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 /* 445 * Free up any temporary buffers allocated which are not needed 446 * after image has been loaded 447 */ 448 kimage_file_post_load_cleanup(image); 449 exchange: 450 image = xchg(dest_image, image); 451 out: 452 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 453 arch_kexec_protect_crashkres(); 454 455 mutex_unlock(&kexec_mutex); 456 kimage_free(image); 457 return ret; 458 } 459 460 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 461 struct kexec_buf *kbuf) 462 { 463 struct kimage *image = kbuf->image; 464 unsigned long temp_start, temp_end; 465 466 temp_end = min(end, kbuf->buf_max); 467 temp_start = temp_end - kbuf->memsz; 468 469 do { 470 /* align down start */ 471 temp_start = temp_start & (~(kbuf->buf_align - 1)); 472 473 if (temp_start < start || temp_start < kbuf->buf_min) 474 return 0; 475 476 temp_end = temp_start + kbuf->memsz - 1; 477 478 /* 479 * Make sure this does not conflict with any of existing 480 * segments 481 */ 482 if (kimage_is_destination_range(image, temp_start, temp_end)) { 483 temp_start = temp_start - PAGE_SIZE; 484 continue; 485 } 486 487 /* We found a suitable memory range */ 488 break; 489 } while (1); 490 491 /* If we are here, we found a suitable memory range */ 492 kbuf->mem = temp_start; 493 494 /* Success, stop navigating through remaining System RAM ranges */ 495 return 1; 496 } 497 498 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 499 struct kexec_buf *kbuf) 500 { 501 struct kimage *image = kbuf->image; 502 unsigned long temp_start, temp_end; 503 504 temp_start = max(start, kbuf->buf_min); 505 506 do { 507 temp_start = ALIGN(temp_start, kbuf->buf_align); 508 temp_end = temp_start + kbuf->memsz - 1; 509 510 if (temp_end > end || temp_end > kbuf->buf_max) 511 return 0; 512 /* 513 * Make sure this does not conflict with any of existing 514 * segments 515 */ 516 if (kimage_is_destination_range(image, temp_start, temp_end)) { 517 temp_start = temp_start + PAGE_SIZE; 518 continue; 519 } 520 521 /* We found a suitable memory range */ 522 break; 523 } while (1); 524 525 /* If we are here, we found a suitable memory range */ 526 kbuf->mem = temp_start; 527 528 /* Success, stop navigating through remaining System RAM ranges */ 529 return 1; 530 } 531 532 static int locate_mem_hole_callback(struct resource *res, void *arg) 533 { 534 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 535 u64 start = res->start, end = res->end; 536 unsigned long sz = end - start + 1; 537 538 /* Returning 0 will take to next memory range */ 539 if (sz < kbuf->memsz) 540 return 0; 541 542 if (end < kbuf->buf_min || start > kbuf->buf_max) 543 return 0; 544 545 /* 546 * Allocate memory top down with-in ram range. Otherwise bottom up 547 * allocation. 548 */ 549 if (kbuf->top_down) 550 return locate_mem_hole_top_down(start, end, kbuf); 551 return locate_mem_hole_bottom_up(start, end, kbuf); 552 } 553 554 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 555 static int kexec_walk_memblock(struct kexec_buf *kbuf, 556 int (*func)(struct resource *, void *)) 557 { 558 int ret = 0; 559 u64 i; 560 phys_addr_t mstart, mend; 561 struct resource res = { }; 562 563 if (kbuf->image->type == KEXEC_TYPE_CRASH) 564 return func(&crashk_res, kbuf); 565 566 if (kbuf->top_down) { 567 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 568 &mstart, &mend, NULL) { 569 /* 570 * In memblock, end points to the first byte after the 571 * range while in kexec, end points to the last byte 572 * in the range. 573 */ 574 res.start = mstart; 575 res.end = mend - 1; 576 ret = func(&res, kbuf); 577 if (ret) 578 break; 579 } 580 } else { 581 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 582 &mstart, &mend, NULL) { 583 /* 584 * In memblock, end points to the first byte after the 585 * range while in kexec, end points to the last byte 586 * in the range. 587 */ 588 res.start = mstart; 589 res.end = mend - 1; 590 ret = func(&res, kbuf); 591 if (ret) 592 break; 593 } 594 } 595 596 return ret; 597 } 598 #else 599 static int kexec_walk_memblock(struct kexec_buf *kbuf, 600 int (*func)(struct resource *, void *)) 601 { 602 return 0; 603 } 604 #endif 605 606 /** 607 * kexec_walk_resources - call func(data) on free memory regions 608 * @kbuf: Context info for the search. Also passed to @func. 609 * @func: Function to call for each memory region. 610 * 611 * Return: The memory walk will stop when func returns a non-zero value 612 * and that value will be returned. If all free regions are visited without 613 * func returning non-zero, then zero will be returned. 614 */ 615 static int kexec_walk_resources(struct kexec_buf *kbuf, 616 int (*func)(struct resource *, void *)) 617 { 618 if (kbuf->image->type == KEXEC_TYPE_CRASH) 619 return walk_iomem_res_desc(crashk_res.desc, 620 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 621 crashk_res.start, crashk_res.end, 622 kbuf, func); 623 else 624 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 625 } 626 627 /** 628 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 629 * @kbuf: Parameters for the memory search. 630 * 631 * On success, kbuf->mem will have the start address of the memory region found. 632 * 633 * Return: 0 on success, negative errno on error. 634 */ 635 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 636 { 637 int ret; 638 639 /* Arch knows where to place */ 640 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 641 return 0; 642 643 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 644 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 645 else 646 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 647 648 return ret == 1 ? 