Loading...
Note: File does not exist in v5.4.
1// 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>
16#include <linux/kfence.h>
17#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{
31 return kasan_check_range((unsigned long)p, size, false, _RET_IP_);
32}
33EXPORT_SYMBOL(__kasan_check_read);
34
35bool __kasan_check_write(const volatile void *p, unsigned int size)
36{
37 return kasan_check_range((unsigned long)p, size, true, _RET_IP_);
38}
39EXPORT_SYMBOL(__kasan_check_write);
40
41#undef memset
42void *memset(void *addr, int c, size_t len)
43{
44 if (!kasan_check_range((unsigned long)addr, len, true, _RET_IP_))
45 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{
54 if (!kasan_check_range((unsigned long)src, len, false, _RET_IP_) ||
55 !kasan_check_range((unsigned long)dest, len, true, _RET_IP_))
56 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{
65 if (!kasan_check_range((unsigned long)src, len, false, _RET_IP_) ||
66 !kasan_check_range((unsigned long)dest, len, true, _RET_IP_))
67 return NULL;
68
69 return __memcpy(dest, src, len);
70}
71
72void kasan_poison(const void *addr, size_t size, u8 value, bool init)
73{
74 void *shadow_start, *shadow_end;
75
76 if (!kasan_arch_is_ready())
77 return;
78
79 /*
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 */
84 addr = kasan_reset_tag(addr);
85
86 /* Skip KFENCE memory if called explicitly outside of sl*b. */
87 if (is_kfence_address(addr))
88 return;
89
90 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);
97
98 __memset(shadow_start, value, shadow_end - shadow_start);
99}
100EXPORT_SYMBOL(kasan_poison);
101
102#ifdef CONFIG_KASAN_GENERIC
103void kasan_poison_last_granule(const void *addr, size_t size)
104{
105 if (!kasan_arch_is_ready())
106 return;
107
108 if (size & KASAN_GRANULE_MASK) {
109 u8 *shadow = (u8 *)kasan_mem_to_shadow(addr + size);
110 *shadow = size & KASAN_GRANULE_MASK;
111 }
112}
113#endif
114
115void kasan_unpoison(const void *addr, size_t size, bool init)
116{
117 u8 tag = get_tag(addr);
118
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 */
124 addr = kasan_reset_tag(addr);
125
126 /*
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 */
131 if (is_kfence_address(addr))
132 return;
133
134 if (WARN_ON((unsigned long)addr & KASAN_GRANULE_MASK))
135 return;
136
137 /* Unpoison all granules that cover the object. */
138 kasan_poison(addr, round_up(size, KASAN_GRANULE_SIZE), tag, false);
139
140 /* Partially poison the last granule for the generic mode. */
141 if (IS_ENABLED(CONFIG_KASAN_GENERIC))
142 kasan_poison_last_granule(addr, size);
143}
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) ||
194 WARN_ON(start_kaddr % KASAN_MEMORY_PER_SHADOW_PAGE))
195 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
257static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
258 void *unused)
259{
260 unsigned long page;
261 pte_t pte;
262
263 if (likely(!pte_none(*ptep)))
264 return 0;
265
266 page = __get_free_page(GFP_KERNEL);
267 if (!page)
268 return -ENOMEM;
269
270 memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
271 pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
272
273 spin_lock(&init_mm.page_table_lock);
274 if (likely(pte_none(*ptep))) {
275 set_pte_at(&init_mm, addr, ptep, pte);
276 page = 0;
277 }
278 spin_unlock(&init_mm.page_table_lock);
279 if (page)
280 free_page(page);
281 return 0;
282}
283
284int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
285{
286 unsigned long shadow_start, shadow_end;
287 int ret;
288
289 if (!is_vmalloc_or_module_addr((void *)addr))
290 return 0;
291
292 shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
293 shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
294 shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
295 shadow_end = ALIGN(shadow_end, PAGE_SIZE);
296
297 ret = apply_to_page_range(&init_mm, shadow_start,
298 shadow_end - shadow_start,
299 kasan_populate_vmalloc_pte, NULL);
300 if (ret)
301 return ret;
302
303 flush_cache_vmap(shadow_start, shadow_end);
304
305 /*
306 * We need to be careful about inter-cpu effects here. Consider:
307 *
308 * CPU#0 CPU#1
309 * WRITE_ONCE(p, vmalloc(100)); while (x = READ_ONCE(p)) ;
310 * p[99] = 1;
311 *
312 * With compiler instrumentation, that ends up looking like this:
313 *
314 * CPU#0 CPU#1
315 * // vmalloc() allocates memory
316 * // let a = area->addr
317 * // we reach kasan_populate_vmalloc
318 * // and call kasan_unpoison:
319 * STORE shadow(a), unpoison_val
320 * ...
