xref: /linux/mm/util.c (revision eb88e6bfbc0a975e08a18c39d1138d3e6cdc00a5)
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 #include <linux/fsnotify.h>
27 
28 #include <linux/uaccess.h>
29 
30 #include <kunit/visibility.h>
31 
32 #include "internal.h"
33 #include "swap.h"
34 
35 /**
36  * kfree_const - conditionally free memory
37  * @x: pointer to the memory
38  *
39  * Function calls kfree only if @x is not in .rodata section.
40  */
kfree_const(const void * x)41 void kfree_const(const void *x)
42 {
43 	if (!is_kernel_rodata((unsigned long)x))
44 		kfree(x);
45 }
46 EXPORT_SYMBOL(kfree_const);
47 
48 /**
49  * __kmemdup_nul - Create a NUL-terminated string from @s, which might be unterminated.
50  * @s: The data to copy
51  * @len: The size of the data, not including the NUL terminator
52  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
53  *
54  * Return: newly allocated copy of @s with NUL-termination or %NULL in
55  * case of error
56  */
__kmemdup_nul(const char * s,size_t len,gfp_t gfp)57 static __always_inline char *__kmemdup_nul(const char *s, size_t len, gfp_t gfp)
58 {
59 	char *buf;
60 
61 	/* '+1' for the NUL terminator */
62 	buf = kmalloc_track_caller(len + 1, gfp);
63 	if (!buf)
64 		return NULL;
65 
66 	memcpy(buf, s, len);
67 	/* Ensure the buf is always NUL-terminated, regardless of @s. */
68 	buf[len] = '\0';
69 	return buf;
70 }
71 
72 /**
73  * kstrdup - allocate space for and copy an existing string
74  * @s: the string to duplicate
75  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
76  *
77  * Return: newly allocated copy of @s or %NULL in case of error
78  */
79 noinline
kstrdup(const char * s,gfp_t gfp)80 char *kstrdup(const char *s, gfp_t gfp)
81 {
82 	return s ? __kmemdup_nul(s, strlen(s), gfp) : NULL;
83 }
84 EXPORT_SYMBOL(kstrdup);
85 
86 /**
87  * kstrdup_const - conditionally duplicate an existing const string
88  * @s: the string to duplicate
89  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
90  *
91  * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
92  * must not be passed to krealloc().
93  *
94  * Return: source string if it is in .rodata section otherwise
95  * fallback to kstrdup.
96  */
kstrdup_const(const char * s,gfp_t gfp)97 const char *kstrdup_const(const char *s, gfp_t gfp)
98 {
99 	if (is_kernel_rodata((unsigned long)s))
100 		return s;
101 
102 	return kstrdup(s, gfp);
103 }
104 EXPORT_SYMBOL(kstrdup_const);
105 
106 /**
107  * kstrndup - allocate space for and copy an existing string
108  * @s: the string to duplicate
109  * @max: read at most @max chars from @s
110  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
111  *
112  * Note: Use kmemdup_nul() instead if the size is known exactly.
113  *
114  * Return: newly allocated copy of @s or %NULL in case of error
115  */
kstrndup(const char * s,size_t max,gfp_t gfp)116 char *kstrndup(const char *s, size_t max, gfp_t gfp)
117 {
118 	return s ? __kmemdup_nul(s, strnlen(s, max), gfp) : NULL;
119 }
120 EXPORT_SYMBOL(kstrndup);
121 
122 /**
123  * kmemdup - duplicate region of memory
124  *
125  * @src: memory region to duplicate
126  * @len: memory region length
127  * @gfp: GFP mask to use
128  *
129  * Return: newly allocated copy of @src or %NULL in case of error,
130  * result is physically contiguous. Use kfree() to free.
131  */
kmemdup_noprof(const void * src,size_t len,gfp_t gfp)132 void *kmemdup_noprof(const void *src, size_t len, gfp_t gfp)
133 {
134 	void *p;
135 
136 	p = kmalloc_node_track_caller_noprof(len, gfp, NUMA_NO_NODE, _RET_IP_);
137 	if (p)
138 		memcpy(p, src, len);
139 	return p;
140 }
141 EXPORT_SYMBOL(kmemdup_noprof);
142 
143 /**
144  * kmemdup_array - duplicate a given array.
145  *
146  * @src: array to duplicate.
147  * @count: number of elements to duplicate from array.
148  * @element_size: size of each element of array.
149  * @gfp: GFP mask to use.
150  *
151  * Return: duplicated array of @src or %NULL in case of error,
152  * result is physically contiguous. Use kfree() to free.
153  */
kmemdup_array(const void * src,size_t count,size_t element_size,gfp_t gfp)154 void *kmemdup_array(const void *src, size_t count, size_t element_size, gfp_t gfp)
155 {
156 	return kmemdup(src, size_mul(element_size, count), gfp);
157 }
158 EXPORT_SYMBOL(kmemdup_array);
159 
160 /**
161  * kvmemdup - duplicate region of memory
162  *
163  * @src: memory region to duplicate
164  * @len: memory region length
165  * @gfp: GFP mask to use
166  *
167  * Return: newly allocated copy of @src or %NULL in case of error,
168  * result may be not physically contiguous. Use kvfree() to free.
