1 /* 2 * Debug helper to dump the current kernel pagetables of the system 3 * so that we can see what the various memory ranges are set to. 4 * 5 * (C) Copyright 2008 Intel Corporation 6 * 7 * Author: Arjan van de Ven <arjan@linux.intel.com> 8 * 9 * This program is free software; you can redistribute it and/or 10 * modify it under the terms of the GNU General Public License 11 * as published by the Free Software Foundation; version 2 12 * of the License. 13 */ 14 15 #include <linux/debugfs.h> 16 #include <linux/mm.h> 17 #include <linux/init.h> 18 #include <linux/sched.h> 19 #include <linux/seq_file.h> 20 21 #include <asm/kasan.h> 22 #include <asm/pgtable.h> 23 24 /* 25 * The dumper groups pagetable entries of the same type into one, and for 26 * that it needs to keep some state when walking, and flush this state 27 * when a "break" in the continuity is found. 28 */ 29 struct pg_state { 30 int level; 31 pgprot_t current_prot; 32 unsigned long start_address; 33 unsigned long current_address; 34 const struct addr_marker *marker; 35 unsigned long lines; 36 bool to_dmesg; 37 bool check_wx; 38 unsigned long wx_pages; 39 }; 40 41 struct addr_marker { 42 unsigned long start_address; 43 const char *name; 44 unsigned long max_lines; 45 }; 46 47 /* indices for address_markers; keep sync'd w/ address_markers below */ 48 enum address_markers_idx { 49 USER_SPACE_NR = 0, 50 #ifdef CONFIG_X86_64 51 KERNEL_SPACE_NR, 52 LOW_KERNEL_NR, 53 VMALLOC_START_NR, 54 VMEMMAP_START_NR, 55 #ifdef CONFIG_KASAN 56 KASAN_SHADOW_START_NR, 57 KASAN_SHADOW_END_NR, 58 #endif 59 # ifdef CONFIG_X86_ESPFIX64 60 ESPFIX_START_NR, 61 # endif 62 HIGH_KERNEL_NR, 63 MODULES_VADDR_NR, 64 MODULES_END_NR, 65 #else 66 KERNEL_SPACE_NR, 67 VMALLOC_START_NR, 68 VMALLOC_END_NR, 69 # ifdef CONFIG_HIGHMEM 70 PKMAP_BASE_NR, 71 # endif 72 FIXADDR_START_NR, 73 #endif 74 }; 75 76 /* Address space markers hints */ 77 static struct addr_marker address_markers[] = { 78 { 0, "User Space" }, 79 #ifdef CONFIG_X86_64 80 { 0x8000000000000000UL, "Kernel Space" }, 81 { 0/* PAGE_OFFSET */, "Low Kernel Mapping" }, 82 { 0/* VMALLOC_START */, "vmalloc() Area" }, 83 { 0/* VMEMMAP_START */, "Vmemmap" }, 84 #ifdef CONFIG_KASAN 85 { KASAN_SHADOW_START, "KASAN shadow" }, 86 { KASAN_SHADOW_END, "KASAN shadow end" }, 87 #endif 88 # ifdef CONFIG_X86_ESPFIX64 89 { ESPFIX_BASE_ADDR, "ESPfix Area", 16 }, 90 # endif 91 # ifdef CONFIG_EFI 92 { EFI_VA_END, "EFI Runtime Services" }, 93 # endif 94 { __START_KERNEL_map, "High Kernel Mapping" }, 95 { MODULES_VADDR, "Modules" }, 96 { MODULES_END, "End Modules" }, 97 #else 98 { PAGE_OFFSET, "Kernel Mapping" }, 99 { 0/* VMALLOC_START */, "vmalloc() Area" }, 100 { 0/*VMALLOC_END*/, "vmalloc() End" }, 101 # ifdef CONFIG_HIGHMEM 102 { 0/*PKMAP_BASE*/, "Persistent kmap() Area" }, 103 # endif 104 { 0/*FIXADDR_START*/, "Fixmap Area" }, 105 #endif 106 { -1, NULL } /* End of list */ 107 }; 108 109 /* Multipliers for offsets within the PTEs */ 110 #define PTE_LEVEL_MULT (PAGE_SIZE) 111 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT) 112 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT) 113 #define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT) 114 115 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \ 116 ({ \ 117 if (to_dmesg) \ 118 printk(KERN_INFO fmt, ##args); \ 119 else \ 120 