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 1349 : void kfree_const(const void *x)
39 : {
40 2698 : if (!is_kernel_rodata((unsigned long)x))
41 853 : kfree(x);
42 1349 : }
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 2320 : char *kstrdup(const char *s, gfp_t gfp)
53 : {
54 : size_t len;
55 : char *buf;
56 :
57 2320 : if (!s)
58 : return NULL;
59 :
60 2318 : len = strlen(s) + 1;
61 2318 : buf = kmalloc_track_caller(len, gfp);
62 2318 : if (buf)
63 2318 : memcpy(buf, s, len);
64 : return buf;
65 : }
66 : EXPORT_SYMBOL(kstrdup);
67 :
68 : /**
69 : * kstrdup_const - conditionally duplicate an existing const string
70 : * @s: the string to duplicate
71 : * @gfp: the GFP mask used in the kmalloc() call when allocating memory
72 : *
73 : * Note: Strings allocated by kstrdup_const should be freed by kfree_const and
74 : * must not be passed to krealloc().
75 : *
76 : * Return: source string if it is in .rodata section otherwise
77 : * fallback to kstrdup.
78 : */
79 9520 : const char *kstrdup_const(const char *s, gfp_t gfp)
80 : {
81 19040 : if (is_kernel_rodata((unsigned long)s))
82 : return s;
83 :
84 2301 : return kstrdup(s, gfp);
85 : }
86 : EXPORT_SYMBOL(kstrdup_const);
87 :
88 : /**
89 : * kstrndup - allocate space for and copy an existing string
90 : * @s: the string to duplicate
91 : * @max: read at most @max chars from @s
92 : * @gfp: the GFP mask used in the kmalloc() call when allocating memory
93 : *
94 : * Note: Use kmemdup_nul() instead if the size is known exactly.
95 : *
96 : * Return: newly allocated copy of @s or %NULL in case of error
97 : */
98 0 : char *kstrndup(const char *s, size_t max, gfp_t gfp)
99 : {
100 : size_t len;
101 : char *buf;
102 :
103 0 : if (!s)
104 : return NULL;
105 :
106 0 : len = strnlen(s, max);
107 0 : buf = kmalloc_track_caller(len+1, gfp);
108 0 : if (buf) {
109 0 : memcpy(buf, s, len);
110 0 : buf[len] = '\0';
111 : }
112 : return buf;
113 : }
114 : EXPORT_SYMBOL(kstrndup);
115 :
116 : /**
117 : * kmemdup - duplicate region of memory
118 : *
119 : * @src: memory region to duplicate
120 : * @len: memory region length
121 : * @gfp: GFP mask to use
122 : *
123 : * Return: newly allocated copy of @src or %NULL in case of error,
124 : * result is physically contiguous. Use kfree() to free.
125 : */
126 10 : void *kmemdup(const void *src, size_t len, gfp_t gfp)
127 : {
128 : void *p;
129 :
130 10 : p = kmalloc_track_caller(len, gfp);
131 10 : if (p)
132 10 : memcpy(p, src, len);
133 10 : return p;
134 : }
135 : EXPORT_SYMBOL(kmemdup);
136 :
137 : /**
138 : * kvmemdup - duplicate region of memory
139 : *
140 : * @src: memory region to duplicate
141 : * @len: memory region length
142 : * @gfp: GFP mask to use
143 : *
144 : * Return: newly allocated copy of @src or %NULL in case of error,
145 : * result may be not physically contiguous. Use kvfree() to free.
146 : */
147 0 : void *kvmemdup(const void *src, size_t len, gfp_t gfp)
148 : {
149 : void *p;
150 :
151 0 : p = kvmalloc(len, gfp);
152 0 : if (p)
153 0 : memcpy(p, src, len);
154 0 : return p;
155 : }
156 : EXPORT_SYMBOL(kvmemdup);
157 :
158 : /**
159 : * kmemdup_nul - Create a NUL-terminated string from unterminated data
160 : * @s: The data to stringify
161 : * @len: The size of the data
162 : * @gfp: the GFP mask used in the kmalloc() call when allocating memory
163 : *
164 : * Return: newly allocated copy of @s with NUL-termination or %NULL in
165 : * case of error
166 : */
167 27 : char *kmemdup_nul(const char *s, size_t len, gfp_t gfp)
168 : {
169 : char *buf;
170 :
171 27 : if (!s)
172 : return NULL;
173 :
174 27 : buf = kmalloc_track_caller(len + 1, gfp);
175 27 : if (buf) {
176 27 : memcpy(buf, s, len);
177 27 : buf[len] = '\0';
178 : }
179 : return buf;
180 : }
181 : EXPORT_SYMBOL(kmemdup_nul);
182 :
183 : /**
184 : * memdup_user - duplicate memory region from user space
185 : *
186 : * @src: source address in user space
187 : * @len: number of bytes to copy
188 : *
189 : * Return: an ERR_PTR() on failure. Result is physically
190 : * contiguous, to be freed by kfree().
