blob: 887e57182716828b7f4f4946fe7145d106ec5bea [file] [log] [blame]
Thomas Garnier0483e1f2016-06-21 17:47:02 -07001/*
2 * This file implements KASLR memory randomization for x86_64. It randomizes
3 * the virtual address space of kernel memory regions (physical memory
4 * mapping, vmalloc & vmemmap) for x86_64. This security feature mitigates
5 * exploits relying on predictable kernel addresses.
6 *
7 * Entropy is generated using the KASLR early boot functions now shared in
8 * the lib directory (originally written by Kees Cook). Randomization is
9 * done on PGD & PUD page table levels to increase possible addresses. The
10 * physical memory mapping code was adapted to support PUD level virtual
11 * addresses. This implementation on the best configuration provides 30,000
12 * possible virtual addresses in average for each memory region. An additional
13 * low memory page is used to ensure each CPU can start with a PGD aligned
14 * virtual address (for realmode).
15 *
16 * The order of each memory region is not changed. The feature looks at
17 * the available space for the regions based on different configuration
18 * options and randomizes the base and space between each. The size of the
19 * physical memory mapping is the available physical memory.
20 */
21
22#include <linux/kernel.h>
23#include <linux/init.h>
24#include <linux/random.h>
25
26#include <asm/pgalloc.h>
27#include <asm/pgtable.h>
28#include <asm/setup.h>
29#include <asm/kaslr.h>
30
31#include "mm_internal.h"
32
33#define TB_SHIFT 40
34
35/*
36 * Virtual address start and end range for randomization. The end changes base
37 * on configuration to have the highest amount of space for randomization.
38 * It increases the possible random position for each randomized region.
39 *
40 * You need to add an if/def entry if you introduce a new memory region
41 * compatible with KASLR. Your entry must be in logical order with memory
42 * layout. For example, ESPFIX is before EFI because its virtual address is
Thomas Garnier25dfe472016-07-27 08:59:56 -070043 * before. You also need to add a BUILD_BUG_ON() in kernel_randomize_memory() to
Thomas Garnier0483e1f2016-06-21 17:47:02 -070044 * ensure that this order is correct and won't be changed.
45 */
Thomas Garnier021182e2016-06-21 17:47:03 -070046static const unsigned long vaddr_start = __PAGE_OFFSET_BASE;
Thomas Garnier25dfe472016-07-27 08:59:56 -070047
48#if defined(CONFIG_X86_ESPFIX64)
49static const unsigned long vaddr_end = ESPFIX_BASE_ADDR;
50#elif defined(CONFIG_EFI)
51static const unsigned long vaddr_end = EFI_VA_START;
52#else
53static const unsigned long vaddr_end = __START_KERNEL_map;
54#endif
Thomas Garnier021182e2016-06-21 17:47:03 -070055
56/* Default values */
57unsigned long page_offset_base = __PAGE_OFFSET_BASE;
58EXPORT_SYMBOL(page_offset_base);
Thomas Garniera95ae272016-06-21 17:47:04 -070059unsigned long vmalloc_base = __VMALLOC_BASE;
60EXPORT_SYMBOL(vmalloc_base);
Thomas Garnier25dfe472016-07-27 08:59:56 -070061unsigned long vmemmap_base = __VMEMMAP_BASE;
62EXPORT_SYMBOL(vmemmap_base);
Thomas Garnier0483e1f2016-06-21 17:47:02 -070063
64/*
65 * Memory regions randomized by KASLR (except modules that use a separate logic
66 * earlier during boot). The list is ordered based on virtual addresses. This
67 * order is kept after randomization.
68 */
69static __initdata struct kaslr_memory_region {
70 unsigned long *base;
71 unsigned long size_tb;
72} kaslr_regions[] = {
Thomas Garnier021182e2016-06-21 17:47:03 -070073 { &page_offset_base, 64/* Maximum */ },
Thomas Garniera95ae272016-06-21 17:47:04 -070074 { &vmalloc_base, VMALLOC_SIZE_TB },
Thomas Garnier25dfe472016-07-27 08:59:56 -070075 { &vmemmap_base, 1 },
Thomas Garnier0483e1f2016-06-21 17:47:02 -070076};
77
78/* Get size in bytes used by the memory region */
79static inline unsigned long get_padding(struct kaslr_memory_region *region)
80{
81 return (region->size_tb << TB_SHIFT);
82}
83
84/*
85 * Apply no randomization if KASLR was disabled at boot or if KASAN
86 * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
87 */
88static inline bool kaslr_memory_enabled(void)
89{
Masahiro Yamadaa5ff1b32016-08-25 15:17:02 -070090 return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
Thomas Garnier0483e1f2016-06-21 17:47:02 -070091}
92
93/* Initialize base and padding for each memory region randomized with KASLR */
94void __init kernel_randomize_memory(void)
95{
96 size_t i;
97 unsigned long vaddr = vaddr_start;
Thomas Garnier021182e2016-06-21 17:47:03 -070098 unsigned long rand, memory_tb;
Thomas Garnier0483e1f2016-06-21 17:47:02 -070099 struct rnd_state rand_state;
100 unsigned long remain_entropy;
101
Thomas Garnier25dfe472016-07-27 08:59:56 -0700102 /*
103 * All these BUILD_BUG_ON checks ensures the memory layout is
104 * consistent with the vaddr_start/vaddr_end variables.