169  */
kvmemdup(const void * src,size_t len,gfp_t gfp)170 void *kvmemdup(const void *src, size_t len, gfp_t gfp)
171 {
172 	void *p;
173 
174 	p = kvmalloc(len, gfp);
175 	if (p)
176 		memcpy(p, src, len);
177 	return p;
178 }
179 EXPORT_SYMBOL(kvmemdup);
180 
181 /**
182  * kmemdup_nul - Create a NUL-terminated string from unterminated data
183  * @s: The data to stringify
184  * @len: The size of the data
185  * @gfp: the GFP mask used in the kmalloc() call when allocating memory
186  *
187  * Return: newly allocated copy of @s with NUL-termination or %NULL in
188  * case of error
189  */
kmemdup_nul(const char * s,size_t len,gfp_t gfp)190 char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
191 {
192 	return s ? __kmemdup_nul(s, len, gfp) : NULL;
193 }
194 EXPORT_SYMBOL(kmemdup_nul);
195 
196 static kmem_buckets *user_buckets __ro_after_init;
197 
init_user_buckets(void)198 static int __init init_user_buckets(void)
199 {
200 	user_buckets = kmem_buckets_create("memdup_user", 0, 0, INT_MAX, NULL);
201 
202 	return 0;
203 }
204 subsys_initcall(init_user_buckets);
205 
206 /**
207  * memdup_user - duplicate memory region from user space
208  *
209  * @src: source address in user space
210  * @len: number of bytes to copy
211  *
212  * Return: an ERR_PTR() on failure.  Result is physically
213  * contiguous, to be freed by kfree().
214  */
memdup_user(const void __user * src,size_t len)215 void *memdup_user(const void __user *src, size_t len)
216 {
217 	void *p;
218 
219 	p = kmem_buckets_alloc_track_caller(user_buckets, len, GFP_USER | __GFP_NOWARN);
220 	if (!p)
221 		return ERR_PTR(-ENOMEM);
222 
223 	if (copy_from_user(p, src, len)) {
224 		kfree(p);
225 		return ERR_PTR(-EFAULT);
226 	}
227 
228 	return p;
229 }
230 EXPORT_SYMBOL(memdup_user);
231 
232 /**
233  * vmemdup_user - duplicate memory region from user space
234  *
235  * @src: source address in user space
236  * @len: number of bytes to copy
237  *
238  * Return: an ERR_PTR() on failure.  Result may be not
239  * physically contiguous.  Use kvfree() to free.
240  */
vmemdup_user(const void __user * src,size_t len)241 void *vmemdup_user(const void __user *src, size_t len)
242 {
243 	void *p;
244 
245 	p = kmem_buckets_valloc(user_buckets, len, GFP_USER);
246 	if (!p)
247 		return ERR_PTR(-ENOMEM);
248 
249 	if (copy_from_user(p, src, len)) {
250 		kvfree(p);
251 		return ERR_PTR(-EFAULT);
252 	}
253 
254 	return p;
255 }
256 EXPORT_SYMBOL(vmemdup_user);
257 
258 /**
259  * strndup_user - duplicate an existing string from user space
260  * @s: The string to duplicate
261  * @n: Maximum number of bytes to copy, including the trailing NUL.
262  *
263  * Return: newly allocated copy of @s or an ERR_PTR() in case of error
264  */
strndup_user(const char __user * s,long n)265 char *strndup_user(const char __user *s, long n)
266 {
267 	char *p;
268 	long length;
269 
270 	length = strnlen_user(s, n);
271 
272 	if (!length)
273 		return ERR_PTR(-EFAULT);
274 
275 	if (length > n)
276 		return ERR_PTR(-EINVAL);
277 
278 	p = memdup_user(s, length);
279 
280 	if (IS_ERR(p))
281 		return p;
282 
283 	p[length - 1] = '\0';
284 
285 	return p;
286 }
287 EXPORT_SYMBOL(strndup_user);
288 
289 /**
290  * memdup_user_nul - duplicate memory region from user space and NUL-terminate
291  *
292  * @src: source address in user space
293  * @len: number of bytes to copy
294  *
295  * Return: an ERR_PTR() on failure.
296  */
memdup_user_nul(const void __user * src,size_t len)297 void *memdup_user_nul(const void __user *src, size_t len)
298 {
299 	char *p;
300 
301 	p = kmem_buckets_alloc_track_caller(user_buckets, len + 1, GFP_USER | __GFP_NOWARN);
302 	if (!p)
303 		return ERR_PTR(-ENOMEM);
304 
305 	if (copy_from_user(p, src, len)) {
306 		kfree(p);
307 		return ERR_PTR(-EFAULT);
308 	}
309 	p[len] = '\0';
310 
311 	return p;
312 }
313 EXPORT_SYMBOL(memdup_user_nul);
314 
315 /* Check if the vma is being used as a stack by this task */
vma_is_stack_for_current(struct vm_area_struct * vma)316 int vma_is_stack_for_current(struct vm_area_struct *vma)
317 {
318 	struct task_struct * __maybe_unused t = current;
319 
320 	return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
321 }
322 
323 /*
324  * Change backing file, only valid to use during initial VMA setup.