191 : */
192 0 : void *memdup_user(const void __user *src, size_t len)
193 : {
194 : void *p;
195 :
196 0 : p = kmalloc_track_caller(len, GFP_USER | __GFP_NOWARN);
197 0 : if (!p)
198 : return ERR_PTR(-ENOMEM);
199 :
200 0 : if (copy_from_user(p, src, len)) {
201 0 : kfree(p);
202 0 : return ERR_PTR(-EFAULT);
203 : }
204 :
205 : return p;
206 : }
207 : EXPORT_SYMBOL(memdup_user);
208 :
209 : /**
210 : * vmemdup_user - duplicate memory region from user space
211 : *
212 : * @src: source address in user space
213 : * @len: number of bytes to copy
214 : *
215 : * Return: an ERR_PTR() on failure. Result may be not
216 : * physically contiguous. Use kvfree() to free.
217 : */
218 0 : void *vmemdup_user(const void __user *src, size_t len)
219 : {
220 : void *p;
221 :
222 0 : p = kvmalloc(len, GFP_USER);
223 0 : if (!p)
224 : return ERR_PTR(-ENOMEM);
225 :
226 0 : if (copy_from_user(p, src, len)) {
227 0 : kvfree(p);
228 0 : return ERR_PTR(-EFAULT);
229 : }
230 :
231 : return p;
232 : }
233 : EXPORT_SYMBOL(vmemdup_user);
234 :
235 : /**
236 : * strndup_user - duplicate an existing string from user space
237 : * @s: The string to duplicate
238 : * @n: Maximum number of bytes to copy, including the trailing NUL.
239 : *
240 : * Return: newly allocated copy of @s or an ERR_PTR() in case of error
241 : */
242 0 : char *strndup_user(const char __user *s, long n)
243 : {
244 : char *p;
245 : long length;
246 :
247 0 : length = strnlen_user(s, n);
248 :
249 0 : if (!length)
250 : return ERR_PTR(-EFAULT);
251 :
252 0 : if (length > n)
253 : return ERR_PTR(-EINVAL);
254 :
255 0 : p = memdup_user(s, length);
256 :
257 0 : if (IS_ERR(p))
258 : return p;
259 :
260 0 : p[length - 1] = '\0';
261 :
262 0 : return p;
263 : }
264 : EXPORT_SYMBOL(strndup_user);
265 :
266 : /**
267 : * memdup_user_nul - duplicate memory region from user space and NUL-terminate
268 : *
269 : * @src: source address in user space
270 : * @len: number of bytes to copy
271 : *
272 : * Return: an ERR_PTR() on failure.
273 : */
274 0 : void *memdup_user_nul(const void __user *src, size_t len)
275 : {
276 : char *p;
277 :
278 : /*
279 : * Always use GFP_KERNEL, since copy_from_user() can sleep and
280 : * cause pagefault, which makes it pointless to use GFP_NOFS
281 : * or GFP_ATOMIC.
282 : */
283 0 : p = kmalloc_track_caller(len + 1, GFP_KERNEL);
284 0 : if (!p)
285 : return ERR_PTR(-ENOMEM);
286 :
287 0 : if (copy_from_user(p, src, len)) {
288 0 : kfree(p);
289 0 : return ERR_PTR(-EFAULT);
290 : }
291 0 : p[len] = '\0';
292 :
293 0 : return p;
294 : }
295 : EXPORT_SYMBOL(memdup_user_nul);
296 :
297 : /* Check if the vma is being used as a stack by this task */
298 0 : int vma_is_stack_for_current(struct vm_area_struct *vma)
299 : {
300 0 : struct task_struct * __maybe_unused t = current;
301 :
302 0 : return (vma->vm_start <= KSTK_ESP(t) && vma->vm_end >= KSTK_ESP(t));
303 : }
304 :
305 : /*
306 : * Change backing file, only valid to use during initial VMA setup.
307 : */
308 0 : void vma_set_file(struct vm_area_struct *vma, struct file *file)
309 : {
310 : /* Changing an anonymous vma with this is illegal */
311 0 : get_file(file);
312 0 : swap(vma->vm_file, file);
313 0 : fput(file);
314 0 : }
315 : EXPORT_SYMBOL(vma_set_file);
316 :
317 : #ifndef STACK_RND_MASK
318 : #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
319 : #endif
320 :
321 0 : unsigned long randomize_stack_top(unsigned long stack_top)
322 : {
323 0 : unsigned long random_variable = 0;
324 :
325 0 : if (current->flags & PF_RANDOMIZE) {
326 0 : random_variable = get_random_long();
327 0 : random_variable &= STACK_RND_MASK;
328 0 : random_variable <<= PAGE_SHIFT;
329 : }
330 : #ifdef CONFIG_STACK_GROWSUP
331 : return PAGE_ALIGN(stack_top) + random_variable;
332 : #else
333 0 : return PAGE_ALIGN(stack_top) - random_variable;
334 : #endif
335 : }
336 :
337 : /**
338 : * randomize_page - Generate a random, page aligned address
339 : * @start: The smallest acceptable address the caller will take.
