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 static int kexec_calculate_store_digests(struct kimage *image); 33 34 /* Maximum size in bytes for kernel/initrd files. */ 35 #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX) 36 37 /* 38 * Currently this is the only default function that is exported as some 39 * architectures need it to do additional handlings. 40 * In the future, other default functions may be exported too if required. 41 */ 42 int kexec_image_probe_default(struct kimage *image, void *buf, 43 unsigned long buf_len) 44 { 45 const struct kexec_file_ops * const *fops; 46 int ret = -ENOEXEC; 47 48 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 49 ret = (*fops)->probe(buf, buf_len); 50 if (!ret) { 51 image->fops = *fops; 52 return ret; 53 } 54 } 55 56 return ret; 57 } 58 59 /* Architectures can provide this probe function */ 60 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, 61 unsigned long buf_len) 62 { 63 return kexec_image_probe_default(image, buf, buf_len); 64 } 65 66 static void *kexec_image_load_default(struct kimage *image) 67 { 68 if (!image->fops || !image->fops->load) 69 return ERR_PTR(-ENOEXEC); 70 71 return image->fops->load(image, image->kernel_buf, 72 image->kernel_buf_len, image->initrd_buf, 73 image->initrd_buf_len, image->cmdline_buf, 74 image->cmdline_buf_len); 75 } 76 77 void * __weak arch_kexec_kernel_image_load(struct kimage *image) 78 { 79 return kexec_image_load_default(image); 80 } 81 82 int kexec_image_post_load_cleanup_default(struct kimage *image) 83 { 84 if (!image->fops || !image->fops->cleanup) 85 return 0; 86 87 return image->fops->cleanup(image->image_loader_data); 88 } 89 90 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) 91 { 92 return kexec_image_post_load_cleanup_default(image); 93 } 94 95 #ifdef CONFIG_KEXEC_SIG 96 static int kexec_image_verify_sig_default(struct kimage *image, void *buf, 97 unsigned long buf_len) 98 { 99 if (!image->fops || !image->fops->verify_sig) { 100 pr_debug("kernel loader does not support signature verification.\n"); 101 return -EKEYREJECTED; 102 } 103 104 return image->fops->verify_sig(buf, buf_len); 105 } 106 107 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, 108 unsigned long buf_len) 109 { 110 return kexec_image_verify_sig_default(image, buf, buf_len); 111 } 112 #endif 113 114 /* 115 * Free up memory used by kernel, initrd, and command line. This is temporary 116 * memory allocation which is not needed any more after these buffers have 117 * been loaded into separate segments and have been copied elsewhere. 118 */ 119 void kimage_file_post_load_cleanup(struct kimage *image) 120 { 121 struct purgatory_info *pi = &image->purgatory_info; 122 123 vfree(image->kernel_buf); 124 image->kernel_buf = NULL; 125 126 vfree(image->initrd_buf); 127 image->initrd_buf = NULL; 128 129 kfree(image->cmdline_buf); 130 image->cmdline_buf = NULL; 131 132 vfree(pi->purgatory_buf); 133 pi->purgatory_buf = NULL; 134 135 vfree(pi->sechdrs); 136 pi->sechdrs = NULL; 137 138 #ifdef CONFIG_IMA_KEXEC 139 vfree(image->ima_buffer); 140 image->ima_buffer = NULL; 141 #endif /* CONFIG_IMA_KEXEC */ 142 143 /* See if architecture has anything to cleanup post load */ 144 arch_kimage_file_post_load_cleanup(image); 145 146 /* 147 * Above call should have called into bootloader to free up 148 * any data stored in kimage->image_loader_data. It should 149 * be ok now to free it up. 150 */ 151 kfree(image->image_loader_data); 152 image->image_loader_data = NULL; 153 } 154 155 #ifdef CONFIG_KEXEC_SIG 156 static int 157 kimage_validate_signature(struct kimage *image) 158 { 159 int ret; 160 161 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, 162 image->kernel_buf_len); 163 if (ret) { 164 165 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) { 166 pr_notice("Enforced kernel signature verification failed (%d).\n", ret); 167 return ret; 168 } 169 170 /* 171 * If IMA is guaranteed to appraise a signature on the kexec 172 * image, permit it even if the kernel is otherwise locked 173 * down. 174 */ 175 if (!ima_appraise_signature(READING_KEXEC_IMAGE) && 176 security_locked_down(LOCKDOWN_KEXEC)) 177 return -EPERM; 178 179 pr_debug("kernel signature verification failed (%d).