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#include "misc.h" 23#include "error.h" 24#include "../string.h" 25 26#include <generated/compile.h> 27#include <linux/module.h> 28#include <linux/uts.h> 29#include <linux/utsname.h> 30#include <linux/ctype.h> 31#include <linux/efi.h> 32#include <generated/utsrelease.h> 33#include <asm/efi.h> 34 35/* Macros used by the included decompressor code below. */ 36#define STATIC 37#include <linux/decompress/mm.h> 38 39#define _SETUP 40#include <asm/setup.h> /* For COMMAND_LINE_SIZE */ 41#undef _SETUP 42 43extern unsigned long get_cmd_line_ptr(void); 44 45/* Simplified build-specific string for starting entropy. */ 46static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@" 47 LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION; 48 49static unsigned long rotate_xor(unsigned long hash, const void *area, 50 size_t size) 51{ 52 size_t i; 53 unsigned long *ptr = (unsigned long *)area; 54 55 for (i = 0; i < size / sizeof(hash); i++) { 56 /* Rotate by odd number of bits and XOR. */ 57 hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7); 58 hash ^= ptr[i]; 59 } 60 61 return hash; 62} 63 64/* Attempt to create a simple but unpredictable starting entropy. */ 65static unsigned long get_boot_seed(void) 66{ 67 unsigned long hash = 0; 68 69 hash = rotate_xor(hash, build_str, sizeof(build_str)); 70 hash = rotate_xor(hash, boot_params, sizeof(*boot_params)); 71 72 return hash; 73} 74 75#define KASLR_COMPRESSED_BOOT 76#include "../../lib/kaslr.c" 77 78 79/* Only supporting at most 4 unusable memmap regions with kaslr */ 80#define MAX_MEMMAP_REGIONS 4 81 82static bool memmap_too_large; 83 84 85/* 86 * Store memory limit: MAXMEM on 64-bit and KERNEL_IMAGE_SIZE on 32-bit. 87 * It may be reduced by "mem=nn[KMG]" or "memmap=nn[KMG]" command line options. 88 */ 89static u64 mem_limit; 90 91/* Number of immovable memory regions */ 92static int num_immovable_mem; 93 94enum mem_avoid_index { 95 MEM_AVOID_ZO_RANGE = 0, 96 MEM_AVOID_INITRD, 97 MEM_AVOID_CMDLINE, 98 MEM_AVOID_BOOTPARAMS, 99 MEM_AVOID_MEMMAP_BEGIN, 100 MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1, 101 MEM_AVOID_MAX, 102}; 103 104static struct mem_vector mem_avoid[MEM_AVOID_MAX]; 105 106static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two) 107{ 108 /* Item one is entirely before item two. */ 109 if (one->start + one->size <= two->start) 110 return false; 111 /* Item one is entirely after item two. */ 112 if (one->start >= two->start + two->size) 113 return false; 114 return true; 115} 116 117char *skip_spaces(const char *str) 118{ 119 while (isspace(*str)) 120 ++str; 121 return (char *)str; 122} 123#include "../../../../lib/ctype.c" 124#include "../../../../lib/cmdline.c" 125 126enum parse_mode { 127 PARSE_MEMMAP, 128 PARSE_EFI, 129}; 130 131static int 132parse_memmap(char *p, u64 *start, u64 *size, enum parse_mode mode) 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 if (mode == PARSE_MEMMAP) { 156 /* 157 * memmap=nn@ss specifies usable region, should 158 * be skipped 159 */ 160 *size = 0; 161 } else { 162 u64 flags; 163 164 /* 165 * efi_fake_mem=nn@ss:attr the attr specifies 166 * flags that might imply a soft-reservation. 167 */ 168 *start = memparse(p + 1, &p); 169 if (p && *p == ':') { 170 p++; 171 if (kstrtoull(p, 0, &flags) < 0) 172 *size = 0; 173 else if (flags & EFI_MEMORY_SP) 174 return 0; 175 } 176 *size = 0; 177 } 178 fallthrough; 179 default: 180 /* 181 * If w/o offset, only size specified, memmap=nn[KMG] has the 182 * same behaviour as mem=nn[KMG]. It limits the max address 183 * system can use. Region above the limit should be avoided. 