1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * kaslr.c 4 * 5 * This contains the routines needed to generate a reasonable level of 6 * entropy to choose a randomized kernel base address offset in support 7 * of Kernel Address Space Layout Randomization (KASLR). Additionally 8 * handles walking the physical memory maps (and tracking memory regions 9 * to avoid) in order to select a physical memory location that can 10 * contain the entire properly aligned running kernel image. 11 * 12 */ 13 14 /* 15 * isspace() in linux/ctype.h is expected by next_args() to filter 16 * out "space/lf/tab". While boot/ctype.h conflicts with linux/ctype.h, 17 * since isdigit() is implemented in both of them. Hence disable it 18 * here. 19 */ 20 #define BOOT_CTYPE_H 21 22 /* 23 * _ctype[] in lib/ctype.c is needed by isspace() of linux/ctype.h. 24 * While both lib/ctype.c and lib/cmdline.c will bring EXPORT_SYMBOL 25 * which is meaningless and will cause compiling error in some cases. 26 * So do not include linux/export.h and define EXPORT_SYMBOL(sym) 27 * as empty. 28 */ 29 #define _LINUX_EXPORT_H 30 #define EXPORT_SYMBOL(sym) 31 32 #include "misc.h" 33 #include "error.h" 34 #include "../string.h" 35 36 #include <generated/compile.h> 37 #include <linux/module.h> 38 #include <linux/uts.h> 39 #include <linux/utsname.h> 40 #include <linux/ctype.h> 41 #include <linux/efi.h> 42 #include <generated/utsrelease.h> 43 #include <asm/efi.h> 44 45 /* Macros used by the included decompressor code below. */ 46 #define STATIC 47 #include <linux/decompress/mm.h> 48 49 extern unsigned long get_cmd_line_ptr(void); 50 51 /* Simplified build-specific string for starting entropy. */ 52 static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@" 53 LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION; 54 55 static unsigned long rotate_xor(unsigned long hash, const void *area, 56 size_t size) 57 { 58 size_t i; 59 unsigned long *ptr = (unsigned long *)area; 60 61 for (i = 0; i < size / sizeof(hash); i++) { 62 /* Rotate by odd number of bits and XOR. */ 63 hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7); 64 hash ^= ptr[i]; 65 } 66 67 return hash; 68 } 69 70 /* Attempt to create a simple but unpredictable starting entropy. */ 71 static unsigned long get_boot_seed(void) 72 { 73 unsigned long hash = 0; 74 75 hash = rotate_xor(hash, build_str, sizeof(build_str)); 76 hash = rotate_xor(hash, boot_params, sizeof(*boot_params)); 77 78 return hash; 79 } 80 81 #define KASLR_COMPRESSED_BOOT 82 #include "../../lib/kaslr.c" 83 84 struct mem_vector { 85 unsigned long long start; 86 unsigned long long size; 87 }; 88 89 /* Only supporting at most 4 unusable memmap regions with kaslr */ 90 #define MAX_MEMMAP_REGIONS 4 91 92 static bool memmap_too_large; 93 94 95 /* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */ 96 unsigned long long mem_limit = ULLONG_MAX; 97 98 99 enum mem_avoid_index { 100 MEM_AVOID_ZO_RANGE = 0, 101 MEM_AVOID_INITRD, 102 MEM_AVOID_CMDLINE, 103 MEM_AVOID_BOOTPARAMS, 104 MEM_AVOID_MEMMAP_BEGIN, 105 MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1, 106 MEM_AVOID_MAX, 107 }; 108 109 static struct mem_vector mem_avoid[MEM_AVOID_MAX]; 110 111 static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two) 112 { 113 /* Item one is entirely before item two. */ 114 if (one->start + one->size <= two->start) 115 return false; 116 /* Item one is entirely after item two. */ 117 if (one->start >= two->start + two->size) 118 return false; 119 return true; 120 } 121 122 char *skip_spaces(const char *str) 123 { 124 while (isspace(*str)) 125 ++str; 126 return (char *)str; 127 } 128 #include "../../../../lib/ctype.c" 129 #include "../../../../lib/cmdline.c" 130 131 static int 132 parse_memmap(char *p, unsigned long long *start, unsigned long long *size) 133 { 134 char *oldp; 135 136 if (!p) 137 return -EINVAL; 138 139 /* We don't care about this option here */ 140 if (!strncmp(p, "exactmap", 8)) 141 return -EINVAL; 142 143 oldp = p; 144 *size = memparse(p, &p); 145 if (p == oldp) 146 return -EINVAL; 147 148 switch (*p) { 149 case '#': 150 case '$': 151 case '!': 152 *start = memparse(p + 1, &p); 153 return 0; 154 case '@': 155 /* memmap=nn@ss specifies usable region, should be skipped */ 156 *size = 0; 157 /* Fall through */ 158 default: 159 /* 160 * If w/o offset, only size specified, memmap=nn[KMG] has the 161 * same behaviour as mem=nn[KMG]. It limits the max address 162 * system can use. Region above the limit should be avoided. 163 */ 164 *start = 0; 165 return 0; 166 } 167 168 return -EINVAL; 169 } 170 171 static void mem_avoid_memmap(char *str) 172 { 173 static int i; 174 175 if (i >= MAX_MEMMAP_REGIONS) 176 return; 177 178 while (str && (i < MAX_MEMMAP_REGIONS)) { 179 int rc; 180 unsigned long long start, size; 181 char *k = strchr(str, ','); 182 183 if (k) 184 *k++ = 0; 185 186 rc = parse_memmap(str, &start, &size); 187 if (rc < 0) 188 break; 189 str = k; 190 191 if (start == 0) { 192 /* Store the specified memory limit if size > 0 */ 193 if (size > 0) 194 mem_limit = size; 195 196 continue; 197 } 198 199 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start; 200 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size; 201 i++; 202 } 203 204 /* More than 4 memmaps, fail kaslr */ 205 if ((i >= MAX_MEMMAP_REGIONS) && str) 206 memmap_too_large = true; 207 } 208 209 static int handle_mem_memmap(void) 210 { 211 char *args = (char *)get_cmd_line_ptr(); 212 size_t len = strlen((char *)args); 213 char *tmp_cmdline; 214 char *param, *val; 215 u64 mem_size; 216 217 if (!strstr(args, "memmap=") && !strstr(args, "mem=")) 218 return 0; 219 220 tmp_cmdline = malloc(len + 1); 221 if (!tmp_cmdline) 222 error("Failed to allocate space for tmp_cmdline"); 223 224 memcpy(tmp_cmdline, args, len); 225 tmp_cmdline[len] = 0; 226 args = tmp_cmdline; 227 228 /* Chew leading spaces */ 229 args = skip_spaces(args); 230 231 while (*args) { 232 args = next_arg(args, ¶m, &val); 233 /* Stop at -- */ 234 if (!val && strcmp(param, "--") == 0) { 235 warn("Only '--' specified in cmdline"); 236 free(tmp_cmdline); 237 return -1; 238 } 239 240 if (!strcmp(param, "memmap")) { 241 mem_avoid_memmap(val); 242 } else if (!strcmp(param, "mem")) { 243 char *p = val; 244 245 if (!strcmp(p, "nopentium")) 246 continue; 247 mem_size = memparse(p, &p); 248 if (mem_size == 0) { 249 free(tmp_cmdline); 250 return -EINVAL; 251 } 252 mem_limit = mem_size; 253 } 254 } 255 256 free(tmp_cmdline); 257 return 0; 258 } 259 260 /* 261 * In theory, KASLR can put the kernel anywhere in the range of [16M, 64T). 262 * The mem_avoid array is used to store the ranges that need to be avoided 263 * when KASLR searches for an appropriate random address. We must avoid any 264 * regions that are unsafe to overlap with during decompression, and other 265 * things like the initrd, cmdline and boot_params. This comment seeks to 266 * explain mem_avoid as clearly as possible since incorrect mem_avoid 267 * memory ranges lead to really hard to debug boot failures. 268 * 269 * The initrd, cmdline, and boot_params are trivial to identify for 270 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and 271 * MEM_AVOID_BOOTPARAMS respectively below. 272 * 273 * What is not obvious how to avoid is the range of memory that is used 274 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover 275 * the compressed kernel (ZO) and its run space, which is used to extract 276 * the uncompressed kernel (VO) and relocs. 277 * 278 * ZO's full run size sits against the end of the decompression buffer, so 279 * we can calculate where text, data, bss, etc of ZO are positioned more 280 * easily. 