Loading...
1#include <linux/mm.h>
2#include <linux/slab.h>
3#include <linux/string.h>
4#include <linux/module.h>
5#include <linux/err.h>
6#include <linux/sched.h>
7#include <asm/uaccess.h>
8
9#include "internal.h"
10
11#define CREATE_TRACE_POINTS
12#include <trace/events/kmem.h>
13
14/**
15 * kstrdup - allocate space for and copy an existing string
16 * @s: the string to duplicate
17 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
18 */
19char *kstrdup(const char *s, gfp_t gfp)
20{
21 size_t len;
22 char *buf;
23
24 if (!s)
25 return NULL;
26
27 len = strlen(s) + 1;
28 buf = kmalloc_track_caller(len, gfp);
29 if (buf)
30 memcpy(buf, s, len);
31 return buf;
32}
33EXPORT_SYMBOL(kstrdup);
34
35/**
36 * kstrndup - allocate space for and copy an existing string
37 * @s: the string to duplicate
38 * @max: read at most @max chars from @s
39 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
40 */
41char *kstrndup(const char *s, size_t max, gfp_t gfp)
42{
43 size_t len;
44 char *buf;
45
46 if (!s)
47 return NULL;
48
49 len = strnlen(s, max);
50 buf = kmalloc_track_caller(len+1, gfp);
51 if (buf) {
52 memcpy(buf, s, len);
53 buf[len] = '\0';
54 }
55 return buf;
56}
57EXPORT_SYMBOL(kstrndup);
58
59/**
60 * kmemdup - duplicate region of memory
61 *
62 * @src: memory region to duplicate
63 * @len: memory region length
64 * @gfp: GFP mask to use
65 */
66void *kmemdup(const void *src, size_t len, gfp_t gfp)
67{
68 void *p;
69
70 p = kmalloc_track_caller(len, gfp);
71 if (p)
72 memcpy(p, src, len);
73 return p;
74}
75EXPORT_SYMBOL(kmemdup);
76
77/**
78 * memdup_user - duplicate memory region from user space
79 *
80 * @src: source address in user space
81 * @len: number of bytes to copy
82 *
83 * Returns an ERR_PTR() on failure.
84 */
85void *memdup_user(const void __user *src, size_t len)
86{
87 void *p;
88
89 /*
90 * Always use GFP_KERNEL, since copy_from_user() can sleep and
91 * cause pagefault, which makes it pointless to use GFP_NOFS
92 * or GFP_ATOMIC.
93 */
94 p = kmalloc_track_caller(len, GFP_KERNEL);
95 if (!p)
96 return ERR_PTR(-ENOMEM);
97
98 if (copy_from_user(p, src, len)) {
99 kfree(p);
100 return ERR_PTR(-EFAULT);
101 }
102
103 return p;
104}
105EXPORT_SYMBOL(memdup_user);
106
107/**
108 * __krealloc - like krealloc() but don't free @p.
109 * @p: object to reallocate memory for.
110 * @new_size: how many bytes of memory are required.
111 * @flags: the type of memory to allocate.
112 *
113 * This function is like krealloc() except it never frees the originally
114 * allocated buffer. Use this if you don't want to free the buffer immediately
115 * like, for example, with RCU.
116 */
117void *__krealloc(const void *p, size_t new_size, gfp_t flags)
118{
119 void *ret;
120 size_t ks = 0;
121
122 if (unlikely(!new_size))
123 return ZERO_SIZE_PTR;
124
125 if (p)
126 ks = ksize(p);
127
128 if (ks >= new_size)
129 return (void *)p;
130
131 ret = kmalloc_track_caller(new_size, flags);
132 if (ret && p)
133 memcpy(ret, p, ks);
134
135 return ret;
136}
137EXPORT_SYMBOL(__krealloc);
138
139/**
140 * krealloc - reallocate memory. The contents will remain unchanged.
141 * @p: object to reallocate memory for.
142 * @new_size: how many bytes of memory are required.
143 * @flags: the type of memory to allocate.
144 *
145 * The contents of the object pointed to are preserved up to the
146 * lesser of the new and old sizes. If @p is %NULL, krealloc()
147 * behaves exactly like kmalloc(). If @size is 0 and @p is not a
148 * %NULL pointer, the object pointed to is freed.
149 */
150void *krealloc(const void *p, size_t new_size, gfp_t flags)
151{
152 void *ret;
153
154 if (unlikely(!new_size)) {
155 kfree(p);
156 return ZERO_SIZE_PTR;
157 }
158
159 ret = __krealloc(p, new_size, flags);
160 if (ret && p != ret)
161 kfree(p);
162
163 return ret;
164}
165EXPORT_SYMBOL(krealloc);
166
167/**
168 * kzfree - like kfree but zero memory
169 * @p: object to free memory of
170 *
171 * The memory of the object @p points to is zeroed before freed.
172 * If @p is %NULL, kzfree() does nothing.
173 *
174 * Note: this function zeroes the whole allocated buffer which can be a good
175 * deal bigger than the requested buffer size passed to kmalloc(). So be
176 * careful when using this function in performance sensitive code.
