1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kexec: kexec_file_load system call 4 * 5 * Copyright (C) 2014 Red Hat Inc. 6 * Authors: 7 * Vivek Goyal <vgoyal@redhat.com> 8 */ 9 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/capability.h> 13 #include <linux/mm.h> 14 #include <linux/file.h> 15 #include <linux/slab.h> 16 #include <linux/kexec.h> 17 #include <linux/memblock.h> 18 #include <linux/mutex.h> 19 #include <linux/list.h> 20 #include <linux/fs.h> 21 #include <linux/ima.h> 22 #include <crypto/hash.h> 23 #include <crypto/sha2.h> 24 #include <linux/elf.h> 25 #include <linux/elfcore.h> 26 #include <linux/kernel.h> 27 #include <linux/kernel_read_file.h> 28 #include <linux/syscalls.h> 29 #include <linux/vmalloc.h> 30 #include "kexec_internal.h" 31 32 #ifdef CONFIG_KEXEC_SIG 33 static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE); 34 35 void set_kexec_sig_enforced(void) 36 { 37 sig_enforce = true; 38 } 39 #endif 40 41 static int kexec_calculate_store_digests(struct kimage *image); 42 43 /* Maximum size in bytes for kernel/initrd files. */ 44 #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX) 45 46 /* 47 * Currently this is the only default function that is exported as some 48 * architectures need it to do additional handlings. 49 * In the future, other default functions may be exported too if required. 50 */ 51 int kexec_image_probe_default(struct kimage *image, void *buf, 52 unsigned long buf_len) 53 { 54 const struct kexec_file_ops * const *fops; 55 int ret = -ENOEXEC; 56 57 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 58 ret = (*fops)->probe(buf, buf_len); 59 if (!ret) { 60 image->fops = *fops; 61 return ret; 62 } 63 } 64 65 return ret; 66 } 67 68 static void *kexec_image_load_default(struct kimage *image) 69 { 70 if (!image->fops || !image->fops->load) 71 return ERR_PTR(-ENOEXEC); 72 73 return image->fops->load(image, image->kernel_buf, 74 image->kernel_buf_len, image->initrd_buf, 75 image->initrd_buf_len, image->cmdline_buf, 76 image->cmdline_buf_len); 77 } 78 79 int kexec_image_post_load_cleanup_default(struct kimage *image) 80 { 81 if (!image->fops || !image->fops->cleanup) 82 return 0; 83 84 return image->fops->cleanup(image->image_loader_data); 85 } 86 87 /* 88 * Free up memory used by kernel, initrd, and command line. This is temporary 89 * memory allocation which is not needed any more after these buffers have 90 * been loaded into separate segments and have been copied elsewhere. 91 */ 92 void kimage_file_post_load_cleanup(struct kimage *image) 93 { 94 struct purgatory_info *pi = &image->purgatory_info; 95 96 vfree(image->kernel_buf); 97 image->kernel_buf = NULL; 98 99 vfree(image->initrd_buf); 100 image->initrd_buf = NULL; 101 102 kfree(image->cmdline_buf); 103 image->cmdline_buf = NULL; 104 105 vfree(pi->purgatory_buf); 106 pi->purgatory_buf = NULL; 107 108 vfree(pi->sechdrs); 109 pi->sechdrs = NULL; 110 111 #ifdef CONFIG_IMA_KEXEC 112 vfree(image->ima_buffer); 113 image->ima_buffer = NULL; 114 #endif /* CONFIG_IMA_KEXEC */ 115 116 /* See if architecture has anything to cleanup post load */ 117 arch_kimage_file_post_load_cleanup(image); 118 119 /* 120 * Above call should have called into bootloader to free up 121 * any data stored in kimage->image_loader_data. It should 122 * be ok now to free it up. 123 */ 124 kfree(image->image_loader_data); 125 image->image_loader_data = NULL; 126 } 127 128 #ifdef CONFIG_KEXEC_SIG 129 #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION 130 int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len) 131 { 132 int ret; 133 134 ret = verify_pefile_signature(kernel, kernel_len, 135 VERIFY_USE_SECONDARY_KEYRING, 136 VERIFYING_KEXEC_PE_SIGNATURE); 137 if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) { 138 ret = verify_pefile_signature(kernel, kernel_len, 139 VERIFY_USE_PLATFORM_KEYRING, 140 VERIFYING_KEXEC_PE_SIGNATURE); 141 } 142 return ret; 143 } 144 #endif 145 146 static int kexec_image_verify_sig(struct kimage *image, void *buf, 147 unsigned long buf_len) 148 { 149 if (!image->fops || !image->fops->verify_sig) { 150 pr_debug("kernel loader does not support signature verification.\n"); 151 return -EKEYREJECTED; 152 } 153 154 return image->fops->verify_sig(buf, buf_len); 155 } 156 157 static int 158 kimage_validate_signature(struct kimage *image) 159 { 160 int ret; 161 162 ret = kexec_image_verify_sig(image, image->kernel_buf, 163 image->kernel_buf_len); 164 if (ret) { 165 166 if (sig_enforce) { 167 pr_notice("Enforced kernel signature verification failed (%d).