281 * 282 * For additional background, the decompression calculations can be found 283 * in header.S, and the memory diagram is based on the one found in misc.c. 284 * 285 * The following conditions are already enforced by the image layouts and 286 * associated code: 287 * - input + input_size >= output + output_size 288 * - kernel_total_size <= init_size 289 * - kernel_total_size <= output_size (see Note below) 290 * - output + init_size >= output + output_size 291 * 292 * (Note that kernel_total_size and output_size have no fundamental 293 * relationship, but output_size is passed to choose_random_location 294 * as a maximum of the two. The diagram is showing a case where 295 * kernel_total_size is larger than output_size, but this case is 296 * handled by bumping output_size.) 297 * 298 * The above conditions can be illustrated by a diagram: 299 * 300 * 0 output input input+input_size output+init_size 301 * | | | | | 302 * | | | | | 303 * |-----|--------|--------|--------------|-----------|--|-------------| 304 * | | | 305 * | | | 306 * output+init_size-ZO_INIT_SIZE output+output_size output+kernel_total_size 307 * 308 * [output, output+init_size) is the entire memory range used for 309 * extracting the compressed image. 310 * 311 * [output, output+kernel_total_size) is the range needed for the 312 * uncompressed kernel (VO) and its run size (bss, brk, etc). 313 * 314 * [output, output+output_size) is VO plus relocs (i.e. the entire 315 * uncompressed payload contained by ZO). This is the area of the buffer 316 * written to during decompression. 317 * 318 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case 319 * range of the copied ZO and decompression code. (i.e. the range 320 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.) 321 * 322 * [input, input+input_size) is the original copied compressed image (ZO) 323 * (i.e. it does not include its run size). This range must be avoided 324 * because it contains the data used for decompression. 325 * 326 * [input+input_size, output+init_size) is [_text, _end) for ZO. This 327 * range includes ZO's heap and stack, and must be avoided since it 328 * performs the decompression. 329 * 330 * Since the above two ranges need to be avoided and they are adjacent, 331 * they can be merged, resulting in: [input, output+init_size) which 332 * becomes the MEM_AVOID_ZO_RANGE below. 333 */ 334 static void mem_avoid_init(unsigned long input, unsigned long input_size, 335 unsigned long output) 336 { 337 unsigned long init_size = boot_params->hdr.init_size; 338 u64 initrd_start, initrd_size; 339 u64 cmd_line, cmd_line_size; 340 char *ptr; 341 342 /* 343 * Avoid the region that is unsafe to overlap during 344 * decompression. 345 */ 346 mem_avoid[MEM_AVOID_ZO_RANGE].start = input; 347 mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input; 348 add_identity_map(mem_avoid[MEM_AVOID_ZO_RANGE].start, 349 mem_avoid[MEM_AVOID_ZO_RANGE].size); 350 351 /* Avoid initrd. */ 352 initrd_start = (u64)boot_params->ext_ramdisk_image << 32; 353 initrd_start |= boot_params->hdr.ramdisk_image; 354 initrd_size = (u64)boot_params->ext_ramdisk_size << 32; 355 initrd_size |= boot_params->hdr.ramdisk_size; 356 mem_avoid[MEM_AVOID_INITRD].start = initrd_start; 357 mem_avoid[MEM_AVOID_INITRD].size = initrd_size; 358 /* No need to set mapping for initrd, it will be handled in VO. */ 359 360 /* Avoid kernel command line. */ 361 cmd_line = (u64)boot_params->ext_cmd_line_ptr << 32; 362 cmd_line |= boot_params->hdr.cmd_line_ptr; 363 /* Calculate size of cmd_line. */ 364 ptr = (char *)(unsigned long)cmd_line; 365 for (cmd_line_size = 0; ptr[cmd_line_size++];) 366 ; 367 mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line; 368 mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size; 369 add_identity_map(mem_avoid[MEM_AVOID_CMDLINE].start, 370 mem_avoid[MEM_AVOID_CMDLINE].size); 371 372 /* Avoid boot parameters. */ 