1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kexec: kexec_file_load system call 4 * 5 * Copyright (C) 2014 Red Hat Inc. 6 * Authors: 7 * Vivek Goyal <vgoyal@redhat.com> 8 */ 9 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/capability.h> 13 #include <linux/mm.h> 14 #include <linux/file.h> 15 #include <linux/slab.h> 16 #include <linux/kexec.h> 17 #include <linux/memblock.h> 18 #include <linux/mutex.h> 19 #include <linux/list.h> 20 #include <linux/fs.h> 21 #include <linux/ima.h> 22 #include <crypto/hash.h> 23 #include <crypto/sha2.h> 24 #include <linux/elf.h> 25 #include <linux/elfcore.h> 26 #include <linux/kernel.h> 27 #include <linux/kernel_read_file.h> 28 #include <linux/syscalls.h> 29 #include <linux/vmalloc.h> 30 #include "kexec_internal.h" 31 32 static int kexec_calculate_store_digests(struct kimage *image); 33 34 /* 35 * Currently this is the only default function that is exported as some 36 * architectures need it to do additional handlings. 37 * In the future, other default functions may be exported too if required. 38 */ 39 int kexec_image_probe_default(struct kimage *image, void *buf, 40 unsigned long buf_len) 41 { 42 const struct kexec_file_ops * const *fops; 43 int ret = -ENOEXEC; 44 45 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 46 ret = (*fops)->probe(buf, buf_len); 47 if (!ret) { 48 image->fops = *fops; 49 return ret; 50 } 51 } 52 53 return ret; 54 } 55 56 /* Architectures can provide this probe function */ 57 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, 58 unsigned long buf_len) 59 { 60 return kexec_image_probe_default(image, buf, buf_len); 61 } 62 63 static void *kexec_image_load_default(struct kimage *image) 64 { 65 if (!image->fops || !image->fops->load) 66 return ERR_PTR(-ENOEXEC); 67 68 return image->fops->load(image, image->kernel_buf, 69 image->kernel_buf_len, image->initrd_buf, 70 image->initrd_buf_len, image->cmdline_buf, 71 image->cmdline_buf_len); 72 } 73 74 void * __weak arch_kexec_kernel_image_load(struct kimage *image) 75 { 76 return kexec_image_load_default(image); 77 } 78 79 int kexec_image_post_load_cleanup_default(struct kimage *image) 80 { 81 if (!image->fops || !image->fops->cleanup) 82 return 0; 83 84 return image->fops->cleanup(image->image_loader_data); 85 } 86 87 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) 88 { 89 return kexec_image_post_load_cleanup_default(image); 90 } 91 92 #ifdef CONFIG_KEXEC_SIG 93 static int kexec_image_verify_sig_default(struct kimage *image, void *buf, 94 unsigned long buf_len) 95 { 96 if (!image->fops || !image->fops->verify_sig) { 97 pr_debug("kernel loader does not support signature verification.\n"); 98 return -EKEYREJECTED; 99 } 100 101 return image->fops->verify_sig(buf, buf_len); 102 } 103 104 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, 105 unsigned long buf_len) 106 { 107 return kexec_image_verify_sig_default(image, buf, buf_len); 108 } 109 #endif 110 111 /* 112 * arch_kexec_apply_relocations_add - apply relocations of type RELA 113 * @pi: Purgatory to be relocated. 114 * @section: Section relocations applying to. 115 * @relsec: Section containing RELAs. 116 * @symtab: Corresponding symtab. 117 * 118 * Return: 0 on success, negative errno on error. 119 */ 120 int __weak 121 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section, 122 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 123 { 124 pr_err("RELA relocation unsupported.\n"); 125 return -ENOEXEC; 126 } 127 128 /* 129 * arch_kexec_apply_relocations - apply relocations of type REL 130 * @pi: Purgatory to be relocated. 131 * @section: Section relocations applying to. 132 * @relsec: Section containing RELs. 133 * @symtab: Corresponding symtab. 134 * 135 * Return: 0 on success, negative errno on error. 136 */ 137 int __weak 138 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section, 139 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 140 { 141 pr_err("REL relocation unsupported.\n"); 142 return -ENOEXEC; 143 } 144 145 /* 146 * Free up memory used by kernel, initrd, and command line. This is temporary 147 * memory allocation which is not needed any more after these buffers have 148 * been loaded into separate segments and have been copied elsewhere. 