\n", ret); 168 return ret; 169 } 170 171 /* 172 * If IMA is guaranteed to appraise a signature on the kexec 173 * image, permit it even if the kernel is otherwise locked 174 * down. 175 */ 176 if (!ima_appraise_signature(READING_KEXEC_IMAGE) && 177 security_locked_down(LOCKDOWN_KEXEC)) 178 return -EPERM; 179 180 pr_debug("kernel signature verification failed (%d).\n", ret); 181 } 182 183 return 0; 184 } 185 #endif 186 187 /* 188 * In file mode list of segments is prepared by kernel. Copy relevant 189 * data from user space, do error checking, prepare segment list 190 */ 191 static int 192 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 193 const char __user *cmdline_ptr, 194 unsigned long cmdline_len, unsigned flags) 195 { 196 ssize_t ret; 197 void *ldata; 198 199 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf, 200 KEXEC_FILE_SIZE_MAX, NULL, 201 READING_KEXEC_IMAGE); 202 if (ret < 0) 203 return ret; 204 image->kernel_buf_len = ret; 205 206 /* Call arch image probe handlers */ 207 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, 208 image->kernel_buf_len); 209 if (ret) 210 goto out; 211 212 #ifdef CONFIG_KEXEC_SIG 213 ret = kimage_validate_signature(image); 214 215 if (ret) 216 goto out; 217 #endif 218 /* It is possible that there no initramfs is being loaded */ 219 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { 220 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf, 221 KEXEC_FILE_SIZE_MAX, NULL, 222 READING_KEXEC_INITRAMFS); 223 if (ret < 0) 224 goto out; 225 image->initrd_buf_len = ret; 226 ret = 0; 227 } 228 229 if (cmdline_len) { 230 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); 231 if (IS_ERR(image->cmdline_buf)) { 232 ret = PTR_ERR(image->cmdline_buf); 233 image->cmdline_buf = NULL; 234 goto out; 235 } 236 237 image->cmdline_buf_len = cmdline_len; 238 239 /* command line should be a string with last byte null */ 240 if (image->cmdline_buf[cmdline_len - 1] != '\0') { 241 ret = -EINVAL; 242 goto out; 243 } 244 245 ima_kexec_cmdline(kernel_fd, image->cmdline_buf, 246 image->cmdline_buf_len - 1); 247 } 248 249 /* IMA needs to pass the measurement list to the next kernel. */ 250 ima_add_kexec_buffer(image); 251 252 /* Call image load handler */ 253 ldata = kexec_image_load_default(image); 254 255 if (IS_ERR(ldata)) { 256 ret = PTR_ERR(ldata); 257 goto out; 258 } 259 260 image->image_loader_data = ldata; 261 out: 262 /* In case of error, free up all allocated memory in this function */ 263 if (ret) 264 kimage_file_post_load_cleanup(image); 265 return ret; 266 } 267 268 static int 269 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, 270 int initrd_fd, const char __user *cmdline_ptr, 271 unsigned long cmdline_len, unsigned long flags) 272 { 273 int ret; 274 struct kimage *image; 275 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; 276 277 image = do_kimage_alloc_init(); 278 if (!image) 279 return -ENOMEM; 280 281 image->file_mode = 1; 282 283 if (kexec_on_panic) { 284 /* Enable special crash kernel control page alloc policy. */ 285 image->control_page = crashk_res.start; 286 image->type = KEXEC_TYPE_CRASH; 287 } 288 289 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 290 cmdline_ptr, cmdline_len, flags); 291 if (ret) 292 goto out_free_image; 293 294 ret = sanity_check_segment_list(image); 295 if (ret) 296 goto out_free_post_load_bufs; 297 298 ret = -ENOMEM; 299 image->control_code_page = kimage_alloc_control_pages(image, 300 get_order(KEXEC_CONTROL_PAGE_SIZE)); 301 if (!image->control_code_page) { 302 pr_err("Could not allocate control_code_buffer\n"); 303 goto out_free_post_load_bufs; 304 } 305 306 if (!kexec_on_panic) { 307 image->swap_page = kimage_alloc_control_pages(image, 0); 308 if (!image->swap_page) { 309 pr_err("Could not allocate swap buffer\n"); 310 goto out_free_control_pages; 311 } 312 } 313 314 *rimage = image; 315 return 0; 316 out_free_control_pages: 317 kimage_free_page_list(&image->control_pages); 318 out_free_post_load_bufs: 319 kimage_file_post_load_cleanup(image); 320 out_free_image: 321 kfree(image); 322 return ret; 323 } 324 325 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 326 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 327 unsigned long, flags) 328 { 329 int image_type = (flags & KEXEC_FILE_ON_CRASH) ? 