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#include <linux/mm.h>
2#include <linux/slab.h>
3#include <linux/string.h>
4#include <linux/compiler.h>
5#include <linux/export.h>
6#include <linux/err.h>
7#include <linux/sched.h>
8#include <linux/security.h>
9#include <linux/swap.h>
10#include <linux/swapops.h>
11#include <linux/mman.h>
12#include <linux/hugetlb.h>
13#include <linux/vmalloc.h>
14
15#include <asm/sections.h>
16#include <asm/uaccess.h>
17
18#include "internal.h"
19
20static inline int is_kernel_rodata(unsigned long addr)
21{
22 return addr >= (unsigned long)__start_rodata &&
23 addr < (unsigned long)__end_rodata;
24}
25
26/**
27 * kfree_const - conditionally free memory
28 * @x: pointer to the memory
29 *
30 * Function calls kfree only if @x is not in .rodata section.
31 */
32void kfree_const(const void *x)
33{
34 if (!is_kernel_rodata((unsigned long)x))
35 kfree(x);
36}
37EXPORT_SYMBOL(kfree_const);
38
39/**
40 * kstrdup - allocate space for and copy an existing string
41 * @s: the string to duplicate
42 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
43 */
44char *kstrdup(const char *s, gfp_t gfp)
45{
46 size_t len;
47 char *buf;
48
49 if (!s)
50 return NULL;
51
52 len = strlen(s) + 1;
53 buf = kmalloc_track_caller(len, gfp);
54 if (buf)
55 memcpy(buf, s, len);
56 return buf;
57}
58EXPORT_SYMBOL(kstrdup);
59
60/**
61 * kstrdup_const - conditionally duplicate an existing const string
62 * @s: the string to duplicate
63 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
64 *
65 * Function returns source string if it is in .rodata section otherwise it
66 * fallbacks to kstrdup.
67 * Strings allocated by kstrdup_const should be freed by kfree_const.
68 */
69const char *kstrdup_const(const char *s, gfp_t gfp)
70{
71 if (is_kernel_rodata((unsigned long)s))
72 return s;
73
74 return kstrdup(s, gfp);
75}
76EXPORT_SYMBOL(kstrdup_const);
77
78/**
79 * kstrndup - allocate space for and copy an existing string
80 * @s: the string to duplicate
81 * @max: read at most @max chars from @s
82 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
83 */
84char *kstrndup(const char *s, size_t max, gfp_t gfp)
85{
86 size_t len;
87 char *buf;
88
89 if (!s)
90 return NULL;
91
92 len = strnlen(s, max);
93 buf = kmalloc_track_caller(len+1, gfp);
94 if (buf) {
95 memcpy(buf, s, len);
96 buf[len] = '\0';
97 }
98 return buf;
99}
100EXPORT_SYMBOL(kstrndup);
101
102/**
103 * kmemdup - duplicate region of memory
104 *
105 * @src: memory region to duplicate
106 * @len: memory region length
107 * @gfp: GFP mask to use
108 */
109void *kmemdup(const void *src, size_t len, gfp_t gfp)
110{
111 void *p;
112
113 p = kmalloc_track_caller(len, gfp);
114 if (p)
115 memcpy(p, src, len);
116 return p;
117}
118EXPORT_SYMBOL(kmemdup);
119
120/**
121 * memdup_user - duplicate memory region from user space
122 *
123 * @src: source address in user space
124 * @len: number of bytes to copy
125 *
126 * Returns an ERR_PTR() on failure.
127 */
128void *memdup_user(const void __user *src, size_t len)
129{
130 void *p;
131
132 /*
133 * Always use GFP_KERNEL, since copy_from_user() can sleep and
134 * cause pagefault, which makes it pointless to use GFP_NOFS
135 * or GFP_ATOMIC.
136 */
137 p = kmalloc_track_caller(len, GFP_KERNEL);
138 if (!p)
139 return ERR_PTR(-ENOMEM);
140
141 if (copy_from_user(p, src, len)) {
142 kfree(p);
143 return ERR_PTR(-EFAULT);
144 }
145
146 return p;
147}
148EXPORT_SYMBOL(memdup_user);
149
150/*
151 * strndup_user - duplicate an existing string from user space
152 * @s: The string to duplicate
153 * @n: Maximum number of bytes to copy, including the trailing NUL.