149 */ 150 void kimage_file_post_load_cleanup(struct kimage *image) 151 { 152 struct purgatory_info *pi = &image->purgatory_info; 153 154 vfree(image->kernel_buf); 155 image->kernel_buf = NULL; 156 157 vfree(image->initrd_buf); 158 image->initrd_buf = NULL; 159 160 kfree(image->cmdline_buf); 161 image->cmdline_buf = NULL; 162 163 vfree(pi->purgatory_buf); 164 pi->purgatory_buf = NULL; 165 166 vfree(pi->sechdrs); 167 pi->sechdrs = NULL; 168 169 /* See if architecture has anything to cleanup post load */ 170 arch_kimage_file_post_load_cleanup(image); 171 172 /* 173 * Above call should have called into bootloader to free up 174 * any data stored in kimage->image_loader_data. It should 175 * be ok now to free it up. 176 */ 177 kfree(image->image_loader_data); 178 image->image_loader_data = NULL; 179 } 180 181 #ifdef CONFIG_KEXEC_SIG 182 static int 183 kimage_validate_signature(struct kimage *image) 184 { 185 int ret; 186 187 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, 188 image->kernel_buf_len); 189 if (ret) { 190 191 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) { 192 pr_notice("Enforced kernel signature verification failed (%d).\n", ret); 193 return ret; 194 } 195 196 /* 197 * If IMA is guaranteed to appraise a signature on the kexec 198 * image, permit it even if the kernel is otherwise locked 199 * down. 200 */ 201 if (!ima_appraise_signature(READING_KEXEC_IMAGE) && 202 security_locked_down(LOCKDOWN_KEXEC)) 203 return -EPERM; 204 205 pr_debug("kernel signature verification failed (%d).\n", ret); 206 } 207 208 return 0; 209 } 210 #endif 211 212 /* 213 * In file mode list of segments is prepared by kernel. Copy relevant 214 * data from user space, do error checking, prepare segment list 215 */ 216 static int 217 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 218 const char __user *cmdline_ptr, 219 unsigned long cmdline_len, unsigned flags) 220 { 221 int ret; 222 void *ldata; 223 224 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf, 225 INT_MAX, NULL, READING_KEXEC_IMAGE); 226 if (ret < 0) 227 return ret; 228 image->kernel_buf_len = ret; 229 230 /* Call arch image probe handlers */ 231 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, 232 image->kernel_buf_len); 233 if (ret) 234 goto out; 235 236 #ifdef CONFIG_KEXEC_SIG 237 ret = kimage_validate_signature(image); 238 239 if (ret) 240 goto out; 241 #endif 242 /* It is possible that there no initramfs is being loaded */ 243 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { 244 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf, 245 INT_MAX, NULL, 246 READING_KEXEC_INITRAMFS); 247 if (ret < 0) 248 goto out; 249 image->initrd_buf_len = ret; 250 ret = 0; 251 } 252 253 if (cmdline_len) { 254 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); 255 if (IS_ERR(image->cmdline_buf)) { 256 ret = PTR_ERR(image->cmdline_buf); 257 image->cmdline_buf = NULL; 258 goto out; 259 } 260 261 image->cmdline_buf_len = cmdline_len; 262 263 /* command line should be a string with last byte null */ 264 if (image->cmdline_buf[cmdline_len - 1] != '\0') { 265 ret = -EINVAL; 266 goto out; 267 } 268 269 ima_kexec_cmdline(kernel_fd, image->cmdline_buf, 270 image->cmdline_buf_len - 1); 271 } 272 273 /* IMA needs to pass the measurement list to the next kernel. */ 274 ima_add_kexec_buffer(image); 275 276 /* Call arch image load handlers */ 277 ldata = arch_kexec_kernel_image_load(image); 278 279 if (IS_ERR(ldata)) { 280 ret = PTR_ERR(ldata); 281 goto out; 282 } 283 284 image->image_loader_data = ldata; 285 out: 286 /* In case of error, free up all allocated memory in this function */ 287 if (ret) 288 kimage_file_post_load_cleanup(image); 289 return ret; 290 } 291 292 static int 293 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, 294 int initrd_fd, const char __user *cmdline_ptr, 295 unsigned long cmdline_len, unsigned long flags) 296 { 297 int ret; 298 struct kimage *image; 299 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; 300 301 image = do_kimage_alloc_init(); 302 if (!image) 303 return -ENOMEM; 304 305 image->file_mode = 1; 306 307 if (kexec_on_panic) { 308 /* Enable special crash kernel control page alloc policy. */ 309 image->control_page = crashk_res.start; 310 image->type = KEXEC_TYPE_CRASH; 311 } 312 313 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 314 cmdline_ptr, cmdline_len, flags); 315 if (ret) 316 goto out_free_image; 317 318 ret = sanity_check_segment_list(image); 319 if (ret) 320 goto out_free_post_load_bufs; 321 322 ret = -ENOMEM; 323 image->control_code_page = kimage_alloc_control_pages(image, 324 get_order(KEXEC_CONTROL_PAGE_SIZE)); 325 if (!image->control_code_page) { 326 pr_err("Could not allocate control_code_buffer\n"); 327 goto out_free_post_load_bufs; 328 } 329 330 if (!kexec_on_panic) { 331 image->swap_page = kimage_alloc_control_pages(image, 0); 332 if (!image->swap_page) { 333 pr_err("Could not allocate swap buffer\n"); 334 goto out_free_control_pages; 335 } 336 } 337 338 *rimage = image; 339 return 0; 340 out_free_control_pages: 341 