373 mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params; 374 mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params); 375 add_identity_map(mem_avoid[MEM_AVOID_BOOTPARAMS].start, 376 mem_avoid[MEM_AVOID_BOOTPARAMS].size); 377 378 /* We don't need to set a mapping for setup_data. */ 379 380 /* Mark the memmap regions we need to avoid */ 381 handle_mem_memmap(); 382 383 #ifdef CONFIG_X86_VERBOSE_BOOTUP 384 /* Make sure video RAM can be used. */ 385 add_identity_map(0, PMD_SIZE); 386 #endif 387 } 388 389 /* 390 * Does this memory vector overlap a known avoided area? If so, record the 391 * overlap region with the lowest address. 392 */ 393 static bool mem_avoid_overlap(struct mem_vector *img, 394 struct mem_vector *overlap) 395 { 396 int i; 397 struct setup_data *ptr; 398 unsigned long earliest = img->start + img->size; 399 bool is_overlapping = false; 400 401 for (i = 0; i < MEM_AVOID_MAX; i++) { 402 if (mem_overlaps(img, &mem_avoid[i]) && 403 mem_avoid[i].start < earliest) { 404 *overlap = mem_avoid[i]; 405 earliest = overlap->start; 406 is_overlapping = true; 407 } 408 } 409 410 /* Avoid all entries in the setup_data linked list. */ 411 ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data; 412 while (ptr) { 413 struct mem_vector avoid; 414 415 avoid.start = (unsigned long)ptr; 416 avoid.size = sizeof(*ptr) + ptr->len; 417 418 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) { 419 *overlap = avoid; 420 earliest = overlap->start; 421 is_overlapping = true; 422 } 423 424 ptr = (struct setup_data *)(unsigned long)ptr->next; 425 } 426 427 return is_overlapping; 428 } 429 430 struct slot_area { 431 unsigned long addr; 432 int num; 433 }; 434 435 #define MAX_SLOT_AREA 100 436 437 static struct slot_area slot_areas[MAX_SLOT_AREA]; 438 439 static unsigned long slot_max; 440 441 static unsigned long slot_area_index; 442 443 static void store_slot_info(struct mem_vector *region, unsigned long image_size) 444 { 445 struct slot_area slot_area; 446 447 if (slot_area_index == MAX_SLOT_AREA) 448 return; 449 450 slot_area.addr = region->start; 451 slot_area.num = (region->size - image_size) / 452 CONFIG_PHYSICAL_ALIGN + 1; 453 454 if (slot_area.num > 0) { 455 slot_areas[slot_area_index++] = slot_area; 456 slot_max += slot_area.num; 457 } 458 } 459 460 static unsigned long slots_fetch_random(void) 461 { 462 unsigned long slot; 463 int i; 464 465 /* Handle case of no slots stored. */ 466 if (slot_max == 0) 467 return 0; 468 469 slot = kaslr_get_random_long("Physical") % slot_max; 470 471 for (i = 0; i < slot_area_index; i++) { 472 if (slot >= slot_areas[i].num) { 473 slot -= slot_areas[i].num; 474 continue; 475 } 476 return slot_areas[i].addr + slot * CONFIG_PHYSICAL_ALIGN; 477 } 478 479 if (i == slot_area_index) 480 debug_putstr("slots_fetch_random() failed!?\n"); 481 return 0; 482 } 483 484 static void process_mem_region(struct mem_vector *entry, 485 unsigned long minimum, 486 unsigned long image_size) 487 { 488 struct mem_vector region, overlap; 489 struct slot_area slot_area; 490 unsigned long start_orig, end; 491 struct mem_vector cur_entry; 492 493 /* On 32-bit, ignore entries entirely above our maximum. */ 494 if (IS_ENABLED(CONFIG_X86_32) && entry->start >= KERNEL_IMAGE_SIZE) 495 return; 496 497 /* Ignore entries entirely below our minimum. */ 498 if (entry->start + entry->size < minimum) 499 return; 500 501 /* Ignore entries above memory limit */ 502 end = min(entry->size + entry->start, mem_limit); 503 if (entry->start >= end) 504 return; 505 cur_entry.start = entry->start; 506 cur_entry.size = end - entry->start; 507 508 region.start = cur_entry.start; 509 region.size = cur_entry.size; 510 511 /* Give up if slot area array is full. */ 512 while (slot_area_index < MAX_SLOT_AREA) { 513 start_orig = region.start; 514 515 /* Potentially raise address to minimum