325  */
vma_set_file(struct vm_area_struct * vma,struct file * file)326 void vma_set_file(struct vm_area_struct *vma, struct file *file)
327 {
328 	/* Changing an anonymous vma with this is illegal */
329 	get_file(file);
330 	swap(vma->vm_file, file);
331 	fput(file);
332 }
333 EXPORT_SYMBOL(vma_set_file);
334 
335 #ifndef STACK_RND_MASK
336 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
337 #endif
338 
randomize_stack_top(unsigned long stack_top)339 unsigned long randomize_stack_top(unsigned long stack_top)
340 {
341 	unsigned long random_variable = 0;
342 
343 	if (current->flags & PF_RANDOMIZE) {
344 		random_variable = get_random_long();
345 		random_variable &= STACK_RND_MASK;
346 		random_variable <<= PAGE_SHIFT;
347 	}
348 #ifdef CONFIG_STACK_GROWSUP
349 	return PAGE_ALIGN(stack_top) + random_variable;
350 #else
351 	return PAGE_ALIGN(stack_top) - random_variable;
352 #endif
353 }
354 
355 /**
356  * randomize_page - Generate a random, page aligned address
357  * @start:	The smallest acceptable address the caller will take.
358  * @range:	The size of the area, starting at @start, within which the
359  *		random address must fall.
360  *
361  * If @start + @range would overflow, @range is capped.
362  *
363  * NOTE: Historical use of randomize_range, which this replaces, presumed that
364  * @start was already page aligned.  We now align it regardless.
365  *
366  * Return: A page aligned address within [start, start + range).  On error,
367  * @start is returned.
368  */
randomize_page(unsigned long start,unsigned long range)369 unsigned long randomize_page(unsigned long start, unsigned long range)
370 {
371 	if (!PAGE_ALIGNED(start)) {
372 		range -= PAGE_ALIGN(start) - start;
373 		start = PAGE_ALIGN(start);
374 	}
375 
376 	if (start > ULONG_MAX - range)
377 		range = ULONG_MAX - start;
378 
379 	range >>= PAGE_SHIFT;
380 
381 	if (range == 0)
382 		return start;
383 
384 	return start + (get_random_long() % range << PAGE_SHIFT);
385 }
386 
387 #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
arch_randomize_brk(struct mm_struct * mm)388 unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
389 {
390 	/* Is the current task 32bit ? */
391 	if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
392 		return randomize_page(mm->brk, SZ_32M);
393 
394 	return randomize_page(mm->brk, SZ_1G);
395 }
396 
arch_mmap_rnd(void)397 unsigned long arch_mmap_rnd(void)
398 {
399 	unsigned long rnd;
400 
401 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
402 	if (is_compat_task())
403 		rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
404 	else
405 #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
406 		rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
407 
408 	return rnd << PAGE_SHIFT;
409 }
410 
mmap_is_legacy(struct rlimit * rlim_stack)411 static int mmap_is_legacy(struct rlimit *rlim_stack)
412 {
413 	if (current->personality & ADDR_COMPAT_LAYOUT)
414 		return 1;
415 
416 	/* On parisc the stack always grows up - so a unlimited stack should
417 	 * not be an indicator to use the legacy memory layout. */
418 	if (rlim_stack->rlim_cur == RLIM_INFINITY &&
419 		!IS_ENABLED(CONFIG_STACK_GROWSUP))
420 		return 1;
421 
422 	return sysctl_legacy_va_layout;
423 }
424 
425 /*
426  * Leave enough space between the mmap area and the stack to honour ulimit in
427  * the face of randomisation.
428  */
429 #define MIN_GAP		(SZ_128M)
430 #define MAX_GAP		(STACK_TOP / 6 * 5)
431 
mmap_base(unsigned long rnd,struct rlimit * rlim_stack)432 static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
433 {
434 #ifdef CONFIG_STACK_GROWSUP
435 	/*
436 	 * For an upwards growing stack the calculation is much simpler.
437 	 * Memory for the maximum stack size is reserved at the top of the
438 	 * task. mmap_base starts directly below the stack and grows
439 	 * downwards.
440 	 */
441 	return PAGE_ALIGN_DOWN(mmap_upper_limit(rlim_stack) - rnd);
442 #else
443 	unsigned long gap = rlim_stack->rlim_cur;
444 	unsigned long pad = stack_guard_gap;
445 
446 	/* Account for stack randomization if necessary */
447 	if (current->flags & PF_RANDOMIZE)
448 		pad += (STACK_RND_MASK << PAGE_SHIFT);
449 
450 	/* Values close to RLIM_INFINITY can overflow. */
451 	if (gap + pad > gap)
452 		gap += pad;
453 
454 	if (gap < MIN_GAP && MIN_GAP < MAX_GAP)
455 		gap = MIN_GAP;
456 	else if (gap > MAX_GAP)
457 		gap = MAX_GAP;
458 
459 	return PAGE_ALIGN(STACK_TOP - gap - rnd);
460 #endif
461 }
462 
arch_pick_mmap_layout(struct mm_struct * mm,struct rlimit * rlim_stack)463 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
464 {
465 	unsigned long random_factor = 0UL;
466 
467 	if (current->flags & PF_RANDOMIZE)
468 		random_factor = arch_mmap_rnd();
469 
470 	if (mmap_is_legacy(rlim_stack)) {
471 		mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
472 		clear_bit(MMF_TOPDOWN, &mm->flags);
473 	} else {
474 		mm->mmap_base = mmap_base(random_factor, rlim_stack);
475 		set_bit(MMF_TOPDOWN, &mm->flags);
476 	}
477 }
478 #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
arch_pick_mmap_layout(struct mm_struct * mm,struct rlimit * rlim_stack)479 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
480 {
481 	mm->mmap_base = TASK_UNMAPPED_BASE;
482 	clear_bit(MMF_TOPDOWN, &mm->flags);
483 }
484 #endif
485 #ifdef CONFIG_MMU
486 EXPORT_SYMBOL_IF_KUNIT(arch_pick_mmap_layout);
487 #endif
488 
489 /**
490  * __account_locked_vm - account locked pages to an mm's locked_vm
491  * @mm:          mm to account against
492  * @pages:       number of pages to account
493  * @inc:         %true if @pages should be considered positive, %false if not
494  * @task:        task used to check RLIMIT_MEMLOCK
495  * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
496  *
497  * Assumes @task and @mm are valid (i.e. at least one reference on each), and
498  * that mmap_lock is held as writer.