177 */
178void kzfree(const void *p)
179{
180 size_t ks;
181 void *mem = (void *)p;
182
183 if (unlikely(ZERO_OR_NULL_PTR(mem)))
184 return;
185 ks = ksize(mem);
186 memset(mem, 0, ks);
187 kfree(mem);
188}
189EXPORT_SYMBOL(kzfree);
190
191/*
192 * strndup_user - duplicate an existing string from user space
193 * @s: The string to duplicate
194 * @n: Maximum number of bytes to copy, including the trailing NUL.
195 */
196char *strndup_user(const char __user *s, long n)
197{
198 char *p;
199 long length;
200
201 length = strnlen_user(s, n);
202
203 if (!length)
204 return ERR_PTR(-EFAULT);
205
206 if (length > n)
207 return ERR_PTR(-EINVAL);
208
209 p = memdup_user(s, length);
210
211 if (IS_ERR(p))
212 return p;
213
214 p[length - 1] = '\0';
215
216 return p;
217}
218EXPORT_SYMBOL(strndup_user);
219
220void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
221 struct vm_area_struct *prev, struct rb_node *rb_parent)
222{
223 struct vm_area_struct *next;
224
225 vma->vm_prev = prev;
226 if (prev) {
227 next = prev->vm_next;
228 prev->vm_next = vma;
229 } else {
230 mm->mmap = vma;
231 if (rb_parent)
232 next = rb_entry(rb_parent,
233 struct vm_area_struct, vm_rb);
234 else
235 next = NULL;
236 }
237 vma->vm_next = next;
238 if (next)
239 next->vm_prev = vma;
240}
241
242#if defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
243void arch_pick_mmap_layout(struct mm_struct *mm)
244{
245 mm->mmap_base = TASK_UNMAPPED_BASE;
246 mm->get_unmapped_area = arch_get_unmapped_area;
247 mm->unmap_area = arch_unmap_area;
248}
249#endif
250
251/*
252 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
253 * back to the regular GUP.
254 * If the architecture not support this function, simply return with no
255 * page pinned
256 */
257int __attribute__((weak)) __get_user_pages_fast(unsigned long start,
258 int nr_pages, int write, struct page **pages)
259{
260 return 0;
261}
262EXPORT_SYMBOL_GPL(__get_user_pages_fast);
263
264/**
265 * get_user_pages_fast() - pin user pages in memory
266 * @start: starting user address
267 * @nr_pages: number of pages from start to pin
268 * @write: whether pages will be written to
269 * @pages: array that receives pointers to the pages pinned.
270 * Should be at least nr_pages long.
271 *
272 * Returns number of pages pinned. This may be fewer than the number
273 * requested. If nr_pages is 0 or negative, returns 0. If no pages
274 * were pinned, returns -errno.
275 *
276 * get_user_pages_fast provides equivalent functionality to get_user_pages,
277 * operating on current and current->mm, with force=0 and vma=NULL. However
278 * unlike get_user_pages, it must be called without mmap_sem held.
279 *
280 * get_user_pages_fast may take mmap_sem and page table locks, so no
281 * assumptions can be made about lack of locking. get_user_pages_fast is to be
282 * implemented in a way that is advantageous (vs get_user_pages()) when the
283 * user memory area is already faulted in and present in ptes. However if the
284 * pages have to be faulted in, it may turn out to be slightly slower so
285 * callers need to carefully consider what to use. On many architectures,
286 * get_user_pages_fast simply falls back to get_user_pages.
287 */
288int __attribute__((weak)) get_user_pages_fast(unsigned long start,
289 int nr_pages, int write, struct page **pages)
290{
291 struct mm_struct *mm = current->mm;
292 int ret;
293
294 down_read(&mm->mmap_sem);
295 ret = get_user_pages(current, mm, start, nr_pages,
296 write, 0, pages, NULL);
297 up_read(&mm->mmap_sem);
298
299 return ret;
300}
301EXPORT_SYMBOL_GPL(get_user_pages_fast);
302
303/* Tracepoints definitions. */
304EXPORT_TRACEPOINT_SYMBOL(kmalloc);
305EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
306EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
307EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
308EXPORT_TRACEPOINT_SYMBOL(kfree);
309EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/mm.h>
3#include <linux/slab.h>
4#include <linux/string.h>
5#include <linux/compiler.h>
6#include <linux/export.h>
7#include <linux/err.h>
8#include <linux/sched.h>
9#include <linux/sched/mm.h>
10#include <linux/sched/signal.h>
11#include <linux/sched/task_stack.h>
12#include <linux/security.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/mman.h>
16#include <linux/hugetlb.h>
17#include <linux/vmalloc.h>
18#include <linux/userfaultfd_k.h>
19#include <linux/elf.h>
20#include <linux/elf-randomize.h>
21#include <linux/personality.h>
22#include <linux/random.h>
23#include <linux/processor.h>
24#include <linux/sizes.h>
25#include <linux/compat.h>
26
27#include <linux/uaccess.h>
28
29#include "internal.h"
30#include "swap.h"
31
32/**
33 * kfree_const - conditionally free memory
34 * @x: pointer to the memory
35 *
36 * Function calls kfree only if @x is not in .rodata section.
