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 if (sz < kbuf->memsz) 544 return 0; 545 546 if (end < kbuf->buf_min || start > kbuf->buf_max) 547 return 0; 548 549 /* 550 * Allocate memory top down with-in ram range. Otherwise bottom up 551 * allocation. 552 */ 553 if (kbuf->top_down) 554 return locate_mem_hole_top_down(start, end, kbuf); 555 return locate_mem_hole_bottom_up(start, end, kbuf); 556 } 557 558 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 559 static int kexec_walk_memblock(struct kexec_buf *kbuf, 560 int (*func)(struct resource *, void *)) 561 { 562 int ret = 0; 563 u64 i; 564 phys_addr_t mstart, mend; 565 struct resource res = { }; 566 567 if (kbuf->image->type == KEXEC_TYPE_CRASH) 568 return func(&crashk_res, kbuf); 569 570 if (kbuf->top_down) { 571 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 572 &mstart, &mend, NULL) { 573 /* 574 * In memblock, end points to the first byte after the 575 * range while in kexec, end points to the last byte 576 * in the range. 577 */ 578 res.start = mstart; 579 res.end = mend - 1; 580 ret = func(&res, kbuf); 581 if (ret) 582 break; 583 } 584 } else { 585 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 586 &mstart, &mend, NULL) { 587 /* 588 * In memblock, end points to the first byte after the 589 * range while in kexec, end points to the last byte 590 * in the range. 591 */ 592 res.start = mstart; 593 res.end = mend - 1; 594 ret = func(&res, kbuf); 595 if (ret) 596 break; 597 } 598 } 599 600 return ret; 601 } 602 #else 603 static int kexec_walk_memblock(struct kexec_buf *kbuf, 604 int (*func)(struct resource *, void *)) 605 { 606 return 0; 607 } 608 #endif 609 610 /** 611 * kexec_walk_resources - call func(data) on free memory regions 612 * @kbuf: Context info for the search. Also passed to @func. 613 * @func: Function to call for each memory region. 614 * 615 * Return: The memory walk will stop when func returns a non-zero value 616 * and that value will be returned. If all free regions are visited without 617 * func returning non-zero, then zero will be returned. 618 */ 619 static int kexec_walk_resources(struct kexec_buf *kbuf, 620 int (*func)(struct resource *, void *)) 621 { 622 if (kbuf->image->type == KEXEC_TYPE_CRASH) 623 return walk_iomem_res_desc(crashk_res.desc, 624 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 625 crashk_res.start, crashk_res.end, 626 kbuf, func); 627 else 628 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 629 } 630 631 /** 632 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 633 * @kbuf: Parameters for the memory search. 634 * 635 * On success, kbuf->mem will have the start address of the memory region found. 636 * 637 * Return: 0 on success, negative errno on error. 638 */ 639 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 640 { 641 int ret; 642 643 /* Arch knows where to place */ 644 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 645 return 0; 646 647 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 648 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 649 else 650 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 651 652 return ret == 1 ? 0 : -EADDRNOTAVAIL; 653 } 654 655 /** 656 * kexec_add_buffer - place a buffer in a kexec segment 657 * @kbuf: Buffer contents and memory parameters. 658 * 659 * This function assumes that kexec_mutex is held. 660 * On successful return, @kbuf->mem will have the physical address of 661 * the buffer in memory. 662 * 663 * Return: 0 on success, negative errno on error. 