\n", ret); 180 } 181 182 return 0; 183 } 184 #endif 185 186 /* 187 * In file mode list of segments is prepared by kernel. Copy relevant 188 * data from user space, do error checking, prepare segment list 189 */ 190 static int 191 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 192 const char __user *cmdline_ptr, 193 unsigned long cmdline_len, unsigned flags) 194 { 195 ssize_t ret; 196 void *ldata; 197 198 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf, 199 KEXEC_FILE_SIZE_MAX, NULL, 200 READING_KEXEC_IMAGE); 201 if (ret < 0) 202 return ret; 203 image->kernel_buf_len = ret; 204 205 /* Call arch image probe handlers */ 206 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, 207 image->kernel_buf_len); 208 if (ret) 209 goto out; 210 211 #ifdef CONFIG_KEXEC_SIG 212 ret = kimage_validate_signature(image); 213 214 if (ret) 215 goto out; 216 #endif 217 /* It is possible that there no initramfs is being loaded */ 218 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { 219 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf, 220 KEXEC_FILE_SIZE_MAX, NULL, 221 READING_KEXEC_INITRAMFS); 222 if (ret < 0) 223 goto out; 224 image->initrd_buf_len = ret; 225 ret = 0; 226 } 227 228 if (cmdline_len) { 229 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); 230 if (IS_ERR(image->cmdline_buf)) { 231 ret = PTR_ERR(image->cmdline_buf); 232 image->cmdline_buf = NULL; 233 goto out; 234 } 235 236 image->cmdline_buf_len = cmdline_len; 237 238 /* command line should be a string with last byte null */ 239 if (image->cmdline_buf[cmdline_len - 1] != '\0') { 240 ret = -EINVAL; 241 goto out; 242 } 243 244 ima_kexec_cmdline(kernel_fd, image->cmdline_buf, 245 image->cmdline_buf_len - 1); 246 } 247 248 /* IMA needs to pass the measurement list to the next kernel. */ 249 ima_add_kexec_buffer(image); 250 251 /* Call arch image load handlers */ 252 ldata = arch_kexec_kernel_image_load(image); 253 254 if (IS_ERR(ldata)) { 255 ret = PTR_ERR(ldata); 256 goto out; 257 } 258 259 image->image_loader_data = ldata; 260 out: 261 /* In case of error, free up all allocated memory in this function */ 262 if (ret) 263 kimage_file_post_load_cleanup(image); 264 return ret; 265 } 266 267 static int 268 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, 269 int initrd_fd, const char __user *cmdline_ptr, 270 unsigned long cmdline_len, unsigned long flags) 271 { 272 int ret; 273 struct kimage *image; 274 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; 275 276 image = do_kimage_alloc_init(); 277 if (!image) 278 return -ENOMEM; 279 280 image->file_mode = 1; 281 282 if (kexec_on_panic) { 283 /* Enable special crash kernel control page alloc policy. */ 284 image->control_page = crashk_res.start; 285 image->type = KEXEC_TYPE_CRASH; 286 } 287 288 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 289 cmdline_ptr, cmdline_len, flags); 290 if (ret) 291 goto out_free_image; 292 293 ret = sanity_check_segment_list(image); 294 if (ret) 295 goto out_free_post_load_bufs; 296 297 ret = -ENOMEM; 298 image->control_code_page = kimage_alloc_control_pages(image, 299 get_order(KEXEC_CONTROL_PAGE_SIZE)); 300 if (!image->control_code_page) { 301 pr_err("Could not allocate control_code_buffer\n"); 302 goto out_free_post_load_bufs; 303 } 304 305 if (!kexec_on_panic) { 306 image->swap_page = kimage_alloc_control_pages(image, 0); 307 if (!image->swap_page) { 308 pr_err("Could not allocate swap buffer\n"); 309 goto out_free_control_pages; 310 } 311 } 312 313 *rimage = image; 314 return 0; 315 out_free_control_pages: 316 kimage_free_page_list(&image->control_pages); 317 out_free_post_load_bufs: 318 kimage_file_post_load_cleanup(image); 319 out_free_image: 320 kfree(image); 321 return ret; 322 } 323 324 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 325 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 326 unsigned long, flags) 327 { 328 int ret = 0, i; 329 struct kimage **dest_image, *image; 330 331 /* We only trust the superuser with rebooting the system. */ 332 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) 333 return -EPERM; 334 335 /* Make sure we have a legal set of flags */ 336 if (flags != (flags & KEXEC_FILE_FLAGS)) 337 return -EINVAL; 338 339 image = NULL; 340 341 if (!mutex_trylock(&kexec_mutex)) 342 return -EBUSY; 343 344 dest_image = &kexec_image; 345 if (flags & KEXEC_FILE_ON_CRASH) { 346 dest_image = &kexec_crash_image; 347 if (kexec_crash_image) 348 arch_kexec_unprotect_crashkres(); 349 } 350 351 if (flags & KEXEC_FILE_UNLOAD) 352 goto exchange; 353 354 /* 355 * In case of crash, new kernel gets loaded in reserved region. It is 356 * same memory where old crash kernel might be loaded. Free any 357 * current crash dump kernel before we corrupt it. 358 */ 