37 */
38void kfree_const(const void *x)
39{
40 if (!is_kernel_rodata((unsigned long)x))
41 kfree(x);
42}
43EXPORT_SYMBOL(kfree_const);
44
45/**
46 * kstrdup - allocate space for and copy an existing string
47 * @s: the string to duplicate
48 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
49 *
50 * Return: newly allocated copy of @s or %NULL in case of error
51 */
52noinline
53char *kstrdup(const char *s, gfp_t gfp)
54{
55 size_t len;
56 char *buf;
57
58 if (!s)
59 return NULL;
60
61 len = strlen(s) + 1;
62 buf = kmalloc_track_caller(len, gfp);
63 if (buf)
64 memcpy(buf, s, len);
65 return buf;
66}
67EXPORT_SYMBOL(kstrdup);
68
69/**
70 * kstrdup_const - conditionally duplicate an existing const string
71 * @s: the string to duplicate
72 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
73 *
74 * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
75 * must not be passed to krealloc().
76 *
77 * Return: source string if it is in .rodata section otherwise
78 * fallback to kstrdup.
79 */
80const char *kstrdup_const(const char *s, gfp_t gfp)
81{
82 if (is_kernel_rodata((unsigned long)s))
83 return s;
84
85 return kstrdup(s, gfp);
86}
87EXPORT_SYMBOL(kstrdup_const);
88
89/**
90 * kstrndup - allocate space for and copy an existing string
91 * @s: the string to duplicate
92 * @max: read at most @max chars from @s
93 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
94 *
95 * Note: Use kmemdup_nul() instead if the size is known exactly.
96 *
97 * Return: newly allocated copy of @s or %NULL in case of error
98 */
99char *kstrndup(const char *s, size_t max, gfp_t gfp)
100{
101 size_t len;
102 char *buf;
103
104 if (!s)
105 return NULL;
106
107 len = strnlen(s, max);
108 buf = kmalloc_track_caller(len+1, gfp);
109 if (buf) {
110 memcpy(buf, s, len);
111 buf[len] = '\0';
112 }
113 return buf;
114}
115EXPORT_SYMBOL(kstrndup);
116
117/**
118 * kmemdup - duplicate region of memory
119 *
120 * @src: memory region to duplicate
121 * @len: memory region length
122 * @gfp: GFP mask to use
123 *
124 * Return: newly allocated copy of @src or %NULL in case of error,
125 * result is physically contiguous. Use kfree() to free.
126 */
127void *kmemdup(const void *src, size_t len, gfp_t gfp)
128{
129 void *p;
130
131 p = kmalloc_track_caller(len, gfp);
132 if (p)
133 memcpy(p, src, len);
134 return p;
135}
136EXPORT_SYMBOL(kmemdup);
137
138/**
139 * kvmemdup - duplicate region of memory
140 *
141 * @src: memory region to duplicate
142 * @len: memory region length
143 * @gfp: GFP mask to use
144 *
145 * Return: newly allocated copy of @src or %NULL in case of error,
146 * result may be not physically contiguous. Use kvfree() to free.
147 */
148void *kvmemdup(const void *src, size_t len, gfp_t gfp)
149{
150 void *p;
151
152 p = kvmalloc(len, gfp);
153 if (p)
154 memcpy(p, src, len);
155 return p;
156}
157EXPORT_SYMBOL(kvmemdup);
158
159/**
160 * kmemdup_nul - Create a NUL-terminated string from unterminated data
161 * @s: The data to stringify
162 * @len: The size of the data
163 * @gfp: the GFP mask used in the kmalloc() call when allocating memory
164 *
165 * Return: newly allocated copy of @s with NUL-termination or %NULL in
166 * case of error
167 */
168char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
169{
170 char *buf;
171
172 if (!s)
173 return NULL;
174
175 buf = kmalloc_track_caller(len + 1, gfp);
176 if (buf) {
177 memcpy(buf, s, len);
178 buf[len] = '\0';
179 }
180 return buf;
181}
182EXPORT_SYMBOL(kmemdup_nul);
183
184/**
185 * memdup_user - duplicate memory region from user space
186 *
187 * @src: source address in user space
188 * @len: number of bytes to copy
189 *
190 * Return: an ERR_PTR() on failure. Result is physically
191 * contiguous, to be freed by kfree().
192 */
193void *memdup_user(const void __user *src, size_t len)
194{
195 void *p;
196
197 p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
198 if (!p)
199 return ERR_PTR(-ENOMEM);
200
201 if (copy_from_user(p, src, len)) {
202 kfree(p);
203 return ERR_PTR(-EFAULT);
204 }
205
206 return p;
207}
208EXPORT_SYMBOL(memdup_user);
209
210/**
211 * vmemdup_user - duplicate memory region from user space
212 *
213 * @src: source address in user space
214 * @len: number of bytes to copy
215 *
216 * Return: an ERR_PTR() on failure. Result may be not
217 * physically contiguous. Use kvfree() to free.
218 */
219void *vmemdup_user(const void __user *src, size_t len)
220{
221 void *p;
222
223 p = kvmalloc(len, GFP_USER);
224 if (!p)
225 return ERR_PTR(-ENOMEM);
226
227 if (copy_from_user(p, src, len)) {
228 kvfree(p);
229 return ERR_PTR(-EFAULT);
230 }
231
232 return p;
233}
234EXPORT_SYMBOL(vmemdup_user);
235
236/**
237 * strndup_user - duplicate an existing string from user space
238 * @s: The string to duplicate
239 * @n: Maximum number of bytes to copy, including the trailing NUL.