330 KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT; 331 struct kimage **dest_image, *image; 332 int ret = 0, i; 333 334 /* We only trust the superuser with rebooting the system. */ 335 if (!kexec_load_permitted(image_type)) 336 return -EPERM; 337 338 /* Make sure we have a legal set of flags */ 339 if (flags != (flags & KEXEC_FILE_FLAGS)) 340 return -EINVAL; 341 342 image = NULL; 343 344 if (!kexec_trylock()) 345 return -EBUSY; 346 347 if (image_type == KEXEC_TYPE_CRASH) { 348 dest_image = &kexec_crash_image; 349 if (kexec_crash_image) 350 arch_kexec_unprotect_crashkres(); 351 } else { 352 dest_image = &kexec_image; 353 } 354 355 if (flags & KEXEC_FILE_UNLOAD) 356 goto exchange; 357 358 /* 359 * In case of crash, new kernel gets loaded in reserved region. It is 360 * same memory where old crash kernel might be loaded. Free any 361 * current crash dump kernel before we corrupt it. 362 */ 363 if (flags & KEXEC_FILE_ON_CRASH) 364 kimage_free(xchg(&kexec_crash_image, NULL)); 365 366 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 367 cmdline_len, flags); 368 if (ret) 369 goto out; 370 371 ret = machine_kexec_prepare(image); 372 if (ret) 373 goto out; 374 375 /* 376 * Some architecture(like S390) may touch the crash memory before 377 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 378 */ 379 ret = kimage_crash_copy_vmcoreinfo(image); 380 if (ret) 381 goto out; 382 383 ret = kexec_calculate_store_digests(image); 384 if (ret) 385 goto out; 386 387 for (i = 0; i < image->nr_segments; i++) { 388 struct kexec_segment *ksegment; 389 390 ksegment = &image->segment[i]; 391 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 392 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 393 ksegment->memsz); 394 395 ret = kimage_load_segment(image, &image->segment[i]); 396 if (ret) 397 goto out; 398 } 399 400 kimage_terminate(image); 401 402 ret = machine_kexec_post_load(image); 403 if (ret) 404 goto out; 405 406 /* 407 * Free up any temporary buffers allocated which are not needed 408 * after image has been loaded 409 */ 410 kimage_file_post_load_cleanup(image); 411 exchange: 412 image = xchg(dest_image, image); 413 out: 414 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 415 arch_kexec_protect_crashkres(); 416 417 kexec_unlock(); 418 kimage_free(image); 419 return ret; 420 } 421 422 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 423 struct kexec_buf *kbuf) 424 { 425 struct kimage *image = kbuf->image; 426 unsigned long temp_start, temp_end; 427 428 temp_end = min(end, kbuf->buf_max); 429 temp_start = temp_end - kbuf->memsz; 430 431 do { 432 /* align down start */ 433 temp_start = temp_start & (~(kbuf->buf_align - 1)); 434 435 if (temp_start < start || temp_start < kbuf->buf_min) 436 return 0; 437 438 temp_end = temp_start + kbuf->memsz - 1; 439 440 /* 441 * Make sure this does not conflict with any of existing 442 * segments 443 */ 444 if (kimage_is_destination_range(image, temp_start, temp_end)) { 445 temp_start = temp_start - PAGE_SIZE; 446 continue; 447 } 448 449 /* We found a suitable memory range */ 450 break; 451 } while (1); 452 453 /* If we are here, we found a suitable memory range */ 454 kbuf->mem = temp_start; 455 456 /* Success, stop navigating through remaining System RAM ranges */ 457 return 1; 458 } 459 460 static int locate_mem_hole_bottom_up(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_start = max(start, kbuf->buf_min); 467 468 do { 469 temp_start = ALIGN(temp_start, kbuf->buf_align); 470 temp_end = temp_start + kbuf->memsz - 1; 471 472 if (temp_end > end || temp_end > kbuf->buf_max) 473 return 0; 474 /* 475 * Make sure this does not conflict with any of existing 476 * segments 477 */ 478 if (kimage_is_destination_range(image, temp_start, temp_end)) { 479 temp_start = temp_start + PAGE_SIZE; 480 continue; 481 } 482 483 /* We found a suitable memory range */ 484 break; 485 } while (1); 486 487 /* If we are here, we found a suitable memory range */ 488 kbuf->mem = temp_start; 489 490 /* Success, stop navigating through remaining System RAM ranges */ 491 return 1; 492 } 493 494 static int locate_mem_hole_callback(struct resource *res, void *arg) 495 { 496 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 497 u64 start = res->start, end = res->end; 498 unsigned long sz = end - start + 1; 499 500 /* Returning 0 will take to next memory range */ 501 502 /* Don't use memory that will be detected and handled by a driver. */ 503 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) 504 return 0; 505 506 if (sz < kbuf->memsz) 507 return 0; 508 509 if (end < kbuf->buf_min || start > kbuf->buf_max) 510 return 0; 511 512 /* 513 * Allocate memory top down with-in ram range. Otherwise bottom up 514 * allocation. 