arch/x86/mm/kaslr.c

   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
  27 #include <asm/pgalloc.h>
  28 #include <asm/pgtable.h>
  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  * Virtual address start and end range for randomization. The end changes base
  38  * on configuration to have the highest amount of space for randomization.
  39  * It increases the possible random position for each randomized region.
  40  *
  41  * You need to add an if/def entry if you introduce a new memory region
  42  * compatible with KASLR. Your entry must be in logical order with memory
  43  * layout. For example, ESPFIX is before EFI because its virtual address is
  44  * before. You also need to add a BUILD_BUG_ON() in kernel_randomize_memory() to
  45  * ensure that this order is correct and won't be changed.
  46  */
  47 static const unsigned long vaddr_start = __PAGE_OFFSET_BASE;
  48
  49 #if defined(CONFIG_X86_ESPFIX64)
  50 static const unsigned long vaddr_end = ESPFIX_BASE_ADDR;
  51 #elif defined(CONFIG_EFI)
  52 static const unsigned long vaddr_end = EFI_VA_END;
  53 #else
  54 static const unsigned long vaddr_end = __START_KERNEL_map;
  55 #endif
  56
  57 /* Default values */
  58 unsigned long page_offset_base = __PAGE_OFFSET_BASE;
  59 EXPORT_SYMBOL(page_offset_base);
  60 unsigned long vmalloc_base = __VMALLOC_BASE;
  61 EXPORT_SYMBOL(vmalloc_base);
  62 unsigned long vmemmap_base = __VMEMMAP_BASE;
  63 EXPORT_SYMBOL(vmemmap_base);
  64
  65 /*
  66  * Memory regions randomized by KASLR (except modules that use a separate logic
  67  * earlier during boot). The list is ordered based on virtual addresses. This
  68  * order is kept after randomization.
  69  */
  70 static __initdata struct kaslr_memory_region {
  71         unsigned long *base;
  72         unsigned long size_tb;
  73 } kaslr_regions[] = {
  74         { &page_offset_base, 1 << (__PHYSICAL_MASK_SHIFT - TB_SHIFT) /* Maximum */ },
  75         { &vmalloc_base, VMALLOC_SIZE_TB },
  76         { &vmemmap_base, 1 },
  77 };
  78
  79 /* Get size in bytes used by the memory region */
  80 static inline unsigned long get_padding(struct kaslr_memory_region *region)
  81 {
  82         return (region->size_tb << TB_SHIFT);
  83 }
  84
  85 /*
  86  * Apply no randomization if KASLR was disabled at boot or if KASAN
  87  * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
  88  */
  89 static inline bool kaslr_memory_enabled(void)
  90 {
  91         return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
  92 }
  93
  94 /* Initialize base and padding for each memory region randomized with KASLR */
  95 void __init kernel_randomize_memory(void)
  96 {
  97         size_t i;
  98         unsigned long vaddr = vaddr_start;
  99         unsigned long rand, memory_tb;
 100         struct rnd_state rand_state;
 101         unsigned long remain_entropy;
 102
 103         /*
 104          * All these BUILD_BUG_ON checks ensures the memory layout is
 105          * consistent with the vaddr_start/vaddr_end variables.
 106          */
 107         BUILD_BUG_ON(vaddr_start >= vaddr_end);
 108         BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_ESPFIX64) &&
 109                      vaddr_end >= EFI_VA_END);
 110         BUILD_BUG_ON((IS_ENABLED(CONFIG_X86_ESPFIX64) ||
 111                       IS_ENABLED(CONFIG_EFI)) &&
 112                      vaddr_end >= __START_KERNEL_map);
 113         BUILD_BUG_ON(vaddr_end > __START_KERNEL_map);
 114
 115         if (!kaslr_memory_enabled())
 116                 return;
 117
 118         /*
 119          * Update Physical memory mapping to available and
 120          * add padding if needed (especially for memory hotplug support).