359 if (flags & KEXEC_FILE_ON_CRASH) 360 kimage_free(xchg(&kexec_crash_image, NULL)); 361 362 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 363 cmdline_len, flags); 364 if (ret) 365 goto out; 366 367 ret = machine_kexec_prepare(image); 368 if (ret) 369 goto out; 370 371 /* 372 * Some architecture(like S390) may touch the crash memory before 373 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 374 */ 375 ret = kimage_crash_copy_vmcoreinfo(image); 376 if (ret) 377 goto out; 378 379 ret = kexec_calculate_store_digests(image); 380 if (ret) 381 goto out; 382 383 for (i = 0; i < image->nr_segments; i++) { 384 struct kexec_segment *ksegment; 385 386 ksegment = &image->segment[i]; 387 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 388 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 389 ksegment->memsz); 390 391 ret = kimage_load_segment(image, &image->segment[i]); 392 if (ret) 393 goto out; 394 } 395 396 kimage_terminate(image); 397 398 ret = machine_kexec_post_load(image); 399 if (ret) 400 goto out; 401 402 /* 403 * Free up any temporary buffers allocated which are not needed 404 * after image has been loaded 405 */ 406 kimage_file_post_load_cleanup(image); 407 exchange: 408 image = xchg(dest_image, image); 409 out: 410 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 411 arch_kexec_protect_crashkres(); 412 413 mutex_unlock(&kexec_mutex); 414 kimage_free(image); 415 return ret; 416 } 417 418 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 419 struct kexec_buf *kbuf) 420 { 421 struct kimage *image = kbuf->image; 422 unsigned long temp_start, temp_end; 423 424 temp_end = min(end, kbuf->buf_max); 425 temp_start = temp_end - kbuf->memsz; 426 427 do { 428 /* align down start */ 429 temp_start = temp_start & (~(kbuf->buf_align - 1)); 430 431 if (temp_start < start || temp_start < kbuf->buf_min) 432 return 0; 433 434 temp_end = temp_start + kbuf->memsz - 1; 435 436 /* 437 * Make sure this does not conflict with any of existing 438 * segments 439 */ 440 if (kimage_is_destination_range(image, temp_start, temp_end)) { 441 temp_start = temp_start - PAGE_SIZE; 442 continue; 443 } 444 445 /* We found a suitable memory range */ 446 break; 447 } while (1); 448 449 /* If we are here, we found a suitable memory range */ 450 kbuf->mem = temp_start; 451 452 /* Success, stop navigating through remaining System RAM ranges */ 453 return 1; 454 } 455 456 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 457 struct kexec_buf *kbuf) 458 { 459 struct kimage *image = kbuf->image; 460 unsigned long temp_start, temp_end; 461 462 temp_start = max(start, kbuf->buf_min); 463 464 do { 465 temp_start = ALIGN(temp_start, kbuf->buf_align); 466 temp_end = temp_start + kbuf->memsz - 1; 467 468 if (temp_end > end || temp_end > kbuf->buf_max) 469 return 0; 470 /* 471 * Make sure this does not conflict with any of existing 472 * segments 473 */ 474 if (kimage_is_destination_range(image, temp_start, temp_end)) { 475 temp_start = temp_start + PAGE_SIZE; 476 continue; 477 } 478 479 /* We found a suitable memory range */ 480 break; 481 } while (1); 482 483 /* If we are here, we found a suitable memory range */ 484 kbuf->mem = temp_start; 485 486 /* Success, stop navigating through remaining System RAM ranges */ 487 return 1; 488 } 489 490 static int locate_mem_hole_callback(struct resource *res, void *arg) 491 { 492 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 493 u64 start = res->start, end = res->end; 494 unsigned long sz = end - start + 1; 495 496 /* Returning 0 will take to next memory range */ 497 498 /* Don't use memory that will be detected and handled by a driver. */ 499 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) 500 return 0; 501 502 if (sz < kbuf->memsz) 503 return 0; 504 505 if (end < kbuf->buf_min || start > kbuf->buf_max) 506 return 0; 507 508 /* 509 * Allocate memory top down with-in ram range. Otherwise bottom up 510 * allocation. 511 */ 512 if (kbuf->top_down) 513 return locate_mem_hole_top_down(start, end, kbuf); 514 return locate_mem_hole_bottom_up(start, end, kbuf); 515 } 516 517 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 518 static int kexec_walk_memblock(struct kexec_buf *kbuf, 519 int (*func)(struct resource *, void *)) 520 { 521 int ret = 0; 522 u64 i; 523 phys_addr_t mstart, mend; 524 struct resource res = { }; 525 526 if (kbuf->image->type == KEXEC_TYPE_CRASH) 527 return func(&crashk_res, kbuf); 528 529 /* 530 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See 531 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in 532 * locate_mem_hole_callback(). 