664 */ 665 int kexec_add_buffer(struct kexec_buf *kbuf) 666 { 667 668 struct kexec_segment *ksegment; 669 int ret; 670 671 /* Currently adding segment this way is allowed only in file mode */ 672 if (!kbuf->image->file_mode) 673 return -EINVAL; 674 675 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 676 return -EINVAL; 677 678 /* 679 * Make sure we are not trying to add buffer after allocating 680 * control pages. All segments need to be placed first before 681 * any control pages are allocated. As control page allocation 682 * logic goes through list of segments to make sure there are 683 * no destination overlaps. 684 */ 685 if (!list_empty(&kbuf->image->control_pages)) { 686 WARN_ON(1); 687 return -EINVAL; 688 } 689 690 /* Ensure minimum alignment needed for segments. */ 691 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 692 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 693 694 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 695 ret = kexec_locate_mem_hole(kbuf); 696 if (ret) 697 return ret; 698 699 /* Found a suitable memory range */ 700 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 701 ksegment->kbuf = kbuf->buffer; 702 ksegment->bufsz = kbuf->bufsz; 703 ksegment->mem = kbuf->mem; 704 ksegment->memsz = kbuf->memsz; 705 kbuf->image->nr_segments++; 706 return 0; 707 } 708 709 /* Calculate and store the digest of segments */ 710 static int kexec_calculate_store_digests(struct kimage *image) 711 { 712 struct crypto_shash *tfm; 713 struct shash_desc *desc; 714 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 715 size_t desc_size, nullsz; 716 char *digest; 717 void *zero_buf; 718 struct kexec_sha_region *sha_regions; 719 struct purgatory_info *pi = &image->purgatory_info; 720 721 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 722 return 0; 723 724 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 725 zero_buf_sz = PAGE_SIZE; 726 727 tfm = crypto_alloc_shash("sha256", 0, 0); 728 if (IS_ERR(tfm)) { 729 ret = PTR_ERR(tfm); 730 goto out; 731 } 732 733 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 734 desc = kzalloc(desc_size, GFP_KERNEL); 735 if (!desc) { 736 ret = -ENOMEM; 737 goto out_free_tfm; 738 } 739 740 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 741 sha_regions = vzalloc(sha_region_sz); 742 if (!sha_regions) 743 goto out_free_desc; 744 745 desc->tfm = tfm; 746 747 ret = crypto_shash_init(desc); 748 if (ret < 0) 749 goto out_free_sha_regions; 750 751 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 752 if (!digest) { 753 ret = -ENOMEM; 754 goto out_free_sha_regions; 755 } 756 757 for (j = i = 0; i < image->nr_segments; i++) { 758 struct kexec_segment *ksegment; 759 760 ksegment = &image->segment[i]; 761 /* 762 * Skip purgatory as it will be modified once we put digest 763 * info in purgatory. 764 */ 765 if (ksegment->kbuf == pi->purgatory_buf) 766 continue; 767 768 ret = crypto_shash_update(desc, ksegment->kbuf, 769 ksegment->bufsz); 770 if (ret) 771 break; 772 773 /* 774 * Assume rest of the buffer is filled with zero and 775 * update digest accordingly. 776 */ 777 nullsz = ksegment->memsz - ksegment->bufsz; 778 while (nullsz) { 779 unsigned long bytes = nullsz; 780 781 if (bytes > zero_buf_sz) 782 bytes = zero_buf_sz; 783 ret = crypto_shash_update(desc, zero_buf, bytes); 784 if (ret) 785 break; 786 nullsz -= bytes; 787 } 788 789 if (ret) 790 break; 791 792 sha_regions[j].start = ksegment->mem; 793 sha_regions[j].len = ksegment->memsz; 794 j++; 795 } 796 797 if (!ret) { 798 ret = crypto_shash_final(desc, digest); 799 if (ret) 800 goto out_free_digest; 801 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 802 sha_regions, sha_region_sz, 0); 803 if (ret) 804 goto out_free_digest; 805 806 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 807 digest, SHA256_DIGEST_SIZE, 0); 808 if (ret) 809 goto out_free_digest; 810 } 811 812 out_free_digest: 813 kfree(digest); 814 out_free_sha_regions: 815 vfree(sha_regions); 816 out_free_desc: 817 kfree(desc); 818 out_free_tfm: 819 kfree(tfm); 820 out: 821 return ret; 822 } 823 824 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 825 /* 826 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 827 * @pi: Purgatory to be loaded. 