location. */ 516 if (region.start < minimum) 517 region.start = minimum; 518 519 /* Potentially raise address to meet alignment needs. */ 520 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN); 521 522 /* Did we raise the address above the passed in memory entry? */ 523 if (region.start > cur_entry.start + cur_entry.size) 524 return; 525 526 /* Reduce size by any delta from the original address. */ 527 region.size -= region.start - start_orig; 528 529 /* On 32-bit, reduce region size to fit within max size. */ 530 if (IS_ENABLED(CONFIG_X86_32) && 531 region.start + region.size > KERNEL_IMAGE_SIZE) 532 region.size = KERNEL_IMAGE_SIZE - region.start; 533 534 /* Return if region can't contain decompressed kernel */ 535 if (region.size < image_size) 536 return; 537 538 /* If nothing overlaps, store the region and return. */ 539 if (!mem_avoid_overlap(®ion, &overlap)) { 540 store_slot_info(®ion, image_size); 541 return; 542 } 543 544 /* Store beginning of region if holds at least image_size. */ 545 if (overlap.start > region.start + image_size) { 546 struct mem_vector beginning; 547 548 beginning.start = region.start; 549 beginning.size = overlap.start - region.start; 550 store_slot_info(&beginning, image_size); 551 } 552 553 /* Return if overlap extends to or past end of region. */ 554 if (overlap.start + overlap.size >= region.start + region.size) 555 return; 556 557 /* Clip off the overlapping region and start over. */ 558 region.size -= overlap.start - region.start + overlap.size; 559 region.start = overlap.start + overlap.size; 560 } 561 } 562 563 #ifdef CONFIG_EFI 564 /* 565 * Returns true if mirror region found (and must have been processed 566 * for slots adding) 567 */ 568 static bool 569 process_efi_entries(unsigned long minimum, unsigned long image_size) 570 { 571 struct efi_info *e = &boot_params->efi_info; 572 bool efi_mirror_found = false; 573 struct mem_vector region; 574 efi_memory_desc_t *md; 575 unsigned long pmap; 576 char *signature; 577 u32 nr_desc; 578 int i; 579 580 signature = (char *)&e->efi_loader_signature; 581 if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) && 582 strncmp(signature, EFI64_LOADER_SIGNATURE, 4)) 583 return false; 584 585 #ifdef CONFIG_X86_32 586 /* Can't handle data above 4GB at this time */ 587 if (e->efi_memmap_hi) { 588 warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n"); 589 return false; 590 } 591 pmap = e->efi_memmap; 592 #else 593 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32)); 594 #endif 595 596 nr_desc = e->efi_memmap_size / e->efi_memdesc_size; 597 for (i = 0; i < nr_desc; i++) { 598 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i); 599 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) { 600 efi_mirror_found = true; 601 break; 602 } 603 } 604 605 for (i = 0; i < nr_desc; i++) { 606 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i); 607 608 /* 609 * Here we are more conservative in picking free memory than 610 * the EFI spec allows: 611 * 612 * According to the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also 613 * free memory and thus available to place the kernel image into, 614 * but in practice there's firmware where using that memory leads 615 * to crashes. 616 * 617 * Only EFI_CONVENTIONAL_MEMORY is guaranteed to be free. 618 */ 619 if (md->type != EFI_CONVENTIONAL_MEMORY) 620 continue; 621 622 if (efi_mirror_found && 623 !(md->attribute & EFI_MEMORY_MORE_RELIABLE)) 624 continue; 625 626 region.start = md->phys_addr; 627 region.size = md->num_pages << EFI_PAGE_SHIFT; 628 process_mem_region(®ion, minimum, image_size); 629 if (slot_area_index == MAX_SLOT_AREA) { 630 debug_putstr("Aborted EFI scan (slot_areas full)!