184 */ 185 *start = 0; 186 return 0; 187 } 188 189 return -EINVAL; 190} 191 192static void mem_avoid_memmap(enum parse_mode mode, char *str) 193{ 194 static int i; 195 196 if (i >= MAX_MEMMAP_REGIONS) 197 return; 198 199 while (str && (i < MAX_MEMMAP_REGIONS)) { 200 int rc; 201 u64 start, size; 202 char *k = strchr(str, ','); 203 204 if (k) 205 *k++ = 0; 206 207 rc = parse_memmap(str, &start, &size, mode); 208 if (rc < 0) 209 break; 210 str = k; 211 212 if (start == 0) { 213 /* Store the specified memory limit if size > 0 */ 214 if (size > 0 && size < mem_limit) 215 mem_limit = size; 216 217 continue; 218 } 219 220 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start; 221 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size; 222 i++; 223 } 224 225 /* More than 4 memmaps, fail kaslr */ 226 if ((i >= MAX_MEMMAP_REGIONS) && str) 227 memmap_too_large = true; 228} 229 230/* Store the number of 1GB huge pages which users specified: */ 231static unsigned long max_gb_huge_pages; 232 233static void parse_gb_huge_pages(char *param, char *val) 234{ 235 static bool gbpage_sz; 236 char *p; 237 238 if (!strcmp(param, "hugepagesz")) { 239 p = val; 240 if (memparse(p, &p) != PUD_SIZE) { 241 gbpage_sz = false; 242 return; 243 } 244 245 if (gbpage_sz) 246 warn("Repeatedly set hugeTLB page size of 1G!\n"); 247 gbpage_sz = true; 248 return; 249 } 250 251 if (!strcmp(param, "hugepages") && gbpage_sz) { 252 p = val; 253 max_gb_huge_pages = simple_strtoull(p, &p, 0); 254 return; 255 } 256} 257 258static void handle_mem_options(void) 259{ 260 char *args = (char *)get_cmd_line_ptr(); 261 size_t len; 262 char *tmp_cmdline; 263 char *param, *val; 264 u64 mem_size; 265 266 if (!args) 267 return; 268 269 len = strnlen(args, COMMAND_LINE_SIZE-1); 270 tmp_cmdline = malloc(len + 1); 271 if (!tmp_cmdline) 272 error("Failed to allocate space for tmp_cmdline"); 273 274 memcpy(tmp_cmdline, args, len); 275 tmp_cmdline[len] = 0; 276 args = tmp_cmdline; 277 278 /* Chew leading spaces */ 279 args = skip_spaces(args); 280 281 while (*args) { 282 args = next_arg(args, ¶m, &val); 283 /* Stop at -- */ 284 if (!val && strcmp(param, "--") == 0) 285 break; 286 287 if (!strcmp(param, "memmap")) { 288 mem_avoid_memmap(PARSE_MEMMAP, val); 289 } else if (IS_ENABLED(CONFIG_X86_64) && strstr(param, "hugepages")) { 290 parse_gb_huge_pages(param, val); 291 } else if (!strcmp(param, "mem")) { 292 char *p = val; 293 294 if (!strcmp(p, "nopentium")) 295 continue; 296 mem_size = memparse(p, &p); 297 if (mem_size == 0) 298 break; 299 300 if (mem_size < mem_limit) 301 mem_limit = mem_size; 302 } else if (!strcmp(param, "efi_fake_mem")) { 303 mem_avoid_memmap(PARSE_EFI, val); 304 } 305 } 306 307 free(tmp_cmdline); 308 return; 309} 310 311/* 312 * In theory, KASLR can put the kernel anywhere in the range of [16M, MAXMEM) 313 * on 64-bit, and [16M, KERNEL_IMAGE_SIZE) on 32-bit. 314 * 315 * The mem_avoid array is used to store the ranges that need to be avoided 316 * when KASLR searches for an appropriate random address. We must avoid any 317 * regions that are unsafe to overlap with during decompression, and other 318 * things like the initrd, cmdline and boot_params. This comment seeks to 319 * explain mem_avoid as clearly as possible since incorrect mem_avoid 320 * memory ranges lead to really hard to debug boot failures. 321 * 322 * The initrd, cmdline, and boot_params are trivial to identify for 323 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and 324 * MEM_AVOID_BOOTPARAMS respectively below. 325 * 326 * What is not obvious how to avoid is the range of memory that is used 327 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover 328 * the compressed kernel (ZO) and its run space, which is used to extract 329 * the uncompressed kernel (VO) and relocs. 330 * 331 * ZO's full run size sits against the end of the decompression buffer, so 332 * we can calculate where text, data, bss, etc of ZO are positioned more 333 * easily. 