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)
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 next = mm->mmap;
284 mm->mmap = vma;
285 }
286 vma->vm_next = next;
287 if (next)
288 next->vm_prev = vma;
289}
290
291void __vma_unlink_list(struct mm_struct *mm, struct vm_area_struct *vma)
292{
293 struct vm_area_struct *prev, *next;
294
295 next = vma->vm_next;
296 prev = vma->vm_prev;
297 if (prev)
298 prev->vm_next = next;
299 else
300 mm->mmap = next;
301 if (next)
302 next->vm_prev = prev;
303}
304
305/* Check if the vma is being used as a stack by this task */
306int vma_is_stack_for_current(struct vm_area_struct *vma)
307{
308 struct task_struct * __maybe_unused t = current;
309
310 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
311}
312
313#ifndef STACK_RND_MASK
314#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
315#endif
316
317unsigned long randomize_stack_top(unsigned long stack_top)
318{
319 unsigned long random_variable = 0;
320
321 if (current->flags & PF_RANDOMIZE) {
322 random_variable = get_random_long();
323 random_variable &= STACK_RND_MASK;
324 random_variable <<= PAGE_SHIFT;
325 }
326#ifdef CONFIG_STACK_GROWSUP
327 return PAGE_ALIGN(stack_top) + random_variable;
328#else
329 return PAGE_ALIGN(stack_top) - random_variable;
330#endif
331}
332
333#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
334unsigned long arch_randomize_brk(struct mm_struct *mm)
335{
336 /* Is the current task 32bit ? */
337 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
338 return randomize_page(mm->brk, SZ_32M);
339
340 return randomize_page(mm->brk, SZ_1G);
341}
342
343unsigned long arch_mmap_rnd(void)
344{
345 unsigned long rnd;
346
347#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
348 if (is_compat_task())
349 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
350 else
351#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
352 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
353
354 return rnd << PAGE_SHIFT;
355}
356
357static int mmap_is_legacy(struct rlimit *rlim_stack)
358{
359 if (current->personality & ADDR_COMPAT_LAYOUT)
360 return 1;
361
362 if (rlim_stack->rlim_cur == RLIM_INFINITY)
363 return 1;
364
365 return sysctl_legacy_va_layout;
366}
367
368/*
369 * Leave enough space between the mmap area and the stack to honour ulimit in
370 * the face of randomisation.
371 */
372#define MIN_GAP (SZ_128M)
373#define MAX_GAP (STACK_TOP / 6 * 5)
374
375static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
376{
377 unsigned long gap = rlim_stack->rlim_cur;
378 unsigned long pad = stack_guard_gap;
379
380 /* Account for stack randomization if necessary */
381 if (current->flags & PF_RANDOMIZE)
382 pad += (STACK_RND_MASK << PAGE_SHIFT);
383
384 /* Values close to RLIM_INFINITY can overflow. */
385 if (gap + pad > gap)
386 gap += pad;
387
388 if (gap < MIN_GAP)
389 gap = MIN_GAP;
390 else if (gap > MAX_GAP)
391 gap = MAX_GAP;
392
393 return PAGE_ALIGN(STACK_TOP - gap - rnd);
394}
395
396void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
397{
398 unsigned long random_factor = 0UL;
399
400 if (current->flags & PF_RANDOMIZE)
401 random_factor = arch_mmap_rnd();
402
403 if (mmap_is_legacy(rlim_stack)) {
404 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
405 mm->get_unmapped_area = arch_get_unmapped_area;
406 } else {
407 mm->mmap_base = mmap_base(random_factor, rlim_stack);
408 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
409 }
410}
411#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
412void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
413{
414 mm->mmap_base = TASK_UNMAPPED_BASE;
415 mm->get_unmapped_area = arch_get_unmapped_area;
416}
417#endif
418
419/**
420 * __account_locked_vm - account locked pages to an mm's locked_vm
421 * @mm: mm to account against
422 * @pages: number of pages to account
423 * @inc: %true if @pages should be considered positive, %false if not
424 * @task: task used to check RLIMIT_MEMLOCK
425 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
426 *
427 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
428 * that mmap_lock is held as writer.
