Line data Source code
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 : */
38 1500 : void kfree_const(const void *x)
39 : {
40 3000 : if (!is_kernel_rodata((unsigned long)x))
41 899 : kfree(x);
42 1500 : }
43 : EXPORT_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 : */
52 : noinline
53 2345 : char *kstrdup(const char *s, gfp_t gfp)
54 : {
55 : size_t len;
56 : char *buf;
57 :
58 2345 : if (!s)
59 : return NULL;
60 :
61 2343 : len = strlen(s) + 1;
62 2343 : buf = kmalloc_track_caller(len, gfp);
63 2343 : if (buf)
64 2343 : memcpy(buf, s, len);
65 : return buf;
66 : }
67 : EXPORT_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 : */
80 9678 : const char *kstrdup_const(const char *s, gfp_t gfp)
81 : {
82 19356 : if (is_kernel_rodata((unsigned long)s))
83 : return s;
84 :
85 2326 : return kstrdup(s, gfp);
86 : }
87 : EXPORT_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 : */
99 0 : char *kstrndup(const char *s, size_t max, gfp_t gfp)
100 : {
101 : size_t len;
102 : char *buf;
103 :
104 0 : if (!s)
105 : return NULL;
106 :
107 0 : len = strnlen(s, max);
108 0 : buf = kmalloc_track_caller(len+1, gfp);
109 0 : if (buf) {
110 0 : memcpy(buf, s, len);
111 0 : buf[len] = '\0';
112 : }
113 : return buf;
114 : }
115 : EXPORT_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 : */
127 10 : void *kmemdup(const void *src, size_t len, gfp_t gfp)
128 : {
129 : void *p;
130 :
131 10 : p = kmalloc_track_caller(len, gfp);
132 10 : if (p)
133 10 : memcpy(p, src, len);
134 10 : return p;
135 : }
136 : EXPORT_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 : */
148 0 : void *kvmemdup(const void *src, size_t len, gfp_t gfp)
149 : {
150 : void *p;
151 :
152 0 : p = kvmalloc(len, gfp);
153 0 : if (p)
154 0 : memcpy(p, src, len);
155 0 : return p;
156 : }
157 : EXPORT_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 : */
168 32 : char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
169 : {
170 : char *buf;
171 :
172 32 : if (!s)
173 : return NULL;
174 :
175 32 : buf = kmalloc_track_caller(len + 1, gfp);
176 32 : if (buf) {
177 32 : memcpy(buf, s, len);
178 32 : buf[len] = '\0';
179 : }
180 : return buf;
181 : }
182 : EXPORT_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 : */
193 0 : void *memdup_user(const void __user *src, size_t len)
194 : {
195 : void *p;
196 :
197 0 : p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
198 0 : if (!p)
199 : return ERR_PTR(-ENOMEM);
200 :
201 0 : if (copy_from_user(p, src, len)) {
202 0 : kfree(p);
203 0 : return ERR_PTR(-EFAULT);
204 : }
205 :
206 : return p;
207 : }
208 : EXPORT_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 : */
219 0 : void *vmemdup_user(const void __user *src, size_t len)
220 : {
221 : void *p;
222 :
223 0 : p = kvmalloc(len, GFP_USER);
224 0 : if (!p)
225 : return ERR_PTR(-ENOMEM);
226 :
227 0 : if (copy_from_user(p, src, len)) {
228 0 : kvfree(p);
229 0 : return ERR_PTR(-EFAULT);
230 : }
231 :
232 : return p;
233 : }
234 : EXPORT_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 : */
243 0 : char *strndup_user(const char __user *s, long n)
244 : {
245 : char *p;
246 : long length;
247 :
248 0 : length = strnlen_user(s, n);
249 :
250 0 : if (!length)
251 : return ERR_PTR(-EFAULT);
252 :
253 0 : if (length > n)
254 : return ERR_PTR(-EINVAL);
255 :
256 0 : p = memdup_user(s, length);
257 :
258 0 : if (IS_ERR(p))
259 : return p;
260 :
261 0 : p[length - 1] = '\0';
262 :
263 0 : return p;
264 : }
265 : EXPORT_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 : */
275 0 : void *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 0 : p = kmalloc_track_caller(len + 1, GFP_KERNEL);
285 0 : if (!p)
286 : return ERR_PTR(-ENOMEM);
287 :
288 0 : if (copy_from_user(p, src, len)) {
289 0 : kfree(p);
290 0 : return ERR_PTR(-EFAULT);
291 : }
292 0 : p[len] = '\0';
293 :
294 0 : return p;
295 : }
296 : EXPORT_SYMBOL(memdup_user_nul);
297 :
298 : /* Check if the vma is being used as a stack by this task */
