blob: 4a4929b29a237f7fd3617817491d173ed7d205b0 [file] [log] [blame]
Andrey Konovalovbb359db2020-12-22 12:00:32 -08001// SPDX-License-Identifier: GPL-2.0
2/*
3 * This file contains KASAN runtime code that manages shadow memory for
4 * generic and software tag-based KASAN modes.
5 *
6 * Copyright (c) 2014 Samsung Electronics Co., Ltd.
7 * Author: Andrey Ryabinin <ryabinin.a.a@gmail.com>
8 *
9 * Some code borrowed from https://github.com/xairy/kasan-prototype by
10 * Andrey Konovalov <andreyknvl@gmail.com>
11 */
12
13#include <linux/init.h>
14#include <linux/kasan.h>
15#include <linux/kernel.h>
Alexander Potapenko2b830522021-02-25 17:19:21 -080016#include <linux/kfence.h>
Andrey Konovalovbb359db2020-12-22 12:00:32 -080017#include <linux/kmemleak.h>
18#include <linux/memory.h>
19#include <linux/mm.h>
20#include <linux/string.h>
21#include <linux/types.h>
22#include <linux/vmalloc.h>
23
24#include <asm/cacheflush.h>
25#include <asm/tlbflush.h>
26
27#include "kasan.h"
28
29bool __kasan_check_read(const volatile void *p, unsigned int size)
30{
Andrey Konovalovf00748b2021-02-24 12:05:05 -080031 return kasan_check_range((unsigned long)p, size, false, _RET_IP_);
Andrey Konovalovbb359db2020-12-22 12:00:32 -080032}
33EXPORT_SYMBOL(__kasan_check_read);
34
35bool __kasan_check_write(const volatile void *p, unsigned int size)
36{
Andrey Konovalovf00748b2021-02-24 12:05:05 -080037 return kasan_check_range((unsigned long)p, size, true, _RET_IP_);
Andrey Konovalovbb359db2020-12-22 12:00:32 -080038}
39EXPORT_SYMBOL(__kasan_check_write);
40
41#undef memset
42void *memset(void *addr, int c, size_t len)
43{
Andrey Konovalovf00748b2021-02-24 12:05:05 -080044 if (!kasan_check_range((unsigned long)addr, len, true, _RET_IP_))
Andrey Konovalovbb359db2020-12-22 12:00:32 -080045 return NULL;
46
47 return __memset(addr, c, len);
48}
49
50#ifdef __HAVE_ARCH_MEMMOVE
51#undef memmove
52void *memmove(void *dest, const void *src, size_t len)
53{
Andrey Konovalovf00748b2021-02-24 12:05:05 -080054 if (!kasan_check_range((unsigned long)src, len, false, _RET_IP_) ||
55 !kasan_check_range((unsigned long)dest, len, true, _RET_IP_))
Andrey Konovalovbb359db2020-12-22 12:00:32 -080056 return NULL;
57
58 return __memmove(dest, src, len);
59}
60#endif
61
62#undef memcpy
63void *memcpy(void *dest, const void *src, size_t len)
64{
Andrey Konovalovf00748b2021-02-24 12:05:05 -080065 if (!kasan_check_range((unsigned long)src, len, false, _RET_IP_) ||
66 !kasan_check_range((unsigned long)dest, len, true, _RET_IP_))
Andrey Konovalovbb359db2020-12-22 12:00:32 -080067 return NULL;
68
69 return __memcpy(dest, src, len);
70}
71
Andrey Konovalovaa5c2192021-04-29 22:59:59 -070072void kasan_poison(const void *addr, size_t size, u8 value, bool init)
Andrey Konovalovbb359db2020-12-22 12:00:32 -080073{
74 void *shadow_start, *shadow_end;
75
Daniel Axtensaf3751f32021-06-28 19:40:42 -070076 if (!kasan_arch_is_ready())
77 return;
78
Andrey Konovalovbb359db2020-12-22 12:00:32 -080079 /*
80 * Perform shadow offset calculation based on untagged address, as
81 * some of the callers (e.g. kasan_poison_object_data) pass tagged
82 * addresses to this function.
