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