533 */ 534 if (kbuf->top_down) { 535 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 536 &mstart, &mend, NULL) { 537 /* 538 * In memblock, end points to the first byte after the 539 * range while in kexec, end points to the last byte 540 * in the range. 541 */ 542 res.start = mstart; 543 res.end = mend - 1; 544 ret = func(&res, kbuf); 545 if (ret) 546 break; 547 } 548 } else { 549 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 550 &mstart, &mend, NULL) { 551 /* 552 * In memblock, end points to the first byte after the 553 * range while in kexec, end points to the last byte 554 * in the range. 555 */ 556 res.start = mstart; 557 res.end = mend - 1; 558 ret = func(&res, kbuf); 559 if (ret) 560 break; 561 } 562 } 563 564 return ret; 565 } 566 #else 567 static int kexec_walk_memblock(struct kexec_buf *kbuf, 568 int (*func)(struct resource *, void *)) 569 { 570 return 0; 571 } 572 #endif 573 574 /** 575 * kexec_walk_resources - call func(data) on free memory regions 576 * @kbuf: Context info for the search. Also passed to @func. 577 * @func: Function to call for each memory region. 578 * 579 * Return: The memory walk will stop when func returns a non-zero value 580 * and that value will be returned. If all free regions are visited without 581 * func returning non-zero, then zero will be returned. 582 */ 583 static int kexec_walk_resources(struct kexec_buf *kbuf, 584 int (*func)(struct resource *, void *)) 585 { 586 if (kbuf->image->type == KEXEC_TYPE_CRASH) 587 return walk_iomem_res_desc(crashk_res.desc, 588 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 589 crashk_res.start, crashk_res.end, 590 kbuf, func); 591 else 592 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 593 } 594 595 /** 596 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 597 * @kbuf: Parameters for the memory search. 598 * 599 * On success, kbuf->mem will have the start address of the memory region found. 600 * 601 * Return: 0 on success, negative errno on error. 602 */ 603 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 604 { 605 int ret; 606 607 /* Arch knows where to place */ 608 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 609 return 0; 610 611 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 612 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 613 else 614 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 615 616 return ret == 1 ? 0 : -EADDRNOTAVAIL; 617 } 618 619 /** 620 * arch_kexec_locate_mem_hole - Find free memory to place the segments. 621 * @kbuf: Parameters for the memory search. 622 * 623 * On success, kbuf->mem will have the start address of the memory region found. 624 * 625 * Return: 0 on success, negative errno on error. 626 */ 627 int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf) 628 { 629 return kexec_locate_mem_hole(kbuf); 630 } 631 632 /** 633 * kexec_add_buffer - place a buffer in a kexec segment 634 * @kbuf: Buffer contents and memory parameters. 635 * 636 * This function assumes that kexec_mutex is held. 637 * On successful return, @kbuf->mem will have the physical address of 638 * the buffer in memory. 639 * 640 * Return: 0 on success, negative errno on error. 641 */ 642 int kexec_add_buffer(struct kexec_buf *kbuf) 643 { 644 struct kexec_segment *ksegment; 645 int ret; 646 647 /* Currently adding segment this way is allowed only in file mode */ 648 if (!kbuf->image->file_mode) 649 return -EINVAL; 650 651 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 652 return -EINVAL; 653 654 /* 655 * Make sure we are not trying to add buffer after allocating 656 * control pages. All segments need to be placed first before 657 * any control pages are allocated. As control page allocation 658 * logic goes through list of segments to make sure there are 659 * no destination overlaps. 