499  *
500  * Return:
501  * * 0       on success
502  * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
503  */
__account_locked_vm(struct mm_struct * mm,unsigned long pages,bool inc,struct task_struct * task,bool bypass_rlim)504 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
505 			struct task_struct *task, bool bypass_rlim)
506 {
507 	unsigned long locked_vm, limit;
508 	int ret = 0;
509 
510 	mmap_assert_write_locked(mm);
511 
512 	locked_vm = mm->locked_vm;
513 	if (inc) {
514 		if (!bypass_rlim) {
515 			limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
516 			if (locked_vm + pages > limit)
517 				ret = -ENOMEM;
518 		}
519 		if (!ret)
520 			mm->locked_vm = locked_vm + pages;
521 	} else {
522 		WARN_ON_ONCE(pages > locked_vm);
523 		mm->locked_vm = locked_vm - pages;
524 	}
525 
526 	pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
527 		 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
528 		 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
529 		 ret ? " - exceeded" : "");
530 
531 	return ret;
532 }
533 EXPORT_SYMBOL_GPL(__account_locked_vm);
534 
535 /**
536  * account_locked_vm - account locked pages to an mm's locked_vm
537  * @mm:          mm to account against, may be NULL
538  * @pages:       number of pages to account
539  * @inc:         %true if @pages should be considered positive, %false if not
540  *
541  * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
542  *
543  * Return:
544  * * 0       on success, or if mm is NULL
545  * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
546  */
account_locked_vm(struct mm_struct * mm,unsigned long pages,bool inc)547 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
548 {
549 	int ret;
550 
551 	if (pages == 0 || !mm)
552 		return 0;
553 
554 	mmap_write_lock(mm);
555 	ret = __account_locked_vm(mm, pages, inc, current,
556 				  capable(CAP_IPC_LOCK));
557 	mmap_write_unlock(mm);
558 
559 	return ret;
560 }
561 EXPORT_SYMBOL_GPL(account_locked_vm);
562 
vm_mmap_pgoff(struct file * file,unsigned long addr,unsigned long len,unsigned long prot,unsigned long flag,unsigned long pgoff)563 unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
564 	unsigned long len, unsigned long prot,
565 	unsigned long flag, unsigned long pgoff)
566 {
567 	unsigned long ret;
568 	struct mm_struct *mm = current->mm;
569 	unsigned long populate;
570 	LIST_HEAD(uf);
571 
572 	ret = security_mmap_file(file, prot, flag);
573 	if (!ret)
574 		ret = fsnotify_mmap_perm(file, prot, pgoff >> PAGE_SHIFT, len);
575 	if (!ret) {
576 		if (mmap_write_lock_killable(mm))
577 			return -EINTR;
578 		ret = do_mmap(file, addr, len, prot, flag, 0, pgoff, &populate,
579 			      &uf);
580 		mmap_write_unlock(mm);
581 		userfaultfd_unmap_complete(mm, &uf);
582 		if (populate)
583 			mm_populate(ret, populate);
584 	}
585 	return ret;
586 }
587 
588 /*
589  * Perform a userland memory mapping into the current process address space. See
590  * the comment for do_mmap() for more details on this operation in general.
591  *
592  * This differs from do_mmap() in that:
593  *
594  * a. An offset parameter is provided rather than pgoff, which is both checked
595  *    for overflow and page alignment.
596  * b. mmap locking is performed on the caller's behalf.
597  * c. Userfaultfd unmap events and memory population are handled.
598  *
599  * This means that this function performs essentially the same work as if
600  * userland were invoking mmap (2).
601  *
602  * Returns either an error, or the address at which the requested mapping has
603  * been performed.
604  */
vm_mmap(struct file * file,unsigned long addr,unsigned long len,unsigned long prot,unsigned long flag,unsigned long offset)605 unsigned long vm_mmap(struct file *file, unsigned long addr,
606 	unsigned long len, unsigned long prot,
607 	unsigned long flag, unsigned long offset)
608 {
609 	if (unlikely(offset + PAGE_ALIGN(len) < offset))
610 		return -EINVAL;
611 	if (unlikely(offset_in_page(offset)))
612 		return -EINVAL;
613 
614 	return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
615 }
616 EXPORT_SYMBOL(vm_mmap);
617 
kmalloc_gfp_adjust(gfp_t flags,size_t size)618 static gfp_t kmalloc_gfp_adjust(gfp_t flags, size_t size)
619 {
620 	/*
621 	 * We want to attempt a large physically contiguous block first because
622 	 * it is less likely to fragment multiple larger blocks and therefore
623 	 * contribute to a long term fragmentation less than vmalloc fallback.
624 	 * However make sure that larger requests are not too disruptive - no
625 	 * OOM killer and no allocation failure warnings as we have a fallback.