321 * STORE shadow(a+99), unpoison_val x = LOAD p
322 * // rest of vmalloc process <data dependency>
323 * STORE p, a LOAD shadow(x+99)
324 *
325 * If there is no barrier between the end of unpoisoning the shadow
326 * and the store of the result to p, the stores could be committed
327 * in a different order by CPU#0, and CPU#1 could erroneously observe
328 * poison in the shadow.
329 *
330 * We need some sort of barrier between the stores.
331 *
332 * In the vmalloc() case, this is provided by a smp_wmb() in
333 * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
334 * get_vm_area() and friends, the caller gets shadow allocated but
335 * doesn't have any pages mapped into the virtual address space that
336 * has been reserved. Mapping those pages in will involve taking and
337 * releasing a page-table lock, which will provide the barrier.
338 */
339
340 return 0;
341}
342
343/*
344 * Poison the shadow for a vmalloc region. Called as part of the
345 * freeing process at the time the region is freed.
346 */
347void kasan_poison_vmalloc(const void *start, unsigned long size)
348{
349 if (!is_vmalloc_or_module_addr(start))
350 return;
351
352 size = round_up(size, KASAN_GRANULE_SIZE);
353 kasan_poison(start, size, KASAN_VMALLOC_INVALID, false);
354}
355
356void kasan_unpoison_vmalloc(const void *start, unsigned long size)
357{
358 if (!is_vmalloc_or_module_addr(start))
359 return;
360
361 kasan_unpoison(start, size, false);
362}
363
364static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
365 void *unused)
366{
367 unsigned long page;
368
369 page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
370
371 spin_lock(&init_mm.page_table_lock);
372
373 if (likely(!pte_none(*ptep))) {
374 pte_clear(&init_mm, addr, ptep);
375 free_page(page);
376 }
377 spin_unlock(&init_mm.page_table_lock);
378
379 return 0;
380}
381
382/*
383 * Release the backing for the vmalloc region [start, end), which
384 * lies within the free region [free_region_start, free_region_end).
385 *
386 * This can be run lazily, long after the region was freed. It runs
387 * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
388 * infrastructure.
389 *
390 * How does this work?
391 * -------------------
392 *
393 * We have a region that is page aligned, labeled as A.
394 * That might not map onto the shadow in a way that is page-aligned:
395 *
396 * start end
397 * v v
398 * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
399 * -------- -------- -------- -------- --------
400 * | | | | |
401 * | | | /-------/ |
402 * \-------\|/------/ |/---------------/
403 * ||| ||
404 * |??AAAAAA|AAAAAAAA|AA??????| < shadow
405 * (1) (2) (3)
406 *
407 * First we align the start upwards and the end downwards, so that the
408 * shadow of the region aligns with shadow page boundaries. In the
409 * example, this gives us the shadow page (2). This is the shadow entirely
410 * covered by this allocation.