660 */ 661 if (!list_empty(&kbuf->image->control_pages)) { 662 WARN_ON(1); 663 return -EINVAL; 664 } 665 666 /* Ensure minimum alignment needed for segments. */ 667 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 668 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 669 670 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 671 ret = arch_kexec_locate_mem_hole(kbuf); 672 if (ret) 673 return ret; 674 675 /* Found a suitable memory range */ 676 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 677 ksegment->kbuf = kbuf->buffer; 678 ksegment->bufsz = kbuf->bufsz; 679 ksegment->mem = kbuf->mem; 680 ksegment->memsz = kbuf->memsz; 681 kbuf->image->nr_segments++; 682 return 0; 683 } 684 685 /* Calculate and store the digest of segments */ 686 static int kexec_calculate_store_digests(struct kimage *image) 687 { 688 struct crypto_shash *tfm; 689 struct shash_desc *desc; 690 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 691 size_t desc_size, nullsz; 692 char *digest; 693 void *zero_buf; 694 struct kexec_sha_region *sha_regions; 695 struct purgatory_info *pi = &image->purgatory_info; 696 697 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 698 return 0; 699 700 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 701 zero_buf_sz = PAGE_SIZE; 702 703 tfm = crypto_alloc_shash("sha256", 0, 0); 704 if (IS_ERR(tfm)) { 705 ret = PTR_ERR(tfm); 706 goto out; 707 } 708 709 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 710 desc = kzalloc(desc_size, GFP_KERNEL); 711 if (!desc) { 712 ret = -ENOMEM; 713 goto out_free_tfm; 714 } 715 716 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 717 sha_regions = vzalloc(sha_region_sz); 718 if (!sha_regions) { 719 ret = -ENOMEM; 720 goto out_free_desc; 721 } 722 723 desc->tfm = tfm; 724 725 ret = crypto_shash_init(desc); 726 if (ret < 0) 727 goto out_free_sha_regions; 728 729 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 730 if (!digest) { 731 ret = -ENOMEM; 732 goto out_free_sha_regions; 733 } 734 735 for (j = i = 0; i < image->nr_segments; i++) { 736 struct kexec_segment *ksegment; 737 738 ksegment = &image->segment[i]; 739 /* 740 * Skip purgatory as it will be modified once we put digest 741 * info in purgatory. 742 */ 743 if (ksegment->kbuf == pi->purgatory_buf) 744 continue; 745 746 ret = crypto_shash_update(desc, ksegment->kbuf, 747 ksegment->bufsz); 748 if (ret) 749 break; 750 751 /* 752 * Assume rest of the buffer is filled with zero and 753 * update digest accordingly. 754 */ 755 nullsz = ksegment->memsz - ksegment->bufsz; 756 while (nullsz) { 757 unsigned long bytes = nullsz; 758 759 if (bytes > zero_buf_sz) 760 bytes = zero_buf_sz; 761 ret = crypto_shash_update(desc, zero_buf, bytes); 762 if (ret) 763 break; 764 nullsz -= bytes; 765 } 766 767 if (ret) 768 break; 769 770 sha_regions[j].start = ksegment->mem; 771 sha_regions[j].len = ksegment->memsz; 772 j++; 773 } 774 775 if (!ret) { 776 ret = crypto_shash_final(desc, digest); 777 if (ret) 778 goto out_free_digest; 779 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 780 sha_regions, sha_region_sz, 0); 781 if (ret) 782 goto out_free_digest; 783 784 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 785 digest, SHA256_DIGEST_SIZE, 0); 786 if (ret) 787 goto out_free_digest; 788 } 789 790 out_free_digest: 791 kfree(digest); 792 out_free_sha_regions: 793 vfree(sha_regions); 794 out_free_desc: 795 kfree(desc); 796 out_free_tfm: 797 kfree(tfm); 798 out: 799 return ret; 800 } 801 802 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 803 /* 804 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 805 * @pi: Purgatory to be loaded. 806 * @kbuf: Buffer to setup. 807 * 808 * Allocates the memory needed for the buffer. Caller is responsible to free 809 * the memory after use. 810 * 811 * Return: 0 on success, negative errno on error. 812 */ 813 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 814 struct kexec_buf *kbuf) 815 { 816 const Elf_Shdr *sechdrs; 817 unsigned long bss_align; 818 unsigned long bss_sz; 819 unsigned long align; 820 int i, ret; 821 822 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 823 kbuf->buf_align = bss_align = 1; 824 kbuf->bufsz = bss_sz = 0; 825 826 for (i = 0; i < pi->ehdr->e_shnum; i++) { 827 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 828 continue; 829 830 align = sechdrs[i].sh_addralign; 831 if (sechdrs[i].sh_type != SHT_NOBITS) { 832 if (kbuf->buf_align < align) 833 kbuf->buf_align = align; 834 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 835 kbuf->bufsz += sechdrs[i].sh_size; 836 } else { 837 if (bss_align < align) 838 bss_align = align; 839 bss_sz = ALIGN(bss_sz, align); 840 bss_sz += sechdrs[i].sh_size; 841 } 842 } 843 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 844 kbuf->memsz = kbuf->bufsz + bss_sz; 845 if (kbuf->buf_align < bss_align) 846 kbuf->buf_align = bss_align; 847 848 kbuf->buffer = vzalloc(kbuf->bufsz); 849 if (!kbuf->buffer) 850 return -ENOMEM; 851 pi->purgatory_buf = kbuf->buffer; 852 853 ret = kexec_add_buffer(kbuf); 854 if (ret) 855 goto out; 856 857 return 0; 858 out: 859 vfree(pi->purgatory_buf); 860 pi->purgatory_buf = NULL; 861 return ret; 862 } 863 864 /* 865 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 866 * @pi: Purgatory to be loaded. 