83 */
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -080084 addr = kasan_reset_tag(addr);
Andrey Konovalovbb359db2020-12-22 12:00:32 -080085
Alexander Potapenko2b830522021-02-25 17:19:21 -080086 /* Skip KFENCE memory if called explicitly outside of sl*b. */
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -080087 if (is_kfence_address(addr))
Alexander Potapenko2b830522021-02-25 17:19:21 -080088 return;
89
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -080090 if (WARN_ON((unsigned long)addr & KASAN_GRANULE_MASK))
91 return;
92 if (WARN_ON(size & KASAN_GRANULE_MASK))
93 return;
94
95 shadow_start = kasan_mem_to_shadow(addr);
96 shadow_end = kasan_mem_to_shadow(addr + size);
Andrey Konovalovbb359db2020-12-22 12:00:32 -080097
98 __memset(shadow_start, value, shadow_end - shadow_start);
99}
Andrey Konovalov573a4802021-02-24 12:05:21 -0800100EXPORT_SYMBOL(kasan_poison);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800101
Andrey Konovalove2db1a92021-02-25 17:19:59 -0800102#ifdef CONFIG_KASAN_GENERIC
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800103void kasan_poison_last_granule(const void *addr, size_t size)
Andrey Konovalove2db1a92021-02-25 17:19:59 -0800104{
Daniel Axtensaf3751f32021-06-28 19:40:42 -0700105 if (!kasan_arch_is_ready())
106 return;
107
Andrey Konovalove2db1a92021-02-25 17:19:59 -0800108 if (size & KASAN_GRANULE_MASK) {
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800109 u8 *shadow = (u8 *)kasan_mem_to_shadow(addr + size);
Andrey Konovalove2db1a92021-02-25 17:19:59 -0800110 *shadow = size & KASAN_GRANULE_MASK;
111 }
112}
113#endif
114
Andrey Konovalovaa5c2192021-04-29 22:59:59 -0700115void kasan_unpoison(const void *addr, size_t size, bool init)
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800116{
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800117 u8 tag = get_tag(addr);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800118
119 /*
120 * Perform shadow offset calculation based on untagged address, as
121 * some of the callers (e.g. kasan_unpoison_object_data) pass tagged
122 * addresses to this function.
123 */
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800124 addr = kasan_reset_tag(addr);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800125
Alexander Potapenko2b830522021-02-25 17:19:21 -0800126 /*
127 * Skip KFENCE memory if called explicitly outside of sl*b. Also note
128 * that calls to ksize(), where size is not a multiple of machine-word
129 * size, would otherwise poison the invalid portion of the word.
130 */
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800131 if (is_kfence_address(addr))
Alexander Potapenko2b830522021-02-25 17:19:21 -0800132 return;
133
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800134 if (WARN_ON((unsigned long)addr & KASAN_GRANULE_MASK))
135 return;
136
137 /* Unpoison all granules that cover the object. */
Andrey Konovalovaa5c2192021-04-29 22:59:59 -0700138 kasan_poison(addr, round_up(size, KASAN_GRANULE_SIZE), tag, false);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800139
Andrey Konovalove2db1a92021-02-25 17:19:59 -0800140 /* Partially poison the last granule for the generic mode. */
141 if (IS_ENABLED(CONFIG_KASAN_GENERIC))
Andrey Konovalovcde8a7e2021-02-25 17:20:27 -0800142 kasan_poison_last_granule(addr, size);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800143}
144
145#ifdef CONFIG_MEMORY_HOTPLUG
146static bool shadow_mapped(unsigned long addr)
147{
148 pgd_t *pgd = pgd_offset_k(addr);
149 p4d_t *p4d;
150 pud_t *pud;
151 pmd_t *pmd;
152 pte_t *pte;
153
154 if (pgd_none(*pgd))
155 return false;
156 p4d = p4d_offset(pgd, addr);
157 if (p4d_none(*p4d))
158 return false;
159 pud = pud_offset(p4d, addr);
160 if (pud_none(*pud))
161 return false;
162
163 /*
164 * We can't use pud_large() or pud_huge(), the first one is
165 * arch-specific, the last one depends on HUGETLB_PAGE. So let's abuse
166 * pud_bad(), if pud is bad then it's bad because it's huge.