515 */ 516 if (kbuf->top_down) 517 return locate_mem_hole_top_down(start, end, kbuf); 518 return locate_mem_hole_bottom_up(start, end, kbuf); 519 } 520 521 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 522 static int kexec_walk_memblock(struct kexec_buf *kbuf, 523 int (*func)(struct resource *, void *)) 524 { 525 int ret = 0; 526 u64 i; 527 phys_addr_t mstart, mend; 528 struct resource res = { }; 529 530 if (kbuf->image->type == KEXEC_TYPE_CRASH) 531 return func(&crashk_res, kbuf); 532 533 /* 534 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See 535 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in 536 * locate_mem_hole_callback(). 537 */ 538 if (kbuf->top_down) { 539 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 540 &mstart, &mend, NULL) { 541 /* 542 * In memblock, end points to the first byte after the 543 * range while in kexec, end points to the last byte 544 * in the range. 545 */ 546 res.start = mstart; 547 res.end = mend - 1; 548 ret = func(&res, kbuf); 549 if (ret) 550 break; 551 } 552 } else { 553 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 554 &mstart, &mend, NULL) { 555 /* 556 * In memblock, end points to the first byte after the 557 * range while in kexec, end points to the last byte 558 * in the range. 559 */ 560 res.start = mstart; 561 res.end = mend - 1; 562 ret = func(&res, kbuf); 563 if (ret) 564 break; 565 } 566 } 567 568 return ret; 569 } 570 #else 571 static int kexec_walk_memblock(struct kexec_buf *kbuf, 572 int (*func)(struct resource *, void *)) 573 { 574 return 0; 575 } 576 #endif 577 578 /** 579 * kexec_walk_resources - call func(data) on free memory regions 580 * @kbuf: Context info for the search. Also passed to @func. 581 * @func: Function to call for each memory region. 582 * 583 * Return: The memory walk will stop when func returns a non-zero value 584 * and that value will be returned. If all free regions are visited without 585 * func returning non-zero, then zero will be returned. 586 */ 587 static int kexec_walk_resources(struct kexec_buf *kbuf, 588 int (*func)(struct resource *, void *)) 589 { 590 if (kbuf->image->type == KEXEC_TYPE_CRASH) 591 return walk_iomem_res_desc(crashk_res.desc, 592 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 593 crashk_res.start, crashk_res.end, 594 kbuf, func); 595 else 596 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 597 } 598 599 /** 600 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 601 * @kbuf: Parameters for the memory search. 602 * 603 * On success, kbuf->mem will have the start address of the memory region found. 604 * 605 * Return: 0 on success, negative errno on error. 606 */ 607 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 608 { 609 int ret; 610 611 /* Arch knows where to place */ 612 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 613 return 0; 614 615 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 616 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 617 else 618 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 619 620 return ret == 1 ? 0 : -EADDRNOTAVAIL; 621 } 622 623 /** 624 * kexec_add_buffer - place a buffer in a kexec segment 625 * @kbuf: Buffer contents and memory parameters. 626 * 627 * This function assumes that kexec_mutex is held. 628 * On successful return, @kbuf->mem will have the physical address of 629 * the buffer in memory. 630 * 631 * Return: 0 on success, negative errno on error. 632 */ 633 int kexec_add_buffer(struct kexec_buf *kbuf) 634 { 635 struct kexec_segment *ksegment; 636 int ret; 637 638 /* Currently adding segment this way is allowed only in file mode */ 639 if (!kbuf->image->file_mode) 640 return -EINVAL; 641 642 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 643 return -EINVAL; 644 645 /* 646 * Make sure we are not trying to add buffer after allocating 647 * control pages. All segments need to be placed first before 648 * any control pages are allocated. As control page allocation 649 * logic goes through list of segments to make sure there are 650 * no destination overlaps. 