299 0 : int vma_is_stack_for_current(struct vm_area_struct *vma)
300 : {
301 0 : struct task_struct * __maybe_unused t = current;
302 :
303 0 : 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 : */
309 0 : void vma_set_file(struct vm_area_struct *vma, struct file *file)
310 : {
311 : /* Changing an anonymous vma with this is illegal */
312 0 : get_file(file);
313 0 : swap(vma->vm_file, file);
314 0 : fput(file);
315 0 : }
316 : EXPORT_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 :
322 0 : unsigned long randomize_stack_top(unsigned long stack_top)
323 : {
324 0 : unsigned long random_variable = 0;
325 :
326 0 : if (current->flags & PF_RANDOMIZE) {
327 0 : random_variable = get_random_long();
328 0 : random_variable &= STACK_RND_MASK;
329 0 : random_variable <<= PAGE_SHIFT;
330 : }
331 : #ifdef CONFIG_STACK_GROWSUP
332 : return PAGE_ALIGN(stack_top) + random_variable;
333 : #else
334 0 : 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 : */
352 0 : unsigned long randomize_page(unsigned long start, unsigned long range)
353 : {
354 0 : if (!PAGE_ALIGNED(start)) {
355 0 : range -= PAGE_ALIGN(start) - start;
356 0 : start = PAGE_ALIGN(start);
357 : }
358 :
359 0 : if (start > ULONG_MAX - range)
360 0 : range = ULONG_MAX - start;
361 :
362 0 : range >>= PAGE_SHIFT;
363 :
364 0 : if (range == 0)
365 : return start;
366 :
367 0 : return start + (get_random_long() % range << PAGE_SHIFT);
368 : }
369 :
370 : #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
371 : unsigned 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 :
380 : unsigned 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 :
394 : static int mmap_is_legacy(struct rlimit *rlim_stack)
395 : {
396 : if (current->personality & ADDR_COMPAT_LAYOUT)
397 : return 1;
398 :
399 : if (rlim_stack->rlim_cur == RLIM_INFINITY)
400 : return 1;
401 :
402 : return sysctl_legacy_va_layout;
403 : }
404 :
405 : /*
406 : * Leave enough space between the mmap area and the stack to honour ulimit in
407 : * the face of randomisation.
408 : */
409 : #define MIN_GAP (SZ_128M)
410 : #define MAX_GAP (STACK_TOP / 6 * 5)
411 :
412 : static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
413 : {
414 : unsigned long gap = rlim_stack->rlim_cur;
415 : unsigned long pad = stack_guard_gap;
416 :
417 : /* Account for stack randomization if necessary */
418 : if (current->flags & PF_RANDOMIZE)
419 : pad += (STACK_RND_MASK << PAGE_SHIFT);
420 :
421 : /* Values close to RLIM_INFINITY can overflow. */
422 : if (gap + pad > gap)
423 : gap += pad;
424 :
425 : if (gap < MIN_GAP)
426 : gap = MIN_GAP;
427 : else if (gap > MAX_GAP)
428 : gap = MAX_GAP;
429 :
430 : return PAGE_ALIGN(STACK_TOP - gap - rnd);
431 : }
432 :
433 : void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
434 : {
435 : unsigned long random_factor = 0UL;
436 :
437 : if (current->flags & PF_RANDOMIZE)
438 : random_factor = arch_mmap_rnd();
439 :
440 : if (mmap_is_legacy(rlim_stack)) {
441 : mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
442 : mm->get_unmapped_area = arch_get_unmapped_area;
443 : } else {
444 : mm->mmap_base = mmap_base(random_factor, rlim_stack);
445 : mm->get_unmapped_area = arch_get_unmapped_area_topdown;
446 : }
447 : }
448 : #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
449 0 : void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
450 : {
451 0 : mm->mmap_base = TASK_UNMAPPED_BASE;
452 0 : mm->get_unmapped_area = arch_get_unmapped_area;
453 0 : }
454 : #endif
455 :
456 : /**
457 : * __account_locked_vm - account locked pages to an mm's locked_vm
458 : * @mm: mm to account against
459 : * @pages: number of pages to account
460 : * @inc: %true if @pages should be considered positive, %false if not
461 : * @task: task used to check RLIMIT_MEMLOCK
462 : * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
463 : *
464 : * Assumes @task and @mm are valid (i.e. at least one reference on each), and
465 : * that mmap_lock is held as writer.