240 *
241 * Return: newly allocated copy of @s or an ERR_PTR() in case of error
242 */
243char *strndup_user(const char __user *s, long n)
244{
245 char *p;
246 long length;
247
248 length = strnlen_user(s, n);
249
250 if (!length)
251 return ERR_PTR(-EFAULT);
252
253 if (length > n)
254 return ERR_PTR(-EINVAL);
255
256 p = memdup_user(s, length);
257
258 if (IS_ERR(p))
259 return p;
260
261 p[length - 1] = '\0';
262
263 return p;
264}
265EXPORT_SYMBOL(strndup_user);
266
267/**
268 * memdup_user_nul - duplicate memory region from user space and NUL-terminate
269 *
270 * @src: source address in user space
271 * @len: number of bytes to copy
272 *
273 * Return: an ERR_PTR() on failure.
274 */
275void *memdup_user_nul(const void __user *src, size_t len)
276{
277 char *p;
278
279 /*
280 * Always use GFP_KERNEL, since copy_from_user() can sleep and
281 * cause pagefault, which makes it pointless to use GFP_NOFS
282 * or GFP_ATOMIC.
283 */
284 p = kmalloc_track_caller(len + 1, GFP_KERNEL);
285 if (!p)
286 return ERR_PTR(-ENOMEM);
287
288 if (copy_from_user(p, src, len)) {
289 kfree(p);
290 return ERR_PTR(-EFAULT);
291 }
292 p[len] = '\0';
293
294 return p;
295}
296EXPORT_SYMBOL(memdup_user_nul);
297
298/* Check if the vma is being used as a stack by this task */
299int vma_is_stack_for_current(struct vm_area_struct *vma)
300{
301 struct task_struct * __maybe_unused t = current;
302
303 return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
304}
305
306/*
307 * Change backing file, only valid to use during initial VMA setup.
308 */
309void vma_set_file(struct vm_area_struct *vma, struct file *file)
310{
311 /* Changing an anonymous vma with this is illegal */
312 get_file(file);
313 swap(vma->vm_file, file);
314 fput(file);
315}
316EXPORT_SYMBOL(vma_set_file);
317
318#ifndef STACK_RND_MASK
319#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
320#endif
321
322unsigned long randomize_stack_top(unsigned long stack_top)
323{
324 unsigned long random_variable = 0;
325
326 if (current->flags & PF_RANDOMIZE) {
327 random_variable = get_random_long();
328 random_variable &= STACK_RND_MASK;
329 random_variable <<= PAGE_SHIFT;
330 }
331#ifdef CONFIG_STACK_GROWSUP
332 return PAGE_ALIGN(stack_top) + random_variable;
333#else
334 return PAGE_ALIGN(stack_top) - random_variable;
335#endif
336}
337
338/**
339 * randomize_page - Generate a random, page aligned address
340 * @start: The smallest acceptable address the caller will take.
341 * @range: The size of the area, starting at @start, within which the
342 * random address must fall.
343 *
344 * If @start + @range would overflow, @range is capped.
345 *
346 * NOTE: Historical use of randomize_range, which this replaces, presumed that
347 * @start was already page aligned. We now align it regardless.
348 *
349 * Return: A page aligned address within [start, start + range). On error,
350 * @start is returned.
351 */
352unsigned long randomize_page(unsigned long start, unsigned long range)
353{
354 if (!PAGE_ALIGNED(start)) {
355 range -= PAGE_ALIGN(start) - start;
356 start = PAGE_ALIGN(start);
357 }
358
359 if (start > ULONG_MAX - range)
360 range = ULONG_MAX - start;
361
362 range >>= PAGE_SHIFT;
363
364 if (range == 0)
365 return start;
366
367 return start + (get_random_long() % range << PAGE_SHIFT);
368}
369
370#ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
371unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
372{
373 /* Is the current task 32bit ? */
374 if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
375 return randomize_page(mm->brk, SZ_32M);
376
377 return randomize_page(mm->brk, SZ_1G);
378}
379
380unsigned long arch_mmap_rnd(void)
381{
382 unsigned long rnd;
383
384#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
385 if (is_compat_task())
386 rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
387 else
388#endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
389 rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
390
391 return rnd << PAGE_SHIFT;
392}
393
394static int mmap_is_legacy(struct rlimit *rlim_stack)
395{
396 if (current->personality & ADDR_COMPAT_LAYOUT)
397 return 1;
398
399 /* On parisc the stack always grows up - so a unlimited stack should
400 * not be an indicator to use the legacy memory layout. */
401 if (rlim_stack->rlim_cur == RLIM_INFINITY &&
402 !IS_ENABLED(CONFIG_STACK_GROWSUP))
403 return 1;
404
405 return sysctl_legacy_va_layout;
406}
407
408/*
409 * Leave enough space between the mmap area and the stack to honour ulimit in
410 * the face of randomisation.
411 */
412#define MIN_GAP (SZ_128M)
413#define MAX_GAP (STACK_TOP / 6 * 5)
414
415static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
416{
417#ifdef CONFIG_STACK_GROWSUP
418 /*
419 * For an upwards growing stack the calculation is much simpler.
420 * Memory for the maximum stack size is reserved at the top of the
421 * task. mmap_base starts directly below the stack and grows
422 * downwards.
