 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
608 * preferable to the vmalloc fallback, due to visible performance drawbacks.
609 *
610 * Return: pointer to the allocated memory of %NULL in case of failure
611 */
612void *kvmalloc_node(size_t size, gfp_t flags, int node)
613{
614 gfp_t kmalloc_flags = flags;
615 void *ret;
616
617 /*
618 * We want to attempt a large physically contiguous block first because
619 * it is less likely to fragment multiple larger blocks and therefore
620 * contribute to a long term fragmentation less than vmalloc fallback.
621 * However make sure that larger requests are not too disruptive - no
622 * OOM killer and no allocation failure warnings as we have a fallback.
623 */
624 if (size > PAGE_SIZE) {
625 kmalloc_flags |= __GFP_NOWARN;
626
627 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
628 kmalloc_flags |= __GFP_NORETRY;
629
630 /* nofail semantic is implemented by the vmalloc fallback */
631 kmalloc_flags &= ~__GFP_NOFAIL;
632 }
633
634 ret = kmalloc_node(size, kmalloc_flags, node);
635
636 /*
637 * It doesn't really make sense to fallback to vmalloc for sub page
638 * requests
639 */
640 if (ret || size <= PAGE_SIZE)
641 return ret;
642
643 /* non-sleeping allocations are not supported by vmalloc */
644 if (!gfpflags_allow_blocking(flags))
645 return NULL;
646
647 /* Don't even allow crazy sizes */
648 if (unlikely(size > INT_MAX)) {
649 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
650 return NULL;
651 }
652
653 /*
654 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
655 * since the callers already cannot assume anything
656 * about the resulting pointer, and cannot play
657 * protection games.
658 */
659 return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
660 flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
661 node, __builtin_return_address(0));
662}
663EXPORT_SYMBOL(kvmalloc_node);
664
665/**
666 * kvfree() - Free memory.
667 * @addr: Pointer to allocated memory.
668 *
669 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
670 * It is slightly more efficient to use kfree() or vfree() if you are certain
671 * that you know which one to use.
672 *
673 * Context: Either preemptible task context or not-NMI interrupt.
674 */
675void kvfree(const void *addr)
676{
677 if (is_vmalloc_addr(addr))
678 vfree(addr);
679 else
680 kfree(addr);
681}
682EXPORT_SYMBOL(kvfree);
683
684/**
685 * kvfree_sensitive - Free a data object containing sensitive information.
686 * @addr: address of the data object to be freed.
687 * @len: length of the data object.
688 *
689 * Use the special memzero_explicit() function to clear the content of a
690 * kvmalloc'ed object containing sensitive data to make sure that the
691 * compiler won't optimize out the data clearing.
692 */
693void kvfree_sensitive(const void *addr, size_t len)
694{
695 if (likely(!ZERO_OR_NULL_PTR(addr))) {
696 memzero_explicit((void *)addr, len);
697 kvfree(addr);
698 }
699}
700EXPORT_SYMBOL(kvfree_sensitive);
701
702void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
703{
704 void *newp;
705
706 if (oldsize >= newsize)
707 return (void *)p;
708 newp = kvmalloc(newsize, flags);
709 if (!newp)
710 return NULL;
711 memcpy(newp, p, oldsize);
712 kvfree(p);
713 return newp;
714}
715EXPORT_SYMBOL(kvrealloc);
716
717/**
718 * __vmalloc_array - allocate memory for a virtually contiguous array.
719 * @n: number of elements.
720 * @size: element size.
721 * @flags: the type of memory to allocate (see kmalloc).
722 */
723void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
724{
725 size_t bytes;
726
727 if (unlikely(check_mul_overflow(n, size, &bytes)))
728 return NULL;
729 return __vmalloc(bytes, flags);
730}
731EXPORT_SYMBOL(__vmalloc_array);
732
733/**
734 * vmalloc_array - allocate memory for a virtually contiguous array.
735 * @n: number of elements.
736 * @size: element size.
737 */
738void *vmalloc_array(size_t n, size_t size)
739{
740 return __vmalloc_array(n, size, GFP_KERNEL);
741}
742EXPORT_SYMBOL(vmalloc_array);
743
744/**
745 * __vcalloc - allocate and zero memory for a virtually contiguous array.