466 : *
467 : * Return:
468 : * * 0 on success
469 : * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
470 : */
471 0 : int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
472 : struct task_struct *task, bool bypass_rlim)
473 : {
474 : unsigned long locked_vm, limit;
475 0 : int ret = 0;
476 :
477 0 : mmap_assert_write_locked(mm);
478 :
479 0 : locked_vm = mm->locked_vm;
480 0 : if (inc) {
481 0 : if (!bypass_rlim) {
482 0 : limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
483 0 : if (locked_vm + pages > limit)
484 0 : ret = -ENOMEM;
485 : }
486 0 : if (!ret)
487 0 : mm->locked_vm = locked_vm + pages;
488 : } else {
489 0 : WARN_ON_ONCE(pages > locked_vm);
490 0 : mm->locked_vm = locked_vm - pages;
491 : }
492 :
493 : pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
494 : (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
495 : locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
496 : ret ? " - exceeded" : "");
497 :
498 0 : return ret;
499 : }
500 : EXPORT_SYMBOL_GPL(__account_locked_vm);
501 :
502 : /**
503 : * account_locked_vm - account locked pages to an mm's locked_vm
504 : * @mm: mm to account against, may be NULL
505 : * @pages: number of pages to account
506 : * @inc: %true if @pages should be considered positive, %false if not
507 : *
508 : * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
509 : *
510 : * Return:
511 : * * 0 on success, or if mm is NULL
512 : * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
513 : */
514 0 : int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
515 : {
516 : int ret;
517 :
518 0 : if (pages == 0 || !mm)
519 : return 0;
520 :
521 0 : mmap_write_lock(mm);
522 0 : ret = __account_locked_vm(mm, pages, inc, current,
523 0 : capable(CAP_IPC_LOCK));
524 0 : mmap_write_unlock(mm);
525 :
526 0 : return ret;
527 : }
528 : EXPORT_SYMBOL_GPL(account_locked_vm);
529 :
530 0 : unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
531 : unsigned long len, unsigned long prot,
532 : unsigned long flag, unsigned long pgoff)
533 : {
534 : unsigned long ret;
535 0 : struct mm_struct *mm = current->mm;
536 : unsigned long populate;
537 0 : LIST_HEAD(uf);
538 :
539 0 : ret = security_mmap_file(file, prot, flag);
540 : if (!ret) {
541 0 : if (mmap_write_lock_killable(mm))
542 : return -EINTR;
543 0 : ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
544 : &uf);
545 0 : mmap_write_unlock(mm);
546 0 : userfaultfd_unmap_complete(mm, &uf);
547 0 : if (populate)
548 0 : mm_populate(ret, populate);
549 : }
550 : return ret;
551 : }
552 :
553 0 : unsigned long vm_mmap(struct file *file, unsigned long addr,
554 : unsigned long len, unsigned long prot,
555 : unsigned long flag, unsigned long offset)
556 : {
557 0 : if (unlikely(offset + PAGE_ALIGN(len) < offset))
558 : return -EINVAL;
559 0 : if (unlikely(offset_in_page(offset)))
560 : return -EINVAL;
561 :
562 0 : return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
563 : }
564 : EXPORT_SYMBOL(vm_mmap);
565 :
566 : /**
567 : * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
568 : * failure, fall back to non-contiguous (vmalloc) allocation.