411 *
412 * Then we have the tricky bits. We want to know if we can free the
413 * partially covered shadow pages - (1) and (3) in the example. For this,
414 * we are given the start and end of the free region that contains this
415 * allocation. Extending our previous example, we could have:
416 *
417 * free_region_start free_region_end
418 * | start end |
419 * v v v v
420 * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
421 * -------- -------- -------- -------- --------
422 * | | | | |
423 * | | | /-------/ |
424 * \-------\|/------/ |/---------------/
425 * ||| ||
426 * |FFAAAAAA|AAAAAAAA|AAF?????| < shadow
427 * (1) (2) (3)
428 *
429 * Once again, we align the start of the free region up, and the end of
430 * the free region down so that the shadow is page aligned. So we can free
431 * page (1) - we know no allocation currently uses anything in that page,
432 * because all of it is in the vmalloc free region. But we cannot free
433 * page (3), because we can't be sure that the rest of it is unused.
434 *
435 * We only consider pages that contain part of the original region for
436 * freeing: we don't try to free other pages from the free region or we'd
437 * end up trying to free huge chunks of virtual address space.
438 *
439 * Concurrency
440 * -----------
441 *
442 * How do we know that we're not freeing a page that is simultaneously
443 * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
444 *
445 * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
446 * at the same time. While we run under free_vmap_area_lock, the population
447 * code does not.
448 *
449 * free_vmap_area_lock instead operates to ensure that the larger range
450 * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
451 * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
452 * no space identified as free will become used while we are running. This
453 * means that so long as we are careful with alignment and only free shadow
454 * pages entirely covered by the free region, we will not run in to any
455 * trouble - any simultaneous allocations will be for disjoint regions.
456 */
457void kasan_release_vmalloc(unsigned long start, unsigned long end,
458 unsigned long free_region_start,
459 unsigned long free_region_end)
460{
461 void *shadow_start, *shadow_end;
462 unsigned long region_start, region_end;
463 unsigned long size;
464
465 region_start = ALIGN(start, KASAN_MEMORY_PER_SHADOW_PAGE);
466 region_end = ALIGN_DOWN(end, KASAN_MEMORY_PER_SHADOW_PAGE);
467
468 free_region_start = ALIGN(free_region_start, KASAN_MEMORY_PER_SHADOW_PAGE);
469
470 if (start != region_start &&
471 free_region_start < region_start)
472 region_start -= KASAN_MEMORY_PER_SHADOW_PAGE;
473
474 free_region_end = ALIGN_DOWN(free_region_end, KASAN_MEMORY_PER_SHADOW_PAGE);
475
476 if (end != region_end &&
477 free_region_end > region_end)
478 region_end += KASAN_MEMORY_PER_SHADOW_PAGE;
479
480 shadow_start = kasan_mem_to_shadow((void *)region_start);
481 shadow_end = kasan_mem_to_shadow((void *)region_end);
482
483 if (shadow_end > shadow_start) {
484 size = shadow_end - shadow_start;
485 apply_to_existing_page_range(&init_mm,
486 (unsigned long)shadow_start,
487 size, kasan_depopulate_vmalloc_pte,
488 NULL);
489 flush_tlb_kernel_range((unsigned long)shadow_start,
490 (unsigned long)shadow_end);
491 }
492}
493
494#else /* CONFIG_KASAN_VMALLOC */
495
496int kasan_module_alloc(void *addr, size_t size)
497{
498 void *ret;
499 size_t scaled_size;
500 size_t shadow_size;
501 unsigned long shadow_start;
502
503 shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
504 scaled_size = (size + KASAN_GRANULE_SIZE - 1) >>
505 KASAN_SHADOW_SCALE_SHIFT;
506 shadow_size = round_up(scaled_size, PAGE_SIZE);
507
508 if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
509 return -EINVAL;
510
511 ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
512 shadow_start + shadow_size,
513 GFP_KERNEL,
514 PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
515 __builtin_return_address(0));
516
517 if (ret) {
518 __memset(ret, KASAN_SHADOW_INIT, shadow_size);
519 find_vm_area(addr)->flags |= VM_KASAN;
520 kmemleak_ignore(ret);
521 return 0;
522 }
523
524 return -ENOMEM;
525}
526
527void kasan_free_shadow(const struct vm_struct *vm)
528{
529 if (vm->flags & VM_KASAN)
530 vfree(kasan_mem_to_shadow(vm->addr));
531}
532
533#endif