423 */
424 return PAGE_ALIGN_DOWN(mmap_upper_limit(rlim_stack) - rnd);
425#else
426 unsigned long gap = rlim_stack->rlim_cur;
427 unsigned long pad = stack_guard_gap;
428
429 /* Account for stack randomization if necessary */
430 if (current->flags & PF_RANDOMIZE)
431 pad += (STACK_RND_MASK << PAGE_SHIFT);
432
433 /* Values close to RLIM_INFINITY can overflow. */
434 if (gap + pad > gap)
435 gap += pad;
436
437 if (gap < MIN_GAP)
438 gap = MIN_GAP;
439 else if (gap > MAX_GAP)
440 gap = MAX_GAP;
441
442 return PAGE_ALIGN(STACK_TOP - gap - rnd);
443#endif
444}
445
446void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
447{
448 unsigned long random_factor = 0UL;
449
450 if (current->flags & PF_RANDOMIZE)
451 random_factor = arch_mmap_rnd();
452
453 if (mmap_is_legacy(rlim_stack)) {
454 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
455 mm->get_unmapped_area = arch_get_unmapped_area;
456 } else {
457 mm->mmap_base = mmap_base(random_factor, rlim_stack);
458 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
459 }
460}
461#elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
462void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
463{
464 mm->mmap_base = TASK_UNMAPPED_BASE;
465 mm->get_unmapped_area = arch_get_unmapped_area;
466}
467#endif
468
469/**
470 * __account_locked_vm - account locked pages to an mm's locked_vm
471 * @mm: mm to account against
472 * @pages: number of pages to account
473 * @inc: %true if @pages should be considered positive, %false if not
474 * @task: task used to check RLIMIT_MEMLOCK
475 * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
476 *
477 * Assumes @task and @mm are valid (i.e. at least one reference on each), and
478 * that mmap_lock is held as writer.
479 *
480 * Return:
481 * * 0 on success
482 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
483 */
484int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
485 struct task_struct *task, bool bypass_rlim)
486{
487 unsigned long locked_vm, limit;
488 int ret = 0;
489
490 mmap_assert_write_locked(mm);
491
492 locked_vm = mm->locked_vm;
493 if (inc) {
494 if (!bypass_rlim) {
495 limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
496 if (locked_vm + pages > limit)
497 ret = -ENOMEM;
498 }
499 if (!ret)
500 mm->locked_vm = locked_vm + pages;
501 } else {
502 WARN_ON_ONCE(pages > locked_vm);
503 mm->locked_vm = locked_vm - pages;
504 }
505
506 pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
507 (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
508 locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
509 ret ? " - exceeded" : "");
510
511 return ret;
512}
513EXPORT_SYMBOL_GPL(__account_locked_vm);
514
515/**
516 * account_locked_vm - account locked pages to an mm's locked_vm
517 * @mm: mm to account against, may be NULL
518 * @pages: number of pages to account
519 * @inc: %true if @pages should be considered positive, %false if not
520 *
521 * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
522 *
523 * Return:
524 * * 0 on success, or if mm is NULL
525 * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
526 */
527int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
528{
529 int ret;
530
531 if (pages == 0 || !mm)
532 return 0;
533
534 mmap_write_lock(mm);
535 ret = __account_locked_vm(mm, pages, inc, current,
536 capable(CAP_IPC_LOCK));
537 mmap_write_unlock(mm);
538
539 return ret;
540}
541EXPORT_SYMBOL_GPL(account_locked_vm);
542
543unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
544 unsigned long len, unsigned long prot,
545 unsigned long flag, unsigned long pgoff)
546{
547 unsigned long ret;
548 struct mm_struct *mm = current->mm;
549 unsigned long populate;
550 LIST_HEAD(uf);
551
552 ret = security_mmap_file(file, prot, flag);
553 if (!ret) {
554 if (mmap_write_lock_killable(mm))
555 return -EINTR;
556 ret = do_mmap(file, addr, len, prot, flag, 0, pgoff, &populate,
557 &uf);
558 mmap_write_unlock(mm);
559 userfaultfd_unmap_complete(mm, &uf);
560 if (populate)
561 mm_populate(ret, populate);
562 }
563 return ret;
564}
565
566unsigned long vm_mmap(struct file *file, unsigned long addr,
567 unsigned long len, unsigned long prot,
568 unsigned long flag, unsigned long offset)
569{
570 if (unlikely(offset + PAGE_ALIGN(len) < offset))
571 return -EINVAL;
572 if (unlikely(offset_in_page(offset)))
573 return -EINVAL;
574
575 return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
576}
577EXPORT_SYMBOL(vm_mmap);
578
579/**
580 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
581 * failure, fall back to non-contiguous (vmalloc) allocation.
582 * @size: size of the request.
583 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
584 * @node: numa node to allocate from
585 *
586 * Uses kmalloc to get the memory but if the allocation fails then falls back
587 * to the vmalloc allocator. Use kvfree for freeing the memory.
588 *
589 * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
590 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
591 * preferable to the vmalloc fallback, due to visible performance drawbacks.
592 *
593 * Return: pointer to the allocated memory of %NULL in case of failure
594 */
595void *kvmalloc_node(size_t size, gfp_t flags, int node)
596{
597 gfp_t kmalloc_flags = flags;
598 void *ret;
599
600 /*
601 * We want to attempt a large physically contiguous block first because
602 * it is less likely to fragment multiple larger blocks and therefore
603 * contribute to a long term fragmentation less than vmalloc fallback.