kimage_free_page_list(&image->control_pages); 342 out_free_post_load_bufs: 343 kimage_file_post_load_cleanup(image); 344 out_free_image: 345 kfree(image); 346 return ret; 347 } 348 349 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 350 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 351 unsigned long, flags) 352 { 353 int ret = 0, i; 354 struct kimage **dest_image, *image; 355 356 /* We only trust the superuser with rebooting the system. */ 357 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) 358 return -EPERM; 359 360 /* Make sure we have a legal set of flags */ 361 if (flags != (flags & KEXEC_FILE_FLAGS)) 362 return -EINVAL; 363 364 image = NULL; 365 366 if (!mutex_trylock(&kexec_mutex)) 367 return -EBUSY; 368 369 dest_image = &kexec_image; 370 if (flags & KEXEC_FILE_ON_CRASH) { 371 dest_image = &kexec_crash_image; 372 if (kexec_crash_image) 373 arch_kexec_unprotect_crashkres(); 374 } 375 376 if (flags & KEXEC_FILE_UNLOAD) 377 goto exchange; 378 379 /* 380 * In case of crash, new kernel gets loaded in reserved region. It is 381 * same memory where old crash kernel might be loaded. Free any 382 * current crash dump kernel before we corrupt it. 383 */ 384 if (flags & KEXEC_FILE_ON_CRASH) 385 kimage_free(xchg(&kexec_crash_image, NULL)); 386 387 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 388 cmdline_len, flags); 389 if (ret) 390 goto out; 391 392 ret = machine_kexec_prepare(image); 393 if (ret) 394 goto out; 395 396 /* 397 * Some architecture(like S390) may touch the crash memory before 398 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 399 */ 400 ret = kimage_crash_copy_vmcoreinfo(image); 401 if (ret) 402 goto out; 403 404 ret = kexec_calculate_store_digests(image); 405 if (ret) 406 goto out; 407 408 for (i = 0; i < image->nr_segments; i++) { 409 struct kexec_segment *ksegment; 410 411 ksegment = &image->segment[i]; 412 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 413 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 414 ksegment->memsz); 415 416 ret = kimage_load_segment(image, &image->segment[i]); 417 if (ret) 418 goto out; 419 } 420 421 kimage_terminate(image); 422 423 ret = machine_kexec_post_load(image); 424 if (ret) 425 goto out; 426 427 /* 428 * Free up any temporary buffers allocated which are not needed 429 * after image has been loaded 430 */ 431 kimage_file_post_load_cleanup(image); 432 exchange: 433 image = xchg(dest_image, image); 434 out: 435 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 436 arch_kexec_protect_crashkres(); 437 438 mutex_unlock(&kexec_mutex); 439 kimage_free(image); 440 return ret; 441 } 442 443 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 444 struct kexec_buf *kbuf) 445 { 446 struct kimage *image = kbuf->image; 447 unsigned long temp_start, temp_end; 448 449 temp_end = min(end, kbuf->buf_max); 450 temp_start = temp_end - kbuf->memsz; 451 452 do { 453 /* align down start */ 454 temp_start = temp_start & (~(kbuf->buf_align - 1)); 455 456 if (temp_start < start || temp_start < kbuf->buf_min) 457 return 0; 458 459 temp_end = temp_start + kbuf->memsz - 1; 460 461 /* 462 * Make sure this does not conflict with any of existing 463 * segments 464 */ 465 if (kimage_is_destination_range(image, temp_start, temp_end)) { 466 temp_start = temp_start - PAGE_SIZE; 467 continue; 468 } 469 470 /* We found a suitable memory range */ 471 break; 472 } while (1); 473 474 /* If we are here, we found a suitable memory range */ 475 kbuf->mem = temp_start; 476 477 /* Success, stop navigating through remaining System RAM ranges */ 478 return 1; 479 } 480 481 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 482 struct kexec_buf *kbuf) 483 { 484 struct kimage *image = kbuf->image; 485 unsigned long temp_start, temp_end; 486 487 temp_start = max(start, kbuf->buf_min); 488 489 do { 490 temp_start = ALIGN(temp_start, kbuf->buf_align); 491 temp_end = temp_start + kbuf->memsz - 1; 492 493 if (temp_end > end || temp_end > kbuf->buf_max) 494 return 0; 495 /* 496 * Make sure this does not conflict with any of existing 497 * segments 498 */ 499 if (kimage_is_destination_range(image, temp_start, temp_end)) { 500 temp_start = temp_start + PAGE_SIZE; 501 continue; 502 } 503 504 /* We found a suitable memory range */ 505 break; 506 } while (1); 507 508 /* If we are here, we found a suitable memory range */ 509 kbuf->mem = temp_start; 510 511 /* Success, stop navigating through remaining System RAM ranges */ 512 return 1; 513 } 514 515 static int locate_mem_hole_callback(struct resource *res, void *arg) 516 { 517 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 518 u64 start = res->start, end = res->end; 519 unsigned long sz = end - start + 1; 520 521 /* Returning 0 will take to next memory range */ 522 523 /* Don't use memory that will be detected and handled by a driver. */ 524 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) 525 return 0; 526 527 if (sz < kbuf->memsz) 528 return 0; 529 530 if (end < kbuf->buf_min || start > kbuf->buf_max) 531 return 0; 532 533 /* 534 * Allocate memory top down with-in ram range. Otherwise bottom up 535 * allocation. 