1 // SPDX-License-Identifier: GPL-2.0 2 #define DISABLE_BRANCH_PROFILING 3 #define pr_fmt(fmt) "kasan: " fmt 4 #include <linux/bootmem.h> 5 #include <linux/kasan.h> 6 #include <linux/kdebug.h> 7 #include <linux/memblock.h> 8 #include <linux/mm.h> 9 #include <linux/sched.h> 10 #include <linux/sched/task.h> 11 #include <linux/vmalloc.h> 12 13 #include <asm/e820/types.h> 14 #include <asm/pgalloc.h> 15 #include <asm/tlbflush.h> 16 #include <asm/sections.h> 17 #include <asm/pgtable.h> 18 #include <asm/cpu_entry_area.h> 19 20 extern struct range pfn_mapped[E820_MAX_ENTRIES]; 21 22 static p4d_t tmp_p4d_table[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 23 24 static __init void *early_alloc(size_t size, int nid) 25 { 26 return memblock_virt_alloc_try_nid_nopanic(size, size, 27 __pa(MAX_DMA_ADDRESS), BOOTMEM_ALLOC_ACCESSIBLE, nid); 28 } 29 30 static void __init kasan_populate_pmd(pmd_t *pmd, unsigned long addr, 31 unsigned long end, int nid) 32 { 33 pte_t *pte; 34 35 if (pmd_none(*pmd)) { 36 void *p; 37 38 if (boot_cpu_has(X86_FEATURE_PSE) && 39 ((end - addr) == PMD_SIZE) && 40 IS_ALIGNED(addr, PMD_SIZE)) { 41 p = early_alloc(PMD_SIZE, nid); 42 if (p && pmd_set_huge(pmd, __pa(p), PAGE_KERNEL)) 43 return; 44 else if (p) 45 memblock_free(__pa(p), PMD_SIZE); 46 } 47 48 p = early_alloc(PAGE_SIZE, nid); 49 pmd_populate_kernel(&init_mm, pmd, p); 50 } 51 52 pte = pte_offset_kernel(pmd, addr); 53 do { 54 pte_t entry; 55 void *p; 56 57 if (!pte_none(*pte)) 58 continue; 59 60 p = early_alloc(PAGE_SIZE, nid); 61 entry = pfn_pte(PFN_DOWN(__pa(p)), PAGE_KERNEL); 62 set_pte_at(&init_mm, addr, pte, entry); 63 } while (pte++, addr += PAGE_SIZE, addr != end); 64 } 65 66 static void __init kasan_populate_pud(pud_t *pud, unsigned long addr, 67 unsigned long end, int nid) 68 { 69 pmd_t *pmd; 70 unsigned long next; 71 72 if (pud_none(*pud)) { 73 void *p; 74 75 if (boot_cpu_has(X86_FEATURE_GBPAGES) && 76 ((end - addr) == PUD_SIZE) && 77 IS_ALIGNED(addr, PUD_SIZE)) { 78 p = early_alloc(PUD_SIZE, nid); 79 if (p && pud_set_huge(pud, __pa(p), PAGE_KERNEL)) 80 return; 81 else if (p) 82 memblock_free(__pa(p), PUD_SIZE); 83 } 84 85 p = early_alloc(PAGE_SIZE, nid); 86 pud_populate(&init_mm, pud, p); 87 } 88 89 pmd = pmd_offset(pud, addr); 90 do { 91 next = pmd_addr_end(addr, end); 92 if (!pmd_large(*pmd)) 93 kasan_populate_pmd(pmd, addr, next, nid); 94 } while (pmd++, addr = next, addr != end); 95 } 96 97 static void __init kasan_populate_p4d(p4d_t *p4d, unsigned long addr, 98 unsigned long end, int nid) 99 { 100 pud_t *pud; 101 unsigned long next; 102 103 if (p4d_none(*p4d)) { 104 void *p = early_alloc(PAGE_SIZE, nid); 105 106 p4d_populate(&init_mm, p4d, p); 107 } 108 109 pud = pud_offset(p4d, addr); 110 do { 111 next = pud_addr_end(addr, end); 112 if (!pud_large(*pud)) 113 kasan_populate_pud(pud, addr, next, nid); 114 } while (pud++, addr = next, addr != end); 115 } 116 117 static void __init kasan_populate_pgd(pgd_t *pgd, unsigned long addr, 118 unsigned long end, int nid) 119 { 120 void *p; 121 p4d_t *p4d; 122 unsigned long next; 123 124 if (pgd_none(*pgd)) { 125 p = early_alloc(PAGE_SIZE, nid); 126 pgd_populate(&init_mm, pgd, p); 127 } 128 129 p4d = p4d_offset(pgd, addr); 130 do { 131 next = p4d_addr_end(addr, end); 132 kasan_populate_p4d(p4d, addr, next, nid); 133 } while (p4d++, addr = next, addr != end); 134 } 135 136 static void __init kasan_populate_shadow(unsigned long addr, unsigned long end, 137 int nid) 138 { 139 pgd_t *pgd; 140 unsigned long next; 141 142 addr = addr & PAGE_MASK; 143 end = round_up(end, PAGE_SIZE); 144 pgd = pgd_offset_k(addr); 145 do { 146 next = pgd_addr_end(addr, end); 147 kasan_populate_pgd(pgd, addr, next, nid); 148 } while (pgd++, addr = next, addr != end); 149 } 150 151 static void __init map_range(struct range *range) 152 { 153 unsigned long start; 154 unsigned long end; 155 156 start = (unsigned long)kasan_mem_to_shadow(pfn_to_kaddr(range->start)); 157 end = (unsigned long)kasan_mem_to_shadow(pfn_to_kaddr(range->end)); 158 159 kasan_populate_shadow(start, end, early_pfn_to_nid(range->start)); 160 } 161 162 static void __init clear_pgds(unsigned long start, 163 unsigned long end) 164 { 165 pgd_t *pgd; 166 /* See comment in kasan_init() */ 167 unsigned long pgd_end = end & PGDIR_MASK; 168 169 for (; start < pgd_end; start += PGDIR_SIZE) { 170 pgd = pgd_offset_k(start); 171 /* 172 * With folded p4d, pgd_clear() is nop, use p4d_clear() 173 * instead. 174 */ 175 if (CONFIG_PGTABLE_LEVELS < 5) 176 p4d_clear(p4d_offset(pgd, start)); 177 else 178 pgd_clear(pgd); 179 } 180 181 pgd = pgd_offset_k(start); 182 for (; start < end; start += P4D_SIZE) 183 p4d_clear(p4d_offset(pgd, start)); 184 } 185 186 static inline p4d_t *early_p4d_offset(pgd_t *pgd, unsigned long addr) 187 { 188 unsigned long p4d; 189 190 if (!IS_ENABLED(CONFIG_X86_5LEVEL)) 191 return (p4d_t *)pgd; 192 193 p4d = __pa_nodebug(pgd_val(*pgd)) & PTE_PFN_MASK; 194 p4d += __START_KERNEL_map - phys_base; 195 return (p4d_t *)p4d + p4d_index(addr); 196 } 197 198 static void __init kasan_early_p4d_populate(pgd_t *pgd, 199 unsigned long addr, 200 unsigned long end) 201 { 202 pgd_t pgd_entry; 203 p4d_t *p4d, p4d_entry; 204 unsigned long next; 205 206 if (pgd_none(*pgd)) { 207 pgd_entry = __pgd(_KERNPG_TABLE | __pa_nodebug(kasan_zero_p4d)); 208 set_pgd(pgd, pgd_entry); 209 } 210 211 p4d = early_p4d_offset(pgd, addr); 212 do { 213 next = p4d_addr_end(addr, end); 214 215 if (!p4d_none(*p4d)) 216 continue; 217 218 p4d_entry = __p4d(_KERNPG_TABLE | __pa_nodebug(kasan_zero_pud)); 219 set_p4d(p4d, p4d_entry); 220 } while (p4d++, addr = next, addr != end && p4d_none(*p4d)); 221 } 222 223 static void __init kasan_map_early_shadow(pgd_t *pgd) 224 { 225 /* See comment in kasan_init() */ 226 unsigned long addr = KASAN_SHADOW_START & PGDIR_MASK; 227 unsigned long end = KASAN_SHADOW_END; 228 unsigned long next; 229 230 pgd += pgd_index(addr); 231 do { 232 next = pgd_addr_end(addr, end); 233 kasan_early_p4d_populate(pgd, addr, next); 234 } while (pgd++, addr = next, addr != end); 235 } 236 237 #ifdef CONFIG_KASAN_INLINE 238 static int kasan_die_handler(struct notifier_block *self, 239 unsigned long val, 240 void *data) 241 { 242 if (val == DIE_GPF) { 243 pr_emerg("CONFIG_KASAN_INLINE enabled\n"); 244 pr_emerg("GPF could be caused by NULL-ptr deref or user memory access\n"); 245 } 246 return NOTIFY_OK; 247 } 248 249 static struct notifier_block kasan_die_notifier = { 250 .notifier_call = kasan_die_handler, 251 }; 252 #endif 253 254 void __init kasan_early_init(void) 255 { 256 int i; 257 pteval_t pte_val = __pa_nodebug(kasan_zero_page) | __PAGE_KERNEL | _PAGE_ENC; 258 pmdval_t pmd_val = __pa_nodebug(kasan_zero_pte) | _KERNPG_TABLE; 259 pudval_t pud_val = __pa_nodebug(kasan_zero_pmd) | _KERNPG_TABLE; 260 p4dval_t p4d_val = __pa_nodebug(kasan_zero_pud) | _KERNPG_TABLE; 261 262 for (i = 0; i < PTRS_PER_PTE; i++) 263 kasan_zero_pte[i] = __pte(pte_val); 264 265 for (i = 0; i < PTRS_PER_PMD; i++) 266 kasan_zero_pmd[i] = __pmd(pmd_val); 267 268 for (i = 0; i < PTRS_PER_PUD; i++) 269 kasan_zero_pud[i] = __pud(pud_val); 