167 */
168 if (pud_bad(*pud))
169 return true;
170 pmd = pmd_offset(pud, addr);
171 if (pmd_none(*pmd))
172 return false;
173
174 if (pmd_bad(*pmd))
175 return true;
176 pte = pte_offset_kernel(pmd, addr);
177 return !pte_none(*pte);
178}
179
180static int __meminit kasan_mem_notifier(struct notifier_block *nb,
181 unsigned long action, void *data)
182{
183 struct memory_notify *mem_data = data;
184 unsigned long nr_shadow_pages, start_kaddr, shadow_start;
185 unsigned long shadow_end, shadow_size;
186
187 nr_shadow_pages = mem_data->nr_pages >> KASAN_SHADOW_SCALE_SHIFT;
188 start_kaddr = (unsigned long)pfn_to_kaddr(mem_data->start_pfn);
189 shadow_start = (unsigned long)kasan_mem_to_shadow((void *)start_kaddr);
190 shadow_size = nr_shadow_pages << PAGE_SHIFT;
191 shadow_end = shadow_start + shadow_size;
192
193 if (WARN_ON(mem_data->nr_pages % KASAN_GRANULE_SIZE) ||
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800194 WARN_ON(start_kaddr % KASAN_MEMORY_PER_SHADOW_PAGE))
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800195 return NOTIFY_BAD;
196
197 switch (action) {
198 case MEM_GOING_ONLINE: {
199 void *ret;
200
201 /*
202 * If shadow is mapped already than it must have been mapped
203 * during the boot. This could happen if we onlining previously
204 * offlined memory.
205 */
206 if (shadow_mapped(shadow_start))
207 return NOTIFY_OK;
208
209 ret = __vmalloc_node_range(shadow_size, PAGE_SIZE, shadow_start,
210 shadow_end, GFP_KERNEL,
211 PAGE_KERNEL, VM_NO_GUARD,
212 pfn_to_nid(mem_data->start_pfn),
213 __builtin_return_address(0));
214 if (!ret)
215 return NOTIFY_BAD;
216
217 kmemleak_ignore(ret);
218 return NOTIFY_OK;
219 }
220 case MEM_CANCEL_ONLINE:
221 case MEM_OFFLINE: {
222 struct vm_struct *vm;
223
224 /*
225 * shadow_start was either mapped during boot by kasan_init()
226 * or during memory online by __vmalloc_node_range().
227 * In the latter case we can use vfree() to free shadow.
228 * Non-NULL result of the find_vm_area() will tell us if
229 * that was the second case.
230 *
231 * Currently it's not possible to free shadow mapped
232 * during boot by kasan_init(). It's because the code
233 * to do that hasn't been written yet. So we'll just
234 * leak the memory.
235 */
236 vm = find_vm_area((void *)shadow_start);
237 if (vm)
238 vfree((void *)shadow_start);
239 }
240 }
241
242 return NOTIFY_OK;
243}
244
245static int __init kasan_memhotplug_init(void)
246{
247 hotplug_memory_notifier(kasan_mem_notifier, 0);
248
249 return 0;
250}
251
252core_initcall(kasan_memhotplug_init);
253#endif
254
255#ifdef CONFIG_KASAN_VMALLOC
256
Kefeng Wang3252b1d2021-11-05 13:39:47 -0700257void __init __weak kasan_populate_early_vm_area_shadow(void *start,
258 unsigned long size)
259{
260}
261
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800262static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
263 void *unused)
264{
265 unsigned long page;
266 pte_t pte;
267
268 if (likely(!pte_none(*ptep)))
269 return 0;
270
271 page = __get_free_page(GFP_KERNEL);
272 if (!page)
273 return -ENOMEM;
274
275 memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
276 pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
277
278 spin_lock(&init_mm.page_table_lock);
279 if (likely(pte_none(*ptep))) {
280 set_pte_at(&init_mm, addr, ptep, pte);
281 page = 0;
282 }
283 spin_unlock(&init_mm.page_table_lock);
284 if (page)
285 free_page(page);
286 return 0;
287}
288
289int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
290{
291 unsigned long shadow_start, shadow_end;
292 int ret;
293
294 if (!is_vmalloc_or_module_addr((void *)addr))
295 return 0;
296
297 shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
298 shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
299 shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
300 shadow_end = ALIGN(shadow_end, PAGE_SIZE);
301
302 ret = apply_to_page_range(&init_mm, shadow_start,
303 shadow_end - shadow_start,
304 kasan_populate_vmalloc_pte, NULL);
305 if (ret)
306 return ret;
307
308 flush_cache_vmap(shadow_start, shadow_end);
309
310 /*
311 * We need to be careful about inter-cpu effects here. Consider:
312 *
313 * CPU#0 CPU#1
314 * WRITE_ONCE(p, vmalloc(100)); while (x = READ_ONCE(p)) ;
315 * p[99] = 1;
316 *
317 * With compiler instrumentation, that ends up looking like this:
318 *
319 * CPU#0 CPU#1
320 * // vmalloc() allocates memory
321 * // let a = area->addr
322 * // we reach kasan_populate_vmalloc
Andrey Konovalovf00748b2021-02-24 12:05:05 -0800323 * // and call kasan_unpoison:
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800324 * STORE shadow(a), unpoison_val
325 * ...