569 : * @size: size of the request.
570 : * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
571 : * @node: numa node to allocate from
572 : *
573 : * Uses kmalloc to get the memory but if the allocation fails then falls back
574 : * to the vmalloc allocator. Use kvfree for freeing the memory.
575 : *
576 : * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
577 : * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
578 : * preferable to the vmalloc fallback, due to visible performance drawbacks.
579 : *
580 : * Return: pointer to the allocated memory of %NULL in case of failure
581 : */
582 0 : void *kvmalloc_node(size_t size, gfp_t flags, int node)
583 : {
584 0 : gfp_t kmalloc_flags = flags;
585 : void *ret;
586 :
587 : /*
588 : * We want to attempt a large physically contiguous block first because
589 : * it is less likely to fragment multiple larger blocks and therefore
590 : * contribute to a long term fragmentation less than vmalloc fallback.
591 : * However make sure that larger requests are not too disruptive - no
592 : * OOM killer and no allocation failure warnings as we have a fallback.
593 : */
594 0 : if (size > PAGE_SIZE) {
595 0 : kmalloc_flags |= __GFP_NOWARN;
596 :
597 0 : if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
598 0 : kmalloc_flags |= __GFP_NORETRY;
599 :
600 : /* nofail semantic is implemented by the vmalloc fallback */
601 0 : kmalloc_flags &= ~__GFP_NOFAIL;
602 : }
603 :
604 0 : ret = kmalloc_node(size, kmalloc_flags, node);
605 :
606 : /*
607 : * It doesn't really make sense to fallback to vmalloc for sub page
608 : * requests
609 : */
610 0 : if (ret || size <= PAGE_SIZE)
611 : return ret;
612 :
613 : /* non-sleeping allocations are not supported by vmalloc */
614 0 : if (!gfpflags_allow_blocking(flags))
615 : return NULL;
616 :
617 : /* Don't even allow crazy sizes */
618 0 : if (unlikely(size > INT_MAX)) {
619 0 : WARN_ON_ONCE(!(flags & __GFP_NOWARN));
620 : return NULL;
621 : }
622 :
623 : /*
624 : * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
625 : * since the callers already cannot assume anything
626 : * about the resulting pointer, and cannot play
627 : * protection games.
628 : */
629 0 : return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
630 0 : flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
631 0 : node, __builtin_return_address(0));
632 : }
633 : EXPORT_SYMBOL(kvmalloc_node);
634 :
635 : /**
636 : * kvfree() - Free memory.
637 : * @addr: Pointer to allocated memory.
638 : *
639 : * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
640 : * It is slightly more efficient to use kfree() or vfree() if you are certain
641 : * that you know which one to use.
642 : *
643 : * Context: Either preemptible task context or not-NMI interrupt.
644 : */
645 265 : void kvfree(const void *addr)
646 : {
647 265 : if (is_vmalloc_addr(addr))
648 1 : vfree(addr);
649 : else
650 264 : kfree(addr);
651 265 : }
652 : EXPORT_SYMBOL(kvfree);
653 :
654 : /**
655 : * kvfree_sensitive - Free a data object containing sensitive information.
656 : * @addr: address of the data object to be freed.
657 : * @len: length of the data object.
658 : *
659 : * Use the special memzero_explicit() function to clear the content of a
660 : * kvmalloc'ed object containing sensitive data to make sure that the
661 : * compiler won't optimize out the data clearing.
