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