334 * 335 * For additional background, the decompression calculations can be found 336 * in header.S, and the memory diagram is based on the one found in misc.c. 337 * 338 * The following conditions are already enforced by the image layouts and 339 * associated code: 340 * - input + input_size >= output + output_size 341 * - kernel_total_size <= init_size 342 * - kernel_total_size <= output_size (see Note below) 343 * - output + init_size >= output + output_size 344 * 345 * (Note that kernel_total_size and output_size have no fundamental 346 * relationship, but output_size is passed to choose_random_location 347 * as a maximum of the two. The diagram is showing a case where 348 * kernel_total_size is larger than output_size, but this case is 349 * handled by bumping output_size.) 350 * 351 * The above conditions can be illustrated by a diagram: 352 * 353 * 0 output input input+input_size output+init_size 354 * | | | | | 355 * | | | | | 356 * |-----|--------|--------|--------------|-----------|--|-------------| 357 * | | | 358 * | | | 359 * output+init_size-ZO_INIT_SIZE output+output_size output+kernel_total_size 360 * 361 * [output, output+init_size) is the entire memory range used for 362 * extracting the compressed image. 363 * 364 * [output, output+kernel_total_size) is the range needed for the 365 * uncompressed kernel (VO) and its run size (bss, brk, etc). 366 * 367 * [output, output+output_size) is VO plus relocs (i.e. the entire 368 * uncompressed payload contained by ZO). This is the area of the buffer 369 * written to during decompression. 370 * 371 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case 372 * range of the copied ZO and decompression code. (i.e. the range 373 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.) 374 * 375 * [input, input+input_size) is the original copied compressed image (ZO) 376 * (i.e. it does not include its run size). This range must be avoided 377 * because it contains the data used for decompression. 378 * 379 * [input+input_size, output+init_size) is [_text, _end) for ZO. This 380 * range includes ZO's heap and stack, and must be avoided since it 381 * performs the decompression. 382 * 383 * Since the above two ranges need to be avoided and they are adjacent, 384 * they can be merged, resulting in: [input, output+init_size) which 385 * becomes the MEM_AVOID_ZO_RANGE below. 386 */ 387static void mem_avoid_init(unsigned long input, unsigned long input_size, 388 unsigned long output) 389{ 390 unsigned long init_size = boot_params->hdr.init_size; 391 u64 initrd_start, initrd_size; 392 unsigned long cmd_line, cmd_line_size; 393 394 /* 395 * Avoid the region that is unsafe to overlap during 396 * decompression. 397 */ 398 mem_avoid[MEM_AVOID_ZO_RANGE].start = input; 399 mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input; 400 401 /* Avoid initrd. */ 402 initrd_start = (u64)boot_params->ext_ramdisk_image << 32; 403 initrd_start |= boot_params->hdr.ramdisk_image; 404 initrd_size = (u64)boot_params->ext_ramdisk_size << 32; 405 initrd_size |= boot_params->hdr.ramdisk_size; 406 mem_avoid[MEM_AVOID_INITRD].start = initrd_start; 407 mem_avoid[MEM_AVOID_INITRD].size = initrd_size; 408 /* No need to set mapping for initrd, it will be handled in VO. */ 409 410 /* Avoid kernel command line. */ 411 cmd_line = get_cmd_line_ptr(); 412 /* Calculate size of cmd_line. */ 413 if (cmd_line) { 414 cmd_line_size = strnlen((char *)cmd_line, COMMAND_LINE_SIZE-1) + 1; 415 mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line; 416 mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size; 417 } 418 419 /* Avoid boot parameters. */ 420 mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params; 421 mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params); 422 423 /* We don't need to set a mapping for setup_data. */ 424 425 /* Mark the memmap regions we need to avoid */ 426 handle_mem_options(); 427 428 /* Enumerate the immovable memory regions */ 429 num_immovable_mem = count_immovable_mem_regions(); 430} 431 432/* 433 * Does this memory vector overlap a known avoided area? If so, record the 434 * overlap region with the lowest address. 