1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * This file implements KASLR memory randomization for x86_64. It randomizes
4 * the virtual address space of kernel memory regions (physical memory
5 * mapping, vmalloc & vmemmap) for x86_64. This security feature mitigates
6 * exploits relying on predictable kernel addresses.
7 *
8 * Entropy is generated using the KASLR early boot functions now shared in
9 * the lib directory (originally written by Kees Cook). Randomization is
10 * done on PGD & P4D/PUD page table levels to increase possible addresses.
11 * The physical memory mapping code was adapted to support P4D/PUD level
12 * virtual addresses. This implementation on the best configuration provides
13 * 30,000 possible virtual addresses in average for each memory region.
14 * An additional low memory page is used to ensure each CPU can start with
15 * a PGD aligned virtual address (for realmode).
16 *
17 * The order of each memory region is not changed. The feature looks at
18 * the available space for the regions based on different configuration
19 * options and randomizes the base and space between each. The size of the
20 * physical memory mapping is the available physical memory.
21 */
22
23 #include <linux/kernel.h>
24 #include <linux/init.h>
25 #include <linux/random.h>
26 #include <linux/memblock.h>
27 #include <linux/pgtable.h>
28
29 #include <asm/setup.h>
30 #include <asm/kaslr.h>
31
32 #include "mm_internal.h"
33
34 #define TB_SHIFT 40
35
36 /*
37 * The end address could depend on more configuration options to make the
38 * highest amount of space for randomization available, but that's too hard
39 * to keep straight and caused issues already.
40 */
41 static const unsigned long vaddr_end = CPU_ENTRY_AREA_BASE;
42
43 /*
44 * Memory regions randomized by KASLR (except modules that use a separate logic
45 * earlier during boot). The list is ordered based on virtual addresses. This
46 * order is kept after randomization.
47 */
48 static __initdata struct kaslr_memory_region {
49 unsigned long *base;
50 unsigned long *end;
51 unsigned long size_tb;
52 } kaslr_regions[] = {
53 {
54 .base = &page_offset_base,
55 .end = &physmem_end,
56 },
57 {
58 .base = &vmalloc_base,
59 },
60 {
61 .base = &vmemmap_base,
62 },
63 };
64
65 /* The end of the possible address space for physical memory */
66 unsigned long physmem_end __ro_after_init;
67
68 /* Get size in bytes used by the memory region */
get_padding(struct kaslr_memory_region * region)69 static inline unsigned long get_padding(struct kaslr_memory_region *region)
70 {
71 return (region->size_tb << TB_SHIFT);
72 }
73
74 /* Initialize base and padding for each memory region randomized with KASLR */
kernel_randomize_memory(void)75 void __init kernel_randomize_memory(void)
76 {
77 size_t i;
78 unsigned long vaddr_start, vaddr;
79 unsigned long rand, memory_tb;
80 struct rnd_state rand_state;
81 unsigned long remain_entropy;
82 unsigned long vmemmap_size;
83
84 vaddr_start = pgtable_l5_enabled() ? __PAGE_OFFSET_BASE_L5 : __PAGE_OFFSET_BASE_L4;
85 vaddr = vaddr_start;
86
87 /*
88 * These BUILD_BUG_ON checks ensure the memory layout is consistent
89 * with the vaddr_start/vaddr_end variables. These checks are very
90 * limited....
91 */
92 BUILD_BUG_ON(vaddr_start >= vaddr_end);
93 BUILD_BUG_ON(vaddr_end != CPU_ENTRY_AREA_BASE);
94 BUILD_BUG_ON(vaddr_end > __START_KERNEL_map);
95
96 /* Preset the end of the possible address space for physical memory */
97 physmem_end = ((1ULL << MAX_PHYSMEM_BITS) - 1);
98 if (!kaslr_memory_enabled())
99 return;
100
101 kaslr_regions[0].size_tb = 1 << (MAX_PHYSMEM_BITS - TB_SHIFT);
102 kaslr_regions[1].size_tb = VMALLOC_SIZE_TB;
103
104 /*
105 * Update Physical memory mapping to available and
106 * add padding if needed (especially for memory hotplug support).
107 */
108 BUG_ON(kaslr_regions[0].base != &page_offset_base);
109 memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
110 CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
111
112 /* Adapt physical memory region size based on available memory */
113 if (memory_tb < kaslr_regions[0].size_tb)
114 kaslr_regions[0].size_tb = memory_tb;
115
116 /*
117 * Calculate the vmemmap region size in TBs, aligned to a TB
118 * boundary.
119 */
120 vmemmap_size = (kaslr_regions[0].size_tb << (TB_SHIFT - PAGE_SHIFT)) *
121 sizeof(struct page);
122 kaslr_regions[2].size_tb = DIV_ROUND_UP(vmemmap_size, 1UL << TB_SHIFT);
123
124 /* Calculate entropy available between regions */
125 remain_entropy = vaddr_end - vaddr_start;
126 for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
127 remain_entropy -= get_padding(&kaslr_regions[i]);
128
129 prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
130
131 for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
132 unsigned long entropy;
133
134 /*
135 * Select a random virtual address using the extra entropy
136 * available.
137 */
138 entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
139 prandom_bytes_state(&rand_state, &rand, sizeof(rand));
140 entropy = (rand % (entropy + 1)) & PUD_MASK;
141 vaddr += entropy;
142 *kaslr_regions[i].base = vaddr;
143
144 /* Calculate the end of the region */
145 vaddr += get_padding(&kaslr_regions[i]);
146 /*
147 * KASLR trims the maximum possible size of the
148 * direct-map. Update the physmem_end boundary.
149 * No rounding required as the region starts
150 * PUD aligned and size is in units of TB.
151 */
152 if (kaslr_regions[i].end)
153 *kaslr_regions[i].end = __pa_nodebug(vaddr - 1);
154
155 /* Add a minimum padding based on randomization alignment. */
156 vaddr = round_up(vaddr + 1, PUD_SIZE);
157 remain_entropy -= entropy;
158 }
159 }
160
init_trampoline_kaslr(void)161 void __meminit init_trampoline_kaslr(void)
162 {
163 pud_t *pud_page_tramp, *pud, *pud_tramp;
164 p4d_t *p4d_page_tramp, *p4d, *p4d_tramp;
165 unsigned long paddr, vaddr;
166 pgd_t *pgd;
167
168 pud_page_tramp = alloc_low_page();
169
170 /*
171 * There are two mappings for the low 1MB area, the direct mapping
172 * and the 1:1 mapping for the real mode trampoline:
173 *
174 * Direct mapping: virt_addr = phys_addr + PAGE_OFFSET
175 * 1:1 mapping: virt_addr = phys_addr
176 */
177 paddr = 0;
178 vaddr = (unsigned long)__va(paddr);
179 pgd = pgd_offset_k(vaddr);
180
181 p4d = p4d_offset(pgd, vaddr);
182 pud = pud_offset(p4d, vaddr);
183
184 pud_tramp = pud_page_tramp + pud_index(paddr);
185 *pud_tramp = *pud;
186
187 if (pgtable_l5_enabled()) {
188 p4d_page_tramp = alloc_low_page();
189
190 p4d_tramp = p4d_page_tramp + p4d_index(paddr);
191
192 set_p4d(p4d_tramp,
193 __p4d(_KERNPG_TABLE | __pa(pud_page_tramp)));
194
195 trampoline_pgd_entry =
196 __pgd(_KERNPG_TABLE | __pa(p4d_page_tramp));
197 } else {
198 trampoline_pgd_entry =
199 __pgd(_KERNPG_TABLE | __pa(pud_page_tramp));
200 }
201 }
202