xref: /openbmc/linux/arch/x86/mm/kaslr.c (revision be709d48)
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 
28 #include <asm/pgalloc.h>
29 #include <asm/pgtable.h>
30 #include <asm/setup.h>
31 #include <asm/kaslr.h>
32 
33 #include "mm_internal.h"
34 
35 #define TB_SHIFT 40
36 
37 /*
38  * The end address could depend on more configuration options to make the
39  * highest amount of space for randomization available, but that's too hard
40  * to keep straight and caused issues already.
41  */
42 static const unsigned long vaddr_end = CPU_ENTRY_AREA_BASE;
43 
44 /*
45  * Memory regions randomized by KASLR (except modules that use a separate logic
46  * earlier during boot). The list is ordered based on virtual addresses. This
47  * order is kept after randomization.
48  */
49 static __initdata struct kaslr_memory_region {
50 	unsigned long *base;
51 	unsigned long size_tb;
52 } kaslr_regions[] = {
53 	{ &page_offset_base, 0 },
54 	{ &vmalloc_base, 0 },
55 	{ &vmemmap_base, 1 },
56 };
57 
58 /* Get size in bytes used by the memory region */
59 static inline unsigned long get_padding(struct kaslr_memory_region *region)
60 {
61 	return (region->size_tb << TB_SHIFT);
62 }
63 
64 /*
65  * Apply no randomization if KASLR was disabled at boot or if KASAN
66  * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
67  */
68 static inline bool kaslr_memory_enabled(void)
69 {
70 	return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
71 }
72 
73 /* Initialize base and padding for each memory region randomized with KASLR */
74 void __init kernel_randomize_memory(void)
75 {
76 	size_t i;
77 	unsigned long vaddr_start, vaddr;
78 	unsigned long rand, memory_tb;
79 	struct rnd_state rand_state;
80 	unsigned long remain_entropy;
81 
82 	vaddr_start = pgtable_l5_enabled() ? __PAGE_OFFSET_BASE_L5 : __PAGE_OFFSET_BASE_L4;
83 	vaddr = vaddr_start;
84 
85 	/*
86 	 * These BUILD_BUG_ON checks ensure the memory layout is consistent
87 	 * with the vaddr_start/vaddr_end variables. These checks are very
88 	 * limited....
89 	 */
90 	BUILD_BUG_ON(vaddr_start >= vaddr_end);
91 	BUILD_BUG_ON(vaddr_end != CPU_ENTRY_AREA_BASE);
92 	BUILD_BUG_ON(vaddr_end > __START_KERNEL_map);
93 
94 	if (!kaslr_memory_enabled())
95 		return;
96 
97 	kaslr_regions[0].size_tb = 1 << (__PHYSICAL_MASK_SHIFT - TB_SHIFT);
98 	kaslr_regions[1].size_tb = VMALLOC_SIZE_TB;
99 
100 	/*
101 	 * Update Physical memory mapping to available and
102 	 * add padding if needed (especially for memory hotplug support).
103 	 */
104 	BUG_ON(kaslr_regions[0].base != &page_offset_base);
105 	memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
106 		CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
107 
108 	/* Adapt phyiscal memory region size based on available memory */
109 	if (memory_tb < kaslr_regions[0].size_tb)
110 		kaslr_regions[0].size_tb = memory_tb;
111 
112 	/* Calculate entropy available between regions */
113 	remain_entropy = vaddr_end - vaddr_start;
114 	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
115 		remain_entropy -= get_padding(&kaslr_regions[i]);
116 
117 	prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
118 
119 	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
120 		unsigned long entropy;
121 
122 		/*
123 		 * Select a random virtual address using the extra entropy
124 		 * available.
125 		 */
126 		entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
127 		prandom_bytes_state(&rand_state, &rand, sizeof(rand));
128 		if (pgtable_l5_enabled())
129 			entropy = (rand % (entropy + 1)) & P4D_MASK;
130 		else
131 			entropy = (rand % (entropy + 1)) & PUD_MASK;
132 		vaddr += entropy;
133 		*kaslr_regions[i].base = vaddr;
134 
135 		/*
136 		 * Jump the region and add a minimum padding based on
137 		 * randomization alignment.
138 		 */
139 		vaddr += get_padding(&kaslr_regions[i]);
140 		if (pgtable_l5_enabled())
141 			vaddr = round_up(vaddr + 1, P4D_SIZE);
142 		else
143 			vaddr = round_up(vaddr + 1, PUD_SIZE);
144 		remain_entropy -= entropy;
145 	}
146 }
147 
148 static void __meminit init_trampoline_pud(void)
149 {
150 	unsigned long paddr, paddr_next;
151 	pgd_t *pgd;
152 	pud_t *pud_page, *pud_page_tramp;
153 	int i;
154 
155 	pud_page_tramp = alloc_low_page();
156 
157 	paddr = 0;
158 	pgd = pgd_offset_k((unsigned long)__va(paddr));
159 	pud_page = (pud_t *) pgd_page_vaddr(*pgd);
160 
161 	for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
162 		pud_t *pud, *pud_tramp;
163 		unsigned long vaddr = (unsigned long)__va(paddr);
164 
165 		pud_tramp = pud_page_tramp + pud_index(paddr);
166 		pud = pud_page + pud_index(vaddr);
167 		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
168 
169 		*pud_tramp = *pud;
170 	}
171 
172 	set_pgd(&trampoline_pgd_entry,
173 		__pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
174 }
175 
176 static void __meminit init_trampoline_p4d(void)
177 {
178 	unsigned long paddr, paddr_next;
179 	pgd_t *pgd;
180 	p4d_t *p4d_page, *p4d_page_tramp;
181 	int i;
182 
183 	p4d_page_tramp = alloc_low_page();
184 
185 	paddr = 0;
186 	pgd = pgd_offset_k((unsigned long)__va(paddr));
187 	p4d_page = (p4d_t *) pgd_page_vaddr(*pgd);
188 
189 	for (i = p4d_index(paddr); i < PTRS_PER_P4D; i++, paddr = paddr_next) {
190 		p4d_t *p4d, *p4d_tramp;
191 		unsigned long vaddr = (unsigned long)__va(paddr);
192 
193 		p4d_tramp = p4d_page_tramp + p4d_index(paddr);
194 		p4d = p4d_page + p4d_index(vaddr);
195 		paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
196 
197 		*p4d_tramp = *p4d;
198 	}
199 
200 	set_pgd(&trampoline_pgd_entry,
201 		__pgd(_KERNPG_TABLE | __pa(p4d_page_tramp)));
202 }
203 
204 /*
205  * Create PGD aligned trampoline table to allow real mode initialization
206  * of additional CPUs. Consume only 1 low memory page.
207  */
208 void __meminit init_trampoline(void)
209 {
210 
211 	if (!kaslr_memory_enabled()) {
212 		init_trampoline_default();
213 		return;
214 	}
215 
216 	if (pgtable_l5_enabled())
217 		init_trampoline_p4d();
218 	else
219 		init_trampoline_pud();
220 }
221