651 */ 652 if (!list_empty(&kbuf->image->control_pages)) { 653 WARN_ON(1); 654 return -EINVAL; 655 } 656 657 /* Ensure minimum alignment needed for segments. */ 658 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 659 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 660 661 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 662 ret = arch_kexec_locate_mem_hole(kbuf); 663 if (ret) 664 return ret; 665 666 /* Found a suitable memory range */ 667 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 668 ksegment->kbuf = kbuf->buffer; 669 ksegment->bufsz = kbuf->bufsz; 670 ksegment->mem = kbuf->mem; 671 ksegment->memsz = kbuf->memsz; 672 kbuf->image->nr_segments++; 673 return 0; 674 } 675 676 /* Calculate and store the digest of segments */ 677 static int kexec_calculate_store_digests(struct kimage *image) 678 { 679 struct crypto_shash *tfm; 680 struct shash_desc *desc; 681 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 682 size_t desc_size, nullsz; 683 char *digest; 684 void *zero_buf; 685 struct kexec_sha_region *sha_regions; 686 struct purgatory_info *pi = &image->purgatory_info; 687 688 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 689 return 0; 690 691 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 692 zero_buf_sz = PAGE_SIZE; 693 694 tfm = crypto_alloc_shash("sha256", 0, 0); 695 if (IS_ERR(tfm)) { 696 ret = PTR_ERR(tfm); 697 goto out; 698 } 699 700 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 701 desc = kzalloc(desc_size, GFP_KERNEL); 702 if (!desc) { 703 ret = -ENOMEM; 704 goto out_free_tfm; 705 } 706 707 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 708 sha_regions = vzalloc(sha_region_sz); 709 if (!sha_regions) { 710 ret = -ENOMEM; 711 goto out_free_desc; 712 } 713 714 desc->tfm = tfm; 715 716 ret = crypto_shash_init(desc); 717 if (ret < 0) 718 goto out_free_sha_regions; 719 720 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 721 if (!digest) { 722 ret = -ENOMEM; 723 goto out_free_sha_regions; 724 } 725 726 for (j = i = 0; i < image->nr_segments; i++) { 727 struct kexec_segment *ksegment; 728 729 ksegment = &image->segment[i]; 730 /* 731 * Skip purgatory as it will be modified once we put digest 732 * info in purgatory. 733 */ 734 if (ksegment->kbuf == pi->purgatory_buf) 735 continue; 736 737 ret = crypto_shash_update(desc, ksegment->kbuf, 738 ksegment->bufsz); 739 if (ret) 740 break; 741 742 /* 743 * Assume rest of the buffer is filled with zero and 744 * update digest accordingly. 745 */ 746 nullsz = ksegment->memsz - ksegment->bufsz; 747 while (nullsz) { 748 unsigned long bytes = nullsz; 749 750 if (bytes > zero_buf_sz) 751 bytes = zero_buf_sz; 752 ret = crypto_shash_update(desc, zero_buf, bytes); 753 if (ret) 754 break; 755 nullsz -= bytes; 756 } 757 758 if (ret) 759 break; 760 761 sha_regions[j].start = ksegment->mem; 762 sha_regions[j].len = ksegment->memsz; 763 j++; 764 } 765 766 if (!ret) { 767 ret = crypto_shash_final(desc, digest); 768 if (ret) 769 goto out_free_digest; 770 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 771 sha_regions, sha_region_sz, 0); 772 if (ret) 773 goto out_free_digest; 774 775 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 776 digest, SHA256_DIGEST_SIZE, 0); 777 if (ret) 778 goto out_free_digest; 779 } 780 781 out_free_digest: 782 kfree(digest); 783 out_free_sha_regions: 784 vfree(sha_regions); 785 out_free_desc: 786 kfree(desc); 787 out_free_tfm: 788 kfree(tfm); 789 out: 790 return ret; 791 } 792 793 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 794 /* 795 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 796 * @pi: Purgatory to be loaded. 797 * @kbuf: Buffer to setup. 798 * 799 * Allocates the memory needed for the buffer. Caller is responsible to free 800 * the memory after use. 801 * 802 * Return: 0 on success, negative errno on error. 803 */ 804 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 805 struct kexec_buf *kbuf) 806 { 807 const Elf_Shdr *sechdrs; 808 unsigned long bss_align; 809 unsigned long bss_sz; 810 unsigned long align; 811 int i, ret; 812 813 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 814 kbuf->buf_align = bss_align = 1; 815 kbuf->bufsz = bss_sz = 0; 816 817 for (i = 0; i < pi->ehdr->e_shnum; i++) { 818 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 819 continue; 820 821 align = sechdrs[i].sh_addralign; 822 if (sechdrs[i].sh_type != SHT_NOBITS) { 823 if (kbuf->buf_align < align) 824 kbuf->buf_align = align; 825 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 826 kbuf->bufsz += sechdrs[i].sh_size; 827 } else { 828 if (bss_align < align) 829 bss_align = align; 830 bss_sz = ALIGN(bss_sz, align); 831 bss_sz += sechdrs[i].sh_size; 832 } 833 } 834 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 835 kbuf->memsz = kbuf->bufsz + bss_sz; 836 if (kbuf->buf_align < bss_align) 837 kbuf->buf_align = bss_align; 838 839 kbuf->buffer = vzalloc(kbuf->bufsz); 840 if (!kbuf->buffer) 841 return -ENOMEM; 842 pi->purgatory_buf = kbuf->buffer; 843 844 ret = kexec_add_buffer(kbuf); 845 if (ret) 846 goto out; 847 848 return 0; 849 out: 850 vfree(pi->purgatory_buf); 851 pi->purgatory_buf = NULL; 852 return ret; 853 } 854 855 /* 856 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 857 * @pi: Purgatory to be loaded. 