429 *
430 * Return:
431 * * 0 on success
432 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
433 */
434int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
435 struct task_struct *task, bool bypass_rlim)
436{
437 unsigned long locked_vm, limit;
438 int ret = 0;
439
440 mmap_assert_write_locked(mm);
441
442 locked_vm = mm->locked_vm;
443 if (inc) {
444 if (!bypass_rlim) {
445 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
446 if (locked_vm + pages > limit)
447 ret = -ENOMEM;
448 }
449 if (!ret)
450 mm->locked_vm = locked_vm + pages;
451 } else {
452 WARN_ON_ONCE(pages > locked_vm);
453 mm->locked_vm = locked_vm - pages;
454 }
455
456 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
457 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
458 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
459 ret ? " - exceeded" : "");
460
461 return ret;
462}
463EXPORT_SYMBOL_GPL(__account_locked_vm);
464
465/**
466 * account_locked_vm - account locked pages to an mm's locked_vm
467 * @mm: mm to account against, may be NULL
468 * @pages: number of pages to account
469 * @inc: %true if @pages should be considered positive, %false if not
470 *
471 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
472 *
473 * Return:
474 * * 0 on success, or if mm is NULL
475 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
476 */
477int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
478{
479 int ret;
480
481 if (pages == 0 || !mm)
482 return 0;
483
484 mmap_write_lock(mm);
485 ret = __account_locked_vm(mm, pages, inc, current,
486 capable(CAP_IPC_LOCK));
487 mmap_write_unlock(mm);
488
489 return ret;
490}
491EXPORT_SYMBOL_GPL(account_locked_vm);
492
493unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
494 unsigned long len, unsigned long prot,
495 unsigned long flag, unsigned long pgoff)
496{
497 unsigned long ret;
498 struct mm_struct *mm = current->mm;
499 unsigned long populate;
500 LIST_HEAD(uf);
501
502 ret = security_mmap_file(file, prot, flag);
503 if (!ret) {
504 if (mmap_write_lock_killable(mm))
505 return -EINTR;
506 ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
507 &uf);
508 mmap_write_unlock(mm);
509 userfaultfd_unmap_complete(mm, &uf);
510 if (populate)
511 mm_populate(ret, populate);
512 }
513 return ret;
514}
515
516unsigned long vm_mmap(struct file *file, unsigned long addr,
517 unsigned long len, unsigned long prot,
518 unsigned long flag, unsigned long offset)
519{
520 if (unlikely(offset + PAGE_ALIGN(len) < offset))
521 return -EINVAL;
522 if (unlikely(offset_in_page(offset)))
523 return -EINVAL;
524
525 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
526}
527EXPORT_SYMBOL(vm_mmap);
528
529/**
530 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
531 * failure, fall back to non-contiguous (vmalloc) allocation.
532 * @size: size of the request.
533 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
534 * @node: numa node to allocate from
535 *
536 * Uses kmalloc to get the memory but if the allocation fails then falls back
537 * to the vmalloc allocator. Use kvfree for freeing the memory.
538 *
539 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
540 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
541 * preferable to the vmalloc fallback, due to visible performance drawbacks.
542 *
543 * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
544 * fall back to vmalloc.
545 *
546 * Return: pointer to the allocated memory of %NULL in case of failure
547 */
548void *kvmalloc_node(size_t size, gfp_t flags, int node)
549{
550 gfp_t kmalloc_flags = flags;
551 void *ret;
552
553 /*
554 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
555 * so the given set of flags has to be compatible.
556 */
557 if ((flags & GFP_KERNEL) != GFP_KERNEL)
558 return kmalloc_node(size, flags, node);
559
560 /*
561 * We want to attempt a large physically contiguous block first because
562 * it is less likely to fragment multiple larger blocks and therefore
563 * contribute to a long term fragmentation less than vmalloc fallback.
564 * However make sure that larger requests are not too disruptive - no
565 * OOM killer and no allocation failure warnings as we have a fallback.
566 */
567 if (size > PAGE_SIZE) {
568 kmalloc_flags |= __GFP_NOWARN;
569
570 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
571 kmalloc_flags |= __GFP_NORETRY;
572 }
573
574 ret = kmalloc_node(size, kmalloc_flags, node);
575
576 /*
577 * It doesn't really make sense to fallback to vmalloc for sub page
578 * requests
579 */
580 if (ret || size <= PAGE_SIZE)
581 return ret;
582
583 return __vmalloc_node(size, 1, flags, node,
584 __builtin_return_address(0));
585}
586EXPORT_SYMBOL(kvmalloc_node);
587
588/**
589 * kvfree() - Free memory.
590 * @addr: Pointer to allocated memory.
591 *
592 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
593 * It is slightly more efficient to use kfree() or vfree() if you are certain
594 * that you know which one to use.
595 *
596 * Context: Either preemptible task context or not-NMI interrupt.
597 */
598void kvfree(const void *addr)
599{
600 if (is_vmalloc_addr(addr))
601 vfree(addr);
602 else
603 kfree(addr);
604}
605EXPORT_SYMBOL(kvfree);
606
607/**
608 * kvfree_sensitive - Free a data object containing sensitive information.
609 * @addr: address of the data object to be freed.
610 * @len: length of the data object.
611 *
612 * Use the special memzero_explicit() function to clear the content of a
613 * kvmalloc'ed object containing sensitive data to make sure that the
614 * compiler won't optimize out the data clearing.
615 */
616void kvfree_sensitive(const void *addr, size_t len)
617{
618 if (likely(!ZERO_OR_NULL_PTR(addr))) {
619 memzero_explicit((void *)addr, len);
620 kvfree(addr);
621 }
622}
623EXPORT_SYMBOL(kvfree_sensitive);
624
625static inline void *__page_rmapping(struct page *page)
626{
627 unsigned long mapping;
628
629 mapping = (unsigned long)page->mapping;
630 mapping &= ~PAGE_MAPPING_FLAGS;
631
632 return (void *)mapping;
633}
634
635/* Neutral page->mapping pointer to address_space or anon_vma or other */
636void *page_rmapping(struct page *page)
637{
638 page = compound_head(page);
639 return __page_rmapping(page);
640}
641
642/*
643 * Return true if this page is mapped into pagetables.