340 : * @range: The size of the area, starting at @start, within which the
341 : * random address must fall.
342 : *
343 : * If @start + @range would overflow, @range is capped.
344 : *
345 : * NOTE: Historical use of randomize_range, which this replaces, presumed that
346 : * @start was already page aligned. We now align it regardless.
347 : *
348 : * Return: A page aligned address within [start, start + range). On error,
349 : * @start is returned.
350 : */
351 0 : unsigned long randomize_page(unsigned long start, unsigned long range)
352 : {
353 0 : if (!PAGE_ALIGNED(start)) {
354 0 : range -= PAGE_ALIGN(start) - start;
355 0 : start = PAGE_ALIGN(start);
356 : }
357 :
358 0 : if (start > ULONG_MAX - range)
359 0 : range = ULONG_MAX - start;
360 :
361 0 : range >>= PAGE_SHIFT;
362 :
363 0 : if (range == 0)
364 : return start;
365 :
366 0 : return start + (get_random_long() % range << PAGE_SHIFT);
367 : }
368 :
369 : #ifdef CONFIG_ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT
370 : unsigned long __weak arch_randomize_brk(struct mm_struct *mm)
371 : {
372 : /* Is the current task 32bit ? */
373 : if (!IS_ENABLED(CONFIG_64BIT) || is_compat_task())
374 : return randomize_page(mm->brk, SZ_32M);
375 :
376 : return randomize_page(mm->brk, SZ_1G);
377 : }
378 :
379 : unsigned long arch_mmap_rnd(void)
380 : {
381 : unsigned long rnd;
382 :
383 : #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
384 : if (is_compat_task())
385 : rnd = get_random_long() & ((1UL << mmap_rnd_compat_bits) - 1);
386 : else
387 : #endif /* CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS */
388 : rnd = get_random_long() & ((1UL << mmap_rnd_bits) - 1);
389 :
390 : return rnd << PAGE_SHIFT;
391 : }
392 :
393 : static int mmap_is_legacy(struct rlimit *rlim_stack)
394 : {
395 : if (current->personality & ADDR_COMPAT_LAYOUT)
396 : return 1;
397 :
398 : if (rlim_stack->rlim_cur == RLIM_INFINITY)
399 : return 1;
400 :
401 : return sysctl_legacy_va_layout;
402 : }
403 :
404 : /*
405 : * Leave enough space between the mmap area and the stack to honour ulimit in
406 : * the face of randomisation.
407 : */
408 : #define MIN_GAP (SZ_128M)
409 : #define MAX_GAP (STACK_TOP / 6 * 5)
410 :
411 : static unsigned long mmap_base(unsigned long rnd, struct rlimit *rlim_stack)
412 : {
413 : unsigned long gap = rlim_stack->rlim_cur;
414 : unsigned long pad = stack_guard_gap;
415 :
416 : /* Account for stack randomization if necessary */
417 : if (current->flags & PF_RANDOMIZE)
418 : pad += (STACK_RND_MASK << PAGE_SHIFT);
419 :
420 : /* Values close to RLIM_INFINITY can overflow. */
421 : if (gap + pad > gap)
422 : gap += pad;
423 :
424 : if (gap < MIN_GAP)
425 : gap = MIN_GAP;
426 : else if (gap > MAX_GAP)
427 : gap = MAX_GAP;
428 :
429 : return PAGE_ALIGN(STACK_TOP - gap - rnd);
430 : }
431 :
432 : void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
433 : {
434 : unsigned long random_factor = 0UL;
435 :
436 : if (current->flags & PF_RANDOMIZE)
437 : random_factor = arch_mmap_rnd();
438 :
439 : if (mmap_is_legacy(rlim_stack)) {
440 : mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
441 : mm->get_unmapped_area = arch_get_unmapped_area;
442 : } else {
443 : mm->mmap_base = mmap_base(random_factor, rlim_stack);
444 : mm->get_unmapped_area = arch_get_unmapped_area_topdown;
445 : }
446 : }
447 : #elif defined(CONFIG_MMU) && !defined(HAVE_ARCH_PICK_MMAP_LAYOUT)
448 0 : void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
449 : {
450 0 : mm->mmap_base = TASK_UNMAPPED_BASE;
451 0 : mm->get_unmapped_area = arch_get_unmapped_area;
452 0 : }
453 : #endif
454 :
455 : /**
456 : * __account_locked_vm - account locked pages to an mm's locked_vm
457 : * @mm: mm to account against
458 : * @pages: number of pages to account
459 : * @inc: %true if @pages should be considered positive, %false if not
460 : * @task: task used to check RLIMIT_MEMLOCK
461 : * @bypass_rlim: %true if checking RLIMIT_MEMLOCK should be skipped
462 : *
463 : * Assumes @task and @mm are valid (i.e. at least one reference on each), and
464 : * that mmap_lock is held as writer.