536 */ 537 if (kbuf->top_down) 538 return locate_mem_hole_top_down(start, end, kbuf); 539 return locate_mem_hole_bottom_up(start, end, kbuf); 540 } 541 542 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 543 static int kexec_walk_memblock(struct kexec_buf *kbuf, 544 int (*func)(struct resource *, void *)) 545 { 546 int ret = 0; 547 u64 i; 548 phys_addr_t mstart, mend; 549 struct resource res = { }; 550 551 if (kbuf->image->type == KEXEC_TYPE_CRASH) 552 return func(&crashk_res, kbuf); 553 554 if (kbuf->top_down) { 555 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 556 &mstart, &mend, NULL) { 557 /* 558 * In memblock, end points to the first byte after the 559 * range while in kexec, end points to the last byte 560 * in the range. 561 */ 562 res.start = mstart; 563 res.end = mend - 1; 564 ret = func(&res, kbuf); 565 if (ret) 566 break; 567 } 568 } else { 569 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 570 &mstart, &mend, NULL) { 571 /* 572 * In memblock, end points to the first byte after the 573 * range while in kexec, end points to the last byte 574 * in the range. 575 */ 576 res.start = mstart; 577 res.end = mend - 1; 578 ret = func(&res, kbuf); 579 if (ret) 580 break; 581 } 582 } 583 584 return ret; 585 } 586 #else 587 static int kexec_walk_memblock(struct kexec_buf *kbuf, 588 int (*func)(struct resource *, void *)) 589 { 590 return 0; 591 } 592 #endif 593 594 /** 595 * kexec_walk_resources - call func(data) on free memory regions 596 * @kbuf: Context info for the search. Also passed to @func. 597 * @func: Function to call for each memory region. 598 * 599 * Return: The memory walk will stop when func returns a non-zero value 600 * and that value will be returned. If all free regions are visited without 601 * func returning non-zero, then zero will be returned. 602 */ 603 static int kexec_walk_resources(struct kexec_buf *kbuf, 604 int (*func)(struct resource *, void *)) 605 { 606 if (kbuf->image->type == KEXEC_TYPE_CRASH) 607 return walk_iomem_res_desc(crashk_res.desc, 608 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 609 crashk_res.start, crashk_res.end, 610 kbuf, func); 611 else 612 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 613 } 614 615 /** 616 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 617 * @kbuf: Parameters for the memory search. 618 * 619 * On success, kbuf->mem will have the start address of the memory region found. 620 * 621 * Return: 0 on success, negative errno on error. 622 */ 623 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 624 { 625 int ret; 626 627 /* Arch knows where to place */ 628 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 629 return 0; 630 631 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 632 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 633 else 634 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 635 636 return ret == 1 ? 0 : -EADDRNOTAVAIL; 637 } 638 639 /** 640 * arch_kexec_locate_mem_hole - Find free memory to place the segments. 641 * @kbuf: Parameters for the memory search. 642 * 643 * On success, kbuf->mem will have the start address of the memory region found. 644 * 645 * Return: 0 on success, negative errno on error. 646 */ 647 int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf) 648 { 649 return kexec_locate_mem_hole(kbuf); 650 } 651 652 /** 653 * kexec_add_buffer - place a buffer in a kexec segment 654 * @kbuf: Buffer contents and memory parameters. 655 * 656 * This function assumes that kexec_mutex is held. 657 * On successful return, @kbuf->mem will have the physical address of 658 * the buffer in memory. 659 * 660 * Return: 0 on success, negative errno on error. 661 */ 662 int kexec_add_buffer(struct kexec_buf *kbuf) 663 { 664 struct kexec_segment *ksegment; 665 int ret; 666 667 /* Currently adding segment this way is allowed only in file mode */ 668 if (!kbuf->image->file_mode) 669 return -EINVAL; 670 671 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 672 return -EINVAL; 673 674 /* 675 * Make sure we are not trying to add buffer after allocating 676 * control pages. All segments need to be placed first before 677 * any control pages are allocated. As control page allocation 678 * logic goes through list of segments to make sure there are 679 * no destination overlaps. 