xref: /linux/mm/kmsan/init.c (revision f9bff0e31881d03badf191d3b0005839391f5f2b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * KMSAN initialization routines.
4  *
5  * Copyright (C) 2017-2021 Google LLC
6  * Author: Alexander Potapenko <glider@google.com>
7  *
8  */
9 
10 #include "kmsan.h"
11 
12 #include <asm/sections.h>
13 #include <linux/mm.h>
14 #include <linux/memblock.h>
15 
16 #include "../internal.h"
17 
18 #define NUM_FUTURE_RANGES 128
19 struct start_end_pair {
20 	u64 start, end;
21 };
22 
23 static struct start_end_pair start_end_pairs[NUM_FUTURE_RANGES] __initdata;
24 static int future_index __initdata;
25 
26 /*
27  * Record a range of memory for which the metadata pages will be created once
28  * the page allocator becomes available.
29  */
30 static void __init kmsan_record_future_shadow_range(void *start, void *end)
31 {
32 	u64 nstart = (u64)start, nend = (u64)end, cstart, cend;
33 	bool merged = false;
34 
35 	KMSAN_WARN_ON(future_index == NUM_FUTURE_RANGES);
36 	KMSAN_WARN_ON((nstart >= nend) || !nstart || !nend);
37 	nstart = ALIGN_DOWN(nstart, PAGE_SIZE);
38 	nend = ALIGN(nend, PAGE_SIZE);
39 
40 	/*
41 	 * Scan the existing ranges to see if any of them overlaps with
42 	 * [start, end). In that case, merge the two ranges instead of
43 	 * creating a new one.
44 	 * The number of ranges is less than 20, so there is no need to organize
45 	 * them into a more intelligent data structure.
46 	 */
47 	for (int i = 0; i < future_index; i++) {
48 		cstart = start_end_pairs[i].start;
49 		cend = start_end_pairs[i].end;
50 		if ((cstart < nstart && cend < nstart) ||
51 		    (cstart > nend && cend > nend))
52 			/* ranges are disjoint - do not merge */
53 			continue;
54 		start_end_pairs[i].start = min(nstart, cstart);
55 		start_end_pairs[i].end = max(nend, cend);
56 		merged = true;
57 		break;
58 	}
59 	if (merged)
60 		return;
61 	start_end_pairs[future_index].start = nstart;
62 	start_end_pairs[future_index].end = nend;
63 	future_index++;
64 }
65 
66 /*
67  * Initialize the shadow for existing mappings during kernel initialization.
68  * These include kernel text/data sections, NODE_DATA and future ranges
69  * registered while creating other data (e.g. percpu).
70  *
71  * Allocations via memblock can be only done before slab is initialized.
72  */
73 void __init kmsan_init_shadow(void)
74 {
75 	const size_t nd_size = roundup(sizeof(pg_data_t), PAGE_SIZE);
76 	phys_addr_t p_start, p_end;
77 	u64 loop;
78 	int nid;
79 
80 	for_each_reserved_mem_range(loop, &p_start, &p_end)
81 		kmsan_record_future_shadow_range(phys_to_virt(p_start),
82 						 phys_to_virt(p_end));
83 	/* Allocate shadow for .data */
84 	kmsan_record_future_shadow_range(_sdata, _edata);
85 
86 	for_each_online_node(nid)
87 		kmsan_record_future_shadow_range(
88 			NODE_DATA(nid), (char *)NODE_DATA(nid) + nd_size);
89 
90 	for (int i = 0; i < future_index; i++)
91 		kmsan_init_alloc_meta_for_range(
92 			(void *)start_end_pairs[i].start,
93 			(void *)start_end_pairs[i].end);
94 }
95 
96 struct metadata_page_pair {
97 	struct page *shadow, *origin;
98 };
99 static struct metadata_page_pair held_back[MAX_ORDER + 1] __initdata;
100 
101 /*
102  * Eager metadata allocation. When the memblock allocator is freeing pages to
103  * pagealloc, we use 2/3 of them as metadata for the remaining 1/3.
104  * We store the pointers to the returned blocks of pages in held_back[] grouped
105  * by their order: when kmsan_memblock_free_pages() is called for the first
106  * time with a certain order, it is reserved as a shadow block, for the second
107  * time - as an origin block. On the third time the incoming block receives its
108  * shadow and origin ranges from the previously saved shadow and origin blocks,
109  * after which held_back[order] can be used again.