465 : *
466 : * Return:
467 : * * 0 on success
468 : * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
469 : */
470 0 : int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
471 : struct task_struct *task, bool bypass_rlim)
472 : {
473 : unsigned long locked_vm, limit;
474 0 : int ret = 0;
475 :
476 0 : mmap_assert_write_locked(mm);
477 :
478 0 : locked_vm = mm->locked_vm;
479 0 : if (inc) {
480 0 : if (!bypass_rlim) {
481 0 : limit = task_rlimit(task, RLIMIT_MEMLOCK) >> PAGE_SHIFT;
482 0 : if (locked_vm + pages > limit)
483 0 : ret = -ENOMEM;
484 : }
485 0 : if (!ret)
486 0 : mm->locked_vm = locked_vm + pages;
487 : } else {
488 0 : WARN_ON_ONCE(pages > locked_vm);
489 0 : mm->locked_vm = locked_vm - pages;
490 : }
491 :
492 : pr_debug("%s: [%d] caller %ps %c%lu %lu/%lu%s\n", __func__, task->pid,
493 : (void *)_RET_IP_, (inc) ? '+' : '-', pages << PAGE_SHIFT,
494 : locked_vm << PAGE_SHIFT, task_rlimit(task, RLIMIT_MEMLOCK),
495 : ret ? " - exceeded" : "");
496 :
497 0 : return ret;
498 : }
499 : EXPORT_SYMBOL_GPL(__account_locked_vm);
500 :
501 : /**
502 : * account_locked_vm - account locked pages to an mm's locked_vm
503 : * @mm: mm to account against, may be NULL
504 : * @pages: number of pages to account
505 : * @inc: %true if @pages should be considered positive, %false if not
506 : *
507 : * Assumes a non-NULL @mm is valid (i.e. at least one reference on it).
508 : *
509 : * Return:
510 : * * 0 on success, or if mm is NULL
511 : * * -ENOMEM if RLIMIT_MEMLOCK would be exceeded.
512 : */
513 0 : int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc)
514 : {
515 : int ret;
516 :
517 0 : if (pages == 0 || !mm)
518 : return 0;
519 :
520 0 : mmap_write_lock(mm);
521 0 : ret = __account_locked_vm(mm, pages, inc, current,
522 0 : capable(CAP_IPC_LOCK));
523 0 : mmap_write_unlock(mm);
524 :
525 0 : return ret;
526 : }
527 : EXPORT_SYMBOL_GPL(account_locked_vm);
528 :
529 0 : unsigned long vm_mmap_pgoff(struct file *file, unsigned long addr,
530 : unsigned long len, unsigned long prot,
531 : unsigned long flag, unsigned long pgoff)
532 : {
533 : unsigned long ret;
534 0 : struct mm_struct *mm = current->mm;
535 : unsigned long populate;
536 0 : LIST_HEAD(uf);
537 :
538 0 : ret = security_mmap_file(file, prot, flag);
539 : if (!ret) {
540 0 : if (mmap_write_lock_killable(mm))
541 : return -EINTR;
542 0 : ret = do_mmap(file, addr, len, prot, flag, pgoff, &populate,
543 : &uf);
544 0 : mmap_write_unlock(mm);
545 0 : userfaultfd_unmap_complete(mm, &uf);
546 0 : if (populate)
547 0 : mm_populate(ret, populate);
548 : }
549 : return ret;
550 : }
551 :
552 0 : unsigned long vm_mmap(struct file *file, unsigned long addr,
553 : unsigned long len, unsigned long prot,
554 : unsigned long flag, unsigned long offset)
555 : {
556 0 : if (unlikely(offset + PAGE_ALIGN(len) < offset))
557 : return -EINVAL;
558 0 : if (unlikely(offset_in_page(offset)))
559 : return -EINVAL;
560 :
561 0 : return vm_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
562 : }
563 : EXPORT_SYMBOL(vm_mmap);
564 :
565 : /**
566 : * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
567 : * failure, fall back to non-contiguous (vmalloc) allocation.
568 : * @size: size of the request.
569 : * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
570 : * @node: numa node to allocate from
571 : *
572 : * Uses kmalloc to get the memory but if the allocation fails then falls back
573 : * to the vmalloc allocator. Use kvfree for freeing the memory.
574 : *
575 : * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
576 : * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
577 : * preferable to the vmalloc fallback, due to visible performance drawbacks.
578 : *
579 : * Return: pointer to the allocated memory of %NULL in case of failure
580 : */
581 0 : void *kvmalloc_node(size_t size, gfp_t flags, int node)
582 : {
583 0 : gfp_t kmalloc_flags = flags;
584 : void *ret;
585 :
586 : /*
587 : * We want to attempt a large physically contiguous block first because
588 : * it is less likely to fragment multiple larger blocks and therefore
589 : * contribute to a long term fragmentation less than vmalloc fallback.
590 : * However make sure that larger requests are not too disruptive - no
591 : * OOM killer and no allocation failure warnings as we have a fallback.