680 */ 681 if (!list_empty(&kbuf->image->control_pages)) { 682 WARN_ON(1); 683 return -EINVAL; 684 } 685 686 /* Ensure minimum alignment needed for segments. */ 687 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 688 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 689 690 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 691 ret = arch_kexec_locate_mem_hole(kbuf); 692 if (ret) 693 return ret; 694 695 /* Found a suitable memory range */ 696 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 697 ksegment->kbuf = kbuf->buffer; 698 ksegment->bufsz = kbuf->bufsz; 699 ksegment->mem = kbuf->mem; 700 ksegment->memsz = kbuf->memsz; 701 kbuf->image->nr_segments++; 702 return 0; 703 } 704 705 /* Calculate and store the digest of segments */ 706 static int kexec_calculate_store_digests(struct kimage *image) 707 { 708 struct crypto_shash *tfm; 709 struct shash_desc *desc; 710 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 711 size_t desc_size, nullsz; 712 char *digest; 713 void *zero_buf; 714 struct kexec_sha_region *sha_regions; 715 struct purgatory_info *pi = &image->purgatory_info; 716 717 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 718 return 0; 719 720 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 721 zero_buf_sz = PAGE_SIZE; 722 723 tfm = crypto_alloc_shash("sha256", 0, 0); 724 if (IS_ERR(tfm)) { 725 ret = PTR_ERR(tfm); 726 goto out; 727 } 728 729 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 730 desc = kzalloc(desc_size, GFP_KERNEL); 731 if (!desc) { 732 ret = -ENOMEM; 733 goto out_free_tfm; 734 } 735 736 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 737 sha_regions = vzalloc(sha_region_sz); 738 if (!sha_regions) 739 goto out_free_desc; 740 741 desc->tfm = tfm; 742 743 ret = crypto_shash_init(desc); 744 if (ret < 0) 745 goto out_free_sha_regions; 746 747 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 748 if (!digest) { 749 ret = -ENOMEM; 750 goto out_free_sha_regions; 751 } 752 753 for (j = i = 0; i < image->nr_segments; i++) { 754 struct kexec_segment *ksegment; 755 756 ksegment = &image->segment[i]; 757 /* 758 * Skip purgatory as it will be modified once we put digest 759 * info in purgatory. 760 */ 761 if (ksegment->kbuf == pi->purgatory_buf) 762 continue; 763 764 ret = crypto_shash_update(desc, ksegment->kbuf, 765 ksegment->bufsz); 766 if (ret) 767 break; 768 769 /* 770 * Assume rest of the buffer is filled with zero and 771 * update digest accordingly. 772 */ 773 nullsz = ksegment->memsz - ksegment->bufsz; 774 while (nullsz) { 775 unsigned long bytes = nullsz; 776 777 if (bytes > zero_buf_sz) 778 bytes = zero_buf_sz; 779 ret = crypto_shash_update(desc, zero_buf, bytes); 780 if (ret) 781 break; 782 nullsz -= bytes; 783 } 784 785 if (ret) 786 break; 787 788 sha_regions[j].start = ksegment->mem; 789 sha_regions[j].len = ksegment->memsz; 790 j++; 791 } 792 793 if (!ret) { 794 ret = crypto_shash_final(desc, digest); 795 if (ret) 796 goto out_free_digest; 797 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 798 sha_regions, sha_region_sz, 0); 799 if (ret) 800 goto out_free_digest; 801 802 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 803 digest, SHA256_DIGEST_SIZE, 0); 804 if (ret) 805 goto out_free_digest; 806 } 807 808 out_free_digest: 809 kfree(digest); 810 out_free_sha_regions: 811 vfree(sha_regions); 812 out_free_desc: 813 kfree(desc); 814 out_free_tfm: 815 kfree(tfm); 816 out: 817 return ret; 818 } 819 820 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 821 /* 822 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 823 * @pi: Purgatory to be loaded. 824 * @kbuf: Buffer to setup. 825 * 826 * Allocates the memory needed for the buffer. Caller is responsible to free 827 * the memory after use. 828 * 829 * Return: 0 on success, negative errno on error. 830 */ 831 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 832 struct kexec_buf *kbuf) 833 { 834 const Elf_Shdr *sechdrs; 835 unsigned long bss_align; 836 unsigned long bss_sz; 837 unsigned long align; 838 int i, ret; 839 840 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 841 kbuf->buf_align = bss_align = 1; 842 kbuf->bufsz = bss_sz = 0; 843 844 for (i = 0; i < pi->ehdr->e_shnum; i++) { 845 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 846 continue; 847 848 align = sechdrs[i].sh_addralign; 849 if (sechdrs[i].sh_type != SHT_NOBITS) { 850 if (kbuf->buf_align < align) 851 kbuf->buf_align = align; 852 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 853 kbuf->bufsz += sechdrs[i].sh_size; 854 } else { 855 if (bss_align < align) 856 bss_align = align; 857 bss_sz = ALIGN(bss_sz, align); 858 bss_sz += sechdrs[i].sh_size; 859 } 860 } 861 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 862 kbuf->memsz = kbuf->bufsz + bss_sz; 863 if (kbuf->buf_align < bss_align) 864 kbuf->buf_align = bss_align; 865 866 kbuf->buffer = vzalloc(kbuf->bufsz); 867 if (!kbuf->buffer) 868 return -ENOMEM; 869 pi->purgatory_buf = kbuf->buffer; 870 871 ret = kexec_add_buffer(kbuf); 872 if (ret) 873 goto out; 874 875 return 0; 876 out: 877 vfree(pi->purgatory_buf); 878 pi->purgatory_buf = NULL; 879 return ret; 880 } 881 882 /* 883 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 884 * @pi: Purgatory to be loaded. 