435 */ 436static bool mem_avoid_overlap(struct mem_vector *img, 437 struct mem_vector *overlap) 438{ 439 int i; 440 struct setup_data *ptr; 441 u64 earliest = img->start + img->size; 442 bool is_overlapping = false; 443 444 for (i = 0; i < MEM_AVOID_MAX; i++) { 445 if (mem_overlaps(img, &mem_avoid[i]) && 446 mem_avoid[i].start < earliest) { 447 *overlap = mem_avoid[i]; 448 earliest = overlap->start; 449 is_overlapping = true; 450 } 451 } 452 453 /* Avoid all entries in the setup_data linked list. */ 454 ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data; 455 while (ptr) { 456 struct mem_vector avoid; 457 458 avoid.start = (unsigned long)ptr; 459 avoid.size = sizeof(*ptr) + ptr->len; 460 461 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) { 462 *overlap = avoid; 463 earliest = overlap->start; 464 is_overlapping = true; 465 } 466 467 if (ptr->type == SETUP_INDIRECT && 468 ((struct setup_indirect *)ptr->data)->type != SETUP_INDIRECT) { 469 avoid.start = ((struct setup_indirect *)ptr->data)->addr; 470 avoid.size = ((struct setup_indirect *)ptr->data)->len; 471 472 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) { 473 *overlap = avoid; 474 earliest = overlap->start; 475 is_overlapping = true; 476 } 477 } 478 479 ptr = (struct setup_data *)(unsigned long)ptr->next; 480 } 481 482 return is_overlapping; 483} 484 485struct slot_area { 486 u64 addr; 487 unsigned long num; 488}; 489 490#define MAX_SLOT_AREA 100 491 492static struct slot_area slot_areas[MAX_SLOT_AREA]; 493static unsigned int slot_area_index; 494static unsigned long slot_max; 495 496static void store_slot_info(struct mem_vector *region, unsigned long image_size) 497{ 498 struct slot_area slot_area; 499 500 if (slot_area_index == MAX_SLOT_AREA) 501 return; 502 503 slot_area.addr = region->start; 504 slot_area.num = 1 + (region->size - image_size) / CONFIG_PHYSICAL_ALIGN; 505 506 slot_areas[slot_area_index++] = slot_area; 507 slot_max += slot_area.num; 508} 509 510/* 511 * Skip as many 1GB huge pages as possible in the passed region 512 * according to the number which users specified: 513 */ 514static void 515process_gb_huge_pages(struct mem_vector *region, unsigned long image_size) 516{ 517 u64 pud_start, pud_end; 518 unsigned long gb_huge_pages; 519 struct mem_vector tmp; 520 521 if (!IS_ENABLED(CONFIG_X86_64) || !max_gb_huge_pages) { 522 store_slot_info(region, image_size); 523 return; 524 } 525 526 /* Are there any 1GB pages in the region? */ 527 pud_start = ALIGN(region->start, PUD_SIZE); 528 pud_end = ALIGN_DOWN(region->start + region->size, PUD_SIZE); 529 530 /* No good 1GB huge pages found: */ 531 if (pud_start >= pud_end) { 532 store_slot_info(region, image_size); 533 return; 534 } 535 536 /* Check if the head part of the region is usable. */ 537 if (pud_start >= region->start + image_size) { 538 tmp.start = region->start; 539 tmp.size = pud_start - region->start; 540 store_slot_info(&tmp, image_size); 541 } 542 543 /* Skip the good 1GB pages. */ 544 gb_huge_pages = (pud_end - pud_start) >> PUD_SHIFT; 545 if (gb_huge_pages > max_gb_huge_pages) { 546 pud_end = pud_start + (max_gb_huge_pages << PUD_SHIFT); 547 max_gb_huge_pages = 0; 548 } else { 549 max_gb_huge_pages -= gb_huge_pages; 550 } 551 552 /* Check if the tail part of the region is usable. */ 553 if (region->start + region->size >= pud_end + image_size) { 554 tmp.start = pud_end; 555 tmp.size = region->start + region->size - pud_end; 556 store_slot_info(&tmp, image_size); 557 } 558} 559 560static u64 slots_fetch_random(void) 561{ 562 unsigned long slot; 563 unsigned int i; 564 565 /* Handle case of no slots stored. */ 566 if (slot_max == 0) 567 return 0; 568 569 slot = kaslr_get_random_long("Physical") % slot_max; 570 571 for (i = 0; i < slot_area_index; i++) { 572 if (slot >= slot_areas[i].num) { 573 slot -= slot_areas[i].num; 574 continue; 575 } 576 return slot_areas[i].addr + ((u64)slot * CONFIG_PHYSICAL_ALIGN); 577 } 578 579 if (i == slot_area_index) 580 debug_putstr("slots_fetch_random() failed!?