154 */
155char *strndup_user(const char __user *s, long n)
156{
157 char *p;
158 long length;
159
160 length = strnlen_user(s, n);
161
162 if (!length)
163 return ERR_PTR(-EFAULT);
164
165 if (length > n)
166 return ERR_PTR(-EINVAL);
167
168 p = memdup_user(s, length);
169
170 if (IS_ERR(p))
171 return p;
172
173 p[length - 1] = '\0';
174
175 return p;
176}
177EXPORT_SYMBOL(strndup_user);
178
179/**
180 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
181 *
182 * @src: source address in user space
183 * @len: number of bytes to copy
184 *
185 * Returns an ERR_PTR() on failure.
186 */
187void *memdup_user_nul(const void __user *src, size_t len)
188{
189 char *p;
190
191 /*
192 * Always use GFP_KERNEL, since copy_from_user() can sleep and
193 * cause pagefault, which makes it pointless to use GFP_NOFS
194 * or GFP_ATOMIC.
195 */
196 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
197 if (!p)
198 return ERR_PTR(-ENOMEM);
199
200 if (copy_from_user(p, src, len)) {
201 kfree(p);
202 return ERR_PTR(-EFAULT);
203 }
204 p[len] = '\0';
205
206 return p;
207}
208EXPORT_SYMBOL(memdup_user_nul);
209
210void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
211 struct vm_area_struct *prev, struct rb_node *rb_parent)
212{
213 struct vm_area_struct *next;
214
215 vma->vm_prev = prev;
216 if (prev) {
217 next = prev->vm_next;
218 prev->vm_next = vma;
219 } else {
220 mm->mmap = vma;
221 if (rb_parent)
222 next = rb_entry(rb_parent,
223 struct vm_area_struct, vm_rb);
224 else
225 next = NULL;
226 }
227 vma->vm_next = next;
228 if (next)
229 next->vm_prev = vma;
230}
231
232/* Check if the vma is being used as a stack by this task */
233int vma_is_stack_for_task(struct vm_area_struct *vma, struct task_struct *t)
234{
235 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
236}
237
238#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
239void arch_pick_mmap_layout(struct mm_struct *mm)
240{
241 mm->mmap_base = TASK_UNMAPPED_BASE;
242 mm->get_unmapped_area = arch_get_unmapped_area;
243}
244#endif
245
246/*
247 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
248 * back to the regular GUP.
249 * If the architecture not support this function, simply return with no
250 * page pinned
251 */
252int __weak __get_user_pages_fast(unsigned long start,
253 int nr_pages, int write, struct page **pages)
254{
255 return 0;
256}
257EXPORT_SYMBOL_GPL(__get_user_pages_fast);
258
259/**
260 * get_user_pages_fast() - pin user pages in memory
261 * @start: starting user address
262 * @nr_pages: number of pages from start to pin
263 * @write: whether pages will be written to
264 * @pages: array that receives pointers to the pages pinned.
265 * Should be at least nr_pages long.
266 *
267 * Returns number of pages pinned. This may be fewer than the number
268 * requested. If nr_pages is 0 or negative, returns 0. If no pages
269 * were pinned, returns -errno.
270 *
271 * get_user_pages_fast provides equivalent functionality to get_user_pages,
272 * operating on current and current->mm, with force=0 and vma=NULL. However
273 * unlike get_user_pages, it must be called without mmap_sem held.
274 *
275 * get_user_pages_fast may take mmap_sem and page table locks, so no
276 * assumptions can be made about lack of locking. get_user_pages_fast is to be
277 * implemented in a way that is advantageous (vs get_user_pages()) when the
278 * user memory area is already faulted in and present in ptes. However if the
279 * pages have to be faulted in, it may turn out to be slightly slower so
280 * callers need to carefully consider what to use. On many architectures,
281 * get_user_pages_fast simply falls back to get_user_pages.