604 * However make sure that larger requests are not too disruptive - no
605 * OOM killer and no allocation failure warnings as we have a fallback.
606 */
607 if (size > PAGE_SIZE) {
608 kmalloc_flags |= __GFP_NOWARN;
609
610 if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
611 kmalloc_flags |= __GFP_NORETRY;
612
613 /* nofail semantic is implemented by the vmalloc fallback */
614 kmalloc_flags &= ~__GFP_NOFAIL;
615 }
616
617 ret = kmalloc_node(size, kmalloc_flags, node);
618
619 /*
620 * It doesn't really make sense to fallback to vmalloc for sub page
621 * requests
622 */
623 if (ret || size <= PAGE_SIZE)
624 return ret;
625
626 /* non-sleeping allocations are not supported by vmalloc */
627 if (!gfpflags_allow_blocking(flags))
628 return NULL;
629
630 /* Don't even allow crazy sizes */
631 if (unlikely(size > INT_MAX)) {
632 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
633 return NULL;
634 }
635
636 /*
637 * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
638 * since the callers already cannot assume anything
639 * about the resulting pointer, and cannot play
640 * protection games.
641 */
642 return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
643 flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
644 node, __builtin_return_address(0));
645}
646EXPORT_SYMBOL(kvmalloc_node);
647
648/**
649 * kvfree() - Free memory.
650 * @addr: Pointer to allocated memory.
651 *
652 * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
653 * It is slightly more efficient to use kfree() or vfree() if you are certain
654 * that you know which one to use.
655 *
656 * Context: Either preemptible task context or not-NMI interrupt.
657 */
658void kvfree(const void *addr)
659{
660 if (is_vmalloc_addr(addr))
661 vfree(addr);
662 else
663 kfree(addr);
664}
665EXPORT_SYMBOL(kvfree);
666
667/**
668 * kvfree_sensitive - Free a data object containing sensitive information.
669 * @addr: address of the data object to be freed.
670 * @len: length of the data object.
671 *
672 * Use the special memzero_explicit() function to clear the content of a
673 * kvmalloc'ed object containing sensitive data to make sure that the
674 * compiler won't optimize out the data clearing.
675 */
676void kvfree_sensitive(const void *addr, size_t len)
677{
678 if (likely(!ZERO_OR_NULL_PTR(addr))) {
679 memzero_explicit((void *)addr, len);
680 kvfree(addr);
681 }
682}
683EXPORT_SYMBOL(kvfree_sensitive);
684
685void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
686{
687 void *newp;
688
689 if (oldsize >= newsize)
690 return (void *)p;
691 newp = kvmalloc(newsize, flags);
692 if (!newp)
693 return NULL;
694 memcpy(newp, p, oldsize);
695 kvfree(p);
696 return newp;
697}
698EXPORT_SYMBOL(kvrealloc);
699
700/**
701 * __vmalloc_array - allocate memory for a virtually contiguous array.
702 * @n: number of elements.
703 * @size: element size.
704 * @flags: the type of memory to allocate (see kmalloc).
705 */
706void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
707{
708 size_t bytes;
709
710 if (unlikely(check_mul_overflow(n, size, &bytes)))
711 return NULL;
712 return __vmalloc(bytes, flags);
713}
714EXPORT_SYMBOL(__vmalloc_array);
715
716/**
717 * vmalloc_array - allocate memory for a virtually contiguous array.
718 * @n: number of elements.
719 * @size: element size.
720 */
721void *vmalloc_array(size_t n, size_t size)
722{
723 return __vmalloc_array(n, size, GFP_KERNEL);
724}
725EXPORT_SYMBOL(vmalloc_array);
726
727/**
728 * __vcalloc - allocate and zero memory for a virtually contiguous array.
729 * @n: number of elements.
730 * @size: element size.
731 * @flags: the type of memory to allocate (see kmalloc).
732 */
733void *__vcalloc(size_t n, size_t size, gfp_t flags)
734{
735 return __vmalloc_array(n, size, flags | __GFP_ZERO);
736}
737EXPORT_SYMBOL(__vcalloc);
738
739/**
740 * vcalloc - allocate and zero memory for a virtually contiguous array.
741 * @n: number of elements.
742 * @size: element size.
743 */
744void *vcalloc(size_t n, size_t size)
745{
746 return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
747}
748EXPORT_SYMBOL(vcalloc);
749
750struct anon_vma *folio_anon_vma(struct folio *folio)
751{
752 unsigned long mapping = (unsigned long)folio->mapping;
753
754 if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
755 return NULL;
756 return (void *)(mapping - PAGE_MAPPING_ANON);
757}
758
759/**
760 * folio_mapping - Find the mapping where this folio is stored.
761 * @folio: The folio.
762 *
763 * For folios which are in the page cache, return the mapping that this
764 * page belongs to. Folios in the swap cache return the swap mapping
765 * this page is stored in (which is different from the mapping for the
766 * swap file or swap device where the data is stored).
767 *
768 * You can call this for folios which aren't in the swap cache or page
769 * cache and it will return NULL.
