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