644 * For compound page it returns true if any subpage of compound page is mapped.
645 */
646bool page_mapped(struct page *page)
647{
648 int i;
649
650 if (likely(!PageCompound(page)))
651 return atomic_read(&page->_mapcount) >= 0;
652 page = compound_head(page);
653 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
654 return true;
655 if (PageHuge(page))
656 return false;
657 for (i = 0; i < compound_nr(page); i++) {
658 if (atomic_read(&page[i]._mapcount) >= 0)
659 return true;
660 }
661 return false;
662}
663EXPORT_SYMBOL(page_mapped);
664
665struct anon_vma *page_anon_vma(struct page *page)
666{
667 unsigned long mapping;
668
669 page = compound_head(page);
670 mapping = (unsigned long)page->mapping;
671 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
672 return NULL;
673 return __page_rmapping(page);
674}
675
676struct address_space *page_mapping(struct page *page)
677{
678 struct address_space *mapping;
679
680 page = compound_head(page);
681
682 /* This happens if someone calls flush_dcache_page on slab page */
683 if (unlikely(PageSlab(page)))
684 return NULL;
685
686 if (unlikely(PageSwapCache(page))) {
687 swp_entry_t entry;
688
689 entry.val = page_private(page);
690 return swap_address_space(entry);
691 }
692
693 mapping = page->mapping;
694 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
695 return NULL;
696
697 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
698}
699EXPORT_SYMBOL(page_mapping);
700
701/*
702 * For file cache pages, return the address_space, otherwise return NULL
703 */
704struct address_space *page_mapping_file(struct page *page)
705{
706 if (unlikely(PageSwapCache(page)))
707 return NULL;
708 return page_mapping(page);
709}
710
711/* Slow path of page_mapcount() for compound pages */
712int __page_mapcount(struct page *page)
713{
714 int ret;
715
716 ret = atomic_read(&page->_mapcount) + 1;
717 /*
718 * For file THP page->_mapcount contains total number of mapping
719 * of the page: no need to look into compound_mapcount.
720 */
721 if (!PageAnon(page) && !PageHuge(page))
722 return ret;
723 page = compound_head(page);
724 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
725 if (PageDoubleMap(page))
726 ret--;
727 return ret;
728}
729EXPORT_SYMBOL_GPL(__page_mapcount);
730
731int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
732int sysctl_overcommit_ratio __read_mostly = 50;
733unsigned long sysctl_overcommit_kbytes __read_mostly;
734int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
735unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
736unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
737
738int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
739 size_t *lenp, loff_t *ppos)
740{
741 int ret;
742
743 ret = proc_dointvec(table, write, buffer, lenp, ppos);
744 if (ret == 0 && write)
745 sysctl_overcommit_kbytes = 0;
746 return ret;
747}
748
749static void sync_overcommit_as(struct work_struct *dummy)
750{
751 percpu_counter_sync(&vm_committed_as);
752}
753
754int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
755 size_t *lenp, loff_t *ppos)
756{
757 struct ctl_table t;
758 int new_policy;
759 int ret;
760
761 /*
762 * The deviation of sync_overcommit_as could be big with loose policy
763 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
764 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
765 * with the strict "NEVER", and to avoid possible race condtion (even
766 * though user usually won't too frequently do the switching to policy
767 * OVERCOMMIT_NEVER), the switch is done in the following order:
768 * 1. changing the batch
769 * 2. sync percpu count on each CPU
770 * 3. switch the policy
771 */
772 if (write) {
773 t = *table;
774 t.data = &new_policy;
775 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
776 if (ret)
777 return ret;
778
779 mm_compute_batch(new_policy);
780 if (new_policy == OVERCOMMIT_NEVER)
781 schedule_on_each_cpu(sync_overcommit_as);
782 sysctl_overcommit_memory = new_policy;
783 } else {
784 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
785 }
786
787 return ret;
788}
789
790int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
791 size_t *lenp, loff_t *ppos)
792{
793 int ret;
794
795 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
796 if (ret == 0 && write)
797 sysctl_overcommit_ratio = 0;
798 return ret;
799}
800
801/*
802 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
803 */
804unsigned long vm_commit_limit(void)
805{
806 unsigned long allowed;
807
808 if (sysctl_overcommit_kbytes)
809 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
810 else
811 allowed = ((totalram_pages() - hugetlb_total_pages())
812 * sysctl_overcommit_ratio / 100);
813 allowed += total_swap_pages;
814
815 return allowed;
816}
817
818/*
819 * Make sure vm_committed_as in one cacheline and not cacheline shared with
820 * other variables. It can be updated by several CPUs frequently.
821 */
822struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
823
824/*
825 * The global memory commitment made in the system can be a metric
826 * that can be used to drive ballooning decisions when Linux is hosted
827 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
828 * balancing memory across competing virtual machines that are hosted.
829 * Several metrics drive this policy engine including the guest reported
830 * memory commitment.