867 * @kbuf: Buffer prepared to store purgatory. 868 * 869 * Allocates the memory needed for the buffer. Caller is responsible to free 870 * the memory after use. 871 * 872 * Return: 0 on success, negative errno on error. 873 */ 874 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 875 struct kexec_buf *kbuf) 876 { 877 unsigned long bss_addr; 878 unsigned long offset; 879 Elf_Shdr *sechdrs; 880 int i; 881 882 /* 883 * The section headers in kexec_purgatory are read-only. In order to 884 * have them modifiable make a temporary copy. 885 */ 886 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 887 if (!sechdrs) 888 return -ENOMEM; 889 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 890 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 891 pi->sechdrs = sechdrs; 892 893 offset = 0; 894 bss_addr = kbuf->mem + kbuf->bufsz; 895 kbuf->image->start = pi->ehdr->e_entry; 896 897 for (i = 0; i < pi->ehdr->e_shnum; i++) { 898 unsigned long align; 899 void *src, *dst; 900 901 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 902 continue; 903 904 align = sechdrs[i].sh_addralign; 905 if (sechdrs[i].sh_type == SHT_NOBITS) { 906 bss_addr = ALIGN(bss_addr, align); 907 sechdrs[i].sh_addr = bss_addr; 908 bss_addr += sechdrs[i].sh_size; 909 continue; 910 } 911 912 offset = ALIGN(offset, align); 913 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 914 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 915 pi->ehdr->e_entry < (sechdrs[i].sh_addr 916 + sechdrs[i].sh_size)) { 917 kbuf->image->start -= sechdrs[i].sh_addr; 918 kbuf->image->start += kbuf->mem + offset; 919 } 920 921 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 922 dst = pi->purgatory_buf + offset; 923 memcpy(dst, src, sechdrs[i].sh_size); 924 925 sechdrs[i].sh_addr = kbuf->mem + offset; 926 sechdrs[i].sh_offset = offset; 927 offset += sechdrs[i].sh_size; 928 } 929 930 return 0; 931 } 932 933 static int kexec_apply_relocations(struct kimage *image) 934 { 935 int i, ret; 936 struct purgatory_info *pi = &image->purgatory_info; 937 const Elf_Shdr *sechdrs; 938 939 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 940 941 for (i = 0; i < pi->ehdr->e_shnum; i++) { 942 const Elf_Shdr *relsec; 943 const Elf_Shdr *symtab; 944 Elf_Shdr *section; 945 946 relsec = sechdrs + i; 947 948 if (relsec->sh_type != SHT_RELA && 949 relsec->sh_type != SHT_REL) 950 continue; 951 952 /* 953 * For section of type SHT_RELA/SHT_REL, 954 * ->sh_link contains section header index of associated 955 * symbol table. And ->sh_info contains section header 956 * index of section to which relocations apply. 957 */ 958 if (relsec->sh_info >= pi->ehdr->e_shnum || 959 relsec->sh_link >= pi->ehdr->e_shnum) 960 return -ENOEXEC; 961 962 section = pi->sechdrs + relsec->sh_info; 963 symtab = sechdrs + relsec->sh_link; 964 965 if (!(section->sh_flags & SHF_ALLOC)) 966 continue; 967 968 /* 969 * symtab->sh_link contain section header index of associated 970 * string table. 971 */ 972 if (symtab->sh_link >= pi->ehdr->e_shnum) 973 /* Invalid section number? */ 974 continue; 975 976 /* 977 * Respective architecture needs to provide support for applying 978 * relocations of type SHT_RELA/SHT_REL. 979 */ 980 if (relsec->sh_type == SHT_RELA) 981 ret = arch_kexec_apply_relocations_add(pi, section, 982 relsec, symtab); 983 else if (relsec->sh_type == SHT_REL) 984 ret = arch_kexec_apply_relocations(pi, section, 985 relsec, symtab); 986 if (ret) 987 return ret; 988 } 989 990 return 0; 991 } 992 993 /* 994 * kexec_load_purgatory - Load and relocate the purgatory object. 995 * @image: Image to add the purgatory to. 996 * @kbuf: Memory parameters to use. 997 * 998 * Allocates the memory needed for image->purgatory_info.sechdrs and 999 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 1000 * to free the memory after use. 1001 * 1002 * Return: 0 on success, negative errno on error. 