828 * @kbuf: Buffer to setup. 829 * 830 * Allocates the memory needed for the buffer. Caller is responsible to free 831 * the memory after use. 832 * 833 * Return: 0 on success, negative errno on error. 834 */ 835 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 836 struct kexec_buf *kbuf) 837 { 838 const Elf_Shdr *sechdrs; 839 unsigned long bss_align; 840 unsigned long bss_sz; 841 unsigned long align; 842 int i, ret; 843 844 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 845 kbuf->buf_align = bss_align = 1; 846 kbuf->bufsz = bss_sz = 0; 847 848 for (i = 0; i < pi->ehdr->e_shnum; i++) { 849 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 850 continue; 851 852 align = sechdrs[i].sh_addralign; 853 if (sechdrs[i].sh_type != SHT_NOBITS) { 854 if (kbuf->buf_align < align) 855 kbuf->buf_align = align; 856 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 857 kbuf->bufsz += sechdrs[i].sh_size; 858 } else { 859 if (bss_align < align) 860 bss_align = align; 861 bss_sz = ALIGN(bss_sz, align); 862 bss_sz += sechdrs[i].sh_size; 863 } 864 } 865 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 866 kbuf->memsz = kbuf->bufsz + bss_sz; 867 if (kbuf->buf_align < bss_align) 868 kbuf->buf_align = bss_align; 869 870 kbuf->buffer = vzalloc(kbuf->bufsz); 871 if (!kbuf->buffer) 872 return -ENOMEM; 873 pi->purgatory_buf = kbuf->buffer; 874 875 ret = kexec_add_buffer(kbuf); 876 if (ret) 877 goto out; 878 879 return 0; 880 out: 881 vfree(pi->purgatory_buf); 882 pi->purgatory_buf = NULL; 883 return ret; 884 } 885 886 /* 887 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 888 * @pi: Purgatory to be loaded. 889 * @kbuf: Buffer prepared to store purgatory. 890 * 891 * Allocates the memory needed for the buffer. Caller is responsible to free 892 * the memory after use. 893 * 894 * Return: 0 on success, negative errno on error. 895 */ 896 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 897 struct kexec_buf *kbuf) 898 { 899 unsigned long bss_addr; 900 unsigned long offset; 901 Elf_Shdr *sechdrs; 902 int i; 903 904 /* 905 * The section headers in kexec_purgatory are read-only. In order to 906 * have them modifiable make a temporary copy. 907 */ 908 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 909 if (!sechdrs) 910 return -ENOMEM; 911 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 912 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 913 pi->sechdrs = sechdrs; 914 915 offset = 0; 916 bss_addr = kbuf->mem + kbuf->bufsz; 917 kbuf->image->start = pi->ehdr->e_entry; 918 919 for (i = 0; i < pi->ehdr->e_shnum; i++) { 920 unsigned long align; 921 void *src, *dst; 922 923 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 924 continue; 925 926 align = sechdrs[i].sh_addralign; 927 if (sechdrs[i].sh_type == SHT_NOBITS) { 928 bss_addr = ALIGN(bss_addr, align); 929 sechdrs[i].sh_addr = bss_addr; 930 bss_addr += sechdrs[i].sh_size; 931 continue; 932 } 933 934 offset = ALIGN(offset, align); 935 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 936 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 937 pi->ehdr->e_entry < (sechdrs[i].sh_addr 938 + sechdrs[i].sh_size)) { 939 kbuf->image->start -= sechdrs[i].sh_addr; 940 kbuf->image->start += kbuf->mem + offset; 941 } 942 943 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 944 dst = pi->purgatory_buf + offset; 945 memcpy(dst, src, sechdrs[i].sh_size); 946 947 sechdrs[i].sh_addr = kbuf->mem + offset; 948 sechdrs[i].sh_offset = offset; 949 offset += sechdrs[i].sh_size; 950 } 951 952 return 0; 953 } 954 955 static int kexec_apply_relocations(struct kimage *image) 956 { 957 int i, ret; 958 struct purgatory_info *pi = &image->purgatory_info; 959 const Elf_Shdr *sechdrs; 960 961 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 962 963 for (i = 0; i < pi->ehdr->e_shnum; i++) { 964 const Elf_Shdr *relsec; 965 const Elf_Shdr *symtab; 966 Elf_Shdr *section; 967 968 relsec = sechdrs + i; 969 970 if (relsec->sh_type != SHT_RELA && 971 relsec->sh_type != SHT_REL) 972 continue; 973 974 /* 975 * For section of type SHT_RELA/SHT_REL, 976 * ->sh_link contains section header index of associated 977 * symbol table. And ->sh_info contains section header 978 * index of section to which relocations apply. 979 */ 980 if (relsec->sh_info >= pi->ehdr->e_shnum || 981 relsec->sh_link >= pi->ehdr->e_shnum) 982 return -ENOEXEC; 983 984 section = pi->sechdrs + relsec->sh_info; 985 symtab = sechdrs + relsec->sh_link; 986 987 if (!(section->sh_flags & SHF_ALLOC)) 988 continue; 989 990 /* 991 * symtab->sh_link contain section header index of associated 992 * string table. 993 */ 994 if (symtab->sh_link >= pi->ehdr->e_shnum) 995 /* Invalid section number? */ 996 continue; 997 998 /* 999 * Respective architecture needs to provide support for applying 1000 * relocations of type SHT_RELA/SHT_REL. 1001 */ 1002 if (relsec->sh_type == SHT_RELA) 1003 ret = arch_kexec_apply_relocations_add(pi, section, 1004 relsec, symtab); 1005 else if (relsec->sh_type == SHT_REL) 1006 ret = arch_kexec_apply_relocations(pi, section, 1007 relsec, symtab); 1008 if (ret) 1009 return ret; 1010 } 1011 1012 return 0; 1013 } 1014 1015 /* 1016 * kexec_load_purgatory - Load and relocate the purgatory object. 1017 * @image: Image to add the purgatory to. 1018 * @kbuf: Memory parameters to use. 1019 * 1020 * Allocates the memory needed for image->purgatory_info.sechdrs and 1021 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 1022 * to free the memory after use. 1023 * 1024 * Return: 0 on success, negative errno on error. 1025 */ 1026 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 1027 { 1028 struct purgatory_info *pi = &image->purgatory_info; 1029 int ret; 1030 1031 if (kexec_purgatory_size <= 0) 1032 return -EINVAL; 1033 1034 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 1035 1036 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 1037 if (ret) 1038 return ret; 1039 1040 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 1041 if (ret) 1042 goto out_free_kbuf; 1043 1044 ret = kexec_apply_relocations(image); 1045 if (ret) 1046 goto out; 1047 1048 return 0; 1049 out: 1050 vfree(pi->sechdrs); 1051 pi->sechdrs = NULL; 1052 out_free_kbuf: 1053 vfree(pi->purgatory_buf); 1054 pi->purgatory_buf = NULL; 1055 return ret; 1056 } 1057 1058 /* 1059 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1060 * @pi: Purgatory to search in. 1061 * @name: Name of the symbol. 1062 * 1063 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1064 */ 1065 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1066 const char *name) 1067 { 1068 const Elf_Shdr *sechdrs; 1069 const Elf_Ehdr *ehdr; 1070 const Elf_Sym *syms; 1071 const char *strtab; 1072 int i, k; 1073 1074 if (!pi->ehdr) 1075 return NULL; 1076 1077 ehdr = pi->ehdr; 1078 sechdrs = (void *)ehdr + ehdr->e_shoff; 1079 1080 for (i = 0; i < ehdr->e_shnum; i++) { 1081 if (sechdrs[i].sh_type != SHT_SYMTAB) 1082 continue; 1083 1084 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1085 /* Invalid strtab section number */ 1086 continue; 1087 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1088 syms = (void *)ehdr + sechdrs[i].sh_offset; 1089 1090 /* Go through symbols for a match */ 1091 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1092 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1093 continue; 1094 1095 if (strcmp(strtab + syms[k].st_name, name) != 0) 1096 continue; 1097 1098 if (syms[k].st_shndx == SHN_UNDEF || 1099 syms[k].st_shndx >= ehdr->e_shnum) { 1100 pr_debug("Symbol: %s has bad section index %d.