if (m) \ 121 seq_printf(m, fmt, ##args); \ 122 }) 123 124 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \ 125 ({ \ 126 if (to_dmesg) \ 127 printk(KERN_CONT fmt, ##args); \ 128 else \ 129 if (m) \ 130 seq_printf(m, fmt, ##args); \ 131 }) 132 133 /* 134 * Print a readable form of a pgprot_t to the seq_file 135 */ 136 static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg) 137 { 138 pgprotval_t pr = pgprot_val(prot); 139 static const char * const level_name[] = 140 { "cr3", "pgd", "pud", "pmd", "pte" }; 141 142 if (!pgprot_val(prot)) { 143 /* Not present */ 144 pt_dump_cont_printf(m, dmsg, " "); 145 } else { 146 if (pr & _PAGE_USER) 147 pt_dump_cont_printf(m, dmsg, "USR "); 148 else 149 pt_dump_cont_printf(m, dmsg, " "); 150 if (pr & _PAGE_RW) 151 pt_dump_cont_printf(m, dmsg, "RW "); 152 else 153 pt_dump_cont_printf(m, dmsg, "ro "); 154 if (pr & _PAGE_PWT) 155 pt_dump_cont_printf(m, dmsg, "PWT "); 156 else 157 pt_dump_cont_printf(m, dmsg, " "); 158 if (pr & _PAGE_PCD) 159 pt_dump_cont_printf(m, dmsg, "PCD "); 160 else 161 pt_dump_cont_printf(m, dmsg, " "); 162 163 /* Bit 7 has a different meaning on level 3 vs 4 */ 164 if (level <= 3 && pr & _PAGE_PSE) 165 pt_dump_cont_printf(m, dmsg, "PSE "); 166 else 167 pt_dump_cont_printf(m, dmsg, " "); 168 if ((level == 4 && pr & _PAGE_PAT) || 169 ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE)) 170 pt_dump_cont_printf(m, dmsg, "PAT "); 171 else 172 pt_dump_cont_printf(m, dmsg, " "); 173 if (pr & _PAGE_GLOBAL) 174 pt_dump_cont_printf(m, dmsg, "GLB "); 175 else 176 pt_dump_cont_printf(m, dmsg, " "); 177 if (pr & _PAGE_NX) 178 pt_dump_cont_printf(m, dmsg, "NX "); 179 else 180 pt_dump_cont_printf(m, dmsg, "x "); 181 } 182 pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]); 183 } 184 185 /* 186 * On 64 bits, sign-extend the 48 bit address to 64 bit 187 */ 188 static unsigned long normalize_addr(unsigned long u) 189 { 190 #ifdef CONFIG_X86_64 191 return (signed long)(u << 16) >> 16; 192 #else 193 return u; 194 #endif 195 } 196 197 /* 198 * This function gets called on a break in a continuous series 199 * of PTE entries; the next one is different so we need to 200 * print what we collected so far. 201 */ 202 static void note_page(struct seq_file *m, struct pg_state *st, 203 pgprot_t new_prot, int level) 204 { 205 pgprotval_t prot, cur; 206 static const char units[] = "BKMGTPE"; 207 208 /* 209 * If we have a "break" in the series, we need to flush the state that 210 * we have now. "break" is either changing perms, levels or 211 * address space marker. 212 */ 213 prot = pgprot_val(new_prot); 214 cur = pgprot_val(st->current_prot); 215 216 if (!st->level) { 217 /* First entry */ 218 st->current_prot = new_prot; 219 st->level = level; 220 st->marker = address_markers; 221 st->lines = 0; 222 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n", 223 st->marker->name); 224 } else if (prot != cur || level != st->level || 225 st->current_address >= st->marker[1].start_address) { 226 const char *unit = units; 227 unsigned long delta; 228 int width = sizeof(unsigned long) * 2; 229 pgprotval_t pr = pgprot_val(st->current_prot); 230 231 if (st->check_wx && (pr & _PAGE_RW) && !