626 	 */
627 	if (size > PAGE_SIZE) {
628 		flags |= __GFP_NOWARN;
629 
630 		if (!(flags & __GFP_RETRY_MAYFAIL))
631 			flags |= __GFP_NORETRY;
632 
633 		/* nofail semantic is implemented by the vmalloc fallback */
634 		flags &= ~__GFP_NOFAIL;
635 	}
636 
637 	return flags;
638 }
639 
640 /**
641  * __kvmalloc_node - attempt to allocate physically contiguous memory, but upon
642  * failure, fall back to non-contiguous (vmalloc) allocation.
643  * @size: size of the request.
644  * @b: which set of kmalloc buckets to allocate from.
645  * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
646  * @node: numa node to allocate from
647  *
648  * Uses kmalloc to get the memory but if the allocation fails then falls back
649  * to the vmalloc allocator. Use kvfree for freeing the memory.
650  *
651  * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
652  * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
653  * preferable to the vmalloc fallback, due to visible performance drawbacks.
654  *
655  * Return: pointer to the allocated memory of %NULL in case of failure
656  */
__kvmalloc_node_noprof(DECL_BUCKET_PARAMS (size,b),gfp_t flags,int node)657 void *__kvmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node)
658 {
659 	void *ret;
660 
661 	/*
662 	 * It doesn't really make sense to fallback to vmalloc for sub page
663 	 * requests
664 	 */
665 	ret = __kmalloc_node_noprof(PASS_BUCKET_PARAMS(size, b),
666 				    kmalloc_gfp_adjust(flags, size),
667 				    node);
668 	if (ret || size <= PAGE_SIZE)
669 		return ret;
670 
671 	/* non-sleeping allocations are not supported by vmalloc */
672 	if (!gfpflags_allow_blocking(flags))
673 		return NULL;
674 
675 	/* Don't even allow crazy sizes */
676 	if (unlikely(size > INT_MAX)) {
677 		WARN_ON_ONCE(!(flags & __GFP_NOWARN));
678 		return NULL;
679 	}
680 
681 	/*
682 	 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
683 	 * since the callers already cannot assume anything
684 	 * about the resulting pointer, and cannot play
685 	 * protection games.
686 	 */
687 	return __vmalloc_node_range_noprof(size, 1, VMALLOC_START, VMALLOC_END,
688 			flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
689 			node, __builtin_return_address(0));
690 }
691 EXPORT_SYMBOL(__kvmalloc_node_noprof);
692 
693 /**
694  * kvfree() - Free memory.
695  * @addr: Pointer to allocated memory.
696  *
697  * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
698  * It is slightly more efficient to use kfree() or vfree() if you are certain
699  * that you know which one to use.
700  *
701  * Context: Either preemptible task context or not-NMI interrupt.
702  */
kvfree(const void * addr)703 void kvfree(const void *addr)
704 {
705 	if (is_vmalloc_addr(addr))
706 		vfree(addr);
707 	else
708 		kfree(addr);
709 }
710 EXPORT_SYMBOL(kvfree);
711 
712 /**
713  * kvfree_sensitive - Free a data object containing sensitive information.
714  * @addr: address of the data object to be freed.
715  * @len: length of the data object.
716  *
717  * Use the special memzero_explicit() function to clear the content of a
718  * kvmalloc'ed object containing sensitive data to make sure that the
719  * compiler won't optimize out the data clearing.
720  */
kvfree_sensitive(const void * addr,size_t len)721 void kvfree_sensitive(const void *addr, size_t len)
722 {
723 	if (likely(!ZERO_OR_NULL_PTR(addr))) {
724 		memzero_explicit((void *)addr, len);
725 		kvfree(addr);
726 	}
727 }
728 EXPORT_SYMBOL(kvfree_sensitive);
729 
730 /**
731  * kvrealloc - reallocate memory; contents remain unchanged
732  * @p: object to reallocate memory for
733  * @size: the size to reallocate
734  * @flags: the flags for the page level allocator
735  *
736  * If @p is %NULL, kvrealloc() behaves exactly like kvmalloc(). If @size is 0
737  * and @p is not a %NULL pointer, the object pointed to is freed.
738  *
739  * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
740  * initial memory allocation, every subsequent call to this API for the same
741  * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
742  * __GFP_ZERO is not fully honored by this API.
743  *
744  * In any case, the contents of the object pointed to are preserved up to the
745  * lesser of the new and old sizes.
746  *
747  * This function must not be called concurrently with itself or kvfree() for the
748  * same memory allocation.
749  *
750  * Return: pointer to the allocated memory or %NULL in case of error
751  */
kvrealloc_noprof(const void * p,size_t size,gfp_t flags)752 void *kvrealloc_noprof(const void *p, size_t size, gfp_t flags)
753 {
754 	void *n;
755 
756 	if (is_vmalloc_addr(p))
757 		return vrealloc_noprof(p, size, flags);
758 
759 	n = krealloc_noprof(p, size, kmalloc_gfp_adjust(flags, size));
760 	if (!n) {
761 		/* We failed to krealloc(), fall back to kvmalloc(). */
762 		n = kvmalloc_noprof(size, flags);
763 		if (!n)
764 			return NULL;
765 
766 		if (p) {
767 			/* We already know that `p` is not a vmalloc address. */
768 			kasan_disable_current();
769 			memcpy(n, kasan_reset_tag(p), ksize(p));
770 			kasan_enable_current();
771 
772 			kfree(p);
773 		}
774 	}
775 
776 	return n;
777 }
778 EXPORT_SYMBOL(kvrealloc_noprof);
779 
780 /**
781  * __vmalloc_array - allocate memory for a virtually contiguous array.
782  * @n: number of elements.
783  * @size: element size.