282 */
283int __weak get_user_pages_fast(unsigned long start,
284 int nr_pages, int write, struct page **pages)
285{
286 return get_user_pages_unlocked(start, nr_pages, write, 0, pages);
287}
288EXPORT_SYMBOL_GPL(get_user_pages_fast);
289
290unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
291 unsigned long len, unsigned long prot,
292 unsigned long flag, unsigned long pgoff)
293{
294 unsigned long ret;
295 struct mm_struct *mm = current->mm;
296 unsigned long populate;
297
298 ret = security_mmap_file(file, prot, flag);
299 if (!ret) {
300 down_write(&mm->mmap_sem);
301 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
302 &populate);
303 up_write(&mm->mmap_sem);
304 if (populate)
305 mm_populate(ret, populate);
306 }
307 return ret;
308}
309
310unsigned long vm_mmap(struct file *file, unsigned long addr,
311 unsigned long len, unsigned long prot,
312 unsigned long flag, unsigned long offset)
313{
314 if (unlikely(offset + PAGE_ALIGN(len) < offset))
315 return -EINVAL;
316 if (unlikely(offset_in_page(offset)))
317 return -EINVAL;
318
319 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
320}
321EXPORT_SYMBOL(vm_mmap);
322
323void kvfree(const void *addr)
324{
325 if (is_vmalloc_addr(addr))
326 vfree(addr);
327 else
328 kfree(addr);
329}
330EXPORT_SYMBOL(kvfree);
331
332static inline void *__page_rmapping(struct page *page)
333{
334 unsigned long mapping;
335
336 mapping = (unsigned long)page->mapping;
337 mapping &= ~PAGE_MAPPING_FLAGS;
338
339 return (void *)mapping;
340}
341
342/* Neutral page->mapping pointer to address_space or anon_vma or other */
343void *page_rmapping(struct page *page)
344{
345 page = compound_head(page);
346 return __page_rmapping(page);
347}
348
349struct anon_vma *page_anon_vma(struct page *page)
350{
351 unsigned long mapping;
352
353 page = compound_head(page);
354 mapping = (unsigned long)page->mapping;
355 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
356 return NULL;
357 return __page_rmapping(page);
358}
359
360struct address_space *page_mapping(struct page *page)
361{
362 struct address_space *mapping;
363
364 page = compound_head(page);
365
366 /* This happens if someone calls flush_dcache_page on slab page */
367 if (unlikely(PageSlab(page)))
368 return NULL;
369
370 if (unlikely(PageSwapCache(page))) {
371 swp_entry_t entry;
372
373 entry.val = page_private(page);
374 return swap_address_space(entry);
375 }
376
377 mapping = page->mapping;
378 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
379 return NULL;
380 return mapping;
381}
382
383/* Slow path of page_mapcount() for compound pages */
384int __page_mapcount(struct page *page)
385{
386 int ret;
387
388 ret = atomic_read(&page->_mapcount) + 1;
389 page = compound_head(page);
390 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
391 if (PageDoubleMap(page))
392 ret--;
393 return ret;
394}
395EXPORT_SYMBOL_GPL(__page_mapcount);
396
397int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
398int sysctl_overcommit_ratio __read_mostly = 50;
399unsigned long sysctl_overcommit_kbytes __read_mostly;
400int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
401unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
402unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
403
404int overcommit_ratio_handler(struct ctl_table *table, int write,
405 void __user *buffer, size_t *lenp,
406 loff_t *ppos)
407{
408 int ret;
409
410 ret = proc_dointvec(table, write, buffer, lenp, ppos);
411 if (ret == 0 && write)
412 sysctl_overcommit_kbytes = 0;
413 return ret;
414}
415
416int overcommit_kbytes_handler(struct ctl_table *table, int write,
417 void __user *buffer, size_t *lenp,
418 loff_t *ppos)
419{
420 int ret;
421
422 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
423 if (ret == 0 && write)
424 sysctl_overcommit_ratio = 0;
425 return ret;
426}
427
428/*
429 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
430 */
431unsigned long vm_commit_limit(void)
432{
433 unsigned long allowed;
434
435 if (sysctl_overcommit_kbytes)
436 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
437 else
438 allowed = ((totalram_pages - hugetlb_total_pages())
439 * sysctl_overcommit_ratio / 100);
440 allowed += total_swap_pages;
441
442 return allowed;
443}
444
445/*
446 * Make sure vm_committed_as in one cacheline and not cacheline shared with
447 * other variables. It can be updated by several CPUs frequently.
448 */
449struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
450
451/*
452 * The global memory commitment made in the system can be a metric
453 * that can be used to drive ballooning decisions when Linux is hosted
454 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
455 * balancing memory across competing virtual machines that are hosted.