\n"); 581 return 0; 582} 583 584static void __process_mem_region(struct mem_vector *entry, 585 unsigned long minimum, 586 unsigned long image_size) 587{ 588 struct mem_vector region, overlap; 589 u64 region_end; 590 591 /* Enforce minimum and memory limit. */ 592 region.start = max_t(u64, entry->start, minimum); 593 region_end = min(entry->start + entry->size, mem_limit); 594 595 /* Give up if slot area array is full. */ 596 while (slot_area_index < MAX_SLOT_AREA) { 597 /* Potentially raise address to meet alignment needs. */ 598 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN); 599 600 /* Did we raise the address above the passed in memory entry? */ 601 if (region.start > region_end) 602 return; 603 604 /* Reduce size by any delta from the original address. */ 605 region.size = region_end - region.start; 606 607 /* Return if region can't contain decompressed kernel */ 608 if (region.size < image_size) 609 return; 610 611 /* If nothing overlaps, store the region and return. */ 612 if (!mem_avoid_overlap(®ion, &overlap)) { 613 process_gb_huge_pages(®ion, image_size); 614 return; 615 } 616 617 /* Store beginning of region if holds at least image_size. */ 618 if (overlap.start >= region.start + image_size) { 619 region.size = overlap.start - region.start; 620 process_gb_huge_pages(®ion, image_size); 621 } 622 623 /* Clip off the overlapping region and start over. */ 624 region.start = overlap.start + overlap.size; 625 } 626} 627 628static bool process_mem_region(struct mem_vector *region, 629 unsigned long minimum, 630 unsigned long image_size) 631{ 632 int i; 633 /* 634 * If no immovable memory found, or MEMORY_HOTREMOVE disabled, 635 * use @region directly. 636 */ 637 if (!num_immovable_mem) { 638 __process_mem_region(region, minimum, image_size); 639 640 if (slot_area_index == MAX_SLOT_AREA) { 641 debug_putstr("Aborted e820/efi memmap scan (slot_areas full)!\n"); 642 return 1; 643 } 644 return 0; 645 } 646 647#if defined(CONFIG_MEMORY_HOTREMOVE) && defined(CONFIG_ACPI) 648 /* 649 * If immovable memory found, filter the intersection between 650 * immovable memory and @region. 651 */ 652 for (i = 0; i < num_immovable_mem; i++) { 653 u64 start, end, entry_end, region_end; 654 struct mem_vector entry; 655 656 if (!mem_overlaps(region, &immovable_mem[i])) 657 continue; 658 659 start = immovable_mem[i].start; 660 end = start + immovable_mem[i].size; 661 region_end = region->start + region->size; 662 663 entry.start = clamp(region->start, start, end); 664 entry_end = clamp(region_end, start, end); 665 entry.size = entry_end - entry.start; 666 667 __process_mem_region(&entry, minimum, image_size); 668 669 if (slot_area_index == MAX_SLOT_AREA) { 670 debug_putstr("Aborted e820/efi memmap scan when walking immovable regions(slot_areas full)!\n"); 671 return 1; 672 } 673 } 674#endif 675 return 0; 676} 677 678#ifdef CONFIG_EFI 679/* 680 * Returns true if we processed the EFI memmap, which we prefer over the E820 681 * table if it is available. 682 */ 683static bool 684process_efi_entries(unsigned long minimum, unsigned long image_size) 685{ 686 struct efi_info *e = &boot_params->efi_info; 687 bool efi_mirror_found = false; 688 struct mem_vector region; 689 efi_memory_desc_t *md; 690 unsigned long pmap; 691 char *signature; 692 u32 nr_desc; 693 int i; 694 695 signature = (char *)&e->efi_loader_signature; 696 if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) && 697 strncmp(signature, EFI64_LOADER_SIGNATURE, 4)) 698 return false; 699 700#ifdef CONFIG_X86_32 701 /* Can't handle data above 4GB at this time */ 702 if (e->efi_memmap_hi) { 703 warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n"); 704 return false; 705 } 706 pmap = e->efi_memmap; 707#else 708 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32)); 709#endif 710 711 nr_desc = e->efi_memmap_size / e->efi_memdesc_size; 712 for (i = 0; i < nr_desc; i++) { 713 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i); 714 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) { 715 efi_mirror_found = true; 716 break; 717 } 718 } 719 720 for (i = 0; i < nr_desc; i++) { 721 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i); 722 723 /* 724 * Here we are more conservative in picking free memory than 725 * the EFI spec allows: 726 * 727 * According to the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also 728 * free memory and thus available to place the kernel image into, 729 * but in practice there's firmware where using that memory leads 730 * to crashes. 