784  * @flags: the type of memory to allocate (see kmalloc).
785  */
__vmalloc_array_noprof(size_t n,size_t size,gfp_t flags)786 void *__vmalloc_array_noprof(size_t n, size_t size, gfp_t flags)
787 {
788 	size_t bytes;
789 
790 	if (unlikely(check_mul_overflow(n, size, &bytes)))
791 		return NULL;
792 	return __vmalloc_noprof(bytes, flags);
793 }
794 EXPORT_SYMBOL(__vmalloc_array_noprof);
795 
796 /**
797  * vmalloc_array - allocate memory for a virtually contiguous array.
798  * @n: number of elements.
799  * @size: element size.
800  */
vmalloc_array_noprof(size_t n,size_t size)801 void *vmalloc_array_noprof(size_t n, size_t size)
802 {
803 	return __vmalloc_array_noprof(n, size, GFP_KERNEL);
804 }
805 EXPORT_SYMBOL(vmalloc_array_noprof);
806 
807 /**
808  * __vcalloc - allocate and zero memory for a virtually contiguous array.
809  * @n: number of elements.
810  * @size: element size.
811  * @flags: the type of memory to allocate (see kmalloc).
812  */
__vcalloc_noprof(size_t n,size_t size,gfp_t flags)813 void *__vcalloc_noprof(size_t n, size_t size, gfp_t flags)
814 {
815 	return __vmalloc_array_noprof(n, size, flags | __GFP_ZERO);
816 }
817 EXPORT_SYMBOL(__vcalloc_noprof);
818 
819 /**
820  * vcalloc - allocate and zero memory for a virtually contiguous array.
821  * @n: number of elements.
822  * @size: element size.
823  */
vcalloc_noprof(size_t n,size_t size)824 void *vcalloc_noprof(size_t n, size_t size)
825 {
826 	return __vmalloc_array_noprof(n, size, GFP_KERNEL | __GFP_ZERO);
827 }
828 EXPORT_SYMBOL(vcalloc_noprof);
829 
folio_anon_vma(const struct folio * folio)830 struct anon_vma *folio_anon_vma(const struct folio *folio)
831 {
832 	unsigned long mapping = (unsigned long)folio->mapping;
833 
834 	if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
835 		return NULL;
836 	return (void *)(mapping - PAGE_MAPPING_ANON);
837 }
838 
839 /**
840  * folio_mapping - Find the mapping where this folio is stored.
841  * @folio: The folio.
842  *
843  * For folios which are in the page cache, return the mapping that this
844  * page belongs to.  Folios in the swap cache return the swap mapping
845  * this page is stored in (which is different from the mapping for the
846  * swap file or swap device where the data is stored).
847  *
848  * You can call this for folios which aren't in the swap cache or page
849  * cache and it will return NULL.
850  */
folio_mapping(struct folio * folio)851 struct address_space *folio_mapping(struct folio *folio)
852 {
853 	struct address_space *mapping;
854 
855 	/* This happens if someone calls flush_dcache_page on slab page */
856 	if (unlikely(folio_test_slab(folio)))
857 		return NULL;
858 
859 	if (unlikely(folio_test_swapcache(folio)))
860 		return swap_address_space(folio->swap);
861 
862 	mapping = folio->mapping;
863 	if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
864 		return NULL;
865 
866 	return mapping;
867 }
868 EXPORT_SYMBOL(folio_mapping);
869 
870 /**
871  * folio_copy - Copy the contents of one folio to another.
872  * @dst: Folio to copy to.
873  * @src: Folio to copy from.
874  *
875  * The bytes in the folio represented by @src are copied to @dst.
876  * Assumes the caller has validated that @dst is at least as large as @src.
877  * Can be called in atomic context for order-0 folios, but if the folio is
878  * larger, it may sleep.
879  */
folio_copy(struct folio * dst,struct folio * src)880 void folio_copy(struct folio *dst, struct folio *src)
881 {
882 	long i = 0;
883 	long nr = folio_nr_pages(src);
884 
885 	for (;;) {
886 		copy_highpage(folio_page(dst, i), folio_page(src, i));
887 		if (++i == nr)
888 			break;
889 		cond_resched();
890 	}
891 }
892 EXPORT_SYMBOL(folio_copy);
893 
folio_mc_copy(struct folio * dst,struct folio * src)894 int folio_mc_copy(struct folio *dst, struct folio *src)
895 {
896 	long nr = folio_nr_pages(src);
897 	long i = 0;
898 
899 	for (;;) {
900 		if (copy_mc_highpage(folio_page(dst, i), folio_page(src, i)))
901 			return -EHWPOISON;
902 		if (++i == nr)
903 			break;
904 		cond_resched();
905 	}
906 
907 	return 0;
908 }
909 EXPORT_SYMBOL(folio_mc_copy);
910 
911 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
912 int sysctl_overcommit_ratio __read_mostly = 50;
913 unsigned long sysctl_overcommit_kbytes __read_mostly;
914 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
915 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
916 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
917 
overcommit_ratio_handler(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)918 int overcommit_ratio_handler(const struct ctl_table *table, int write, void *buffer,
919 		size_t *lenp, loff_t *ppos)
920 {
921 	int ret;
922 
923 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
924 	if (ret == 0 && write)
925 		sysctl_overcommit_kbytes = 0;
926 	return ret;
927 }
928 
sync_overcommit_as(struct work_struct * dummy)929 static void sync_overcommit_as(struct work_struct *dummy)
930 {
931 	percpu_counter_sync(&vm_committed_as);
932 }
933 
overcommit_policy_handler(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)934 int overcommit_policy_handler(const struct ctl_table *table, int write, void *buffer,
935 		size_t *lenp, loff_t *ppos)
936 {
937 	struct ctl_table t;
938 	int new_policy = -1;
939 	int ret;
940 
941 	/*