456 * Several metrics drive this policy engine including the guest reported
457 * memory commitment.
458 */
459unsigned long vm_memory_committed(void)
460{
461 return percpu_counter_read_positive(&vm_committed_as);
462}
463EXPORT_SYMBOL_GPL(vm_memory_committed);
464
465/*
466 * Check that a process has enough memory to allocate a new virtual
467 * mapping. 0 means there is enough memory for the allocation to
468 * succeed and -ENOMEM implies there is not.
469 *
470 * We currently support three overcommit policies, which are set via the
471 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
472 *
473 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
474 * Additional code 2002 Jul 20 by Robert Love.
475 *
476 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
477 *
478 * Note this is a helper function intended to be used by LSMs which
479 * wish to use this logic.
480 */
481int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
482{
483 long free, allowed, reserve;
484
485 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
486 -(s64)vm_committed_as_batch * num_online_cpus(),
487 "memory commitment underflow");
488
489 vm_acct_memory(pages);
490
491 /*
492 * Sometimes we want to use more memory than we have
493 */
494 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
495 return 0;
496
497 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
498 free = global_page_state(NR_FREE_PAGES);
499 free += global_page_state(NR_FILE_PAGES);
500
501 /*
502 * shmem pages shouldn't be counted as free in this
503 * case, they can't be purged, only swapped out, and
504 * that won't affect the overall amount of available
505 * memory in the system.
506 */
507 free -= global_page_state(NR_SHMEM);
508
509 free += get_nr_swap_pages();
510
511 /*
512 * Any slabs which are created with the
513 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
514 * which are reclaimable, under pressure. The dentry
515 * cache and most inode caches should fall into this
516 */
517 free += global_page_state(NR_SLAB_RECLAIMABLE);
518
519 /*
520 * Leave reserved pages. The pages are not for anonymous pages.
521 */
522 if (free <= totalreserve_pages)
523 goto error;
524 else
525 free -= totalreserve_pages;
526
527 /*
528 * Reserve some for root
529 */
530 if (!cap_sys_admin)
531 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
532
533 if (free > pages)
534 return 0;
535
536 goto error;
537 }
538
539 allowed = vm_commit_limit();
540 /*
541 * Reserve some for root
542 */
543 if (!cap_sys_admin)
544 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
545
546 /*
547 * Don't let a single process grow so big a user can't recover
548 */
549 if (mm) {
550 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
551 allowed -= min_t(long, mm->total_vm / 32, reserve);
552 }
553
554 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
555 return 0;
556error:
557 vm_unacct_memory(pages);
558
559 return -ENOMEM;
560}
561
562/**
563 * get_cmdline() - copy the cmdline value to a buffer.
564 * @task: the task whose cmdline value to copy.
565 * @buffer: the buffer to copy to.
566 * @buflen: the length of the buffer. Larger cmdline values are truncated
567 * to this length.
568 * Returns the size of the cmdline field copied. Note that the copy does
569 * not guarantee an ending NULL byte.
570 */
571int get_cmdline(struct task_struct *task, char *buffer, int buflen)
572{
573 int res = 0;
574 unsigned int len;
575 struct mm_struct *mm = get_task_mm(task);
576 unsigned long arg_start, arg_end, env_start, env_end;
577 if (!mm)
578 goto out;
579 if (!mm->arg_end)
580 goto out_mm; /* Shh! No looking before we're done */
581
582 down_read(&mm->mmap_sem);
583 arg_start = mm->arg_start;
584 arg_end = mm->arg_end;
585 env_start = mm->env_start;
586 env_end = mm->env_end;
587 up_read(&mm->mmap_sem);
588
589 len = arg_end - arg_start;
590
591 if (len > buflen)
592 len = buflen;
593
594 res = access_process_vm(task, arg_start, buffer, len, 0);
595
596 /*
597 * If the nul at the end of args has been overwritten, then
598 * assume application is using setproctitle(3).
599 */
600 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
601 len = strnlen(buffer, res);
602 if (len < res) {
603 res = len;
604 } else {
605 len = env_end - env_start;
606 if (len > buflen - res)
607 len = buflen - res;
608 res += access_process_vm(task, env_start,
609 buffer+res, len, 0);
610 res = strnlen(buffer, res);
611 }
612 }
613out_mm:
614 mmput(mm);
615out:
616 return res;
617}