770 */
771struct address_space *folio_mapping(struct folio *folio)
772{
773 struct address_space *mapping;
774
775 /* This happens if someone calls flush_dcache_page on slab page */
776 if (unlikely(folio_test_slab(folio)))
777 return NULL;
778
779 if (unlikely(folio_test_swapcache(folio)))
780 return swap_address_space(folio->swap);
781
782 mapping = folio->mapping;
783 if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
784 return NULL;
785
786 return mapping;
787}
788EXPORT_SYMBOL(folio_mapping);
789
790/**
791 * folio_copy - Copy the contents of one folio to another.
792 * @dst: Folio to copy to.
793 * @src: Folio to copy from.
794 *
795 * The bytes in the folio represented by @src are copied to @dst.
796 * Assumes the caller has validated that @dst is at least as large as @src.
797 * Can be called in atomic context for order-0 folios, but if the folio is
798 * larger, it may sleep.
799 */
800void folio_copy(struct folio *dst, struct folio *src)
801{
802 long i = 0;
803 long nr = folio_nr_pages(src);
804
805 for (;;) {
806 copy_highpage(folio_page(dst, i), folio_page(src, i));
807 if (++i == nr)
808 break;
809 cond_resched();
810 }
811}
812EXPORT_SYMBOL(folio_copy);
813
814int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
815int sysctl_overcommit_ratio __read_mostly = 50;
816unsigned long sysctl_overcommit_kbytes __read_mostly;
817int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
818unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
819unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
820
821int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
822 size_t *lenp, loff_t *ppos)
823{
824 int ret;
825
826 ret = proc_dointvec(table, write, buffer, lenp, ppos);
827 if (ret == 0 && write)
828 sysctl_overcommit_kbytes = 0;
829 return ret;
830}
831
832static void sync_overcommit_as(struct work_struct *dummy)
833{
834 percpu_counter_sync(&vm_committed_as);
835}
836
837int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
838 size_t *lenp, loff_t *ppos)
839{
840 struct ctl_table t;
841 int new_policy = -1;
842 int ret;
843
844 /*
845 * The deviation of sync_overcommit_as could be big with loose policy
846 * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
847 * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
848 * with the strict "NEVER", and to avoid possible race condition (even
849 * though user usually won't too frequently do the switching to policy
850 * OVERCOMMIT_NEVER), the switch is done in the following order:
851 * 1. changing the batch
852 * 2. sync percpu count on each CPU
853 * 3. switch the policy
854 */
855 if (write) {
856 t = *table;
857 t.data = &new_policy;
858 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
859 if (ret || new_policy == -1)
860 return ret;
861
862 mm_compute_batch(new_policy);
863 if (new_policy == OVERCOMMIT_NEVER)
864 schedule_on_each_cpu(sync_overcommit_as);
865 sysctl_overcommit_memory = new_policy;
866 } else {
867 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
868 }
869
870 return ret;
871}
872
873int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
874 size_t *lenp, loff_t *ppos)
875{
876 int ret;
877
878 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
879 if (ret == 0 && write)
880 sysctl_overcommit_ratio = 0;
881 return ret;
882}
883
884/*
885 * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
886 */
887unsigned long vm_commit_limit(void)
888{
889 unsigned long allowed;
890
891 if (sysctl_overcommit_kbytes)
892 allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
893 else
894 allowed = ((totalram_pages() - hugetlb_total_pages())
895 * sysctl_overcommit_ratio / 100);
896 allowed += total_swap_pages;
897
898 return allowed;
899}
900
901/*
902 * Make sure vm_committed_as in one cacheline and not cacheline shared with
903 * other variables. It can be updated by several CPUs frequently.
904 */
905struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
906
907/*
908 * The global memory commitment made in the system can be a metric
909 * that can be used to drive ballooning decisions when Linux is hosted
910 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
911 * balancing memory across competing virtual machines that are hosted.
912 * Several metrics drive this policy engine including the guest reported
913 * memory commitment.
914 *
915 * The time cost of this is very low for small platforms, and for big
916 * platform like a 2S/36C/72T Skylake server, in worst case where
917 * vm_committed_as's spinlock is under severe contention, the time cost
918 * could be about 30~40 microseconds.
919 */
920unsigned long vm_memory_committed(void)
921{
922 return percpu_counter_sum_positive(&vm_committed_as);
923}
924EXPORT_SYMBOL_GPL(vm_memory_committed);
925
926/*
927 * Check that a process has enough memory to allocate a new virtual
928 * mapping. 0 means there is enough memory for the allocation to
929 * succeed and -ENOMEM implies there is not.
930 *
931 * We currently support three overcommit policies, which are set via the
932 * vm.overcommit_memory sysctl. See Documentation/mm/overcommit-accounting.rst
933 *
934 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
935 * Additional code 2002 Jul 20 by Robert Love.
936 *
937 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
938 *
939 * Note this is a helper function intended to be used by LSMs which
940 * wish to use this logic.
941 */
942int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
943{
944 long allowed;
945
946 vm_acct_memory(pages);
947
948 /*
949 * Sometimes we want to use more memory than we have
950 */
951 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
952 return 0;
953
954 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
955 if (pages > totalram_pages() + total_swap_pages)
956 goto error;
957 return 0;
958 }
959
960 allowed = vm_commit_limit();
961 /*
962 * Reserve some for root
963 */
964 if (!cap_sys_admin)
965 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
966
967 /*
968 * Don't let a single process grow so big a user can't recover
969 */
970 if (mm) {
971 long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
972
973 allowed -= min_t(long, mm->total_vm / 32, reserve);
974 }
975
976 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
977 return 0;
978error:
979 pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
980 __func__, current->pid, current->comm);
981 vm_unacct_memory(pages);
982
983 return -ENOMEM;
984}
985
986/**
987 * get_cmdline() - copy the cmdline value to a buffer.