1003 */ 1004 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 1005 { 1006 struct purgatory_info *pi = &image->purgatory_info; 1007 int ret; 1008 1009 if (kexec_purgatory_size <= 0) 1010 return -EINVAL; 1011 1012 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 1013 1014 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 1015 if (ret) 1016 return ret; 1017 1018 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 1019 if (ret) 1020 goto out_free_kbuf; 1021 1022 ret = kexec_apply_relocations(image); 1023 if (ret) 1024 goto out; 1025 1026 return 0; 1027 out: 1028 vfree(pi->sechdrs); 1029 pi->sechdrs = NULL; 1030 out_free_kbuf: 1031 vfree(pi->purgatory_buf); 1032 pi->purgatory_buf = NULL; 1033 return ret; 1034 } 1035 1036 /* 1037 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1038 * @pi: Purgatory to search in. 1039 * @name: Name of the symbol. 1040 * 1041 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1042 */ 1043 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1044 const char *name) 1045 { 1046 const Elf_Shdr *sechdrs; 1047 const Elf_Ehdr *ehdr; 1048 const Elf_Sym *syms; 1049 const char *strtab; 1050 int i, k; 1051 1052 if (!pi->ehdr) 1053 return NULL; 1054 1055 ehdr = pi->ehdr; 1056 sechdrs = (void *)ehdr + ehdr->e_shoff; 1057 1058 for (i = 0; i < ehdr->e_shnum; i++) { 1059 if (sechdrs[i].sh_type != SHT_SYMTAB) 1060 continue; 1061 1062 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1063 /* Invalid strtab section number */ 1064 continue; 1065 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1066 syms = (void *)ehdr + sechdrs[i].sh_offset; 1067 1068 /* Go through symbols for a match */ 1069 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1070 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1071 continue; 1072 1073 if (strcmp(strtab + syms[k].st_name, name) != 0) 1074 continue; 1075 1076 if (syms[k].st_shndx == SHN_UNDEF || 1077 syms[k].st_shndx >= ehdr->e_shnum) { 1078 pr_debug("Symbol: %s has bad section index %d.\n", 1079 name, syms[k].st_shndx); 1080 return NULL; 1081 } 1082 1083 /* Found the symbol we are looking for */ 1084 return &syms[k]; 1085 } 1086 } 1087 1088 return NULL; 1089 } 1090 1091 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1092 { 1093 struct purgatory_info *pi = &image->purgatory_info; 1094 const Elf_Sym *sym; 1095 Elf_Shdr *sechdr; 1096 1097 sym = kexec_purgatory_find_symbol(pi, name); 1098 if (!sym) 1099 return ERR_PTR(-EINVAL); 1100 1101 sechdr = &pi->sechdrs[sym->st_shndx]; 1102 1103 /* 1104 * Returns the address where symbol will finally be loaded after 1105 * kexec_load_segment() 1106 */ 1107 return (void *)(sechdr->sh_addr + sym->st_value); 1108 } 1109 1110 /* 1111 * Get or set value of a symbol. If "get_value" is true, symbol value is 1112 * returned in buf otherwise symbol value is set based on value in buf. 1113 */ 1114 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1115 void *buf, unsigned int size, bool get_value) 1116 { 1117 struct purgatory_info *pi = &image->purgatory_info; 1118 const Elf_Sym *sym; 1119 Elf_Shdr *sec; 1120 char *sym_buf; 1121 1122 sym = kexec_purgatory_find_symbol(pi, name); 1123 if (!sym) 1124 return -EINVAL; 1125 1126 if (sym->st_size != size) { 1127 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1128 name, (unsigned long)sym->st_size, size); 1129 return -EINVAL; 1130 } 1131 1132 sec = pi->sechdrs + sym->st_shndx; 1133 1134 if (sec->sh_type == SHT_NOBITS) { 1135 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1136 get_value ? "get" : "set"); 1137 return -EINVAL; 1138 } 1139 1140 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1141 1142 if (get_value) 1143 memcpy((void *)buf, sym_buf, size); 1144 else 1145 memcpy((void *)sym_buf, buf, size); 1146 1147 return 0; 1148 } 1149 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1150 1151 int crash_exclude_mem_range(struct crash_mem *mem, 1152 unsigned long long mstart, unsigned long long mend) 1153 { 1154 int i, j; 1155 unsigned long long start, end, p_start, p_end; 1156 struct crash_mem_range temp_range = {0, 0}; 1157 1158 for (i = 0; i < mem->nr_ranges; i++) { 1159 start = mem->ranges[i].start; 1160 end = mem->ranges[i].end; 1161 p_start = mstart; 1162 p_end = mend; 1163 1164 if (mstart > end || mend < start) 1165 continue; 1166 1167 /* Truncate any area outside of range */ 1168 if (mstart < start) 1169 p_start = start; 1170 if (mend > end) 1171 p_end = end; 1172 1173 /* Found completely overlapping range */ 1174 if (p_start == start && p_end == end) { 1175 mem->ranges[i].start = 0; 1176 mem->ranges[i].end = 0; 1177 if (i < mem->nr_ranges - 1) { 1178 /* Shift rest of the ranges to left */ 1179 for (j = i; j < mem->nr_ranges - 1; j++) { 1180 mem->ranges[j].start = 1181 mem->ranges[j+1].start; 1182 mem->ranges[j].end = 1183 mem->ranges[j+1].end; 1184 } 1185 1186 /* 1187 * Continue to check if there are another overlapping ranges 1188 * from the current position because of shifting the above 1189 * mem ranges. 