885 * @kbuf: Buffer prepared to store purgatory. 886 * 887 * Allocates the memory needed for the buffer. Caller is responsible to free 888 * the memory after use. 889 * 890 * Return: 0 on success, negative errno on error. 891 */ 892 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 893 struct kexec_buf *kbuf) 894 { 895 unsigned long bss_addr; 896 unsigned long offset; 897 Elf_Shdr *sechdrs; 898 int i; 899 900 /* 901 * The section headers in kexec_purgatory are read-only. In order to 902 * have them modifiable make a temporary copy. 903 */ 904 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 905 if (!sechdrs) 906 return -ENOMEM; 907 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 908 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 909 pi->sechdrs = sechdrs; 910 911 offset = 0; 912 bss_addr = kbuf->mem + kbuf->bufsz; 913 kbuf->image->start = pi->ehdr->e_entry; 914 915 for (i = 0; i < pi->ehdr->e_shnum; i++) { 916 unsigned long align; 917 void *src, *dst; 918 919 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 920 continue; 921 922 align = sechdrs[i].sh_addralign; 923 if (sechdrs[i].sh_type == SHT_NOBITS) { 924 bss_addr = ALIGN(bss_addr, align); 925 sechdrs[i].sh_addr = bss_addr; 926 bss_addr += sechdrs[i].sh_size; 927 continue; 928 } 929 930 offset = ALIGN(offset, align); 931 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 932 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 933 pi->ehdr->e_entry < (sechdrs[i].sh_addr 934 + sechdrs[i].sh_size)) { 935 kbuf->image->start -= sechdrs[i].sh_addr; 936 kbuf->image->start += kbuf->mem + offset; 937 } 938 939 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 940 dst = pi->purgatory_buf + offset; 941 memcpy(dst, src, sechdrs[i].sh_size); 942 943 sechdrs[i].sh_addr = kbuf->mem + offset; 944 sechdrs[i].sh_offset = offset; 945 offset += sechdrs[i].sh_size; 946 } 947 948 return 0; 949 } 950 951 static int kexec_apply_relocations(struct kimage *image) 952 { 953 int i, ret; 954 struct purgatory_info *pi = &image->purgatory_info; 955 const Elf_Shdr *sechdrs; 956 957 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 958 959 for (i = 0; i < pi->ehdr->e_shnum; i++) { 960 const Elf_Shdr *relsec; 961 const Elf_Shdr *symtab; 962 Elf_Shdr *section; 963 964 relsec = sechdrs + i; 965 966 if (relsec->sh_type != SHT_RELA && 967 relsec->sh_type != SHT_REL) 968 continue; 969 970 /* 971 * For section of type SHT_RELA/SHT_REL, 972 * ->sh_link contains section header index of associated 973 * symbol table. And ->sh_info contains section header 974 * index of section to which relocations apply. 975 */ 976 if (relsec->sh_info >= pi->ehdr->e_shnum || 977 relsec->sh_link >= pi->ehdr->e_shnum) 978 return -ENOEXEC; 979 980 section = pi->sechdrs + relsec->sh_info; 981 symtab = sechdrs + relsec->sh_link; 982 983 if (!(section->sh_flags & SHF_ALLOC)) 984 continue; 985 986 /* 987 * symtab->sh_link contain section header index of associated 988 * string table. 989 */ 990 if (symtab->sh_link >= pi->ehdr->e_shnum) 991 /* Invalid section number? */ 992 continue; 993 994 /* 995 * Respective architecture needs to provide support for applying 996 * relocations of type SHT_RELA/SHT_REL. 997 */ 998 if (relsec->sh_type == SHT_RELA) 999 ret = arch_kexec_apply_relocations_add(pi, section, 1000 relsec, symtab); 1001 else if (relsec->sh_type == SHT_REL) 1002 ret = arch_kexec_apply_relocations(pi, section, 1003 relsec, symtab); 1004 if (ret) 1005 return ret; 1006 } 1007 1008 return 0; 1009 } 1010 1011 /* 1012 * kexec_load_purgatory - Load and relocate the purgatory object. 1013 * @image: Image to add the purgatory to. 1014 * @kbuf: Memory parameters to use. 1015 * 1016 * Allocates the memory needed for image->purgatory_info.sechdrs and 1017 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 1018 * to free the memory after use. 1019 * 1020 * Return: 0 on success, negative errno on error. 