988 * @task: the task whose cmdline value to copy.
989 * @buffer: the buffer to copy to.
990 * @buflen: the length of the buffer. Larger cmdline values are truncated
991 * to this length.
992 *
993 * Return: the size of the cmdline field copied. Note that the copy does
994 * not guarantee an ending NULL byte.
995 */
996int get_cmdline(struct task_struct *task, char *buffer, int buflen)
997{
998 int res = 0;
999 unsigned int len;
1000 struct mm_struct *mm = get_task_mm(task);
1001 unsigned long arg_start, arg_end, env_start, env_end;
1002 if (!mm)
1003 goto out;
1004 if (!mm->arg_end)
1005 goto out_mm; /* Shh! No looking before we're done */
1006
1007 spin_lock(&mm->arg_lock);
1008 arg_start = mm->arg_start;
1009 arg_end = mm->arg_end;
1010 env_start = mm->env_start;
1011 env_end = mm->env_end;
1012 spin_unlock(&mm->arg_lock);
1013
1014 len = arg_end - arg_start;
1015
1016 if (len > buflen)
1017 len = buflen;
1018
1019 res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1020
1021 /*
1022 * If the nul at the end of args has been overwritten, then
1023 * assume application is using setproctitle(3).
1024 */
1025 if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1026 len = strnlen(buffer, res);
1027 if (len < res) {
1028 res = len;
1029 } else {
1030 len = env_end - env_start;
1031 if (len > buflen - res)
1032 len = buflen - res;
1033 res += access_process_vm(task, env_start,
1034 buffer+res, len,
1035 FOLL_FORCE);
1036 res = strnlen(buffer, res);
1037 }
1038 }
1039out_mm:
1040 mmput(mm);
1041out:
1042 return res;
1043}
1044
1045int __weak memcmp_pages(struct page *page1, struct page *page2)
1046{
1047 char *addr1, *addr2;
1048 int ret;
1049
1050 addr1 = kmap_local_page(page1);
1051 addr2 = kmap_local_page(page2);
1052 ret = memcmp(addr1, addr2, PAGE_SIZE);
1053 kunmap_local(addr2);
1054 kunmap_local(addr1);
1055 return ret;
1056}
1057
1058#ifdef CONFIG_PRINTK
1059/**
1060 * mem_dump_obj - Print available provenance information
1061 * @object: object for which to find provenance information.
1062 *
1063 * This function uses pr_cont(), so that the caller is expected to have
1064 * printed out whatever preamble is appropriate. The provenance information
1065 * depends on the type of object and on how much debugging is enabled.
1066 * For example, for a slab-cache object, the slab name is printed, and,
1067 * if available, the return address and stack trace from the allocation
1068 * and last free path of that object.
1069 */
1070void mem_dump_obj(void *object)
1071{
1072 const char *type;
1073
1074 if (kmem_dump_obj(object))
1075 return;
1076
1077 if (vmalloc_dump_obj(object))
1078 return;
1079
1080 if (is_vmalloc_addr(object))
1081 type = "vmalloc memory";
1082 else if (virt_addr_valid(object))
1083 type = "non-slab/vmalloc memory";
1084 else if (object == NULL)
1085 type = "NULL pointer";
1086 else if (object == ZERO_SIZE_PTR)
1087 type = "zero-size pointer";
1088 else
1089 type = "non-paged memory";
1090
1091 pr_cont(" %s\n", type);
1092}
1093EXPORT_SYMBOL_GPL(mem_dump_obj);
1094#endif
1095
1096/*
1097 * A driver might set a page logically offline -- PageOffline() -- and
1098 * turn the page inaccessible in the hypervisor; after that, access to page
1099 * content can be fatal.
1100 *
1101 * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1102 * pages after checking PageOffline(); however, these PFN walkers can race
1103 * with drivers that set PageOffline().
1104 *
1105 * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1106 * synchronize with such drivers, achieving that a page cannot be set
1107 * PageOffline() while frozen.
1108 *
1109 * page_offline_begin()/page_offline_end() is used by drivers that care about
1110 * such races when setting a page PageOffline().
1111 */
1112static DECLARE_RWSEM(page_offline_rwsem);
1113
1114void page_offline_freeze(void)
1115{
1116 down_read(&page_offline_rwsem);
1117}
1118
1119void page_offline_thaw(void)
1120{
1121 up_read(&page_offline_rwsem);
1122}
1123
1124void page_offline_begin(void)
1125{
1126 down_write(&page_offline_rwsem);
1127}
1128EXPORT_SYMBOL(page_offline_begin);
1129
1130void page_offline_end(void)
1131{
1132 up_write(&page_offline_rwsem);
1133}
1134EXPORT_SYMBOL(page_offline_end);
1135
1136#ifndef flush_dcache_folio
1137void flush_dcache_folio(struct folio *folio)
1138{
1139 long i, nr = folio_nr_pages(folio);
1140
1141 for (i = 0; i < nr; i++)
1142 flush_dcache_page(folio_page(folio, i));
1143}
1144EXPORT_SYMBOL(flush_dcache_folio);
1145#endif