0 : -EADDRNOTAVAIL; 649 } 650 651 /** 652 * kexec_add_buffer - place a buffer in a kexec segment 653 * @kbuf: Buffer contents and memory parameters. 654 * 655 * This function assumes that kexec_mutex is held. 656 * On successful return, @kbuf->mem will have the physical address of 657 * the buffer in memory. 658 * 659 * Return: 0 on success, negative errno on error. 660 */ 661 int kexec_add_buffer(struct kexec_buf *kbuf) 662 { 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 = 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; 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 1180 if (mstart > end || mend < start) 1181 continue; 1182 1183 /* Truncate any area outside of range */ 1184 if (mstart < start) 1185 mstart = start; 1186 if (mend > end) 1187 mend = end; 1188 1189 /* Found completely overlapping range */ 1190 if (mstart == start && mend == end) { 1191 mem->ranges[i].start = 0; 1192 mem->ranges[i].end = 0; 1193 if (i < mem->nr_ranges - 1) { 1194 /* Shift rest of the ranges to left */ 1195 for (j = i; j < mem->nr_ranges - 1; j++) { 1196 mem->ranges[j].start = 1197 mem->ranges[j+1].start; 1198 mem->ranges[j].end = 1199 mem->ranges[j+1].end; 1200 } 1201 } 1202 mem->nr_ranges--; 1203 return 0; 1204 } 1205 1206 if (mstart > start && mend < end) { 1207 /* Split original range */ 1208 mem->ranges[i].end = mstart - 1; 1209 temp_range.start = mend + 1; 1210 temp_range.end = end; 1211 } else if (mstart != start) 1212 mem->ranges[i].end = mstart - 1; 1213 else 1214 mem->ranges[i].start = mend + 1; 1215 break; 1216 } 1217 1218 /* If a split happened, add the split to array */ 1219 if (!temp_range.end) 1220 return 0; 1221 1222 /* Split happened */ 1223 if (i == mem->max_nr_ranges - 1) 1224 return -ENOMEM; 1225 1226 /* Location where new range should go */ 1227 j = i + 1; 1228 if (j < mem->nr_ranges) { 1229 /* Move over all ranges one slot towards the end */ 1230 for (i = mem->nr_ranges - 1; i >= j; i--) 1231 mem->ranges[i + 1] = mem->ranges[i]; 1232 } 1233 1234 mem->ranges[j].start = temp_range.start; 1235 mem->ranges[j].end = temp_range.end; 1236 mem->nr_ranges++; 1237 return 0; 1238 } 1239 1240 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, 1241 void **addr, unsigned long *sz) 1242 { 1243 Elf64_Ehdr *ehdr; 1244 Elf64_Phdr *phdr; 1245 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1246 unsigned char *buf; 1247 unsigned int cpu, i; 1248 unsigned long long notes_addr; 1249 unsigned long mstart, mend; 1250 1251 /* extra phdr for vmcoreinfo elf note */ 1252 nr_phdr = nr_cpus + 1; 1253 nr_phdr += mem->nr_ranges; 1254 1255 /* 1256 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1257 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1258 * I think this is required by tools like gdb. So same physical 1259 * memory will be mapped in two elf headers. One will contain kernel 1260 * text virtual addresses and other will have __va(physical) addresses. 1261 */ 1262 1263 nr_phdr++; 1264 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1265 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1266 1267 buf = vzalloc(elf_sz); 1268 if (!buf) 1269 return -ENOMEM; 1270 1271 ehdr = (Elf64_Ehdr *)buf; 1272 phdr = (Elf64_Phdr *)(ehdr + 1); 1273 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1274 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1275 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1276 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1277 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1278 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1279 ehdr->e_type = ET_CORE; 1280 ehdr->e_machine = ELF_ARCH; 1281 ehdr->e_version = EV_CURRENT; 1282 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1283 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1284 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1285 1286 /* Prepare one phdr of type PT_NOTE for each present cpu */ 1287 for_each_present_cpu(cpu) { 1288 phdr->p_type = PT_NOTE; 1289 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1290 phdr->p_offset = phdr->p_paddr = notes_addr; 1291 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1292 (ehdr->e_phnum)++; 1293 phdr++; 1294 } 1295 1296 /* Prepare one PT_NOTE header for vmcoreinfo */ 1297 phdr->p_type = PT_NOTE; 1298 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1299 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1300 (ehdr->e_phnum)++; 1301 phdr++; 1302 1303 /* Prepare PT_LOAD type program header for kernel text region */ 1304 if (kernel_map) { 1305 phdr->p_type = PT_LOAD; 1306 phdr->p_flags = PF_R|PF_W|PF_X; 1307 phdr->p_vaddr = (unsigned long) _text; 1308 phdr->p_filesz = phdr->p_memsz = _end - _text; 1309 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1310 ehdr->e_phnum++; 1311 phdr++; 1312 } 1313 1314 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1315 for (i = 0; i < mem->nr_ranges; i++) { 1316 mstart = mem->ranges[i].start; 1317 mend = mem->ranges[i].end; 1318 1319 phdr->p_type = PT_LOAD; 1320 phdr->p_flags = PF_R|PF_W|PF_X; 1321 phdr->p_offset = mstart; 1322 1323 phdr->p_paddr = mstart; 1324 phdr->p_vaddr = (unsigned long) __va(mstart); 1325 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1326 phdr->p_align = 0; 1327 ehdr->e_phnum++; 1328 phdr++; 1329 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", 1330 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1331 ehdr->e_phnum, phdr->p_offset); 1332 } 1333 1334 *addr = buf; 1335 *sz = elf_sz; 1336 return 0; 1337 } 1338