831 *
832 * The time cost of this is very low for small platforms, and for big
833 * platform like a 2S/36C/72T Skylake server, in worst case where
834 * vm_committed_as's spinlock is under severe contention, the time cost
835 * could be about 30~40 microseconds.
836 */
837unsigned long vm_memory_committed(void)
838{
839 return percpu_counter_sum_positive(&vm_committed_as);
840}
841EXPORT_SYMBOL_GPL(vm_memory_committed);
842
843/*
844 * Check that a process has enough memory to allocate a new virtual
845 * mapping. 0 means there is enough memory for the allocation to
846 * succeed and -ENOMEM implies there is not.
847 *
848 * We currently support three overcommit policies, which are set via the
849 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting.rst
850 *
851 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
852 * Additional code 2002 Jul 20 by Robert Love.
853 *
854 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
855 *
856 * Note this is a helper function intended to be used by LSMs which
857 * wish to use this logic.
858 */
859int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
860{
861 long allowed;
862
863 vm_acct_memory(pages);
864
865 /*
866 * Sometimes we want to use more memory than we have
867 */
868 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
869 return 0;
870
871 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
872 if (pages > totalram_pages() + total_swap_pages)
873 goto error;
874 return 0;
875 }
876
877 allowed = vm_commit_limit();
878 /*
879 * Reserve some for root
880 */
881 if (!cap_sys_admin)
882 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
883
884 /*
885 * Don't let a single process grow so big a user can't recover
886 */
887 if (mm) {
888 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
889
890 allowed -= min_t(long, mm->total_vm / 32, reserve);
891 }
892
893 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
894 return 0;
895error:
896 vm_unacct_memory(pages);
897
898 return -ENOMEM;
899}
900
901/**
902 * get_cmdline() - copy the cmdline value to a buffer.
903 * @task: the task whose cmdline value to copy.
904 * @buffer: the buffer to copy to.
905 * @buflen: the length of the buffer. Larger cmdline values are truncated
906 * to this length.
907 *
908 * Return: the size of the cmdline field copied. Note that the copy does
909 * not guarantee an ending NULL byte.
910 */
911int get_cmdline(struct task_struct *task, char *buffer, int buflen)
912{
913 int res = 0;
914 unsigned int len;
915 struct mm_struct *mm = get_task_mm(task);
916 unsigned long arg_start, arg_end, env_start, env_end;
917 if (!mm)
918 goto out;
919 if (!mm->arg_end)
920 goto out_mm; /* Shh! No looking before we're done */
921
922 spin_lock(&mm->arg_lock);
923 arg_start = mm->arg_start;
924 arg_end = mm->arg_end;
925 env_start = mm->env_start;
926 env_end = mm->env_end;
927 spin_unlock(&mm->arg_lock);
928
929 len = arg_end - arg_start;
930
931 if (len > buflen)
932 len = buflen;
933
934 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
935
936 /*
937 * If the nul at the end of args has been overwritten, then
938 * assume application is using setproctitle(3).
939 */
940 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
941 len = strnlen(buffer, res);
942 if (len < res) {
943 res = len;
944 } else {
945 len = env_end - env_start;
946 if (len > buflen - res)
947 len = buflen - res;
948 res += access_process_vm(task, env_start,
949 buffer+res, len,
950 FOLL_FORCE);
951 res = strnlen(buffer, res);
952 }
953 }
954out_mm:
955 mmput(mm);
956out:
957 return res;
958}
959
960int memcmp_pages(struct page *page1, struct page *page2)
961{
962 char *addr1, *addr2;
963 int ret;
964
965 addr1 = kmap_atomic(page1);
966 addr2 = kmap_atomic(page2);
967 ret = memcmp(addr1, addr2, PAGE_SIZE);
968 kunmap_atomic(addr2);
969 kunmap_atomic(addr1);
970 return ret;
971}
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/sched/mm.h>
9#include <linux/sched/task_stack.h>
10#include <linux/security.h>
11#include <linux/swap.h>
12#include <linux/swapops.h>
13#include <linux/mman.h>
14#include <linux/hugetlb.h>
15#include <linux/vmalloc.h>
16#include <linux/userfaultfd_k.h>
17
18#include <asm/sections.h>
19#include <linux/uaccess.h>
20
21#include "internal.h"
22
23static inline int is_kernel_rodata(unsigned long addr)
24{
25 return addr >= (unsigned long)__start_rodata &&
26 addr < (unsigned long)__end_rodata;
27}
28
29/**
30 * kfree_const - conditionally free memory
31 * @x: pointer to the memory
32 *
33 * Function calls kfree only if @x is not in .rodata section.
34 */
35void kfree_const(const void *x)
36{
37 if (!is_kernel_rodata((unsigned long)x))
38 kfree(x);
39}
40EXPORT_SYMBOL(kfree_const);
41
42/**
43 * kstrdup - allocate space for and copy an existing string
44 * @s: the string to duplicate
45 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
46 */
47char *kstrdup(const char *s, gfp_t gfp)
48{
49 size_t len;
50 char *buf;
51
52 if (!s)
53 return NULL;
54
55 len = strlen(s) + 1;
56 buf = kmalloc_track_caller(len, gfp);
57 if (buf)
58 memcpy(buf, s, len);
59 return buf;
60}
61EXPORT_SYMBOL(kstrdup);
62
63/**
64 * kstrdup_const - conditionally duplicate an existing const string
65 * @s: the string to duplicate
66 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
67 *
68 * Function returns source string if it is in .rodata section otherwise it
69 * fallbacks to kstrdup.