592 : */
593 0 : if (size > PAGE_SIZE) {
594 0 : kmalloc_flags |= __GFP_NOWARN;
595 :
596 0 : if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
597 0 : kmalloc_flags |= __GFP_NORETRY;
598 :
599 : /* nofail semantic is implemented by the vmalloc fallback */
600 0 : kmalloc_flags &= ~__GFP_NOFAIL;
601 : }
602 :
603 0 : ret = kmalloc_node(size, kmalloc_flags, node);
604 :
605 : /*
606 : * It doesn't really make sense to fallback to vmalloc for sub page
607 : * requests
608 : */
609 0 : if (ret || size <= PAGE_SIZE)
610 : return ret;
611 :
612 : /* non-sleeping allocations are not supported by vmalloc */
613 0 : if (!gfpflags_allow_blocking(flags))
614 : return NULL;
615 :
616 : /* Don't even allow crazy sizes */
617 0 : if (unlikely(size > INT_MAX)) {
618 0 : WARN_ON_ONCE(!(flags & __GFP_NOWARN));
619 : return NULL;
620 : }
621 :
622 : /*
623 : * kvmalloc() can always use VM_ALLOW_HUGE_VMAP,
624 : * since the callers already cannot assume anything
625 : * about the resulting pointer, and cannot play
626 : * protection games.
627 : */
628 0 : return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
629 0 : flags, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
630 0 : node, __builtin_return_address(0));
631 : }
632 : EXPORT_SYMBOL(kvmalloc_node);
633 :
634 : /**
635 : * kvfree() - Free memory.
636 : * @addr: Pointer to allocated memory.
637 : *
638 : * kvfree frees memory allocated by any of vmalloc(), kmalloc() or kvmalloc().
639 : * It is slightly more efficient to use kfree() or vfree() if you are certain
640 : * that you know which one to use.
641 : *
642 : * Context: Either preemptible task context or not-NMI interrupt.
643 : */
644 259 : void kvfree(const void *addr)
645 : {
646 259 : if (is_vmalloc_addr(addr))
647 1 : vfree(addr);
648 : else
649 258 : kfree(addr);
650 259 : }
651 : EXPORT_SYMBOL(kvfree);
652 :
653 : /**
654 : * kvfree_sensitive - Free a data object containing sensitive information.
655 : * @addr: address of the data object to be freed.
656 : * @len: length of the data object.
657 : *
658 : * Use the special memzero_explicit() function to clear the content of a
659 : * kvmalloc'ed object containing sensitive data to make sure that the
660 : * compiler won't optimize out the data clearing.
661 : */
662 0 : void kvfree_sensitive(const void *addr, size_t len)
663 : {
664 0 : if (likely(!ZERO_OR_NULL_PTR(addr))) {
665 0 : memzero_explicit((void *)addr, len);
666 0 : kvfree(addr);
667 : }
668 0 : }
669 : EXPORT_SYMBOL(kvfree_sensitive);
670 :
671 0 : void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
672 : {
673 : void *newp;
674 :
675 0 : if (oldsize >= newsize)
676 : return (void *)p;
677 0 : newp = kvmalloc(newsize, flags);
678 0 : if (!newp)
679 : return NULL;
680 0 : memcpy(newp, p, oldsize);
681 0 : kvfree(p);
682 0 : return newp;
683 : }
684 : EXPORT_SYMBOL(kvrealloc);
685 :
686 : /**
687 : * __vmalloc_array - allocate memory for a virtually contiguous array.
688 : * @n: number of elements.
689 : * @size: element size.
690 : * @flags: the type of memory to allocate (see kmalloc).
691 : */
692 0 : void *__vmalloc_array(size_t n, size_t size, gfp_t flags)
693 : {
694 : size_t bytes;
695 :
696 0 : if (unlikely(check_mul_overflow(n, size, &bytes)))
697 : return NULL;
698 0 : return __vmalloc(bytes, flags);
699 : }
700 : EXPORT_SYMBOL(__vmalloc_array);
701 :
702 : /**
703 : * vmalloc_array - allocate memory for a virtually contiguous array.
704 : * @n: number of elements.
705 : * @size: element size.
706 : */
707 0 : void *vmalloc_array(size_t n, size_t size)
708 : {
709 0 : return __vmalloc_array(n, size, GFP_KERNEL);
710 : }
711 : EXPORT_SYMBOL(vmalloc_array);
712 :
713 : /**
714 : * __vcalloc - allocate and zero memory for a virtually contiguous array.
715 : * @n: number of elements.
716 : * @size: element size.
717 : * @flags: the type of memory to allocate (see kmalloc).
718 : */
719 0 : void *__vcalloc(size_t n, size_t size, gfp_t flags)
720 : {
721 0 : return __vmalloc_array(n, size, flags | __GFP_ZERO);
722 : }
723 : EXPORT_SYMBOL(__vcalloc);
724 :
725 : /**
726 : * vcalloc - allocate and zero memory for a virtually contiguous array.
727 : * @n: number of elements.
728 : * @size: element size.