942 	 * The deviation of sync_overcommit_as could be big with loose policy
943 	 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
944 	 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
945 	 * with the strict "NEVER", and to avoid possible race condition (even
946 	 * though user usually won't too frequently do the switching to policy
947 	 * OVERCOMMIT_NEVER), the switch is done in the following order:
948 	 *	1. changing the batch
949 	 *	2. sync percpu count on each CPU
950 	 *	3. switch the policy
951 	 */
952 	if (write) {
953 		t = *table;
954 		t.data = &new_policy;
955 		ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
956 		if (ret || new_policy == -1)
957 			return ret;
958 
959 		mm_compute_batch(new_policy);
960 		if (new_policy == OVERCOMMIT_NEVER)
961 			schedule_on_each_cpu(sync_overcommit_as);
962 		sysctl_overcommit_memory = new_policy;
963 	} else {
964 		ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
965 	}
966 
967 	return ret;
968 }
969 
overcommit_kbytes_handler(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)970 int overcommit_kbytes_handler(const struct ctl_table *table, int write, void *buffer,
971 		size_t *lenp, loff_t *ppos)
972 {
973 	int ret;
974 
975 	ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
976 	if (ret == 0 && write)
977 		sysctl_overcommit_ratio = 0;
978 	return ret;
979 }
980 
981 /*
982  * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
983  */
vm_commit_limit(void)984 unsigned long vm_commit_limit(void)
985 {
986 	unsigned long allowed;
987 
988 	if (sysctl_overcommit_kbytes)
989 		allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
990 	else
991 		allowed = ((totalram_pages() - hugetlb_total_pages())
992 			   * sysctl_overcommit_ratio / 100);
993 	allowed += total_swap_pages;
994 
995 	return allowed;
996 }
997 
998 /*
999  * Make sure vm_committed_as in one cacheline and not cacheline shared with
1000  * other variables. It can be updated by several CPUs frequently.
1001  */
1002 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
1003 
1004 /*
1005  * The global memory commitment made in the system can be a metric
1006  * that can be used to drive ballooning decisions when Linux is hosted
1007  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
1008  * balancing memory across competing virtual machines that are hosted.
1009  * Several metrics drive this policy engine including the guest reported
1010  * memory commitment.
1011  *
1012  * The time cost of this is very low for small platforms, and for big
1013  * platform like a 2S/36C/72T Skylake server, in worst case where
1014  * vm_committed_as's spinlock is under severe contention, the time cost
1015  * could be about 30~40 microseconds.
1016  */
vm_memory_committed(void)1017 unsigned long vm_memory_committed(void)
1018 {
1019 	return percpu_counter_sum_positive(&vm_committed_as);
1020 }
1021 EXPORT_SYMBOL_GPL(vm_memory_committed);
1022 
1023 /*
1024  * Check that a process has enough memory to allocate a new virtual
1025  * mapping. 0 means there is enough memory for the allocation to
1026  * succeed and -ENOMEM implies there is not.
1027  *
1028  * We currently support three overcommit policies, which are set via the
1029  * vm.overcommit_memory sysctl.  See Documentation/mm/overcommit-accounting.rst
1030  *
1031  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1032  * Additional code 2002 Jul 20 by Robert Love.
1033  *
1034  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1035  *
1036  * Note this is a helper function intended to be used by LSMs which
1037  * wish to use this logic.
1038  */
__vm_enough_memory(struct mm_struct * mm,long pages,int cap_sys_admin)1039 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1040 {
1041 	long allowed;
1042 	unsigned long bytes_failed;
1043 
1044 	vm_acct_memory(pages);
1045 
1046 	/*
1047 	 * Sometimes we want to use more memory than we have
1048 	 */
1049 	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1050 		return 0;
1051 
1052 	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1053 		if (pages > totalram_pages() + total_swap_pages)
1054 			goto error;
1055 		return 0;
1056 	}
1057 
1058 	allowed = vm_commit_limit();
1059 	/*
1060 	 * Reserve some for root
1061 	 */
1062 	if (!cap_sys_admin)
1063 		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1064 
1065 	/*
1066 	 * Don't let a single process grow so big a user can't recover
1067 	 */
1068 	if (mm) {
1069 		long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1070 
1071 		allowed -= min_t(long, mm->total_vm / 32, reserve);
1072 	}
1073 
1074 	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1075 		return 0;
1076 error:
1077 	bytes_failed = pages << PAGE_SHIFT;
1078 	pr_warn_ratelimited("%s: pid: %d, comm: %s, bytes: %lu not enough memory for the allocation\n",
1079 			    __func__, current->pid, current->comm, bytes_failed);
1080 	vm_unacct_memory(pages);
1081 
1082 	return -ENOMEM;
1083 }
1084 
1085 /**
1086  * get_cmdline() - copy the cmdline value to a buffer.
1087  * @task:     the task whose cmdline value to copy.
1088  * @buffer:   the buffer to copy to.
1089  * @buflen:   the length of the buffer. Larger cmdline values are truncated
1090  *            to this length.
1091  *
1092  * Return: the size of the cmdline field copied. Note that the copy does
1093  * not guarantee an ending NULL byte.