70 * Strings allocated by kstrdup_const should be freed by kfree_const.
71 */
72const char *kstrdup_const(const char *s, gfp_t gfp)
73{
74 if (is_kernel_rodata((unsigned long)s))
75 return s;
76
77 return kstrdup(s, gfp);
78}
79EXPORT_SYMBOL(kstrdup_const);
80
81/**
82 * kstrndup - allocate space for and copy an existing string
83 * @s: the string to duplicate
84 * @max: read at most @max chars from @s
85 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
86 *
87 * Note: Use kmemdup_nul() instead if the size is known exactly.
88 */
89char *kstrndup(const char *s, size_t max, gfp_t gfp)
90{
91 size_t len;
92 char *buf;
93
94 if (!s)
95 return NULL;
96
97 len = strnlen(s, max);
98 buf = kmalloc_track_caller(len+1, gfp);
99 if (buf) {
100 memcpy(buf, s, len);
101 buf[len] = '\0';
102 }
103 return buf;
104}
105EXPORT_SYMBOL(kstrndup);
106
107/**
108 * kmemdup - duplicate region of memory
109 *
110 * @src: memory region to duplicate
111 * @len: memory region length
112 * @gfp: GFP mask to use
113 */
114void *kmemdup(const void *src, size_t len, gfp_t gfp)
115{
116 void *p;
117
118 p = kmalloc_track_caller(len, gfp);
119 if (p)
120 memcpy(p, src, len);
121 return p;
122}
123EXPORT_SYMBOL(kmemdup);
124
125/**
126 * kmemdup_nul - Create a NUL-terminated string from unterminated data
127 * @s: The data to stringify
128 * @len: The size of the data
129 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
130 */
131char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
132{
133 char *buf;
134
135 if (!s)
136 return NULL;
137
138 buf = kmalloc_track_caller(len + 1, gfp);
139 if (buf) {
140 memcpy(buf, s, len);
141 buf[len] = '\0';
142 }
143 return buf;
144}
145EXPORT_SYMBOL(kmemdup_nul);
146
147/**
148 * memdup_user - duplicate memory region from user space
149 *
150 * @src: source address in user space
151 * @len: number of bytes to copy
152 *
153 * Returns an ERR_PTR() on failure. Result is physically
154 * contiguous, to be freed by kfree().
155 */
156void *memdup_user(const void __user *src, size_t len)
157{
158 void *p;
159
160 p = kmalloc_track_caller(len, GFP_USER);
161 if (!p)
162 return ERR_PTR(-ENOMEM);
163
164 if (copy_from_user(p, src, len)) {
165 kfree(p);
166 return ERR_PTR(-EFAULT);
167 }
168
169 return p;
170}
171EXPORT_SYMBOL(memdup_user);
172
173/**
174 * vmemdup_user - duplicate memory region from user space
175 *
176 * @src: source address in user space
177 * @len: number of bytes to copy
178 *
179 * Returns an ERR_PTR() on failure. Result may be not
180 * physically contiguous. Use kvfree() to free.
181 */
182void *vmemdup_user(const void __user *src, size_t len)
183{
184 void *p;
185
186 p = kvmalloc(len, GFP_USER);
187 if (!p)
188 return ERR_PTR(-ENOMEM);
189
190 if (copy_from_user(p, src, len)) {
191 kvfree(p);
192 return ERR_PTR(-EFAULT);
193 }
194
195 return p;
196}
197EXPORT_SYMBOL(vmemdup_user);
198
199/*
200 * strndup_user - duplicate an existing string from user space
201 * @s: The string to duplicate
202 * @n: Maximum number of bytes to copy, including the trailing NUL.
203 */
204char *strndup_user(const char __user *s, long n)
205{
206 char *p;
207 long length;
208
209 length = strnlen_user(s, n);
210
211 if (!length)
212 return ERR_PTR(-EFAULT);
213
214 if (length > n)
215 return ERR_PTR(-EINVAL);
216
217 p = memdup_user(s, length);
218
219 if (IS_ERR(p))
220 return p;
221
222 p[length - 1] = '\0';
223
224 return p;
225}
226EXPORT_SYMBOL(strndup_user);
227
228/**
229 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
230 *
231 * @src: source address in user space
232 * @len: number of bytes to copy
233 *
234 * Returns an ERR_PTR() on failure.
235 */
236void *memdup_user_nul(const void __user *src, size_t len)
237{
238 char *p;
239
240 /*
241 * Always use GFP_KERNEL, since copy_from_user() can sleep and
242 * cause pagefault, which makes it pointless to use GFP_NOFS
243 * or GFP_ATOMIC.