662 : */
663 0 : void kvfree_sensitive(const void *addr, size_t len)
664 : {
665 0 : if (likely(!ZERO_OR_NULL_PTR(addr))) {
666 0 : memzero_explicit((void *)addr, len);
667 0 : kvfree(addr);
668 : }
669 0 : }
670 : EXPORT_SYMBOL(kvfree_sensitive);
671 :
672 0 : void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
673 : {
674 : void *newp;
675 :
676 0 : if (oldsize >= newsize)
677 : return (void *)p;
678 0 : newp = kvmalloc(newsize, flags);
679 0 : if (!newp)
680 : return NULL;
681 0 : memcpy(newp, p, oldsize);
682 0 : kvfree(p);
683 0 : return newp;
684 : }
685 : EXPORT_SYMBOL(kvrealloc);
686 :
687 : /**
688 : * __vmalloc_array - allocate memory for a virtually contiguous array.
689 : * @n: number of elements.
690 : * @size: element size.
691 : * @flags: the type of memory to allocate (see kmalloc).
692 : */
693 0 : void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
694 : {
695 : size_t bytes;
696 :
697 0 : if (unlikely(check_mul_overflow(n, size, &bytes)))
698 : return NULL;
699 0 : return __vmalloc(bytes, flags);
700 : }
701 : EXPORT_SYMBOL(__vmalloc_array);
702 :
703 : /**
704 : * vmalloc_array - allocate memory for a virtually contiguous array.
705 : * @n: number of elements.
706 : * @size: element size.
707 : */
708 0 : void *vmalloc_array(size_t n, size_t size)
709 : {
710 0 : return __vmalloc_array(n, size, GFP_KERNEL);
711 : }
712 : EXPORT_SYMBOL(vmalloc_array);
713 :
714 : /**
715 : * __vcalloc - allocate and zero memory for a virtually contiguous array.
716 : * @n: number of elements.
717 : * @size: element size.
718 : * @flags: the type of memory to allocate (see kmalloc).
719 : */
720 0 : void *__vcalloc(size_t n, size_t size, gfp_t flags)
721 : {
722 0 : return __vmalloc_array(n, size, flags | __GFP_ZERO);
723 : }
724 : EXPORT_SYMBOL(__vcalloc);
725 :
726 : /**
727 : * vcalloc - allocate and zero memory for a virtually contiguous array.
728 : * @n: number of elements.
729 : * @size: element size.
730 : */
731 0 : void *vcalloc(size_t n, size_t size)
732 : {
733 0 : return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
734 : }
735 : EXPORT_SYMBOL(vcalloc);
736 :
737 : /* Neutral page->mapping pointer to address_space or anon_vma or other */
738 0 : void *page_rmapping(struct page *page)
739 : {
740 0 : return folio_raw_mapping(page_folio(page));
741 : }
742 :
743 0 : struct anon_vma *folio_anon_vma(struct folio *folio)
744 : {
745 0 : unsigned long mapping = (unsigned long)folio->mapping;
746 :
747 0 : if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
748 : return NULL;
749 0 : return (void *)(mapping - PAGE_MAPPING_ANON);
750 : }
751 :
752 : /**
753 : * folio_mapping - Find the mapping where this folio is stored.
754 : * @folio: The folio.
755 : *
756 : * For folios which are in the page cache, return the mapping that this
757 : * page belongs to. Folios in the swap cache return the swap mapping
758 : * this page is stored in (which is different from the mapping for the
759 : * swap file or swap device where the data is stored).
760 : *
761 : * You can call this for folios which aren't in the swap cache or page
762 : * cache and it will return NULL.
763 : */
764 0 : struct address_space *folio_mapping(struct folio *folio)
765 : {
766 : struct address_space *mapping;
767 :
768 : /* This happens if someone calls flush_dcache_page on slab page */
769 0 : if (unlikely(folio_test_slab(folio)))
770 : return NULL;
771 :
772 0 : if (unlikely(folio_test_swapcache(folio)))
773 0 : return swap_address_space(folio_swap_entry(folio));
774 :
775 0 : mapping = folio->mapping;
776 0 : if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
777 : return NULL;
778 :
779 0 : return mapping;
780 : }
781 : EXPORT_SYMBOL(folio_mapping);
782 :
783 : /**
784 : * folio_copy - Copy the contents of one folio to another.
785 : * @dst: Folio to copy to.