110  *
111  * At the very end there may be leftover blocks in held_back[]. They are
112  * collected later by kmsan_memblock_discard().
113  */
114 bool kmsan_memblock_free_pages(struct page *page, unsigned int order)
115 {
116 	struct page *shadow, *origin;
117 
118 	if (!held_back[order].shadow) {
119 		held_back[order].shadow = page;
120 		return false;
121 	}
122 	if (!held_back[order].origin) {
123 		held_back[order].origin = page;
124 		return false;
125 	}
126 	shadow = held_back[order].shadow;
127 	origin = held_back[order].origin;
128 	kmsan_setup_meta(page, shadow, origin, order);
129 
130 	held_back[order].shadow = NULL;
131 	held_back[order].origin = NULL;
132 	return true;
133 }
134 
135 #define MAX_BLOCKS 8
136 struct smallstack {
137 	struct page *items[MAX_BLOCKS];
138 	int index;
139 	int order;
140 };
141 
142 static struct smallstack collect = {
143 	.index = 0,
144 	.order = MAX_ORDER,
145 };
146 
147 static void smallstack_push(struct smallstack *stack, struct page *pages)
148 {
149 	KMSAN_WARN_ON(stack->index == MAX_BLOCKS);
150 	stack->items[stack->index] = pages;
151 	stack->index++;
152 }
153 #undef MAX_BLOCKS
154 
155 static struct page *smallstack_pop(struct smallstack *stack)
156 {
157 	struct page *ret;
158 
159 	KMSAN_WARN_ON(stack->index == 0);
160 	stack->index--;
161 	ret = stack->items[stack->index];
162 	stack->items[stack->index] = NULL;
163 	return ret;
164 }
165 
166 static void do_collection(void)
167 {
168 	struct page *page, *shadow, *origin;
169 
170 	while (collect.index >= 3) {
171 		page = smallstack_pop(&collect);
172 		shadow = smallstack_pop(&collect);
173 		origin = smallstack_pop(&collect);
174 		kmsan_setup_meta(page, shadow, origin, collect.order);
175 		__free_pages_core(page, collect.order);
176 	}
177 }
178 
179 static void collect_split(void)
180 {
181 	struct smallstack tmp = {
182 		.order = collect.order - 1,
183 		.index = 0,
184 	};
185 	struct page *page;
186 
187 	if (!collect.order)
188 		return;
189 	while (collect.index) {
190 		page = smallstack_pop(&collect);
191 		smallstack_push(&tmp, &page[0]);
192 		smallstack_push(&tmp, &page[1 << tmp.order]);
193 	}
194 	__memcpy(&collect, &tmp, sizeof(tmp));
195 }
196 
197 /*
198  * Memblock is about to go away. Split the page blocks left over in held_back[]
199  * and return 1/3 of that memory to the system.
200  */
201 static void kmsan_memblock_discard(void)
202 {
203 	/*
204 	 * For each order=N:
205 	 *  - push held_back[N].shadow and .origin to @collect;
206 	 *  - while there are >= 3 elements in @collect, do garbage collection:
207 	 *    - pop 3 ranges from @collect;
208 	 *    - use two of them as shadow and origin for the third one;
209 	 *    - repeat;
210 	 *  - split each remaining element from @collect into 2 ranges of
211 	 *    order=N-1,
212 	 *  - repeat.
213 	 */
214 	collect.order = MAX_ORDER;
215 	for (int i = MAX_ORDER; i >= 0; i--) {
216 		if (held_back[i].shadow)
217 			smallstack_push(&collect, held_back[i].shadow);
218 		if (held_back[i].origin)
219 			smallstack_push(&collect, held_back[i].origin);
220 		held_back[i].shadow = NULL;
221 		held_back[i].origin = NULL;
222 		do_collection();
223 		collect_split();
224 	}
225 }
226 
227 void __init kmsan_init_runtime(void)
228 {
229 	/* Assuming current is init_task */
230 	kmsan_internal_task_create(current);
231 	kmsan_memblock_discard();
232 	pr_info("Starting KernelMemorySanitizer\n");
233 	pr_info("ATTENTION: KMSAN is a debugging tool! Do not use it on production machines!\n");
234 	kmsan_enabled = true;
235 }
236