244 */
245 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
246 if (!p)
247 return ERR_PTR(-ENOMEM);
248
249 if (copy_from_user(p, src, len)) {
250 kfree(p);
251 return ERR_PTR(-EFAULT);
252 }
253 p[len] = '\0';
254
255 return p;
256}
257EXPORT_SYMBOL(memdup_user_nul);
258
259void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
260 struct vm_area_struct *prev, struct rb_node *rb_parent)
261{
262 struct vm_area_struct *next;
263
264 vma->vm_prev = prev;
265 if (prev) {
266 next = prev->vm_next;
267 prev->vm_next = vma;
268 } else {
269 mm->mmap = vma;
270 if (rb_parent)
271 next = rb_entry(rb_parent,
272 struct vm_area_struct, vm_rb);
273 else
274 next = NULL;
275 }
276 vma->vm_next = next;
277 if (next)
278 next->vm_prev = vma;
279}
280
281/* Check if the vma is being used as a stack by this task */
282int vma_is_stack_for_current(struct vm_area_struct *vma)
283{
284 struct task_struct * __maybe_unused t = current;
285
286 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
287}
288
289#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
290void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
291{
292 mm->mmap_base = TASK_UNMAPPED_BASE;
293 mm->get_unmapped_area = arch_get_unmapped_area;
294}
295#endif
296
297/*
298 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
299 * back to the regular GUP.
300 * Note a difference with get_user_pages_fast: this always returns the
301 * number of pages pinned, 0 if no pages were pinned.
302 * If the architecture does not support this function, simply return with no
303 * pages pinned.
304 */
305int __weak __get_user_pages_fast(unsigned long start,
306 int nr_pages, int write, struct page **pages)
307{
308 return 0;
309}
310EXPORT_SYMBOL_GPL(__get_user_pages_fast);
311
312/**
313 * get_user_pages_fast() - pin user pages in memory
314 * @start: starting user address
315 * @nr_pages: number of pages from start to pin
316 * @write: whether pages will be written to
317 * @pages: array that receives pointers to the pages pinned.
318 * Should be at least nr_pages long.
319 *
320 * Returns number of pages pinned. This may be fewer than the number
321 * requested. If nr_pages is 0 or negative, returns 0. If no pages
322 * were pinned, returns -errno.
323 *
324 * get_user_pages_fast provides equivalent functionality to get_user_pages,
325 * operating on current and current->mm, with force=0 and vma=NULL. However
326 * unlike get_user_pages, it must be called without mmap_sem held.
327 *
328 * get_user_pages_fast may take mmap_sem and page table locks, so no
329 * assumptions can be made about lack of locking. get_user_pages_fast is to be
330 * implemented in a way that is advantageous (vs get_user_pages()) when the
331 * user memory area is already faulted in and present in ptes. However if the
332 * pages have to be faulted in, it may turn out to be slightly slower so
333 * callers need to carefully consider what to use. On many architectures,
334 * get_user_pages_fast simply falls back to get_user_pages.
335 */
336int __weak get_user_pages_fast(unsigned long start,
337 int nr_pages, int write, struct page **pages)
338{
339 return get_user_pages_unlocked(start, nr_pages, pages,
340 write ? FOLL_WRITE : 0);
341}
342EXPORT_SYMBOL_GPL(get_user_pages_fast);
343
344unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
345 unsigned long len, unsigned long prot,
346 unsigned long flag, unsigned long pgoff)
347{
348 unsigned long ret;
349 struct mm_struct *mm = current->mm;
350 unsigned long populate;
351 LIST_HEAD(uf);
352
353 ret = security_mmap_file(file, prot, flag);
354 if (!ret) {
355 if (down_write_killable(&mm->mmap_sem))
356 return -EINTR;
357 ret = do_mmap_pgoff(file, addr, len, prot, flag, pgoff,
358 &populate, &uf);
359 up_write(&mm->mmap_sem);
360 userfaultfd_unmap_complete(mm, &uf);
361 if (populate)
362 mm_populate(ret, populate);
363 }
364 return ret;
365}
366
367unsigned long vm_mmap(struct file *file, unsigned long addr,
368 unsigned long len, unsigned long prot,
369 unsigned long flag, unsigned long offset)
370{
371 if (unlikely(offset + PAGE_ALIGN(len) < offset))
372 return -EINVAL;
373 if (unlikely(offset_in_page(offset)))
374 return -EINVAL;
375
376 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
377}
378EXPORT_SYMBOL(vm_mmap);
379
380/**
381 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
382 * failure, fall back to non-contiguous (vmalloc) allocation.
383 * @size: size of the request.
384 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
385 * @node: numa node to allocate from
386 *
387 * Uses kmalloc to get the memory but if the allocation fails then falls back
388 * to the vmalloc allocator. Use kvfree for freeing the memory.
389 *
390 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
391 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
392 * preferable to the vmalloc fallback, due to visible performance drawbacks.
393 *
394 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
395 */
396void *kvmalloc_node(size_t size, gfp_t flags, int node)
397{
398 gfp_t kmalloc_flags = flags;
399 void *ret;
400
401 /*
402 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
403 * so the given set of flags has to be compatible.