 121          */
 122         BUG_ON(kaslr_regions[0].base != &page_offset_base);
 123         memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
 124                 CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
 125
 126         /* Adapt phyiscal memory region size based on available memory */
 127         if (memory_tb < kaslr_regions[0].size_tb)
 128                 kaslr_regions[0].size_tb = memory_tb;
 129
 130         /* Calculate entropy available between regions */
 131         remain_entropy = vaddr_end - vaddr_start;
 132         for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
 133                 remain_entropy -= get_padding(&kaslr_regions[i]);
 134
 135         prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
 136
 137         for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
 138                 unsigned long entropy;
 139
 140                 /*
 141                  * Select a random virtual address using the extra entropy
 142                  * available.
 143                  */
 144                 entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
 145                 prandom_bytes_state(&rand_state, &rand, sizeof(rand));
 146                 if (IS_ENABLED(CONFIG_X86_5LEVEL))
 147                         entropy = (rand % (entropy + 1)) & P4D_MASK;
 148                 else
 149                         entropy = (rand % (entropy + 1)) & PUD_MASK;
 150                 vaddr += entropy;
 151                 *kaslr_regions[i].base = vaddr;
 152
 153                 /*
 154                  * Jump the region and add a minimum padding based on
 155                  * randomization alignment.
 156                  */
 157                 vaddr += get_padding(&kaslr_regions[i]);
 158                 if (IS_ENABLED(CONFIG_X86_5LEVEL))
 159                         vaddr = round_up(vaddr + 1, P4D_SIZE);
 160                 else
 161                         vaddr = round_up(vaddr + 1, PUD_SIZE);
 162                 remain_entropy -= entropy;
 163         }
 164 }
 165
 166 static void __meminit init_trampoline_pud(void)
 167 {
 168         unsigned long paddr, paddr_next;
 169         pgd_t *pgd;
 170         pud_t *pud_page, *pud_page_tramp;
 171         int i;
 172
 173         pud_page_tramp = alloc_low_page();
 174
 175         paddr = 0;
 176         pgd = pgd_offset_k((unsigned long)__va(paddr));
 177         pud_page = (pud_t *) pgd_page_vaddr(*pgd);
 178
 179         for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
 180                 pud_t *pud, *pud_tramp;
 181                 unsigned long vaddr = (unsigned long)__va(paddr);
 182
 183                 pud_tramp = pud_page_tramp + pud_index(paddr);
 184                 pud = pud_page + pud_index(vaddr);
 185                 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
 186
 187                 *pud_tramp = *pud;
 188         }
 189
 190         set_pgd(&trampoline_pgd_entry,
 191                 __pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
 192 }
 193
 194 static void __meminit init_trampoline_p4d(void)
 195 {
 196         unsigned long paddr, paddr_next;
 197         pgd_t *pgd;
 198         p4d_t *p4d_page, *p4d_page_tramp;
 199         int i;
 200
 201         p4d_page_tramp = alloc_low_page();
 202
 203         paddr = 0;
 204         pgd = pgd_offset_k((unsigned long)__va(paddr));
 205         p4d_page = (p4d_t *) pgd_page_vaddr(*pgd);
 206
 207         for (i = p4d_index(paddr); i < PTRS_PER_P4D; i++, paddr = paddr_next) {
 208                 p4d_t *p4d, *p4d_tramp;
 209                 unsigned long vaddr = (unsigned long)__va(paddr);
 210
 211                 p4d_tramp = p4d_page_tramp + p4d_index(paddr);
 212                 p4d = p4d_page + p4d_index(vaddr);
 213                 paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
 214
 215                 *p4d_tramp = *p4d;
 216         }
 217
 218         set_pgd(&trampoline_pgd_entry,
 219                 __pgd(_KERNPG_TABLE | __pa(p4d_page_tramp)));
 220 }
 221
 222 /*
 223  * Create PGD aligned trampoline table to allow real mode initialization
 224  * of additional CPUs. Consume only 1 low memory page.
 225  */
 226 void __meminit init_trampoline(void)
 227 {
 228
 229         if (!kaslr_memory_enabled()) {
 230                 init_trampoline_default();
 231                 return;
 232         }
 233
 234         if (IS_ENABLED(CONFIG_X86_5LEVEL))
 235                 init_trampoline_p4d();
 236         else
 237                 init_trampoline_pud();
 238 }