786 : * @src: Folio to copy from.
787 : *
788 : * The bytes in the folio represented by @src are copied to @dst.
789 : * Assumes the caller has validated that @dst is at least as large as @src.
790 : * Can be called in atomic context for order-0 folios, but if the folio is
791 : * larger, it may sleep.
792 : */
793 0 : void folio_copy(struct folio *dst, struct folio *src)
794 : {
795 0 : long i = 0;
796 0 : long nr = folio_nr_pages(src);
797 :
798 : for (;;) {
799 0 : copy_highpage(folio_page(dst, i), folio_page(src, i));
800 0 : if (++i == nr)
801 : break;
802 0 : cond_resched();
803 : }
804 0 : }
805 :
806 : int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
807 : int sysctl_overcommit_ratio __read_mostly = 50;
808 : unsigned long sysctl_overcommit_kbytes __read_mostly;
809 : int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
810 : unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
811 : unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
812 :
813 0 : int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
814 : size_t *lenp, loff_t *ppos)
815 : {
816 : int ret;
817 :
818 0 : ret = proc_dointvec(table, write, buffer, lenp, ppos);
819 0 : if (ret == 0 && write)
820 0 : sysctl_overcommit_kbytes = 0;
821 0 : return ret;
822 : }
823 :
824 0 : static void sync_overcommit_as(struct work_struct *dummy)
825 : {
826 0 : percpu_counter_sync(&vm_committed_as);
827 0 : }
828 :
829 0 : int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
830 : size_t *lenp, loff_t *ppos)
831 : {
832 : struct ctl_table t;
833 0 : int new_policy = -1;
834 : int ret;
835 :
836 : /*
837 : * The deviation of sync_overcommit_as could be big with loose policy
838 : * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
839 : * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
840 : * with the strict "NEVER", and to avoid possible race condition (even
841 : * though user usually won't too frequently do the switching to policy
842 : * OVERCOMMIT_NEVER), the switch is done in the following order:
843 : * 1. changing the batch
844 : * 2. sync percpu count on each CPU
845 : * 3. switch the policy
846 : */
847 0 : if (write) {
848 0 : t = *table;
849 0 : t.data = &new_policy;
850 0 : ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
851 0 : if (ret || new_policy == -1)
852 : return ret;
853 :
854 0 : mm_compute_batch(new_policy);
855 0 : if (new_policy == OVERCOMMIT_NEVER)
856 0 : schedule_on_each_cpu(sync_overcommit_as);
857 0 : sysctl_overcommit_memory = new_policy;
858 : } else {
859 0 : ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
860 : }
861 :
862 : return ret;
863 : }
864 :
865 0 : int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
866 : size_t *lenp, loff_t *ppos)
867 : {
868 : int ret;
869 :
870 0 : ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
871 0 : if (ret == 0 && write)
872 0 : sysctl_overcommit_ratio = 0;
873 0 : return ret;
874 : }
875 :
876 : /*
877 : * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
878 : */
879 0 : unsigned long vm_commit_limit(void)
880 : {
881 : unsigned long allowed;
882 :
883 0 : if (sysctl_overcommit_kbytes)
884 0 : allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
885 : else
886 0 : allowed = ((totalram_pages() - hugetlb_total_pages())
887 0 : * sysctl_overcommit_ratio / 100);
888 0 : allowed += total_swap_pages;
889 :
890 0 : return allowed;
891 : }
892 :
893 : /*
894 : * Make sure vm_committed_as in one cacheline and not cacheline shared with
895 : * other variables. It can be updated by several CPUs frequently.
896 : */
897 : struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
898 :
899 : /*
900 : * The global memory commitment made in the system can be a metric
901 : * that can be used to drive ballooning decisions when Linux is hosted
902 : * as a guest. On Hyper-V, the host implements a policy engine for dynamically
903 : * balancing memory across competing virtual machines that are hosted.
904 : * Several metrics drive this policy engine including the guest reported
905 : * memory commitment.