1021 */ 1022 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 1023 { 1024 struct purgatory_info *pi = &image->purgatory_info; 1025 int ret; 1026 1027 if (kexec_purgatory_size <= 0) 1028 return -EINVAL; 1029 1030 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 1031 1032 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 1033 if (ret) 1034 return ret; 1035 1036 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 1037 if (ret) 1038 goto out_free_kbuf; 1039 1040 ret = kexec_apply_relocations(image); 1041 if (ret) 1042 goto out; 1043 1044 return 0; 1045 out: 1046 vfree(pi->sechdrs); 1047 pi->sechdrs = NULL; 1048 out_free_kbuf: 1049 vfree(pi->purgatory_buf); 1050 pi->purgatory_buf = NULL; 1051 return ret; 1052 } 1053 1054 /* 1055 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1056 * @pi: Purgatory to search in. 1057 * @name: Name of the symbol. 1058 * 1059 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1060 */ 1061 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1062 const char *name) 1063 { 1064 const Elf_Shdr *sechdrs; 1065 const Elf_Ehdr *ehdr; 1066 const Elf_Sym *syms; 1067 const char *strtab; 1068 int i, k; 1069 1070 if (!pi->ehdr) 1071 return NULL; 1072 1073 ehdr = pi->ehdr; 1074 sechdrs = (void *)ehdr + ehdr->e_shoff; 1075 1076 for (i = 0; i < ehdr->e_shnum; i++) { 1077 if (sechdrs[i].sh_type != SHT_SYMTAB) 1078 continue; 1079 1080 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1081 /* Invalid strtab section number */ 1082 continue; 1083 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1084 syms = (void *)ehdr + sechdrs[i].sh_offset; 1085 1086 /* Go through symbols for a match */ 1087 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1088 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1089 continue; 1090 1091 if (strcmp(strtab + syms[k].st_name, name) != 0) 1092 continue; 1093 1094 if (syms[k].st_shndx == SHN_UNDEF || 1095 syms[k].st_shndx >= ehdr->e_shnum) { 1096 pr_debug("Symbol: %s has bad section index %d.\n", 1097 name, syms[k].st_shndx); 1098 return NULL; 1099 } 1100 1101 /* Found the symbol we are looking for */ 1102 return &syms[k]; 1103 } 1104 } 1105 1106 return NULL; 1107 } 1108 1109 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1110 { 1111 struct purgatory_info *pi = &image->purgatory_info; 1112 const Elf_Sym *sym; 1113 Elf_Shdr *sechdr; 1114 1115 sym = kexec_purgatory_find_symbol(pi, name); 1116 if (!sym) 1117 return ERR_PTR(-EINVAL); 1118 1119 sechdr = &pi->sechdrs[sym->st_shndx]; 1120 1121 /* 1122 * Returns the address where symbol will finally be loaded after 1123 * kexec_load_segment() 1124 */ 1125 return (void *)(sechdr->sh_addr + sym->st_value); 1126 } 1127 1128 /* 1129 * Get or set value of a symbol. If "get_value" is true, symbol value is 1130 * returned in buf otherwise symbol value is set based on value in buf. 1131 */ 1132 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1133 void *buf, unsigned int size, bool get_value) 1134 { 1135 struct purgatory_info *pi = &image->purgatory_info; 1136 const Elf_Sym *sym; 1137 Elf_Shdr *sec; 1138 char *sym_buf; 1139 1140 sym = kexec_purgatory_find_symbol(pi, name); 1141 if (!sym) 1142 return -EINVAL; 1143 1144 if (sym->st_size != size) { 1145 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1146 name, (unsigned long)sym->st_size, size); 1147 return -EINVAL; 1148 } 1149 1150 sec = pi->sechdrs + sym->st_shndx; 1151 1152 if (sec->sh_type == SHT_NOBITS) { 1153 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1154 get_value ? "get" : "set"); 1155 return -EINVAL; 1156 } 1157 1158 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1159 1160 if (get_value) 1161 memcpy((void *)buf, sym_buf, size); 1162 else 1163 memcpy((void *)sym_buf, buf, size); 1164 1165 return 0; 1166 } 1167 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1168 1169 int crash_exclude_mem_range(struct crash_mem *mem, 1170 unsigned long long mstart, unsigned long long mend) 1171 { 1172 int i, j; 1173 unsigned long long start, end, p_start, p_end; 1174 struct crash_mem_range temp_range = {0, 0}; 1175 1176 for (i = 0; i < mem->nr_ranges; i++) { 1177 start = mem->ranges[i].start; 1178 end = mem->ranges[i].end; 1179 p_start = mstart; 1180 p_end = mend; 1181 1182 if (mstart > end || mend < start) 1183 continue; 1184 1185 /* Truncate any area outside of range */ 1186 if (mstart < start) 1187 p_start = start; 1188 if (mend > end) 1189 p_end = end; 1190 1191 /* Found completely overlapping range */ 1192 if (p_start == start && p_end == end) { 1193 mem->ranges[i].start = 0; 1194 mem->ranges[i].end = 0; 1195 if (i < mem->nr_ranges - 1) { 1196 /* Shift rest of the ranges to left */ 1197 for (j = i; j < mem->nr_ranges - 1; j++) { 1198 mem->ranges[j].start = 1199 mem->ranges[j+1].start; 1200 mem->ranges[j].end = 1201 mem->ranges[j+1].end; 1202 } 1203 1204 /* 1205 * Continue to check if there are another overlapping ranges 1206 * from the current position because of shifting the above 1207 * mem ranges. 