\n"); 631 break; 632 } 633 } 634 return true; 635 } 636 #else 637 static inline bool 638 process_efi_entries(unsigned long minimum, unsigned long image_size) 639 { 640 return false; 641 } 642 #endif 643 644 static void process_e820_entries(unsigned long minimum, 645 unsigned long image_size) 646 { 647 int i; 648 struct mem_vector region; 649 struct boot_e820_entry *entry; 650 651 /* Verify potential e820 positions, appending to slots list. */ 652 for (i = 0; i < boot_params->e820_entries; i++) { 653 entry = &boot_params->e820_table[i]; 654 /* Skip non-RAM entries. */ 655 if (entry->type != E820_TYPE_RAM) 656 continue; 657 region.start = entry->addr; 658 region.size = entry->size; 659 process_mem_region(®ion, minimum, image_size); 660 if (slot_area_index == MAX_SLOT_AREA) { 661 debug_putstr("Aborted e820 scan (slot_areas full)!\n"); 662 break; 663 } 664 } 665 } 666 667 static unsigned long find_random_phys_addr(unsigned long minimum, 668 unsigned long image_size) 669 { 670 /* Check if we had too many memmaps. */ 671 if (memmap_too_large) { 672 debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n"); 673 return 0; 674 } 675 676 /* Make sure minimum is aligned. */ 677 minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN); 678 679 if (process_efi_entries(minimum, image_size)) 680 return slots_fetch_random(); 681 682 process_e820_entries(minimum, image_size); 683 return slots_fetch_random(); 684 } 685 686 static unsigned long find_random_virt_addr(unsigned long minimum, 687 unsigned long image_size) 688 { 689 unsigned long slots, random_addr; 690 691 /* Make sure minimum is aligned. */ 692 minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN); 693 /* Align image_size for easy slot calculations. */ 694 image_size = ALIGN(image_size, CONFIG_PHYSICAL_ALIGN); 695 696 /* 697 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots 698 * that can hold image_size within the range of minimum to 699 * KERNEL_IMAGE_SIZE? 700 */ 701 slots = (KERNEL_IMAGE_SIZE - minimum - image_size) / 702 CONFIG_PHYSICAL_ALIGN + 1; 703 704 random_addr = kaslr_get_random_long("Virtual") % slots; 705 706 return random_addr * CONFIG_PHYSICAL_ALIGN + minimum; 707 } 708 709 /* 710 * Since this function examines addresses much more numerically, 711 * it takes the input and output pointers as 'unsigned long'. 712 */ 713 void choose_random_location(unsigned long input, 714 unsigned long input_size, 715 unsigned long *output, 716 unsigned long output_size, 717 unsigned long *virt_addr) 718 { 719 unsigned long random_addr, min_addr; 720 721 if (cmdline_find_option_bool("nokaslr")) { 722 warn("KASLR disabled: 'nokaslr' on cmdline."); 723 return; 724 } 725 726 boot_params->hdr.loadflags |= KASLR_FLAG; 727 728 /* Prepare to add new identity pagetables on demand. */ 729 initialize_identity_maps(); 730 731 /* Record the various known unsafe memory ranges. */ 732 mem_avoid_init(input, input_size, *output); 733 734 /* 735 * Low end of the randomization range should be the 736 * smaller of 512M or the initial kernel image 737 * location: 738 */ 739 min_addr = min(*output, 512UL << 20); 740 741 /* Walk available memory entries to find a random address. */ 742 random_addr = find_random_phys_addr(min_addr, output_size); 743 if (!random_addr) { 744 warn("Physical KASLR disabled: no suitable memory region!"); 745 } else { 746 /* Update the new physical address location. */ 747 if (*output != random_addr) { 748 add_identity_map(random_addr, output_size); 749 *output = random_addr; 750 } 751 752 /* 753 * This loads the identity mapping page table. 754 * This should only be done if a new physical address 755 * is found for the kernel, otherwise we should keep 756 * the old page table to make it be like the "nokaslr" 757 * case. 758 */ 759 finalize_identity_maps(); 760 } 761 762 763 /* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */ 764 if (IS_ENABLED(CONFIG_X86_64)) 765 random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size); 766 *virt_addr = random_addr; 767 } 768