858 * @kbuf: Buffer prepared to store purgatory. 859 * 860 * Allocates the memory needed for the buffer. Caller is responsible to free 861 * the memory after use. 862 * 863 * Return: 0 on success, negative errno on error. 864 */ 865 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 866 struct kexec_buf *kbuf) 867 { 868 unsigned long bss_addr; 869 unsigned long offset; 870 Elf_Shdr *sechdrs; 871 int i; 872 873 /* 874 * The section headers in kexec_purgatory are read-only. In order to 875 * have them modifiable make a temporary copy. 876 */ 877 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 878 if (!sechdrs) 879 return -ENOMEM; 880 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 881 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 882 pi->sechdrs = sechdrs; 883 884 offset = 0; 885 bss_addr = kbuf->mem + kbuf->bufsz; 886 kbuf->image->start = pi->ehdr->e_entry; 887 888 for (i = 0; i < pi->ehdr->e_shnum; i++) { 889 unsigned long align; 890 void *src, *dst; 891 892 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 893 continue; 894 895 align = sechdrs[i].sh_addralign; 896 if (sechdrs[i].sh_type == SHT_NOBITS) { 897 bss_addr = ALIGN(bss_addr, align); 898 sechdrs[i].sh_addr = bss_addr; 899 bss_addr += sechdrs[i].sh_size; 900 continue; 901 } 902 903 offset = ALIGN(offset, align); 904 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 905 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 906 pi->ehdr->e_entry < (sechdrs[i].sh_addr 907 + sechdrs[i].sh_size)) { 908 kbuf->image->start -= sechdrs[i].sh_addr; 909 kbuf->image->start += kbuf->mem + offset; 910 } 911 912 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 913 dst = pi->purgatory_buf + offset; 914 memcpy(dst, src, sechdrs[i].sh_size); 915 916 sechdrs[i].sh_addr = kbuf->mem + offset; 917 sechdrs[i].sh_offset = offset; 918 offset += sechdrs[i].sh_size; 919 } 920 921 return 0; 922 } 923 924 static int kexec_apply_relocations(struct kimage *image) 925 { 926 int i, ret; 927 struct purgatory_info *pi = &image->purgatory_info; 928 const Elf_Shdr *sechdrs; 929 930 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 931 932 for (i = 0; i < pi->ehdr->e_shnum; i++) { 933 const Elf_Shdr *relsec; 934 const Elf_Shdr *symtab; 935 Elf_Shdr *section; 936 937 relsec = sechdrs + i; 938 939 if (relsec->sh_type != SHT_RELA && 940 relsec->sh_type != SHT_REL) 941 continue; 942 943 /* 944 * For section of type SHT_RELA/SHT_REL, 945 * ->sh_link contains section header index of associated 946 * symbol table. And ->sh_info contains section header 947 * index of section to which relocations apply. 948 */ 949 if (relsec->sh_info >= pi->ehdr->e_shnum || 950 relsec->sh_link >= pi->ehdr->e_shnum) 951 return -ENOEXEC; 952 953 section = pi->sechdrs + relsec->sh_info; 954 symtab = sechdrs + relsec->sh_link; 955 956 if (!(section->sh_flags & SHF_ALLOC)) 957 continue; 958 959 /* 960 * symtab->sh_link contain section header index of associated 961 * string table. 962 */ 963 if (symtab->sh_link >= pi->ehdr->e_shnum) 964 /* Invalid section number? */ 965 continue; 966 967 /* 968 * Respective architecture needs to provide support for applying 969 * relocations of type SHT_RELA/SHT_REL. 970 */ 971 if (relsec->sh_type == SHT_RELA) 972 ret = arch_kexec_apply_relocations_add(pi, section, 973 relsec, symtab); 974 else if (relsec->sh_type == SHT_REL) 975 ret = arch_kexec_apply_relocations(pi, section, 976 relsec, symtab); 977 if (ret) 978 return ret; 979 } 980 981 return 0; 982 } 983 984 /* 985 * kexec_load_purgatory - Load and relocate the purgatory object. 986 * @image: Image to add the purgatory to. 987 * @kbuf: Memory parameters to use. 988 * 989 * Allocates the memory needed for image->purgatory_info.sechdrs and 990 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 991 * to free the memory after use. 992 * 993 * Return: 0 on success, negative errno on error. 