(pr & _PAGE_NX)) { 232 WARN_ONCE(1, 233 "x86/mm: Found insecure W+X mapping at address %p/%pS\n", 234 (void *)st->start_address, 235 (void *)st->start_address); 236 st->wx_pages += (st->current_address - 237 st->start_address) / PAGE_SIZE; 238 } 239 240 /* 241 * Now print the actual finished series 242 */ 243 if (!st->marker->max_lines || 244 st->lines < st->marker->max_lines) { 245 pt_dump_seq_printf(m, st->to_dmesg, 246 "0x%0*lx-0x%0*lx ", 247 width, st->start_address, 248 width, st->current_address); 249 250 delta = st->current_address - st->start_address; 251 while (!(delta & 1023) && unit[1]) { 252 delta >>= 10; 253 unit++; 254 } 255 pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", 256 delta, *unit); 257 printk_prot(m, st->current_prot, st->level, 258 st->to_dmesg); 259 } 260 st->lines++; 261 262 /* 263 * We print markers for special areas of address space, 264 * such as the start of vmalloc space etc. 265 * This helps in the interpretation. 266 */ 267 if (st->current_address >= st->marker[1].start_address) { 268 if (st->marker->max_lines && 269 st->lines > st->marker->max_lines) { 270 unsigned long nskip = 271 st->lines - st->marker->max_lines; 272 pt_dump_seq_printf(m, st->to_dmesg, 273 "... %lu entr%s skipped ... \n", 274 nskip, 275 nskip == 1 ? "y" : "ies"); 276 } 277 st->marker++; 278 st->lines = 0; 279 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n", 280 st->marker->name); 281 } 282 283 st->start_address = st->current_address; 284 st->current_prot = new_prot; 285 st->level = level; 286 } 287 } 288 289 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr, 290 unsigned long P) 291 { 292 int i; 293 pte_t *start; 294 pgprotval_t prot; 295 296 start = (pte_t *) pmd_page_vaddr(addr); 297 for (i = 0; i < PTRS_PER_PTE; i++) { 298 prot = pte_flags(*start); 299 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT); 300 note_page(m, st, __pgprot(prot), 4); 301 start++; 302 } 303 } 304 305 #if PTRS_PER_PMD > 1 306 307 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr, 308 unsigned long P) 309 { 310 int i; 311 pmd_t *start; 312 pgprotval_t prot; 313 314 start = (pmd_t *) pud_page_vaddr(addr); 315 for (i = 0; i < PTRS_PER_PMD; i++) { 316 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT); 317 if (!pmd_none(*start)) { 318 if (pmd_large(*start) || !pmd_present(*start)) { 319 prot = pmd_flags(*start); 320 note_page(m, st, __pgprot(prot), 3); 321 } else { 322 walk_pte_level(m, st, *start, 323 P + i * PMD_LEVEL_MULT); 324 } 325 } else 326 note_page(m, st, __pgprot(0), 3); 327 start++; 328 } 329 } 330 331 #else 332 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p) 333 #define pud_large(a) pmd_large(__pmd(pud_val(a))) 334 #define pud_none(a) pmd_none(__pmd(pud_val(a))) 335 #endif 336 337 #if PTRS_PER_PUD > 1 338 339 /* 340 * This is an optimization for CONFIG_DEBUG_WX=y + CONFIG_KASAN=y 341 * KASAN fills page tables with the same values. Since there is no 342 * point in checking page table more than once we just skip repeated 343 * entries. This saves us dozens of seconds during boot. 344 */ 345 static bool pud_already_checked(pud_t *prev_pud, pud_t *pud, bool checkwx) 346 { 347 return checkwx && prev_pud && (pud_val(*prev_pud) == pud_val(*pud)); 348 } 349 350 static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr, 351 unsigned long P) 352 { 353 int i; 354 pud_t *start; 355 pgprotval_t prot; 356 pud_t *prev_pud = NULL; 357 358 start = (pud_t *) pgd_page_vaddr(addr); 359 360 for (i = 0; i < PTRS_PER_PUD; i++) { 361 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT); 362 if (!pud_none(*start) && 363 !pud_already_checked(prev_pud, start, st->check_wx)) { 364 if (pud_large(*start) || !pud_present(*start)) { 365 prot = pud_flags(*start); 366 note_page(m, st, __pgprot(prot), 2); 367 } else { 368 walk_pmd_level(m, st, *start, 369 P + i * PUD_LEVEL_MULT); 370 } 371 } else 372 note_page(m, st, __pgprot(0), 2); 373 374 prev_pud = start; 375 start++; 376 } 377 } 378 379 #else 380 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p) 381 #define pgd_large(a) pud_large(__pud(pgd_val(a))) 382 #define pgd_none(a) pud_none(__pud(pgd_val(a))) 383 #endif 384 385 static inline bool is_hypervisor_range(int idx) 386 { 387 #ifdef CONFIG_X86_64 388 /* 389 * ffff800000000000 - ffff87ffffffffff is reserved for 390 * the hypervisor. 391 */ 392 return (idx >= pgd_index(__PAGE_OFFSET) - 16) && 393 (idx < pgd_index(__PAGE_OFFSET)); 394 #else 395 return false; 396 #endif 397 } 398 399 static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd, 400 bool checkwx) 401 { 402 #ifdef CONFIG_X86_64 403 pgd_t *start = (pgd_t *) &init_level4_pgt; 404 #else 405 pgd_t *start = swapper_pg_dir; 406 #endif 407 pgprotval_t prot; 408 int i; 409 struct pg_state st = {}; 410 411 if (pgd) { 412 start = pgd; 413 st.to_dmesg = true; 414 } 415 416 st.check_wx = checkwx; 417 if (checkwx) 418 st.wx_pages = 0; 419 420 for (i = 0; i < PTRS_PER_PGD; i++) { 421 st.current_address = normalize_addr(i * PGD_LEVEL_MULT); 422 if (!pgd_none(*start) && !is_hypervisor_range(i)) { 423 if (pgd_large(*start) || !pgd_present(*start)) { 424 prot = pgd_flags(*start); 425 note_page(m, &st, __pgprot(prot), 1); 426 } else { 427 walk_pud_level(m, &st, *start, 428 i * PGD_LEVEL_MULT); 429 } 430 } else 431 note_page(m, &st, __pgprot(0), 1); 432 433 cond_resched(); 434 start++; 435 } 436 437 /* Flush out the last page */ 438 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT); 439 note_page(m, &st, __pgprot(0), 0); 440 if (!checkwx) 441 return; 442 if (st.wx_pages) 443 pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n", 444 st.wx_pages); 445 else 446 pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n"); 447 } 448 449 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd) 450 { 451 ptdump_walk_pgd_level_core(m, pgd, false); 452 } 453 EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level); 454 455 void ptdump_walk_pgd_level_checkwx(void) 456 { 457 ptdump_walk_pgd_level_core(NULL, NULL, true); 458 } 459 460 static int __init pt_dump_init(void) 461 { 462 /* 463 * Various markers are not compile-time constants, so assign them 464 * here. 465 */ 466 #ifdef CONFIG_X86_64 467 address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET; 468 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START; 469 address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START; 470 #endif 471 #ifdef CONFIG_X86_32 472 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START; 473 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END; 474 # ifdef CONFIG_HIGHMEM 475 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE; 476 # endif 477 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START; 478 #endif 479 480 return 0; 481 } 482 __initcall(pt_dump_init); 483