1190 */ 1191 i--; 1192 mem->nr_ranges--; 1193 continue; 1194 } 1195 mem->nr_ranges--; 1196 return 0; 1197 } 1198 1199 if (p_start > start && p_end < end) { 1200 /* Split original range */ 1201 mem->ranges[i].end = p_start - 1; 1202 temp_range.start = p_end + 1; 1203 temp_range.end = end; 1204 } else if (p_start != start) 1205 mem->ranges[i].end = p_start - 1; 1206 else 1207 mem->ranges[i].start = p_end + 1; 1208 break; 1209 } 1210 1211 /* If a split happened, add the split to array */ 1212 if (!temp_range.end) 1213 return 0; 1214 1215 /* Split happened */ 1216 if (i == mem->max_nr_ranges - 1) 1217 return -ENOMEM; 1218 1219 /* Location where new range should go */ 1220 j = i + 1; 1221 if (j < mem->nr_ranges) { 1222 /* Move over all ranges one slot towards the end */ 1223 for (i = mem->nr_ranges - 1; i >= j; i--) 1224 mem->ranges[i + 1] = mem->ranges[i]; 1225 } 1226 1227 mem->ranges[j].start = temp_range.start; 1228 mem->ranges[j].end = temp_range.end; 1229 mem->nr_ranges++; 1230 return 0; 1231 } 1232 1233 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map, 1234 void **addr, unsigned long *sz) 1235 { 1236 Elf64_Ehdr *ehdr; 1237 Elf64_Phdr *phdr; 1238 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1239 unsigned char *buf; 1240 unsigned int cpu, i; 1241 unsigned long long notes_addr; 1242 unsigned long mstart, mend; 1243 1244 /* extra phdr for vmcoreinfo ELF note */ 1245 nr_phdr = nr_cpus + 1; 1246 nr_phdr += mem->nr_ranges; 1247 1248 /* 1249 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1250 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1251 * I think this is required by tools like gdb. So same physical 1252 * memory will be mapped in two ELF headers. One will contain kernel 1253 * text virtual addresses and other will have __va(physical) addresses. 1254 */ 1255 1256 nr_phdr++; 1257 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1258 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1259 1260 buf = vzalloc(elf_sz); 1261 if (!buf) 1262 return -ENOMEM; 1263 1264 ehdr = (Elf64_Ehdr *)buf; 1265 phdr = (Elf64_Phdr *)(ehdr + 1); 1266 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1267 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1268 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1269 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1270 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1271 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1272 ehdr->e_type = ET_CORE; 1273 ehdr->e_machine = ELF_ARCH; 1274 ehdr->e_version = EV_CURRENT; 1275 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1276 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1277 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1278 1279 /* Prepare one phdr of type PT_NOTE for each present CPU */ 1280 for_each_present_cpu(cpu) { 1281 phdr->p_type = PT_NOTE; 1282 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1283 phdr->p_offset = phdr->p_paddr = notes_addr; 1284 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1285 (ehdr->e_phnum)++; 1286 phdr++; 1287 } 1288 1289 /* Prepare one PT_NOTE header for vmcoreinfo */ 1290 phdr->p_type = PT_NOTE; 1291 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1292 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1293 (ehdr->e_phnum)++; 1294 phdr++; 1295 1296 /* Prepare PT_LOAD type program header for kernel text region */ 1297 if (need_kernel_map) { 1298 phdr->p_type = PT_LOAD; 1299 phdr->p_flags = PF_R|PF_W|PF_X; 1300 phdr->p_vaddr = (unsigned long) _text; 1301 phdr->p_filesz = phdr->p_memsz = _end - _text; 1302 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1303 ehdr->e_phnum++; 1304 phdr++; 1305 } 1306 1307 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1308 for (i = 0; i < mem->nr_ranges; i++) { 1309 mstart = mem->ranges[i].start; 1310 mend = mem->ranges[i].end; 1311 1312 phdr->p_type = PT_LOAD; 1313 phdr->p_flags = PF_R|PF_W|PF_X; 1314 phdr->p_offset = mstart; 1315 1316 phdr->p_paddr = mstart; 1317 phdr->p_vaddr = (unsigned long) __va(mstart); 1318 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1319 phdr->p_align = 0; 1320 ehdr->e_phnum++; 1321 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", 1322 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1323 ehdr->e_phnum, phdr->p_offset); 1324 phdr++; 1325 } 1326 1327 *addr = buf; 1328 *sz = elf_sz; 1329 return 0; 1330 } 1331