746 * @n: number of elements.
747 * @size: element size.
748 * @flags: the type of memory to allocate (see kmalloc).
749 */
750void *__vcalloc(size_t n, size_t size, gfp_t flags)
751{
752 return __vmalloc_array(n, size, flags | __GFP_ZERO);
753}
754EXPORT_SYMBOL(__vcalloc);
755
756/**
757 * vcalloc - allocate and zero memory for a virtually contiguous array.
758 * @n: number of elements.
759 * @size: element size.
760 */
761void *vcalloc(size_t n, size_t size)
762{
763 return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
764}
765EXPORT_SYMBOL(vcalloc);
766
767struct anon_vma *folio_anon_vma(struct folio *folio)
768{
769 unsigned long mapping = (unsigned long)folio->mapping;
770
771 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
772 return NULL;
773 return (void *)(mapping - PAGE_MAPPING_ANON);
774}
775
776/**
777 * folio_mapping - Find the mapping where this folio is stored.
778 * @folio: The folio.
779 *
780 * For folios which are in the page cache, return the mapping that this
781 * page belongs to. Folios in the swap cache return the swap mapping
782 * this page is stored in (which is different from the mapping for the
783 * swap file or swap device where the data is stored).
784 *
785 * You can call this for folios which aren't in the swap cache or page
786 * cache and it will return NULL.
787 */
788struct address_space *folio_mapping(struct folio *folio)
789{
790 struct address_space *mapping;
791
792 /* This happens if someone calls flush_dcache_page on slab page */
793 if (unlikely(folio_test_slab(folio)))
794 return NULL;
795
796 if (unlikely(folio_test_swapcache(folio)))
797 return swap_address_space(folio->swap);
798
799 mapping = folio->mapping;
800 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
801 return NULL;
802
803 return mapping;
804}
805EXPORT_SYMBOL(folio_mapping);
806
807/**
808 * folio_copy - Copy the contents of one folio to another.
809 * @dst: Folio to copy to.
810 * @src: Folio to copy from.
811 *
812 * The bytes in the folio represented by @src are copied to @dst.
813 * Assumes the caller has validated that @dst is at least as large as @src.
814 * Can be called in atomic context for order-0 folios, but if the folio is
815 * larger, it may sleep.
816 */
817void folio_copy(struct folio *dst, struct folio *src)
818{
819 long i = 0;
820 long nr = folio_nr_pages(src);
821
822 for (;;) {
823 copy_highpage(folio_page(dst, i), folio_page(src, i));
824 if (++i == nr)
825 break;
826 cond_resched();
827 }
828}
829EXPORT_SYMBOL(folio_copy);
830
831int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
832int sysctl_overcommit_ratio __read_mostly = 50;
833unsigned long sysctl_overcommit_kbytes __read_mostly;
834int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
835unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
836unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
837
838int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
839 size_t *lenp, loff_t *ppos)
840{
841 int ret;
842
843 ret = proc_dointvec(table, write, buffer, lenp, ppos);
844 if (ret == 0 && write)
845 sysctl_overcommit_kbytes = 0;
846 return ret;
847}
848
849static void sync_overcommit_as(struct work_struct *dummy)
850{
851 percpu_counter_sync(&vm_committed_as);
852}
853
854int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
855 size_t *lenp, loff_t *ppos)
856{
857 struct ctl_table t;
858 int new_policy = -1;
859 int ret;
860
861 /*
862 * The deviation of sync_overcommit_as could be big with loose policy
863 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
864 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
865 * with the strict "NEVER", and to avoid possible race condition (even
866 * though user usually won't too frequently do the switching to policy
867 * OVERCOMMIT_NEVER), the switch is done in the following order:
868 * 1. changing the batch
869 * 2. sync percpu count on each CPU
870 * 3. switch the policy
871 */
872 if (write) {
873 t = *table;
874 t.data = &new_policy;
875 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
876 if (ret || new_policy == -1)
877 return ret;
878
879 mm_compute_batch(new_policy);
880 if (new_policy == OVERCOMMIT_NEVER)
881 schedule_on_each_cpu(sync_overcommit_as);
882 sysctl_overcommit_memory = new_policy;
883 } else {
884 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
885 }
886
887 return ret;
888}
889
890int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
891 size_t *lenp, loff_t *ppos)
892{
893 int ret;
894
895 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
896 if (ret == 0 && write)
897 sysctl_overcommit_ratio = 0;
898 return ret;
899}
900
901/*
902 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
903 */
904unsigned long vm_commit_limit(void)
905{
906 unsigned long allowed;
907
908 if (sysctl_overcommit_kbytes)
909 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
910 else
911 allowed = ((totalram_pages() - hugetlb_total_pages())
912 * sysctl_overcommit_ratio / 100);
913 allowed += total_swap_pages;
914
915 return allowed;
916}
917
918/*
919 * Make sure vm_committed_as in one cacheline and not cacheline shared with
920 * other variables. It can be updated by several CPUs frequently.