1094  */
get_cmdline(struct task_struct * task,char * buffer,int buflen)1095 int get_cmdline(struct task_struct *task, char *buffer, int buflen)
1096 {
1097 	int res = 0;
1098 	unsigned int len;
1099 	struct mm_struct *mm = get_task_mm(task);
1100 	unsigned long arg_start, arg_end, env_start, env_end;
1101 	if (!mm)
1102 		goto out;
1103 	if (!mm->arg_end)
1104 		goto out_mm;	/* Shh! No looking before we're done */
1105 
1106 	spin_lock(&mm->arg_lock);
1107 	arg_start = mm->arg_start;
1108 	arg_end = mm->arg_end;
1109 	env_start = mm->env_start;
1110 	env_end = mm->env_end;
1111 	spin_unlock(&mm->arg_lock);
1112 
1113 	len = arg_end - arg_start;
1114 
1115 	if (len > buflen)
1116 		len = buflen;
1117 
1118 	res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1119 
1120 	/*
1121 	 * If the nul at the end of args has been overwritten, then
1122 	 * assume application is using setproctitle(3).
1123 	 */
1124 	if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1125 		len = strnlen(buffer, res);
1126 		if (len < res) {
1127 			res = len;
1128 		} else {
1129 			len = env_end - env_start;
1130 			if (len > buflen - res)
1131 				len = buflen - res;
1132 			res += access_process_vm(task, env_start,
1133 						 buffer+res, len,
1134 						 FOLL_FORCE);
1135 			res = strnlen(buffer, res);
1136 		}
1137 	}
1138 out_mm:
1139 	mmput(mm);
1140 out:
1141 	return res;
1142 }
1143 
memcmp_pages(struct page * page1,struct page * page2)1144 int __weak memcmp_pages(struct page *page1, struct page *page2)
1145 {
1146 	char *addr1, *addr2;
1147 	int ret;
1148 
1149 	addr1 = kmap_local_page(page1);
1150 	addr2 = kmap_local_page(page2);
1151 	ret = memcmp(addr1, addr2, PAGE_SIZE);
1152 	kunmap_local(addr2);
1153 	kunmap_local(addr1);
1154 	return ret;
1155 }
1156 
1157 #ifdef CONFIG_PRINTK
1158 /**
1159  * mem_dump_obj - Print available provenance information
1160  * @object: object for which to find provenance information.
1161  *
1162  * This function uses pr_cont(), so that the caller is expected to have
1163  * printed out whatever preamble is appropriate.  The provenance information
1164  * depends on the type of object and on how much debugging is enabled.
1165  * For example, for a slab-cache object, the slab name is printed, and,
1166  * if available, the return address and stack trace from the allocation
1167  * and last free path of that object.
1168  */
mem_dump_obj(void * object)1169 void mem_dump_obj(void *object)
1170 {
1171 	const char *type;
1172 
1173 	if (kmem_dump_obj(object))
1174 		return;
1175 
1176 	if (vmalloc_dump_obj(object))
1177 		return;
1178 
1179 	if (is_vmalloc_addr(object))
1180 		type = "vmalloc memory";
1181 	else if (virt_addr_valid(object))
1182 		type = "non-slab/vmalloc memory";
1183 	else if (object == NULL)
1184 		type = "NULL pointer";
1185 	else if (object == ZERO_SIZE_PTR)
1186 		type = "zero-size pointer";
1187 	else
1188 		type = "non-paged memory";
1189 
1190 	pr_cont(" %s\n", type);
1191 }
1192 EXPORT_SYMBOL_GPL(mem_dump_obj);
1193 #endif
1194 
1195 /*
1196  * A driver might set a page logically offline -- PageOffline() -- and
1197  * turn the page inaccessible in the hypervisor; after that, access to page
1198  * content can be fatal.
1199  *
1200  * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1201  * pages after checking PageOffline(); however, these PFN walkers can race
1202  * with drivers that set PageOffline().
1203  *
1204  * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1205  * synchronize with such drivers, achieving that a page cannot be set
1206  * PageOffline() while frozen.
1207  *
1208  * page_offline_begin()/page_offline_end() is used by drivers that care about
1209  * such races when setting a page PageOffline().
1210  */
1211 static DECLARE_RWSEM(page_offline_rwsem);
1212 
page_offline_freeze(void)1213 void page_offline_freeze(void)
1214 {
1215 	down_read(&page_offline_rwsem);
1216 }
1217 
page_offline_thaw(void)1218 void page_offline_thaw(void)
1219 {
1220 	up_read(&page_offline_rwsem);
1221 }
1222 
page_offline_begin(void)1223 void page_offline_begin(void)
1224 {
1225 	down_write(&page_offline_rwsem);
1226 }
1227 EXPORT_SYMBOL(page_offline_begin);
1228 
page_offline_end(void)1229 void page_offline_end(void)
1230 {
1231 	up_write(&page_offline_rwsem);
1232 }
1233 EXPORT_SYMBOL(page_offline_end);
1234 
1235 #ifndef flush_dcache_folio
flush_dcache_folio(struct folio * folio)1236 void flush_dcache_folio(struct folio *folio)
1237 {
1238 	long i, nr = folio_nr_pages(folio);
1239 
1240 	for (i = 0; i < nr; i++)
1241 		flush_dcache_page(folio_page(folio, i));
1242 }
1243 EXPORT_SYMBOL(flush_dcache_folio);
1244 #endif
1245