1208 */ 1209 i--; 1210 mem->nr_ranges--; 1211 continue; 1212 } 1213 mem->nr_ranges--; 1214 return 0; 1215 } 1216 1217 if (p_start > start && p_end < end) { 1218 /* Split original range */ 1219 mem->ranges[i].end = p_start - 1; 1220 temp_range.start = p_end + 1; 1221 temp_range.end = end; 1222 } else if (p_start != start) 1223 mem->ranges[i].end = p_start - 1; 1224 else 1225 mem->ranges[i].start = p_end + 1; 1226 break; 1227 } 1228 1229 /* If a split happened, add the split to array */ 1230 if (!temp_range.end) 1231 return 0; 1232 1233 /* Split happened */ 1234 if (i == mem->max_nr_ranges - 1) 1235 return -ENOMEM; 1236 1237 /* Location where new range should go */ 1238 j = i + 1; 1239 if (j < mem->nr_ranges) { 1240 /* Move over all ranges one slot towards the end */ 1241 for (i = mem->nr_ranges - 1; i >= j; i--) 1242 mem->ranges[i + 1] = mem->ranges[i]; 1243 } 1244 1245 mem->ranges[j].start = temp_range.start; 1246 mem->ranges[j].end = temp_range.end; 1247 mem->nr_ranges++; 1248 return 0; 1249 } 1250 1251 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, 1252 void **addr, unsigned long *sz) 1253 { 1254 Elf64_Ehdr *ehdr; 1255 Elf64_Phdr *phdr; 1256 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1257 unsigned char *buf; 1258 unsigned int cpu, i; 1259 unsigned long long notes_addr; 1260 unsigned long mstart, mend; 1261 1262 /* extra phdr for vmcoreinfo ELF note */ 1263 nr_phdr = nr_cpus + 1; 1264 nr_phdr += mem->nr_ranges; 1265 1266 /* 1267 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1268 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1269 * I think this is required by tools like gdb. So same physical 1270 * memory will be mapped in two ELF headers. One will contain kernel 1271 * text virtual addresses and other will have __va(physical) addresses. 1272 */ 1273 1274 nr_phdr++; 1275 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1276 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1277 1278 buf = vzalloc(elf_sz); 1279 if (!buf) 1280 return -ENOMEM; 1281 1282 ehdr = (Elf64_Ehdr *)buf; 1283 phdr = (Elf64_Phdr *)(ehdr + 1); 1284 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1285 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1286 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1287 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1288 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1289 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1290 ehdr->e_type = ET_CORE; 1291 ehdr->e_machine = ELF_ARCH; 1292 ehdr->e_version = EV_CURRENT; 1293 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1294 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1295 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1296 1297 /* Prepare one phdr of type PT_NOTE for each present CPU */ 1298 for_each_present_cpu(cpu) { 1299 phdr->p_type = PT_NOTE; 1300 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1301 phdr->p_offset = phdr->p_paddr = notes_addr; 1302 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1303 (ehdr->e_phnum)++; 1304 phdr++; 1305 } 1306 1307 /* Prepare one PT_NOTE header for vmcoreinfo */ 1308 phdr->p_type = PT_NOTE; 1309 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1310 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1311 (ehdr->e_phnum)++; 1312 phdr++; 1313 1314 /* Prepare PT_LOAD type program header for kernel text region */ 1315 if (kernel_map) { 1316 phdr->p_type = PT_LOAD; 1317 phdr->p_flags = PF_R|PF_W|PF_X; 1318 phdr->p_vaddr = (unsigned long) _text; 1319 phdr->p_filesz = phdr->p_memsz = _end - _text; 1320 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1321 ehdr->e_phnum++; 1322 phdr++; 1323 } 1324 1325 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1326 for (i = 0; i < mem->nr_ranges; i++) { 1327 mstart = mem->ranges[i].start; 1328 mend = mem->ranges[i].end; 1329 1330 phdr->p_type = PT_LOAD; 1331 phdr->p_flags = PF_R|PF_W|PF_X; 1332 phdr->p_offset = mstart; 1333 1334 phdr->p_paddr = mstart; 1335 phdr->p_vaddr = (unsigned long) __va(mstart); 1336 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1337 phdr->p_align = 0; 1338 ehdr->e_phnum++; 1339 pr_debug("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n", 1340 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1341 ehdr->e_phnum, phdr->p_offset); 1342 phdr++; 1343 } 1344 1345 *addr = buf; 1346 *sz = elf_sz; 1347 return 0; 1348 } 1349