906 : *
907 : * The time cost of this is very low for small platforms, and for big
908 : * platform like a 2S/36C/72T Skylake server, in worst case where
909 : * vm_committed_as's spinlock is under severe contention, the time cost
910 : * could be about 30~40 microseconds.
911 : */
912 0 : unsigned long vm_memory_committed(void)
913 : {
914 0 : return percpu_counter_sum_positive(&vm_committed_as);
915 : }
916 : EXPORT_SYMBOL_GPL(vm_memory_committed);
917 :
918 : /*
919 : * Check that a process has enough memory to allocate a new virtual
920 : * mapping. 0 means there is enough memory for the allocation to
921 : * succeed and -ENOMEM implies there is not.
922 : *
923 : * We currently support three overcommit policies, which are set via the
924 : * vm.overcommit_memory sysctl. See Documentation/mm/overcommit-accounting.rst
925 : *
926 : * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
927 : * Additional code 2002 Jul 20 by Robert Love.
928 : *
929 : * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
930 : *
931 : * Note this is a helper function intended to be used by LSMs which
932 : * wish to use this logic.
933 : */
934 0 : int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
935 : {
936 : long allowed;
937 :
938 0 : vm_acct_memory(pages);
939 :
940 : /*
941 : * Sometimes we want to use more memory than we have
942 : */
943 0 : if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
944 : return 0;
945 :
946 0 : if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
947 0 : if (pages > totalram_pages() + total_swap_pages)
948 : goto error;
949 : return 0;
950 : }
951 :
952 0 : allowed = vm_commit_limit();
953 : /*
954 : * Reserve some for root
955 : */
956 0 : if (!cap_sys_admin)
957 0 : allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
958 :
959 : /*
960 : * Don't let a single process grow so big a user can't recover
961 : */
962 0 : if (mm) {
963 0 : long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
964 :
965 0 : allowed -= min_t(long, mm->total_vm / 32, reserve);
966 : }
967 :
968 0 : if (percpu_counter_read_positive(&vm_committed_as) < allowed)
969 : return 0;
970 : error:
971 0 : pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
972 : __func__, current->pid, current->comm);
973 0 : vm_unacct_memory(pages);
974 :
975 0 : return -ENOMEM;
976 : }
977 :
978 : /**
979 : * get_cmdline() - copy the cmdline value to a buffer.
980 : * @task: the task whose cmdline value to copy.
981 : * @buffer: the buffer to copy to.
982 : * @buflen: the length of the buffer. Larger cmdline values are truncated
983 : * to this length.
984 : *
985 : * Return: the size of the cmdline field copied. Note that the copy does
986 : * not guarantee an ending NULL byte.
987 : */
988 0 : int get_cmdline(struct task_struct *task, char *buffer, int buflen)
989 : {
990 0 : int res = 0;
991 : unsigned int len;
992 0 : struct mm_struct *mm = get_task_mm(task);
993 : unsigned long arg_start, arg_end, env_start, env_end;
994 0 : if (!mm)
995 : goto out;
996 0 : if (!mm->arg_end)
997 : goto out_mm; /* Shh! No looking before we're done */
998 :
999 0 : spin_lock(&mm->arg_lock);
1000 0 : arg_start = mm->arg_start;
1001 0 : arg_end = mm->arg_end;
1002 0 : env_start = mm->env_start;
1003 0 : env_end = mm->env_end;
1004 0 : spin_unlock(&mm->arg_lock);
1005 :
1006 0 : len = arg_end - arg_start;
1007 :
1008 0 : if (len > buflen)
1009 0 : len = buflen;
1010 :
1011 0 : res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1012 :
1013 : /*
1014 : * If the nul at the end of args has been overwritten, then
1015 : * assume application is using setproctitle(3).
