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  * 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 size_tb;
  51 } kaslr_regions[] = {
  52         { &page_offset_base, 0 },
  53         { &vmalloc_base, 0 },
  54         { &vmemmap_base, 1 },
  55 };
  56
  57 /* Get size in bytes used by the memory region */
  58 static inline unsigned long get_padding(struct kaslr_memory_region *region)
  59 {
  60         return (region->size_tb << TB_SHIFT);
  61 }
  62
  63 /*
  64  * Apply no randomization if KASLR was disabled at boot or if KASAN
  65  * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
  66  */
  67 static inline bool kaslr_memory_enabled(void)
  68 {
  69         return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
  70 }
  71
  72 /* Initialize base and padding for each memory region randomized with KASLR */
  73 void __init kernel_randomize_memory(void)
  74 {
  75         size_t i;
  76         unsigned long vaddr_start, vaddr;
  77         unsigned long rand, memory_tb;
  78         struct rnd_state rand_state;
  79         unsigned long remain_entropy;
  80
  81         vaddr_start = pgtable_l5_enabled ? __PAGE_OFFSET_BASE_L5 : __PAGE_OFFSET_BASE_L4;
  82         vaddr = vaddr_start;
  83
  84         /*
  85          * These BUILD_BUG_ON checks ensure the memory layout is consistent
  86          * with the vaddr_start/vaddr_end variables. These checks are very
  87          * limited....
  88          */
  89         BUILD_BUG_ON(vaddr_start >= vaddr_end);
  90         BUILD_BUG_ON(vaddr_end != CPU_ENTRY_AREA_BASE);
  91         BUILD_BUG_ON(vaddr_end > __START_KERNEL_map);
  92
  93         if (!kaslr_memory_enabled())
  94                 return;
  95
  96         kaslr_regions[0].size_tb = 1 << (__PHYSICAL_MASK_SHIFT - TB_SHIFT);
  97         kaslr_regions[1].size_tb = VMALLOC_SIZE_TB;
  98
  99         /*
 100          * Update Physical memory mapping to available and
 101          * add padding if needed (especially for memory hotplug support).
 102          */
 103         BUG_ON(kaslr_regions[0].base != &page_offset_base);
 104         memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
 105                 CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
 106
 107         /* Adapt phyiscal memory region size based on available memory */
 108         if (memory_tb < kaslr_regions[0].size_tb)
 109                 kaslr_regions[0].size_tb = memory_tb;
 110
 111         /* Calculate entropy available between regions */
 112         remain_entropy = vaddr_end - vaddr_start;
 113         for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
 114                 remain_entropy -= get_padding(&kaslr_regions[i]);
 115
 116         prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
 117
 118         for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
 119                 unsigned long entropy;
 120
 121                 /*
 122                  * Select a random virtual address using the extra entropy
 123                  * available.
 124                  */
 125                 entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
 126                 prandom_bytes_state(&rand_state, &rand, sizeof(rand));
 127                 if (IS_ENABLED(CONFIG_X86_5LEVEL))
 128                         entropy = (rand % (entropy + 1)) & P4D_MASK;
 129                 else
 130                         entropy = (rand % (entropy + 1)) & PUD_MASK;
 131                 vaddr += entropy;
 132                 *kaslr_regions[i].base = vaddr;
 133
 134                 /*
 135                  * Jump the region and add a minimum padding based on
 136                  * randomization alignment.
 137                  */
 138                 vaddr += get_padding(&kaslr_regions[i]);
 139                 if (IS_ENABLED(CONFIG_X86_5LEVEL))
 140                         vaddr = round_up(vaddr + 1, P4D_SIZE);
 141                 else
 142                         vaddr = round_up(vaddr + 1, PUD_SIZE);
 143                 remain_entropy -= entropy;
 144         }
 145 }
 146
 147 static void __meminit init_trampoline_pud(void)
 148 {
 149         unsigned long paddr, paddr_next;
 150         pgd_t *pgd;
 151         pud_t *pud_page, *pud_page_tramp;
 152         int i;
 153
 154         pud_page_tramp = alloc_low_page();
 155
 156         paddr = 0;
 157         pgd = pgd_offset_k((unsigned long)__va(paddr));
 158         pud_page = (pud_t *) pgd_page_vaddr(*pgd);
 159
 160         for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
 161                 pud_t *pud, *pud_tramp;
 162                 unsigned long vaddr = (unsigned long)__va(paddr);
 163
 164                 pud_tramp = pud_page_tramp + pud_index(paddr);
 165                 pud = pud_page + pud_index(vaddr);
 166                 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
 167
 168                 *pud_tramp = *pud;
 169         }
 170
 171         set_pgd(&trampoline_pgd_entry,
 172                 __pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
 173 }
 174
 175 static void __meminit init_trampoline_p4d(void)
 176 {
 177         unsigned long paddr, paddr_next;
 178         pgd_t *pgd;
 179         p4d_t *p4d_page, *p4d_page_tramp;
 180         int i;
 181
 182         p4d_page_tramp = alloc_low_page();
 183
 184         paddr = 0;
 185         pgd = pgd_offset_k((unsigned long)__va(paddr));
 186         p4d_page = (p4d_t *) pgd_page_vaddr(*pgd);
 187
 188         for (i = p4d_index(paddr); i < PTRS_PER_P4D; i++, paddr = paddr_next) {
 189                 p4d_t *p4d, *p4d_tramp;
 190                 unsigned long vaddr = (unsigned long)__va(paddr);
 191
 192                 p4d_tramp = p4d_page_tramp + p4d_index(paddr);
 193                 p4d = p4d_page + p4d_index(vaddr);
 194                 paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
 195
 196                 *p4d_tramp = *p4d;
 197         }
 198
 199         set_pgd(&trampoline_pgd_entry,
 200                 __pgd(_KERNPG_TABLE | __pa(p4d_page_tramp)));
 201 }
 202
 203 /*
 204  * Create PGD aligned trampoline table to allow real mode initialization
 205  * of additional CPUs. Consume only 1 low memory page.
 206  */
 207 void __meminit init_trampoline(void)
 208 {
 209
 210         if (!kaslr_memory_enabled()) {
 211                 init_trampoline_default();
 212                 return;
 213         }
 214
 215         if (IS_ENABLED(CONFIG_X86_5LEVEL))
 216                 init_trampoline_p4d();
 217         else
 218                 init_trampoline_pud();
 219 }