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