404 */
405 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
406
407 /*
408 * We want to attempt a large physically contiguous block first because
409 * it is less likely to fragment multiple larger blocks and therefore
410 * contribute to a long term fragmentation less than vmalloc fallback.
411 * However make sure that larger requests are not too disruptive - no
412 * OOM killer and no allocation failure warnings as we have a fallback.
413 */
414 if (size > PAGE_SIZE) {
415 kmalloc_flags |= __GFP_NOWARN;
416
417 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
418 kmalloc_flags |= __GFP_NORETRY;
419 }
420
421 ret = kmalloc_node(size, kmalloc_flags, node);
422
423 /*
424 * It doesn't really make sense to fallback to vmalloc for sub page
425 * requests
426 */
427 if (ret || size <= PAGE_SIZE)
428 return ret;
429
430 return __vmalloc_node_flags_caller(size, node, flags,
431 __builtin_return_address(0));
432}
433EXPORT_SYMBOL(kvmalloc_node);
434
435void kvfree(const void *addr)
436{
437 if (is_vmalloc_addr(addr))
438 vfree(addr);
439 else
440 kfree(addr);
441}
442EXPORT_SYMBOL(kvfree);
443
444static inline void *__page_rmapping(struct page *page)
445{
446 unsigned long mapping;
447
448 mapping = (unsigned long)page->mapping;
449 mapping &= ~PAGE_MAPPING_FLAGS;
450
451 return (void *)mapping;
452}
453
454/* Neutral page->mapping pointer to address_space or anon_vma or other */
455void *page_rmapping(struct page *page)
456{
457 page = compound_head(page);
458 return __page_rmapping(page);
459}
460
461/*
462 * Return true if this page is mapped into pagetables.
463 * For compound page it returns true if any subpage of compound page is mapped.
464 */
465bool page_mapped(struct page *page)
466{
467 int i;
468
469 if (likely(!PageCompound(page)))
470 return atomic_read(&page->_mapcount) >= 0;
471 page = compound_head(page);
472 if (atomic_read(compound_mapcount_ptr(page)) >= 0)
473 return true;
474 if (PageHuge(page))
475 return false;
476 for (i = 0; i < hpage_nr_pages(page); i++) {
477 if (atomic_read(&page[i]._mapcount) >= 0)
478 return true;
479 }
480 return false;
481}
482EXPORT_SYMBOL(page_mapped);
483
484struct anon_vma *page_anon_vma(struct page *page)
485{
486 unsigned long mapping;
487
488 page = compound_head(page);
489 mapping = (unsigned long)page->mapping;
490 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
491 return NULL;
492 return __page_rmapping(page);
493}
494
495struct address_space *page_mapping(struct page *page)
496{
497 struct address_space *mapping;
498
499 page = compound_head(page);
500
501 /* This happens if someone calls flush_dcache_page on slab page */
502 if (unlikely(PageSlab(page)))
503 return NULL;
504
505 if (unlikely(PageSwapCache(page))) {
506 swp_entry_t entry;
507
508 entry.val = page_private(page);
509 return swap_address_space(entry);
510 }
511
512 mapping = page->mapping;
513 if ((unsigned long)mapping & PAGE_MAPPING_ANON)
514 return NULL;
515
516 return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
517}
518EXPORT_SYMBOL(page_mapping);
519
520/*
521 * For file cache pages, return the address_space, otherwise return NULL
522 */
523struct address_space *page_mapping_file(struct page *page)
524{
525 if (unlikely(PageSwapCache(page)))
526 return NULL;
527 return page_mapping(page);
528}
529
530/* Slow path of page_mapcount() for compound pages */
531int __page_mapcount(struct page *page)
532{
533 int ret;
534
535 ret = atomic_read(&page->_mapcount) + 1;
536 /*
537 * For file THP page->_mapcount contains total number of mapping
538 * of the page: no need to look into compound_mapcount.
539 */
540 if (!PageAnon(page) && !PageHuge(page))
541 return ret;
542 page = compound_head(page);
543 ret += atomic_read(compound_mapcount_ptr(page)) + 1;
544 if (PageDoubleMap(page))
545 ret--;
546 return ret;
547}
548EXPORT_SYMBOL_GPL(__page_mapcount);
549
550int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
551int sysctl_overcommit_ratio __read_mostly = 50;
552unsigned long sysctl_overcommit_kbytes __read_mostly;
553int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
554unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
555unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
556
557int overcommit_ratio_handler(struct ctl_table *table, int write,
558 void __user *buffer, size_t *lenp,
559 loff_t *ppos)
560{
561 int ret;
562
563 ret = proc_dointvec(table, write, buffer, lenp, ppos);
564 if (ret == 0 && write)
565 sysctl_overcommit_kbytes = 0;
566 return ret;
567}
568
569int overcommit_kbytes_handler(struct ctl_table *table, int write,
570 void __user *buffer, size_t *lenp,
571 loff_t *ppos)
572{
573 int ret;
574
575 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
576 if (ret == 0 && write)
577 sysctl_overcommit_ratio = 0;
578 return ret;
579}
580
581/*
582 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
583 */
584unsigned long vm_commit_limit(void)
585{
586 unsigned long allowed;
587
588 if (sysctl_overcommit_kbytes)
589 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
590 else
591 allowed = ((totalram_pages - hugetlb_total_pages())
592 * sysctl_overcommit_ratio / 100);
593 allowed += total_swap_pages;
594
595 return allowed;
596}
597
598/*
599 * Make sure vm_committed_as in one cacheline and not cacheline shared with
600 * other variables. It can be updated by several CPUs frequently.