1016 : */
1017 0 : if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1018 0 : len = strnlen(buffer, res);
1019 0 : if (len < res) {
1020 0 : res = len;
1021 : } else {
1022 0 : len = env_end - env_start;
1023 0 : if (len > buflen - res)
1024 0 : len = buflen - res;
1025 0 : res += access_process_vm(task, env_start,
1026 0 : buffer+res, len,
1027 : FOLL_FORCE);
1028 0 : res = strnlen(buffer, res);
1029 : }
1030 : }
1031 : out_mm:
1032 0 : mmput(mm);
1033 : out:
1034 0 : return res;
1035 : }
1036 :
1037 0 : int __weak memcmp_pages(struct page *page1, struct page *page2)
1038 : {
1039 : char *addr1, *addr2;
1040 : int ret;
1041 :
1042 0 : addr1 = kmap_atomic(page1);
1043 0 : addr2 = kmap_atomic(page2);
1044 0 : ret = memcmp(addr1, addr2, PAGE_SIZE);
1045 0 : kunmap_atomic(addr2);
1046 0 : kunmap_atomic(addr1);
1047 0 : return ret;
1048 : }
1049 :
1050 : #ifdef CONFIG_PRINTK
1051 : /**
1052 : * mem_dump_obj - Print available provenance information
1053 : * @object: object for which to find provenance information.
1054 : *
1055 : * This function uses pr_cont(), so that the caller is expected to have
1056 : * printed out whatever preamble is appropriate. The provenance information
1057 : * depends on the type of object and on how much debugging is enabled.
1058 : * For example, for a slab-cache object, the slab name is printed, and,
1059 : * if available, the return address and stack trace from the allocation
1060 : * and last free path of that object.
1061 : */
1062 0 : void mem_dump_obj(void *object)
1063 : {
1064 : const char *type;
1065 :
1066 0 : if (kmem_valid_obj(object)) {
1067 0 : kmem_dump_obj(object);
1068 0 : return;
1069 : }
1070 :
1071 0 : if (vmalloc_dump_obj(object))
1072 : return;
1073 :
1074 0 : if (virt_addr_valid(object))
1075 : type = "non-slab/vmalloc memory";
1076 0 : else if (object == NULL)
1077 : type = "NULL pointer";
1078 0 : else if (object == ZERO_SIZE_PTR)
1079 : type = "zero-size pointer";
1080 : else
1081 0 : type = "non-paged memory";
1082 :
1083 0 : pr_cont(" %s\n", type);
1084 : }
1085 : EXPORT_SYMBOL_GPL(mem_dump_obj);
1086 : #endif
1087 :
1088 : /*
1089 : * A driver might set a page logically offline -- PageOffline() -- and
1090 : * turn the page inaccessible in the hypervisor; after that, access to page
1091 : * content can be fatal.
1092 : *
1093 : * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1094 : * pages after checking PageOffline(); however, these PFN walkers can race
1095 : * with drivers that set PageOffline().
1096 : *
1097 : * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1098 : * synchronize with such drivers, achieving that a page cannot be set
1099 : * PageOffline() while frozen.
1100 : *
1101 : * page_offline_begin()/page_offline_end() is used by drivers that care about
1102 : * such races when setting a page PageOffline().
1103 : */
1104 : static DECLARE_RWSEM(page_offline_rwsem);
1105 :
1106 0 : void page_offline_freeze(void)
1107 : {
1108 0 : down_read(&page_offline_rwsem);
1109 0 : }
1110 :
1111 0 : void page_offline_thaw(void)
1112 : {
1113 0 : up_read(&page_offline_rwsem);
1114 0 : }
1115 :
1116 0 : void page_offline_begin(void)
1117 : {
1118 0 : down_write(&page_offline_rwsem);
1119 0 : }
1120 : EXPORT_SYMBOL(page_offline_begin);
1121 :
1122 0 : void page_offline_end(void)
1123 : {
1124 0 : up_write(&page_offline_rwsem);
1125 0 : }
1126 : EXPORT_SYMBOL(page_offline_end);
1127 :
1128 : #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_FOLIO
1129 : void flush_dcache_folio(struct folio *folio)
1130 : {
1131 : long i, nr = folio_nr_pages(folio);
1132 :
1133 : for (i = 0; i < nr; i++)
1134 : flush_dcache_page(folio_page(folio, i));
1135 : }
1136 : EXPORT_SYMBOL(flush_dcache_folio);
1137 : #endif
|
