xref: /openbmc/linux/arch/x86/mm/kaslr.c (revision db65c49e)
1 /*
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
43  * before. You also need to add a BUILD_BUG_ON in kernel_randomize_memory to
44  * ensure that this order is correct and won't be changed.
45  */
46 static const unsigned long vaddr_start = __PAGE_OFFSET_BASE;
47 static const unsigned long vaddr_end = VMEMMAP_START;
48 
49 /* Default values */
50 unsigned long page_offset_base = __PAGE_OFFSET_BASE;
51 EXPORT_SYMBOL(page_offset_base);
52 unsigned long vmalloc_base = __VMALLOC_BASE;
53 EXPORT_SYMBOL(vmalloc_base);
54 
55 /*
56  * Memory regions randomized by KASLR (except modules that use a separate logic
57  * earlier during boot). The list is ordered based on virtual addresses. This
58  * order is kept after randomization.
59  */
60 static __initdata struct kaslr_memory_region {
61 	unsigned long *base;
62 	unsigned long size_tb;
63 } kaslr_regions[] = {
64 	{ &page_offset_base, 64/* Maximum */ },
65 	{ &vmalloc_base, VMALLOC_SIZE_TB },
66 };
67 
68 /* Get size in bytes used by the memory region */
69 static inline unsigned long get_padding(struct kaslr_memory_region *region)
70 {
71 	return (region->size_tb << TB_SHIFT);
72 }
73 
74 /*
75  * Apply no randomization if KASLR was disabled at boot or if KASAN
76  * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
77  */
78 static inline bool kaslr_memory_enabled(void)
79 {
80 	return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
81 }
82 
83 /* Initialize base and padding for each memory region randomized with KASLR */
84 void __init kernel_randomize_memory(void)
85 {
86 	size_t i;
87 	unsigned long vaddr = vaddr_start;
88 	unsigned long rand, memory_tb;
89 	struct rnd_state rand_state;
90 	unsigned long remain_entropy;
91 
92 	if (!kaslr_memory_enabled())
93 		return;
94 
95 	/*
96 	 * Update Physical memory mapping to available and
97 	 * add padding if needed (especially for memory hotplug support).
98 	 */
99 	BUG_ON(kaslr_regions[0].base != &page_offset_base);
100 	memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
101 		CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
102 
103 	/* Adapt phyiscal memory region size based on available memory */
104 	if (memory_tb < kaslr_regions[0].size_tb)
105 		kaslr_regions[0].size_tb = memory_tb;
106 
107 	/* Calculate entropy available between regions */
108 	remain_entropy = vaddr_end - vaddr_start;
109 	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
110 		remain_entropy -= get_padding(&kaslr_regions[i]);
111 
112 	prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
113 
114 	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
115 		unsigned long entropy;
116 
117 		/*
118 		 * Select a random virtual address using the extra entropy
119 		 * available.
120 		 */
121 		entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
122 		prandom_bytes_state(&rand_state, &rand, sizeof(rand));
123 		entropy = (rand % (entropy + 1)) & PUD_MASK;
124 		vaddr += entropy;
125 		*kaslr_regions[i].base = vaddr;
126 
127 		/*
128 		 * Jump the region and add a minimum padding based on
129 		 * randomization alignment.
130 		 */
131 		vaddr += get_padding(&kaslr_regions[i]);
132 		vaddr = round_up(vaddr + 1, PUD_SIZE);
133 		remain_entropy -= entropy;
134 	}
135 }
136 
137 /*
138  * Create PGD aligned trampoline table to allow real mode initialization
139  * of additional CPUs. Consume only 1 low memory page.
140  */
141 void __meminit init_trampoline(void)
142 {
143 	unsigned long paddr, paddr_next;
144 	pgd_t *pgd;
145 	pud_t *pud_page, *pud_page_tramp;
146 	int i;
147 
148 	if (!kaslr_memory_enabled()) {
149 		init_trampoline_default();
150 		return;
151 	}
152 
153 	pud_page_tramp = alloc_low_page();
154 
155 	paddr = 0;
156 	pgd = pgd_offset_k((unsigned long)__va(paddr));
157 	pud_page = (pud_t *) pgd_page_vaddr(*pgd);
158 
159 	for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
160 		pud_t *pud, *pud_tramp;
161 		unsigned long vaddr = (unsigned long)__va(paddr);
162 
163 		pud_tramp = pud_page_tramp + pud_index(paddr);
164 		pud = pud_page + pud_index(vaddr);
165 		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
166 
167 		*pud_tramp = *pud;
168 	}
169 
170 	set_pgd(&trampoline_pgd_entry,
171 		__pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
172 }
173