729 : */
730 0 : void *vcalloc(size_t n, size_t size)
731 : {
732 0 : return __vmalloc_array(n, size, GFP_KERNEL | __GFP_ZERO);
733 : }
734 : EXPORT_SYMBOL(vcalloc);
735 :
736 : /* Neutral page->mapping pointer to address_space or anon_vma or other */
737 0 : void *page_rmapping(struct page *page)
738 : {
739 0 : return folio_raw_mapping(page_folio(page));
740 : }
741 :
742 0 : struct anon_vma *folio_anon_vma(struct folio *folio)
743 : {
744 0 : unsigned long mapping = (unsigned long)folio->mapping;
745 :
746 0 : if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
747 : return NULL;
748 0 : return (void *)(mapping - PAGE_MAPPING_ANON);
749 : }
750 :
751 : /**
752 : * folio_mapping - Find the mapping where this folio is stored.
753 : * @folio: The folio.
754 : *
755 : * For folios which are in the page cache, return the mapping that this
756 : * page belongs to. Folios in the swap cache return the swap mapping
757 : * this page is stored in (which is different from the mapping for the
758 : * swap file or swap device where the data is stored).
759 : *
760 : * You can call this for folios which aren't in the swap cache or page
761 : * cache and it will return NULL.
762 : */
763 0 : struct address_space *folio_mapping(struct folio *folio)
764 : {
765 : struct address_space *mapping;
766 :
767 : /* This happens if someone calls flush_dcache_page on slab page */
768 0 : if (unlikely(folio_test_slab(folio)))
769 : return NULL;
770 :
771 0 : if (unlikely(folio_test_swapcache(folio)))
772 0 : return swap_address_space(folio_swap_entry(folio));
773 :
774 0 : mapping = folio->mapping;
775 0 : if ((unsigned long)mapping & PAGE_MAPPING_FLAGS)
776 : return NULL;
777 :
778 0 : return mapping;
779 : }
780 : EXPORT_SYMBOL(folio_mapping);
781 :
782 : /**
783 : * folio_copy - Copy the contents of one folio to another.
784 : * @dst: Folio to copy to.
785 : * @src: Folio to copy from.
786 : *
787 : * The bytes in the folio represented by @src are copied to @dst.
788 : * Assumes the caller has validated that @dst is at least as large as @src.
789 : * Can be called in atomic context for order-0 folios, but if the folio is
790 : * larger, it may sleep.
791 : */
792 0 : void folio_copy(struct folio *dst, struct folio *src)
793 : {
794 0 : long i = 0;
795 0 : long nr = folio_nr_pages(src);
796 :
797 : for (;;) {
798 0 : copy_highpage(folio_page(dst, i), folio_page(src, i));
799 0 : if (++i == nr)
800 : break;
801 0 : cond_resched();
802 : }
803 0 : }
804 :
805 : int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;
806 : int sysctl_overcommit_ratio __read_mostly = 50;
807 : unsigned long sysctl_overcommit_kbytes __read_mostly;
808 : int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
809 : unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
810 : unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
811 :
812 0 : int overcommit_ratio_handler(struct ctl_table *table, int write, void *buffer,
813 : size_t *lenp, loff_t *ppos)
814 : {
815 : int ret;
816 :
817 0 : ret = proc_dointvec(table, write, buffer, lenp, ppos);
818 0 : if (ret == 0 && write)
819 0 : sysctl_overcommit_kbytes = 0;
820 0 : return ret;
821 : }
822 :
823 0 : static void sync_overcommit_as(struct work_struct *dummy)
824 : {
825 0 : percpu_counter_sync(&vm_committed_as);
826 0 : }
827 :
828 0 : int overcommit_policy_handler(struct ctl_table *table, int write, void *buffer,
829 : size_t *lenp, loff_t *ppos)
830 : {
831 : struct ctl_table t;
832 0 : int new_policy = -1;
833 : int ret;
834 :
835 : /*
836 : * The deviation of sync_overcommit_as could be big with loose policy
837 : * like OVERCOMMIT_ALWAYS/OVERCOMMIT_GUESS. When changing policy to
838 : * strict OVERCOMMIT_NEVER, we need to reduce the deviation to comply
839 : * with the strict "NEVER", and to avoid possible race condition (even
840 : * though user usually won't too frequently do the switching to policy
841 : * OVERCOMMIT_NEVER), the switch is done in the following order:
842 : * 1. changing the batch
843 : * 2. sync percpu count on each CPU
844 : * 3. switch the policy
845 : */
846 0 : if (write) {
847 0 : t = *table;
848 0 : t.data = &new_policy;
849 0 : ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
850 0 : if (ret || new_policy == -1)
851 : return ret;
852 :
853 0 : mm_compute_batch(new_policy);
854 0 : if (new_policy == OVERCOMMIT_NEVER)
855 0 : schedule_on_each_cpu(sync_overcommit_as);
856 0 : sysctl_overcommit_memory = new_policy;
857 : } else {
858 0 : ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
859 : }
860 :
861 : return ret;
862 : }
863 :
864 0 : int overcommit_kbytes_handler(struct ctl_table *table, int write, void *buffer,
865 : size_t *lenp, loff_t *ppos)
866 : {
867 : int ret;
868 :
869 0 : ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
870 0 : if (ret == 0 && write)
871 0 : sysctl_overcommit_ratio = 0;
872 0 : return ret;
873 : }
874 :
875 : /*
876 : * Committed memory limit enforced when OVERCOMMIT_NEVER policy is used
877 : */
878 0 : unsigned long vm_commit_limit(void)
879 : {
880 : unsigned long allowed;
881 :
882 0 : if (sysctl_overcommit_kbytes)
883 0 : allowed = sysctl_overcommit_kbytes >> (PAGE_SHIFT - 10);
884 : else
885 0 : allowed = ((totalram_pages() - hugetlb_total_pages())
886 0 : * sysctl_overcommit_ratio / 100);
887 0 : allowed += total_swap_pages;
888 :
889 0 : return allowed;
890 : }
891 :
892 : /*
893 : * Make sure vm_committed_as in one cacheline and not cacheline shared with
894 : * other variables. It can be updated by several CPUs frequently.