994 */ 995 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 996 { 997 struct purgatory_info *pi = &image->purgatory_info; 998 int ret; 999 1000 if (kexec_purgatory_size <= 0) 1001 return -EINVAL; 1002 1003 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 1004 1005 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 1006 if (ret) 1007 return ret; 1008 1009 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 1010 if (ret) 1011 goto out_free_kbuf; 1012 1013 ret = kexec_apply_relocations(image); 1014 if (ret) 1015 goto out; 1016 1017 return 0; 1018 out: 1019 vfree(pi->sechdrs); 1020 pi->sechdrs = NULL; 1021 out_free_kbuf: 1022 vfree(pi->purgatory_buf); 1023 pi->purgatory_buf = NULL; 1024 return ret; 1025 } 1026 1027 /* 1028 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1029 * @pi: Purgatory to search in. 1030 * @name: Name of the symbol. 1031 * 1032 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1033 */ 1034 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1035 const char *name) 1036 { 1037 const Elf_Shdr *sechdrs; 1038 const Elf_Ehdr *ehdr; 1039 const Elf_Sym *syms; 1040 const char *strtab; 1041 int i, k; 1042 1043 if (!pi->ehdr) 1044 return NULL; 1045 1046 ehdr = pi->ehdr; 1047 sechdrs = (void *)ehdr + ehdr->e_shoff; 1048 1049 for (i = 0; i < ehdr->e_shnum; i++) { 1050 if (sechdrs[i].sh_type != SHT_SYMTAB) 1051 continue; 1052 1053 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1054 /* Invalid strtab section number */ 1055 continue; 1056 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1057 syms = (void *)ehdr + sechdrs[i].sh_offset; 1058 1059 /* Go through symbols for a match */ 1060 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1061 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1062 continue; 1063 1064 if (strcmp(strtab + syms[k].st_name, name) != 0) 1065 continue; 1066 1067 if (syms[k].st_shndx == SHN_UNDEF || 1068 syms[k].st_shndx >= ehdr->e_shnum) { 1069 pr_debug("Symbol: %s has bad section index %d.\n", 1070 name, syms[k].st_shndx); 1071 return NULL; 1072 } 1073 1074 /* Found the symbol we are looking for */ 1075 return &syms[k]; 1076 } 1077 } 1078 1079 return NULL; 1080 } 1081 1082 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1083 { 1084 struct purgatory_info *pi = &image->purgatory_info; 1085 const Elf_Sym *sym; 1086 Elf_Shdr *sechdr; 1087 1088 sym = kexec_purgatory_find_symbol(pi, name); 1089 if (!sym) 1090 return ERR_PTR(-EINVAL); 1091 1092 sechdr = &pi->sechdrs[sym->st_shndx]; 1093 1094 /* 1095 * Returns the address where symbol will finally be loaded after 1096 * kexec_load_segment() 1097 */ 1098 return (void *)(sechdr->sh_addr + sym->st_value); 1099 } 1100 1101 /* 1102 * Get or set value of a symbol. If "get_value" is true, symbol value is 1103 * returned in buf otherwise symbol value is set based on value in buf. 1104 */ 1105 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1106 void *buf, unsigned int size, bool get_value) 1107 { 1108 struct purgatory_info *pi = &image->purgatory_info; 1109 const Elf_Sym *sym; 1110 Elf_Shdr *sec; 1111 char *sym_buf; 1112 1113 sym = kexec_purgatory_find_symbol(pi, name); 1114 if (!sym) 1115 return -EINVAL; 1116 1117 if (sym->st_size != size) { 1118 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1119 name, (unsigned long)sym->st_size, size); 1120 return -EINVAL; 1121 } 1122 1123 sec = pi->sechdrs + sym->st_shndx; 1124 1125 if (sec->sh_type == SHT_NOBITS) { 1126 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1127 get_value ? "get" : "set"); 1128 return -EINVAL; 1129 } 1130 1131 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1132 1133 if (get_value) 1134 memcpy((void *)buf, sym_buf, size); 1135 else 1136 memcpy((void *)sym_buf, buf, size); 1137 1138 return 0; 1139 } 1140 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1141 1142 int crash_exclude_mem_range(struct crash_mem *mem, 1143 unsigned long long mstart, unsigned long long mend) 1144 { 1145 int i, j; 1146 unsigned long long start, end, p_start, p_end; 1147 struct range temp_range = {0, 0}; 1148 1149 for (i = 0; i < mem->nr_ranges; i++) { 1150 start = mem->ranges[i].start; 1151 end = mem->ranges[i].end; 1152 p_start = mstart; 1153 p_end = mend; 1154 1155 if (mstart > end || mend < start) 1156 continue; 1157 1158 /* Truncate any area outside of range */ 1159 if (mstart < start) 1160 p_start = start; 1161 if (mend > end) 1162 p_end = end; 1163 1164 /* Found completely overlapping range */ 1165 if (p_start == start && p_end == end) { 1166 mem->ranges[i].start = 0; 1167 mem->ranges[i].end = 0; 1168 if (i < mem->nr_ranges - 1) { 1169 /* Shift rest of the ranges to left */ 1170 for (j = i; j < mem->nr_ranges - 1; j++) { 1171 mem->ranges[j].start = 1172 mem->ranges[j+1].start; 1173 mem->ranges[j].end = 1174 mem->ranges[j+1].end; 1175 } 1176 1177 /* 1178 * Continue to check if there are another overlapping ranges 1179 * from the current position because of shifting the above 1180 * mem ranges. 