105 */
106 BUILD_BUG_ON(vaddr_start >= vaddr_end);
Masahiro Yamadac0a0aba2016-10-27 17:46:38 -0700107 BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_ESPFIX64) &&
Thomas Garnier25dfe472016-07-27 08:59:56 -0700108 vaddr_end >= EFI_VA_START);
Masahiro Yamadac0a0aba2016-10-27 17:46:38 -0700109 BUILD_BUG_ON((IS_ENABLED(CONFIG_X86_ESPFIX64) ||
110 IS_ENABLED(CONFIG_EFI)) &&
Thomas Garnier25dfe472016-07-27 08:59:56 -0700111 vaddr_end >= __START_KERNEL_map);
112 BUILD_BUG_ON(vaddr_end > __START_KERNEL_map);
113
Thomas Garnier0483e1f2016-06-21 17:47:02 -0700114 if (!kaslr_memory_enabled())
115 return;
116
Thomas Garnier90397a42016-06-21 17:47:06 -0700117 /*
118 * Update Physical memory mapping to available and
119 * add padding if needed (especially for memory hotplug support).
120 */
Thomas Garnier021182e2016-06-21 17:47:03 -0700121 BUG_ON(kaslr_regions[0].base != &page_offset_base);
Thomas Garnierc7d23612016-08-09 10:11:04 -0700122 memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
Thomas Garnier90397a42016-06-21 17:47:06 -0700123 CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
Thomas Garnier021182e2016-06-21 17:47:03 -0700124
125 /* Adapt phyiscal memory region size based on available memory */
126 if (memory_tb < kaslr_regions[0].size_tb)
127 kaslr_regions[0].size_tb = memory_tb;
128
Thomas Garnier0483e1f2016-06-21 17:47:02 -0700129 /* Calculate entropy available between regions */
130 remain_entropy = vaddr_end - vaddr_start;
131 for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
132 remain_entropy -= get_padding(&kaslr_regions[i]);
133
134 prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
135
136 for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
137 unsigned long entropy;
138
139 /*
140 * Select a random virtual address using the extra entropy
141 * available.
142 */
143 entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
144 prandom_bytes_state(&rand_state, &rand, sizeof(rand));
145 entropy = (rand % (entropy + 1)) & PUD_MASK;
146 vaddr += entropy;
147 *kaslr_regions[i].base = vaddr;
148
149 /*
150 * Jump the region and add a minimum padding based on
151 * randomization alignment.
152 */
153 vaddr += get_padding(&kaslr_regions[i]);
154 vaddr = round_up(vaddr + 1, PUD_SIZE);
155 remain_entropy -= entropy;
156 }
157}
158
159/*
160 * Create PGD aligned trampoline table to allow real mode initialization
161 * of additional CPUs. Consume only 1 low memory page.
162 */
163void __meminit init_trampoline(void)
164{
165 unsigned long paddr, paddr_next;
166 pgd_t *pgd;
167 pud_t *pud_page, *pud_page_tramp;
168 int i;
169
170 if (!kaslr_memory_enabled()) {
171 init_trampoline_default();
172 return;
173 }
174
175 pud_page_tramp = alloc_low_page();
176
177 paddr = 0;
178 pgd = pgd_offset_k((unsigned long)__va(paddr));
179 pud_page = (pud_t *) pgd_page_vaddr(*pgd);
180
181 for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
182 pud_t *pud, *pud_tramp;
183 unsigned long vaddr = (unsigned long)__va(paddr);
184
185 pud_tramp = pud_page_tramp + pud_index(paddr);
186 pud = pud_page + pud_index(vaddr);
187 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
188
189 *pud_tramp = *pud;
190 }
191
192 set_pgd(&trampoline_pgd_entry,
193 __pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
194}