601 */
602struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
603
604/*
605 * The global memory commitment made in the system can be a metric
606 * that can be used to drive ballooning decisions when Linux is hosted
607 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
608 * balancing memory across competing virtual machines that are hosted.
609 * Several metrics drive this policy engine including the guest reported
610 * memory commitment.
611 */
612unsigned long vm_memory_committed(void)
613{
614 return percpu_counter_read_positive(&vm_committed_as);
615}
616EXPORT_SYMBOL_GPL(vm_memory_committed);
617
618/*
619 * Check that a process has enough memory to allocate a new virtual
620 * mapping. 0 means there is enough memory for the allocation to
621 * succeed and -ENOMEM implies there is not.
622 *
623 * We currently support three overcommit policies, which are set via the
624 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
625 *
626 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
627 * Additional code 2002 Jul 20 by Robert Love.
628 *
629 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
630 *
631 * Note this is a helper function intended to be used by LSMs which
632 * wish to use this logic.
633 */
634int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
635{
636 long free, allowed, reserve;
637
638 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
639 -(s64)vm_committed_as_batch * num_online_cpus(),
640 "memory commitment underflow");
641
642 vm_acct_memory(pages);
643
644 /*
645 * Sometimes we want to use more memory than we have
646 */
647 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
648 return 0;
649
650 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
651 free = global_zone_page_state(NR_FREE_PAGES);
652 free += global_node_page_state(NR_FILE_PAGES);
653
654 /*
655 * shmem pages shouldn't be counted as free in this
656 * case, they can't be purged, only swapped out, and
657 * that won't affect the overall amount of available
658 * memory in the system.
659 */
660 free -= global_node_page_state(NR_SHMEM);
661
662 free += get_nr_swap_pages();
663
664 /*
665 * Any slabs which are created with the
666 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
667 * which are reclaimable, under pressure. The dentry
668 * cache and most inode caches should fall into this
669 */
670 free += global_node_page_state(NR_SLAB_RECLAIMABLE);
671
672 /*
673 * Part of the kernel memory, which can be released
674 * under memory pressure.
675 */
676 free += global_node_page_state(
677 NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
678
679 /*
680 * Leave reserved pages. The pages are not for anonymous pages.
681 */
682 if (free <= totalreserve_pages)
683 goto error;
684 else
685 free -= totalreserve_pages;
686
687 /*
688 * Reserve some for root
689 */
690 if (!cap_sys_admin)
691 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
692
693 if (free > pages)
694 return 0;
695
696 goto error;
697 }
698
699 allowed = vm_commit_limit();
700 /*
701 * Reserve some for root
702 */
703 if (!cap_sys_admin)
704 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
705
706 /*
707 * Don't let a single process grow so big a user can't recover
708 */
709 if (mm) {
710 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
711 allowed -= min_t(long, mm->total_vm / 32, reserve);
712 }
713
714 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
715 return 0;
716error:
717 vm_unacct_memory(pages);
718
719 return -ENOMEM;
720}
721
722/**
723 * get_cmdline() - copy the cmdline value to a buffer.
724 * @task: the task whose cmdline value to copy.
725 * @buffer: the buffer to copy to.
726 * @buflen: the length of the buffer. Larger cmdline values are truncated
727 * to this length.
728 * Returns the size of the cmdline field copied. Note that the copy does
729 * not guarantee an ending NULL byte.
730 */
731int get_cmdline(struct task_struct *task, char *buffer, int buflen)
732{
733 int res = 0;
734 unsigned int len;
735 struct mm_struct *mm = get_task_mm(task);
736 unsigned long arg_start, arg_end, env_start, env_end;
737 if (!mm)
738 goto out;
739 if (!mm->arg_end)
740 goto out_mm; /* Shh! No looking before we're done */
741
742 down_read(&mm->mmap_sem);
743 arg_start = mm->arg_start;
744 arg_end = mm->arg_end;
745 env_start = mm->env_start;
746 env_end = mm->env_end;
747 up_read(&mm->mmap_sem);
748
749 len = arg_end - arg_start;
750
751 if (len > buflen)
752 len = buflen;
753
754 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
755
756 /*
757 * If the nul at the end of args has been overwritten, then
758 * assume application is using setproctitle(3).
759 */
760 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
761 len = strnlen(buffer, res);
762 if (len < res) {
763 res = len;
764 } else {
765 len = env_end - env_start;
766 if (len > buflen - res)
767 len = buflen - res;
768 res += access_process_vm(task, env_start,
769 buffer+res, len,
770 FOLL_FORCE);
771 res = strnlen(buffer, res);
772 }
773 }
774out_mm:
775 mmput(mm);
776out:
777 return res;
778}