895 : */
896 : struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
897 :
898 : /*
899 : * The global memory commitment made in the system can be a metric
900 : * that can be used to drive ballooning decisions when Linux is hosted
901 : * as a guest. On Hyper-V, the host implements a policy engine for dynamically
902 : * balancing memory across competing virtual machines that are hosted.
903 : * Several metrics drive this policy engine including the guest reported
904 : * memory commitment.
905 : *
906 : * The time cost of this is very low for small platforms, and for big
907 : * platform like a 2S/36C/72T Skylake server, in worst case where
908 : * vm_committed_as's spinlock is under severe contention, the time cost
909 : * could be about 30~40 microseconds.
910 : */
911 0 : unsigned long vm_memory_committed(void)
912 : {
913 0 : return percpu_counter_sum_positive(&vm_committed_as);
914 : }
915 : EXPORT_SYMBOL_GPL(vm_memory_committed);
916 :
917 : /*
918 : * Check that a process has enough memory to allocate a new virtual
919 : * mapping. 0 means there is enough memory for the allocation to
920 : * succeed and -ENOMEM implies there is not.
921 : *
922 : * We currently support three overcommit policies, which are set via the
923 : * vm.overcommit_memory sysctl. See Documentation/mm/overcommit-accounting.rst
924 : *
925 : * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
926 : * Additional code 2002 Jul 20 by Robert Love.
927 : *
928 : * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
929 : *
930 : * Note this is a helper function intended to be used by LSMs which
931 : * wish to use this logic.
932 : */
933 0 : int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
934 : {
935 : long allowed;
936 :
937 0 : vm_acct_memory(pages);
938 :
939 : /*
940 : * Sometimes we want to use more memory than we have
941 : */
942 0 : if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
943 : return 0;
944 :
945 0 : if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
946 0 : if (pages > totalram_pages() + total_swap_pages)
947 : goto error;
948 : return 0;
949 : }
950 :
951 0 : allowed = vm_commit_limit();
952 : /*
953 : * Reserve some for root
954 : */
955 0 : if (!cap_sys_admin)
956 0 : allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
957 :
958 : /*
959 : * Don't let a single process grow so big a user can't recover
960 : */
961 0 : if (mm) {
962 0 : long reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
963 :
964 0 : allowed -= min_t(long, mm->total_vm / 32, reserve);
965 : }
966 :
967 0 : if (percpu_counter_read_positive(&vm_committed_as) < allowed)
968 : return 0;
969 : error:
970 0 : pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
971 : __func__, current->pid, current->comm);
972 0 : vm_unacct_memory(pages);
973 :
974 0 : return -ENOMEM;
975 : }
976 :
977 : /**
978 : * get_cmdline() - copy the cmdline value to a buffer.
979 : * @task: the task whose cmdline value to copy.
980 : * @buffer: the buffer to copy to.
981 : * @buflen: the length of the buffer. Larger cmdline values are truncated
982 : * to this length.
983 : *
984 : * Return: the size of the cmdline field copied. Note that the copy does
985 : * not guarantee an ending NULL byte.
986 : */
987 0 : int get_cmdline(struct task_struct *task, char *buffer, int buflen)
988 : {
989 0 : int res = 0;
990 : unsigned int len;
991 0 : struct mm_struct *mm = get_task_mm(task);
992 : unsigned long arg_start, arg_end, env_start, env_end;
993 0 : if (!mm)
994 : goto out;
995 0 : if (!mm->arg_end)
996 : goto out_mm; /* Shh! No looking before we're done */
997 :
998 0 : spin_lock(&mm->arg_lock);
999 0 : arg_start = mm->arg_start;
1000 0 : arg_end = mm->arg_end;
1001 0 : env_start = mm->env_start;
1002 0 : env_end = mm->env_end;
1003 0 : spin_unlock(&mm->arg_lock);
1004 :
1005 0 : len = arg_end - arg_start;
1006 :
1007 0 : if (len > buflen)
1008 0 : len = buflen;
1009 :
1010 0 : res = access_process_vm(task, arg_start, buffer, len, FOLL_FORCE);
1011 :
1012 : /*
1013 : * If the nul at the end of args has been overwritten, then
1014 : * assume application is using setproctitle(3).