270 271 for (i = 0; IS_ENABLED(CONFIG_X86_5LEVEL) && i < PTRS_PER_P4D; i++) 272 kasan_zero_p4d[i] = __p4d(p4d_val); 273 274 kasan_map_early_shadow(early_top_pgt); 275 kasan_map_early_shadow(init_top_pgt); 276 } 277 278 void __init kasan_init(void) 279 { 280 int i; 281 void *shadow_cpu_entry_begin, *shadow_cpu_entry_end; 282 283 #ifdef CONFIG_KASAN_INLINE 284 register_die_notifier(&kasan_die_notifier); 285 #endif 286 287 memcpy(early_top_pgt, init_top_pgt, sizeof(early_top_pgt)); 288 289 /* 290 * We use the same shadow offset for 4- and 5-level paging to 291 * facilitate boot-time switching between paging modes. 292 * As result in 5-level paging mode KASAN_SHADOW_START and 293 * KASAN_SHADOW_END are not aligned to PGD boundary. 294 * 295 * KASAN_SHADOW_START doesn't share PGD with anything else. 296 * We claim whole PGD entry to make things easier. 297 * 298 * KASAN_SHADOW_END lands in the last PGD entry and it collides with 299 * bunch of things like kernel code, modules, EFI mapping, etc. 300 * We need to take extra steps to not overwrite them. 301 */ 302 if (IS_ENABLED(CONFIG_X86_5LEVEL)) { 303 void *ptr; 304 305 ptr = (void *)pgd_page_vaddr(*pgd_offset_k(KASAN_SHADOW_END)); 306 memcpy(tmp_p4d_table, (void *)ptr, sizeof(tmp_p4d_table)); 307 set_pgd(&early_top_pgt[pgd_index(KASAN_SHADOW_END)], 308 __pgd(__pa(tmp_p4d_table) | _KERNPG_TABLE)); 309 } 310 311 load_cr3(early_top_pgt); 312 __flush_tlb_all(); 313 314 clear_pgds(KASAN_SHADOW_START & PGDIR_MASK, KASAN_SHADOW_END); 315 316 kasan_populate_zero_shadow((void *)(KASAN_SHADOW_START & PGDIR_MASK), 317 kasan_mem_to_shadow((void *)PAGE_OFFSET)); 318 319 for (i = 0; i < E820_MAX_ENTRIES; i++) { 320 if (pfn_mapped[i].end == 0) 321 break; 322 323 map_range(&pfn_mapped[i]); 324 } 325 326 shadow_cpu_entry_begin = (void *)CPU_ENTRY_AREA_BASE; 327 shadow_cpu_entry_begin = kasan_mem_to_shadow(shadow_cpu_entry_begin); 328 shadow_cpu_entry_begin = (void *)round_down((unsigned long)shadow_cpu_entry_begin, 329 PAGE_SIZE); 330 331 shadow_cpu_entry_end = (void *)(CPU_ENTRY_AREA_BASE + 332 CPU_ENTRY_AREA_MAP_SIZE); 333 shadow_cpu_entry_end = kasan_mem_to_shadow(shadow_cpu_entry_end); 334 shadow_cpu_entry_end = (void *)round_up((unsigned long)shadow_cpu_entry_end, 335 PAGE_SIZE); 336 337 kasan_populate_zero_shadow( 338 kasan_mem_to_shadow((void *)PAGE_OFFSET + MAXMEM), 339 shadow_cpu_entry_begin); 340 341 kasan_populate_shadow((unsigned long)shadow_cpu_entry_begin, 342 (unsigned long)shadow_cpu_entry_end, 0); 343 344 kasan_populate_zero_shadow(shadow_cpu_entry_end, 345 kasan_mem_to_shadow((void *)__START_KERNEL_map)); 346 347 kasan_populate_shadow((unsigned long)kasan_mem_to_shadow(_stext), 348 (unsigned long)kasan_mem_to_shadow(_end), 349 early_pfn_to_nid(__pa(_stext))); 350 351 kasan_populate_zero_shadow(kasan_mem_to_shadow((void *)MODULES_END), 352 (void *)KASAN_SHADOW_END); 353 354 load_cr3(init_top_pgt); 355 __flush_tlb_all(); 356 357 /* 358 * kasan_zero_page has been used as early shadow memory, thus it may 359 * contain some garbage. Now we can clear and write protect it, since 360 * after the TLB flush no one should write to it. 361 */ 362 memset(kasan_zero_page, 0, PAGE_SIZE); 363 for (i = 0; i < PTRS_PER_PTE; i++) { 364 pte_t pte = __pte(__pa(kasan_zero_page) | __PAGE_KERNEL_RO | _PAGE_ENC); 365 set_pte(&kasan_zero_pte[i], pte); 366 } 367 /* Flush TLBs again to be sure that write protection applied. */ 368 __flush_tlb_all(); 369 370 init_task.kasan_depth = 0; 371 pr_info("KernelAddressSanitizer initialized\n"); 372 } 373