326 * STORE shadow(a+99), unpoison_val x = LOAD p
327 * // rest of vmalloc process <data dependency>
328 * STORE p, a LOAD shadow(x+99)
329 *
Ingo Molnarf0953a12021-05-06 18:06:47 -0700330 * If there is no barrier between the end of unpoisoning the shadow
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800331 * and the store of the result to p, the stores could be committed
332 * in a different order by CPU#0, and CPU#1 could erroneously observe
333 * poison in the shadow.
334 *
335 * We need some sort of barrier between the stores.
336 *
337 * In the vmalloc() case, this is provided by a smp_wmb() in
338 * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
339 * get_vm_area() and friends, the caller gets shadow allocated but
340 * doesn't have any pages mapped into the virtual address space that
341 * has been reserved. Mapping those pages in will involve taking and
342 * releasing a page-table lock, which will provide the barrier.
343 */
344
345 return 0;
346}
347
348/*
349 * Poison the shadow for a vmalloc region. Called as part of the
350 * freeing process at the time the region is freed.
351 */
352void kasan_poison_vmalloc(const void *start, unsigned long size)
353{
354 if (!is_vmalloc_or_module_addr(start))
355 return;
356
357 size = round_up(size, KASAN_GRANULE_SIZE);
Andrey Konovalovaa5c2192021-04-29 22:59:59 -0700358 kasan_poison(start, size, KASAN_VMALLOC_INVALID, false);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800359}
360
361void kasan_unpoison_vmalloc(const void *start, unsigned long size)
362{
363 if (!is_vmalloc_or_module_addr(start))
364 return;
365
Andrey Konovalovaa5c2192021-04-29 22:59:59 -0700366 kasan_unpoison(start, size, false);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800367}
368
369static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
370 void *unused)
371{
372 unsigned long page;
373
374 page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
375
376 spin_lock(&init_mm.page_table_lock);
377
378 if (likely(!pte_none(*ptep))) {
379 pte_clear(&init_mm, addr, ptep);
380 free_page(page);
381 }
382 spin_unlock(&init_mm.page_table_lock);
383
384 return 0;
385}
386
387/*
388 * Release the backing for the vmalloc region [start, end), which
389 * lies within the free region [free_region_start, free_region_end).
390 *
391 * This can be run lazily, long after the region was freed. It runs
392 * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
393 * infrastructure.
394 *
395 * How does this work?
396 * -------------------
397 *
Ingo Molnarf0953a12021-05-06 18:06:47 -0700398 * We have a region that is page aligned, labeled as A.
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800399 * That might not map onto the shadow in a way that is page-aligned:
400 *
401 * start end
402 * v v
403 * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
404 * -------- -------- -------- -------- --------
405 * | | | | |
406 * | | | /-------/ |
407 * \-------\|/------/ |/---------------/
408 * ||| ||
409 * |??AAAAAA|AAAAAAAA|AA??????| < shadow
410 * (1) (2) (3)
411 *
412 * First we align the start upwards and the end downwards, so that the
413 * shadow of the region aligns with shadow page boundaries. In the
414 * example, this gives us the shadow page (2). This is the shadow entirely
415 * covered by this allocation.