731 * 732 * Only EFI_CONVENTIONAL_MEMORY is guaranteed to be free. 733 */ 734 if (md->type != EFI_CONVENTIONAL_MEMORY) 735 continue; 736 737 if (efi_soft_reserve_enabled() && 738 (md->attribute & EFI_MEMORY_SP)) 739 continue; 740 741 if (efi_mirror_found && 742 !(md->attribute & EFI_MEMORY_MORE_RELIABLE)) 743 continue; 744 745 region.start = md->phys_addr; 746 region.size = md->num_pages << EFI_PAGE_SHIFT; 747 if (process_mem_region(®ion, minimum, image_size)) 748 break; 749 } 750 return true; 751} 752#else 753static inline bool 754process_efi_entries(unsigned long minimum, unsigned long image_size) 755{ 756 return false; 757} 758#endif 759 760static void process_e820_entries(unsigned long minimum, 761 unsigned long image_size) 762{ 763 int i; 764 struct mem_vector region; 765 struct boot_e820_entry *entry; 766 767 /* Verify potential e820 positions, appending to slots list. */ 768 for (i = 0; i < boot_params->e820_entries; i++) { 769 entry = &boot_params->e820_table[i]; 770 /* Skip non-RAM entries. */ 771 if (entry->type != E820_TYPE_RAM) 772 continue; 773 region.start = entry->addr; 774 region.size = entry->size; 775 if (process_mem_region(®ion, minimum, image_size)) 776 break; 777 } 778} 779 780static unsigned long find_random_phys_addr(unsigned long minimum, 781 unsigned long image_size) 782{ 783 u64 phys_addr; 784 785 /* Bail out early if it's impossible to succeed. */ 786 if (minimum + image_size > mem_limit) 787 return 0; 788 789 /* Check if we had too many memmaps. */ 790 if (memmap_too_large) { 791 debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n"); 792 return 0; 793 } 794 795 if (!process_efi_entries(minimum, image_size)) 796 process_e820_entries(minimum, image_size); 797 798 phys_addr = slots_fetch_random(); 799 800 /* Perform a final check to make sure the address is in range. */ 801 if (phys_addr < minimum || phys_addr + image_size > mem_limit) { 802 warn("Invalid physical address chosen!\n"); 803 return 0; 804 } 805 806 return (unsigned long)phys_addr; 807} 808 809static unsigned long find_random_virt_addr(unsigned long minimum, 810 unsigned long image_size) 811{ 812 unsigned long slots, random_addr; 813 814 /* 815 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots 816 * that can hold image_size within the range of minimum to 817 * KERNEL_IMAGE_SIZE? 818 */ 819 slots = 1 + (KERNEL_IMAGE_SIZE - minimum - image_size) / CONFIG_PHYSICAL_ALIGN; 820 821 random_addr = kaslr_get_random_long("Virtual") % slots; 822 823 return random_addr * CONFIG_PHYSICAL_ALIGN + minimum; 824} 825 826/* 827 * Since this function examines addresses much more numerically, 828 * it takes the input and output pointers as 'unsigned long'. 829 */ 830void choose_random_location(unsigned long input, 831 unsigned long input_size, 832 unsigned long *output, 833 unsigned long output_size, 834 unsigned long *virt_addr) 835{ 836 unsigned long random_addr, min_addr; 837 838 if (cmdline_find_option_bool("nokaslr")) { 839 warn("KASLR disabled: 'nokaslr' on cmdline."); 840 return; 841 } 842 843 boot_params->hdr.loadflags |= KASLR_FLAG; 844 845 if (IS_ENABLED(CONFIG_X86_32)) 846 mem_limit = KERNEL_IMAGE_SIZE; 847 else 848 mem_limit = MAXMEM; 849 850 /* Record the various known unsafe memory ranges. */ 851 mem_avoid_init(input, input_size, *output); 852 853 /* 854 * Low end of the randomization range should be the 855 * smaller of 512M or the initial kernel image 856 * location: 857 */ 858 min_addr = min(*output, 512UL << 20); 859 /* Make sure minimum is aligned. */ 860 min_addr = ALIGN(min_addr, CONFIG_PHYSICAL_ALIGN); 861 862 /* Walk available memory entries to find a random address. */ 863 random_addr = find_random_phys_addr(min_addr, output_size); 864 if (!random_addr) { 865 warn("Physical KASLR disabled: no suitable memory region!"); 866 } else { 867 /* Update the new physical address location. */ 868 if (*output != random_addr) 869 *output = random_addr; 870 } 871 872 873 /* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */ 874 if (IS_ENABLED(CONFIG_X86_64)) 875 random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size); 876 *virt_addr = random_addr; 877} 878