1015 : */
1016 0 : if (res > 0 && buffer[res-1] != '\0' && len < buflen) {
1017 0 : len = strnlen(buffer, res);
1018 0 : if (len < res) {
1019 0 : res = len;
1020 : } else {
1021 0 : len = env_end - env_start;
1022 0 : if (len > buflen - res)
1023 0 : len = buflen - res;
1024 0 : res += access_process_vm(task, env_start,
1025 0 : buffer+res, len,
1026 : FOLL_FORCE);
1027 0 : res = strnlen(buffer, res);
1028 : }
1029 : }
1030 : out_mm:
1031 0 : mmput(mm);
1032 : out:
1033 0 : return res;
1034 : }
1035 :
1036 0 : int __weak memcmp_pages(struct page *page1, struct page *page2)
1037 : {
1038 : char *addr1, *addr2;
1039 : int ret;
1040 :
1041 0 : addr1 = kmap_atomic(page1);
1042 0 : addr2 = kmap_atomic(page2);
1043 0 : ret = memcmp(addr1, addr2, PAGE_SIZE);
1044 0 : kunmap_atomic(addr2);
1045 0 : kunmap_atomic(addr1);
1046 0 : return ret;
1047 : }
1048 :
1049 : #ifdef CONFIG_PRINTK
1050 : /**
1051 : * mem_dump_obj - Print available provenance information
1052 : * @object: object for which to find provenance information.
1053 : *
1054 : * This function uses pr_cont(), so that the caller is expected to have
1055 : * printed out whatever preamble is appropriate. The provenance information
1056 : * depends on the type of object and on how much debugging is enabled.
1057 : * For example, for a slab-cache object, the slab name is printed, and,
1058 : * if available, the return address and stack trace from the allocation
1059 : * and last free path of that object.
1060 : */
1061 0 : void mem_dump_obj(void *object)
1062 : {
1063 : const char *type;
1064 :
1065 0 : if (kmem_valid_obj(object)) {
1066 0 : kmem_dump_obj(object);
1067 0 : return;
1068 : }
1069 :
1070 0 : if (vmalloc_dump_obj(object))
1071 : return;
1072 :
1073 0 : if (virt_addr_valid(object))
1074 : type = "non-slab/vmalloc memory";
1075 0 : else if (object == NULL)
1076 : type = "NULL pointer";
1077 0 : else if (object == ZERO_SIZE_PTR)
1078 : type = "zero-size pointer";
1079 : else
1080 0 : type = "non-paged memory";
1081 :
1082 0 : pr_cont(" %s\n", type);
1083 : }
1084 : EXPORT_SYMBOL_GPL(mem_dump_obj);
1085 : #endif
1086 :
1087 : /*
1088 : * A driver might set a page logically offline -- PageOffline() -- and
1089 : * turn the page inaccessible in the hypervisor; after that, access to page
1090 : * content can be fatal.
1091 : *
1092 : * Some special PFN walkers -- i.e., /proc/kcore -- read content of random
1093 : * pages after checking PageOffline(); however, these PFN walkers can race
1094 : * with drivers that set PageOffline().
1095 : *
1096 : * page_offline_freeze()/page_offline_thaw() allows for a subsystem to
1097 : * synchronize with such drivers, achieving that a page cannot be set
1098 : * PageOffline() while frozen.
1099 : *
1100 : * page_offline_begin()/page_offline_end() is used by drivers that care about
1101 : * such races when setting a page PageOffline().
1102 : */
1103 : static DECLARE_RWSEM(page_offline_rwsem);
1104 :
1105 0 : void page_offline_freeze(void)
1106 : {
1107 0 : down_read(&page_offline_rwsem);
1108 0 : }
1109 :
1110 0 : void page_offline_thaw(void)
1111 : {
1112 0 : up_read(&page_offline_rwsem);
1113 0 : }
1114 :
1115 0 : void page_offline_begin(void)
1116 : {
1117 0 : down_write(&page_offline_rwsem);
1118 0 : }
1119 : EXPORT_SYMBOL(page_offline_begin);
1120 :
1121 0 : void page_offline_end(void)
1122 : {
1123 0 : up_write(&page_offline_rwsem);
1124 0 : }
1125 : EXPORT_SYMBOL(page_offline_end);
1126 :
1127 : #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_FOLIO
1128 : void flush_dcache_folio(struct folio *folio)
1129 : {
1130 : long i, nr = folio_nr_pages(folio);
1131 :
1132 : for (i = 0; i < nr; i++)
1133 : flush_dcache_page(folio_page(folio, i));
1134 : }
1135 : EXPORT_SYMBOL(flush_dcache_folio);
1136 : #endif
|