416 *
417 * Then we have the tricky bits. We want to know if we can free the
418 * partially covered shadow pages - (1) and (3) in the example. For this,
419 * we are given the start and end of the free region that contains this
420 * allocation. Extending our previous example, we could have:
421 *
422 * free_region_start free_region_end
423 * | start end |
424 * v v v v
425 * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
426 * -------- -------- -------- -------- --------
427 * | | | | |
428 * | | | /-------/ |
429 * \-------\|/------/ |/---------------/
430 * ||| ||
431 * |FFAAAAAA|AAAAAAAA|AAF?????| < shadow
432 * (1) (2) (3)
433 *
434 * Once again, we align the start of the free region up, and the end of
435 * the free region down so that the shadow is page aligned. So we can free
436 * page (1) - we know no allocation currently uses anything in that page,
437 * because all of it is in the vmalloc free region. But we cannot free
438 * page (3), because we can't be sure that the rest of it is unused.
439 *
440 * We only consider pages that contain part of the original region for
441 * freeing: we don't try to free other pages from the free region or we'd
442 * end up trying to free huge chunks of virtual address space.
443 *
444 * Concurrency
445 * -----------
446 *
447 * How do we know that we're not freeing a page that is simultaneously
448 * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
449 *
450 * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
451 * at the same time. While we run under free_vmap_area_lock, the population
452 * code does not.
453 *
454 * free_vmap_area_lock instead operates to ensure that the larger range
455 * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
456 * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
457 * no space identified as free will become used while we are running. This
458 * means that so long as we are careful with alignment and only free shadow
459 * pages entirely covered by the free region, we will not run in to any
460 * trouble - any simultaneous allocations will be for disjoint regions.
461 */
462void kasan_release_vmalloc(unsigned long start, unsigned long end,
463 unsigned long free_region_start,
464 unsigned long free_region_end)
465{
466 void *shadow_start, *shadow_end;
467 unsigned long region_start, region_end;
468 unsigned long size;
469
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800470 region_start = ALIGN(start, KASAN_MEMORY_PER_SHADOW_PAGE);
471 region_end = ALIGN_DOWN(end, KASAN_MEMORY_PER_SHADOW_PAGE);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800472
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800473 free_region_start = ALIGN(free_region_start, KASAN_MEMORY_PER_SHADOW_PAGE);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800474
475 if (start != region_start &&
476 free_region_start < region_start)
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800477 region_start -= KASAN_MEMORY_PER_SHADOW_PAGE;
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800478
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800479 free_region_end = ALIGN_DOWN(free_region_end, KASAN_MEMORY_PER_SHADOW_PAGE);
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800480
481 if (end != region_end &&
482 free_region_end > region_end)
Andrey Konovalovaffc3f02020-12-22 12:00:35 -0800483 region_end += KASAN_MEMORY_PER_SHADOW_PAGE;
Andrey Konovalovbb359db2020-12-22 12:00:32 -0800484
485 shadow_start = kasan_mem_to_shadow((void *)region_start);
486 shadow_end = kasan_mem_to_shadow((void *)region_end);
487
488 if (shadow_end > shadow_start) {
489 size = shadow_end - shadow_start;
490 apply_to_existing_page_range(&init_mm,
491 (unsigned long)shadow_start,
492 size, kasan_depopulate_vmalloc_pte,
493 NULL);
494 flush_tlb_kernel_range((unsigned long)shadow_start,
495 (unsigned long)shadow_end);
496 }
497}
498
499#else /* CONFIG_KASAN_VMALLOC */
500
501int kasan_module_alloc(void *addr, size_t size)
502{
503 void *ret;
504 size_t scaled_size;
505 size_t shadow_size;
506 unsigned long shadow_start;
507
508 shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
509 scaled_size = (size + KASAN_GRANULE_SIZE - 1) >>
510 KASAN_SHADOW_SCALE_SHIFT;
511 shadow_size = round_up(scaled_size, PAGE_SIZE);
512
513 if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
514 return -EINVAL;
515
516 ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
517 shadow_start + shadow_size,
518 GFP_KERNEL,
519 PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
520 __builtin_return_address(0));
521
522 if (ret) {
523 __memset(ret, KASAN_SHADOW_INIT, shadow_size);
524 find_vm_area(addr)->flags |= VM_KASAN;
525 kmemleak_ignore(ret);
526 return 0;
527 }
528
529 return -ENOMEM;
530}
531
532void kasan_free_shadow(const struct vm_struct *vm)
533{
534 if (vm->flags & VM_KASAN)
535 vfree(kasan_mem_to_shadow(vm->addr));
536}
537
538#endif