\n", 1101 name, syms[k].st_shndx); 1102 return NULL; 1103 } 1104 1105 /* Found the symbol we are looking for */ 1106 return &syms[k]; 1107 } 1108 } 1109 1110 return NULL; 1111 } 1112 1113 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1114 { 1115 struct purgatory_info *pi = &image->purgatory_info; 1116 const Elf_Sym *sym; 1117 Elf_Shdr *sechdr; 1118 1119 sym = kexec_purgatory_find_symbol(pi, name); 1120 if (!sym) 1121 return ERR_PTR(-EINVAL); 1122 1123 sechdr = &pi->sechdrs[sym->st_shndx]; 1124 1125 /* 1126 * Returns the address where symbol will finally be loaded after 1127 * kexec_load_segment() 1128 */ 1129 return (void *)(sechdr->sh_addr + sym->st_value); 1130 } 1131 1132 /* 1133 * Get or set value of a symbol. If "get_value" is true, symbol value is 1134 * returned in buf otherwise symbol value is set based on value in buf. 1135 */ 1136 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1137 void *buf, unsigned int size, bool get_value) 1138 { 1139 struct purgatory_info *pi = &image->purgatory_info; 1140 const Elf_Sym *sym; 1141 Elf_Shdr *sec; 1142 char *sym_buf; 1143 1144 sym = kexec_purgatory_find_symbol(pi, name); 1145 if (!sym) 1146 return -EINVAL; 1147 1148 if (sym->st_size != size) { 1149 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1150 name, (unsigned long)sym->st_size, size); 1151 return -EINVAL; 1152 } 1153 1154 sec = pi->sechdrs + sym->st_shndx; 1155 1156 if (sec->sh_type == SHT_NOBITS) { 1157 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1158 get_value ? "get" : "set"); 1159 return -EINVAL; 1160 } 1161 1162 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1163 1164 if (get_value) 1165 memcpy((void *)buf, sym_buf, size); 1166 else 1167 memcpy((void *)sym_buf, buf, size); 1168 1169 return 0; 1170 } 1171 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1172 1173 int crash_exclude_mem_range(struct crash_mem *mem, 1174 unsigned long long mstart, unsigned long long mend) 1175 { 1176 int i, j; 1177 unsigned long long start, end; 1178 struct crash_mem_range temp_range = {0, 0}; 1179 1180 for (i = 0; i < mem->nr_ranges; i++) { 1181 start = mem->ranges[i].start; 1182 end = mem->ranges[i].end; 1183 1184 if (mstart > end || mend < start) 1185 continue; 1186 1187 /* Truncate any area outside of range */ 1188 if (mstart < start) 1189 mstart = start; 1190 if (mend > end) 1191 mend = end; 1192 1193 /* Found completely overlapping range */ 1194 if (mstart == start && mend == end) { 1195 mem->ranges[i].start = 0; 1196 mem->ranges[i].end = 0; 1197 if (i < mem->nr_ranges - 1) { 1198 /* Shift rest of the ranges to left */ 1199 for (j = i; j < mem->nr_ranges - 1; j++) { 1200 mem->ranges[j].start = 1201 mem->ranges[j+1].start; 1202 mem->ranges[j].end = 1203 mem->ranges[j+1].end; 1204 } 1205 } 1206 mem->nr_ranges--; 1207 return 0; 1208 } 1209 1210 if (mstart > start && mend < end) { 1211 /* Split original range */ 1212 mem->ranges[i].end = mstart - 1; 1213 temp_range.start = mend + 1; 1214 temp_range.end = end; 1215 } else if (mstart != start) 1216 mem->ranges[i].end = mstart - 1; 1217 else 1218 mem->ranges[i].start = mend + 1; 1219 break; 1220 } 1221 1222 /* If a split happened, add the split to array */ 1223 if (!temp_range.end) 1224 return 0; 1225 1226 /* Split happened */ 1227 if (i == mem->max_nr_ranges - 1) 1228 return -ENOMEM; 1229 1230 /* Location where new range should go */ 1231 j = i + 1; 1232 if (j < mem->nr_ranges) { 1233 /* Move over all ranges one slot towards the end */ 1234 for (i = mem->nr_ranges - 1; i >= j; i--) 1235 mem->ranges[i + 1] = mem->ranges[i]; 1236 } 1237 1238 mem->ranges[j].start = temp_range.start; 1239 mem->ranges[j].end = temp_range.end; 1240 mem->nr_ranges++; 1241 return 0; 1242 } 1243 1244 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, 1245 void **addr, unsigned long *sz) 1246 { 1247 Elf64_Ehdr *ehdr; 1248 Elf64_Phdr *phdr; 1249 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1250 unsigned char *buf; 1251 unsigned int cpu, i; 1252 unsigned long long notes_addr; 1253 unsigned long mstart, mend; 1254 1255 /* extra phdr for vmcoreinfo elf note */ 1256 nr_phdr = nr_cpus + 1; 1257 nr_phdr += mem->nr_ranges; 1258 1259 /* 1260 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1261 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1262 * I think this is required by tools like gdb. So same physical 1263 * memory will be mapped in two elf headers. One will contain kernel 1264 * text virtual addresses and other will have __va(physical) addresses. 1265 */ 1266 1267 nr_phdr++; 1268 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1269 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1270 1271 buf = vzalloc(elf_sz); 1272 if (!buf) 1273 return -ENOMEM; 1274 1275 ehdr = (Elf64_Ehdr *)buf; 1276 phdr = (Elf64_Phdr *)(ehdr + 1); 1277 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1278 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1279 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1280 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1281 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1282 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1283 ehdr->e_type = ET_CORE; 1284 ehdr->e_machine = ELF_ARCH; 1285 ehdr->e_version = EV_CURRENT; 1286 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1287 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1288 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1289 1290 /* Prepare one phdr of type PT_NOTE for each present cpu */ 1291 for_each_present_cpu(cpu) { 1292 phdr->p_type = PT_NOTE; 1293 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1294 phdr->p_offset = phdr->p_paddr = notes_addr; 1295 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1296 (ehdr->e_phnum)++; 1297 phdr++; 1298 } 1299 1300 /* Prepare one PT_NOTE header for vmcoreinfo */ 1301 phdr->p_type = PT_NOTE; 1302 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1303 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1304 (ehdr->e_phnum)++; 1305 phdr++; 1306 1307 /* Prepare PT_LOAD type program header for kernel text region */ 1308 if (kernel_map) { 1309 phdr->p_type = PT_LOAD; 1310 phdr->p_flags = PF_R|PF_W|PF_X; 1311 phdr->p_vaddr = (unsigned long) _text; 1312 phdr->p_filesz = phdr->p_memsz = _end - _text; 1313 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1314 ehdr->e_phnum++; 1315 phdr++; 1316 } 1317 1318 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1319 for (i = 0; i < mem->nr_ranges; i++) { 1320 mstart = mem->ranges[i].start; 1321 mend = mem->ranges[i].end; 1322 1323 phdr->p_type = PT_LOAD; 1324 phdr->p_flags = PF_R|PF_W|PF_X; 1325 phdr->p_offset = mstart; 1326 1327 phdr->p_paddr = mstart; 1328 phdr->p_vaddr = (unsigned long) __va(mstart); 1329 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1330 phdr->p_align = 0; 1331 ehdr->e_phnum++; 1332 phdr++; 1333 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", 1334 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1335 ehdr->e_phnum, phdr->p_offset); 1336 } 1337 1338 *addr = buf; 1339 *sz = elf_sz; 1340 return 0; 1341 } 1342