921 */
922struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
923
924/*
925 * The global memory commitment made in the system can be a metric
926 * that can be used to drive ballooning decisions when Linux is hosted
927 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
928 * balancing memory across competing virtual machines that are hosted.
929 * Several metrics drive this policy engine including the guest reported
930 * memory commitment.
931 *
932 * The time cost of this is very low for small platforms, and for big
933 * platform like a 2S/36C/72T Skylake server, in worst case where
934 * vm_committed_as's spinlock is under severe contention, the time cost
935 * could be about 30~40 microseconds.
936 */
937unsigned long vm_memory_committed(void)
938{
939 return percpu_counter_sum_positive(&vm_committed_as);
940}
941EXPORT_SYMBOL_GPL(vm_memory_committed);
942
943/*
944 * Check that a process has enough memory to allocate a new virtual
945 * mapping. 0 means there is enough memory for the allocation to
946 * succeed and -ENOMEM implies there is not.
947 *
948 * We currently support three overcommit policies, which are set via the
949 * vm.overcommit_memory sysctl. See Documentation/mm/overcommit-accounting.rst
950 *
951 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
952 * Additional code 2002 Jul 20 by Robert Love.
953 *
954 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
955 *
956 * Note this is a helper function intended to be used by LSMs which
957 * wish to use this logic.
958 */
959int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
960{
961 long allowed;
962 unsigned long bytes_failed;
963
964 vm_acct_memory(pages);
965
966 /*
967 * Sometimes we want to use more memory than we have
968 */
969 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
970 return 0;
971
972 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
973 if (pages > totalram_pages() + total_swap_pages)
974 goto error;
975 return 0;
976 }
977
978 allowed = vm_commit_limit();
979 /*
980 * Reserve some for root
981 */
982 if (!cap_sys_admin)
983 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
984
985 /*
986 * Don't let a single process grow so big a user can't recover
987 */
988 if (mm) {
989 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
990
991 allowed -= min_t(long, mm->total_vm / 32, reserve);
992 }
993
994 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
995 return 0;
996error:
997 bytes_failed = pages << PAGE_SHIFT;
998 pr_warn_ratelimited("%s: pid: %d, comm: %s, bytes: %lu not enough memory for the allocation\n",
999 __func__, current->pid, current->comm, bytes_failed);
1000 vm_unacct_memory(pages);
1001
1002 return -ENOMEM;
1003}
1004
1005/**
1006 * get_cmdline() - copy the cmdline value to a buffer.
1007 * @task: the task whose cmdline value to copy.
1008 * @buffer: the buffer to copy to.
1009 * @buflen: the length of the buffer. Larger cmdline values are truncated
1010 * to this length.
1011 *
1012 * Return: the size of the cmdline field copied. Note that the copy does
1013 * not guarantee an ending NULL byte.
1014 */
1015int get_cmdline(struct task_struct *task, char *buffer, int buflen)
1016{
1017 int res = 0;
1018 unsigned int len;
1019 struct mm_struct *mm = get_task_mm(task);
1020 unsigned long arg_start, arg_end, env_start, env_end;
1021 if (!mm)
1022 goto out;
1023 if (!mm->arg_end)
1024 goto out_mm; /* Shh! No looking before we're done */
1025
1026 spin_lock(&mm->arg_lock);
1027 arg_start = mm->arg_start;
1028 arg_end = mm->arg_end;
1029 env_start = mm->env_start;
1030 env_end = mm->env_end;
1031 spin_unlock(&mm->arg_lock);
1032
1033 len = arg_end - arg_start;
1034
1035 if (len > buflen)
1036 len = buflen;
1037
1038 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1039
1040 /*
1041 * If the nul at the end of args has been overwritten, then
1042 * assume application is using setproctitle(3).