1181 */ 1182 i--; 1183 mem->nr_ranges--; 1184 continue; 1185 } 1186 mem->nr_ranges--; 1187 return 0; 1188 } 1189 1190 if (p_start > start && p_end < end) { 1191 /* Split original range */ 1192 mem->ranges[i].end = p_start - 1; 1193 temp_range.start = p_end + 1; 1194 temp_range.end = end; 1195 } else if (p_start != start) 1196 mem->ranges[i].end = p_start - 1; 1197 else 1198 mem->ranges[i].start = p_end + 1; 1199 break; 1200 } 1201 1202 /* If a split happened, add the split to array */ 1203 if (!temp_range.end) 1204 return 0; 1205 1206 /* Split happened */ 1207 if (i == mem->max_nr_ranges - 1) 1208 return -ENOMEM; 1209 1210 /* Location where new range should go */ 1211 j = i + 1; 1212 if (j < mem->nr_ranges) { 1213 /* Move over all ranges one slot towards the end */ 1214 for (i = mem->nr_ranges - 1; i >= j; i--) 1215 mem->ranges[i + 1] = mem->ranges[i]; 1216 } 1217 1218 mem->ranges[j].start = temp_range.start; 1219 mem->ranges[j].end = temp_range.end; 1220 mem->nr_ranges++; 1221 return 0; 1222 } 1223 1224 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map, 1225 void **addr, unsigned long *sz) 1226 { 1227 Elf64_Ehdr *ehdr; 1228 Elf64_Phdr *phdr; 1229 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1230 unsigned char *buf; 1231 unsigned int cpu, i; 1232 unsigned long long notes_addr; 1233 unsigned long mstart, mend; 1234 1235 /* extra phdr for vmcoreinfo ELF note */ 1236 nr_phdr = nr_cpus + 1; 1237 nr_phdr += mem->nr_ranges; 1238 1239 /* 1240 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1241 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1242 * I think this is required by tools like gdb. So same physical 1243 * memory will be mapped in two ELF headers. One will contain kernel 1244 * text virtual addresses and other will have __va(physical) addresses. 1245 */ 1246 1247 nr_phdr++; 1248 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1249 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1250 1251 buf = vzalloc(elf_sz); 1252 if (!buf) 1253 return -ENOMEM; 1254 1255 ehdr = (Elf64_Ehdr *)buf; 1256 phdr = (Elf64_Phdr *)(ehdr + 1); 1257 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1258 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1259 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1260 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1261 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1262 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1263 ehdr->e_type = ET_CORE; 1264 ehdr->e_machine = ELF_ARCH; 1265 ehdr->e_version = EV_CURRENT; 1266 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1267 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1268 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1269 1270 /* Prepare one phdr of type PT_NOTE for each present CPU */ 1271 for_each_present_cpu(cpu) { 1272 phdr->p_type = PT_NOTE; 1273 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1274 phdr->p_offset = phdr->p_paddr = notes_addr; 1275 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1276 (ehdr->e_phnum)++; 1277 phdr++; 1278 } 1279 1280 /* Prepare one PT_NOTE header for vmcoreinfo */ 1281 phdr->p_type = PT_NOTE; 1282 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1283 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1284 (ehdr->e_phnum)++; 1285 phdr++; 1286 1287 /* Prepare PT_LOAD type program header for kernel text region */ 1288 if (need_kernel_map) { 1289 phdr->p_type = PT_LOAD; 1290 phdr->p_flags = PF_R|PF_W|PF_X; 1291 phdr->p_vaddr = (unsigned long) _text; 1292 phdr->p_filesz = phdr->p_memsz = _end - _text; 1293 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1294 ehdr->e_phnum++; 1295 phdr++; 1296 } 1297 1298 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1299 for (i = 0; i < mem->nr_ranges; i++) { 1300 mstart = mem->ranges[i].start; 1301 mend = mem->ranges[i].end; 1302 1303 phdr->p_type = PT_LOAD; 1304 phdr->p_flags = PF_R|PF_W|PF_X; 1305 phdr->p_offset = mstart; 1306 1307 phdr->p_paddr = mstart; 1308 phdr->p_vaddr = (unsigned long) __va(mstart); 1309 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1310 phdr->p_align = 0; 1311 ehdr->e_phnum++; 1312 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", 1313 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1314 ehdr->e_phnum, phdr->p_offset); 1315 phdr++; 1316 } 1317 1318 *addr = buf; 1319 *sz = elf_sz; 1320 return 0; 1321 } 1322