1043 */
1044 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1045 len = strnlen(buffer, res);
1046 if (len < res) {
1047 res = len;
1048 } else {
1049 len = env_end - env_start;
1050 if (len > buflen - res)
1051 len = buflen - res;
1052 res += access_process_vm(task, env_start,
1053 buffer+res, len,
1054 FOLL_FORCE);
1055 res = strnlen(buffer, res);
1056 }
1057 }
1058out_mm:
1059 mmput(mm);
1060out:
1061 return res;
1062}
1063
1064int __weak memcmp_pages(struct page *page1, struct page *page2)
1065{
1066 char *addr1, *addr2;
1067 int ret;
1068
1069 addr1 = kmap_local_page(page1);
1070 addr2 = kmap_local_page(page2);
1071 ret = memcmp(addr1, addr2, PAGE_SIZE);
1072 kunmap_local(addr2);
1073 kunmap_local(addr1);
1074 return ret;
1075}
1076
1077#ifdef CONFIG_PRINTK
1078/**
1079 * mem_dump_obj - Print available provenance information
1080 * @object: object for which to find provenance information.
1081 *
1082 * This function uses pr_cont(), so that the caller is expected to have
1083 * printed out whatever preamble is appropriate. The provenance information
1084 * depends on the type of object and on how much debugging is enabled.
1085 * For example, for a slab-cache object, the slab name is printed, and,
1086 * if available, the return address and stack trace from the allocation
1087 * and last free path of that object.
1088 */
1089void mem_dump_obj(void *object)
1090{
1091 const char *type;
1092
1093 if (kmem_dump_obj(object))
1094 return;
1095
1096 if (vmalloc_dump_obj(object))
1097 return;
1098
1099 if (is_vmalloc_addr(object))
1100 type = "vmalloc memory";
1101 else if (virt_addr_valid(object))
1102 type = "non-slab/vmalloc memory";
1103 else if (object == NULL)
1104 type = "NULL pointer";
1105 else if (object == ZERO_SIZE_PTR)
1106 type = "zero-size pointer";
1107 else
1108 type = "non-paged memory";
1109
1110 pr_cont(" %s\n", type);
1111}
1112EXPORT_SYMBOL_GPL(mem_dump_obj);
1113#endif
1114
1115/*
1116 * A driver might set a page logically offline -- PageOffline() -- and
1117 * turn the page inaccessible in the hypervisor; after that, access to page
1118 * content can be fatal.
1119 *
1120 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1121 * pages after checking PageOffline(); however, these PFN walkers can race
1122 * with drivers that set PageOffline().
1123 *
1124 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1125 * synchronize with such drivers, achieving that a page cannot be set
1126 * PageOffline() while frozen.
1127 *
1128 * page_offline_begin()/page_offline_end() is used by drivers that care about
1129 * such races when setting a page PageOffline().
1130 */
1131static DECLARE_RWSEM(page_offline_rwsem);
1132
1133void page_offline_freeze(void)
1134{
1135 down_read(&page_offline_rwsem);
1136}
1137
1138void page_offline_thaw(void)
1139{
1140 up_read(&page_offline_rwsem);
1141}
1142
1143void page_offline_begin(void)
1144{
1145 down_write(&page_offline_rwsem);
1146}
1147EXPORT_SYMBOL(page_offline_begin);
1148
1149void page_offline_end(void)
1150{
1151 up_write(&page_offline_rwsem);
1152}
1153EXPORT_SYMBOL(page_offline_end);
1154
1155#ifndef flush_dcache_folio
1156void flush_dcache_folio(struct folio *folio)
1157{
1158 long i, nr = folio_nr_pages(folio);
1159
1160 for (i = 0; i < nr; i++)
1161 flush_dcache_page(folio_page(folio, i));
1162}
1163EXPORT_SYMBOL(flush_dcache_folio);
1164#endif