1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kexec: kexec_file_load system call 4 * 5 * Copyright (C) 2014 Red Hat Inc. 6 * Authors: 7 * Vivek Goyal <vgoyal@redhat.com> 8 */ 9 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/capability.h> 13 #include <linux/mm.h> 14 #include <linux/file.h> 15 #include <linux/slab.h> 16 #include <linux/kexec.h> 17 #include <linux/memblock.h> 18 #include <linux/mutex.h> 19 #include <linux/list.h> 20 #include <linux/fs.h> 21 #include <linux/ima.h> 22 #include <crypto/hash.h> 23 #include <crypto/sha.h> 24 #include <linux/elf.h> 25 #include <linux/elfcore.h> 26 #include <linux/kernel.h> 27 #include <linux/syscalls.h> 28 #include <linux/vmalloc.h> 29 #include "kexec_internal.h" 30 31 static int kexec_calculate_store_digests(struct kimage *image); 32 33 /* 34 * Currently this is the only default function that is exported as some 35 * architectures need it to do additional handlings. 36 * In the future, other default functions may be exported too if required. 37 */ 38 int kexec_image_probe_default(struct kimage *image, void *buf, 39 unsigned long buf_len) 40 { 41 const struct kexec_file_ops * const *fops; 42 int ret = -ENOEXEC; 43 44 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 45 ret = (*fops)->probe(buf, buf_len); 46 if (!ret) { 47 image->fops = *fops; 48 return ret; 49 } 50 } 51 52 return ret; 53 } 54 55 /* Architectures can provide this probe function */ 56 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, 57 unsigned long buf_len) 58 { 59 return kexec_image_probe_default(image, buf, buf_len); 60 } 61 62 static void *kexec_image_load_default(struct kimage *image) 63 { 64 if (!image->fops || !image->fops->load) 65 return ERR_PTR(-ENOEXEC); 66 67 return image->fops->load(image, image->kernel_buf, 68 image->kernel_buf_len, image->initrd_buf, 69 image->initrd_buf_len, image->cmdline_buf, 70 image->cmdline_buf_len); 71 } 72 73 void * __weak arch_kexec_kernel_image_load(struct kimage *image) 74 { 75 return kexec_image_load_default(image); 76 } 77 78 int kexec_image_post_load_cleanup_default(struct kimage *image) 79 { 80 if (!image->fops || !image->fops->cleanup) 81 return 0; 82 83 return image->fops->cleanup(image->image_loader_data); 84 } 85 86 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) 87 { 88 return kexec_image_post_load_cleanup_default(image); 89 } 90 91 #ifdef CONFIG_KEXEC_VERIFY_SIG 92 static int kexec_image_verify_sig_default(struct kimage *image, void *buf, 93 unsigned long buf_len) 94 { 95 if (!image->fops || !image->fops->verify_sig) { 96 pr_debug("kernel loader does not support signature verification.\n"); 97 return -EKEYREJECTED; 98 } 99 100 return image->fops->verify_sig(buf, buf_len); 101 } 102 103 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, 104 unsigned long buf_len) 105 { 106 return kexec_image_verify_sig_default(image, buf, buf_len); 107 } 108 #endif 109 110 /* 111 * arch_kexec_apply_relocations_add - apply relocations of type RELA 112 * @pi: Purgatory to be relocated. 113 * @section: Section relocations applying to. 114 * @relsec: Section containing RELAs. 115 * @symtab: Corresponding symtab. 116 * 117 * Return: 0 on success, negative errno on error. 118 */ 119 int __weak 120 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section, 121 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 122 { 123 pr_err("RELA relocation unsupported.\n"); 124 return -ENOEXEC; 125 } 126 127 /* 128 * arch_kexec_apply_relocations - apply relocations of type REL 129 * @pi: Purgatory to be relocated. 130 * @section: Section relocations applying to. 131 * @relsec: Section containing RELs. 132 * @symtab: Corresponding symtab. 133 * 134 * Return: 0 on success, negative errno on error. 135 */ 136 int __weak 137 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section, 138 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 139 { 140 pr_err("REL relocation unsupported.\n"); 141 return -ENOEXEC; 142 } 143 144 /* 145 * Free up memory used by kernel, initrd, and command line. This is temporary 146 * memory allocation which is not needed any more after these buffers have 147 * been loaded into separate segments and have been copied elsewhere. 148 */ 149 void kimage_file_post_load_cleanup(struct kimage *image) 150 { 151 struct purgatory_info *pi = &image->purgatory_info; 152 153 vfree(image->kernel_buf); 154 image->kernel_buf = NULL; 155 156 vfree(image->initrd_buf); 157 image->initrd_buf = NULL; 158 159 kfree(image->cmdline_buf); 160 image->cmdline_buf = NULL; 161 162 vfree(pi->purgatory_buf); 163 pi->purgatory_buf = NULL; 164 165 vfree(pi->sechdrs); 166 pi->sechdrs = NULL; 167 168 /* See if architecture has anything to cleanup post load */ 169 arch_kimage_file_post_load_cleanup(image); 170 171 /* 172 * Above call should have called into bootloader to free up 173 * any data stored in kimage->image_loader_data. It should 174 * be ok now to free it up. 175 */ 176 kfree(image->image_loader_data); 177 image->image_loader_data = NULL; 178 } 179 180 /* 181 * In file mode list of segments is prepared by kernel. Copy relevant 182 * data from user space, do error checking, prepare segment list 183 */ 184 static int 185 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 186 const char __user *cmdline_ptr, 187 unsigned long cmdline_len, unsigned flags) 188 { 189 int ret = 0; 190 void *ldata; 191 loff_t size; 192 193 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf, 194 &size, INT_MAX, READING_KEXEC_IMAGE); 195 if (ret) 196 return ret; 197 image->kernel_buf_len = size; 198 199 /* IMA needs to pass the measurement list to the next kernel. */ 200 ima_add_kexec_buffer(image); 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_VERIFY_SIG 209 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, 210 image->kernel_buf_len); 211 if (ret) { 212 pr_debug("kernel signature verification failed.\n"); 213 goto out; 214 } 215 pr_debug("kernel signature verification successful.\n"); 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, &image->initrd_buf, 220 &size, INT_MAX, 221 READING_KEXEC_INITRAMFS); 222 if (ret) 223 goto out; 224 image->initrd_buf_len = size; 225 } 226 227 if (cmdline_len) { 228 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); 229 if (IS_ERR(image->cmdline_buf)) { 230 ret = PTR_ERR(image->cmdline_buf); 231 image->cmdline_buf = NULL; 232 goto out; 233 } 234 235 image->cmdline_buf_len = cmdline_len; 236 237 /* command line should be a string with last byte null */ 238 if (image->cmdline_buf[cmdline_len - 1] != '\0') { 239 ret = -EINVAL; 240 goto out; 241 } 242 } 243 244 /* Call arch image load handlers */ 245 ldata = arch_kexec_kernel_image_load(image); 246 247 if (IS_ERR(ldata)) { 248 ret = PTR_ERR(ldata); 249 goto out; 250 } 251 252 image->image_loader_data = ldata; 253 out: 254 /* In case of error, free up all allocated memory in this function */ 255 if (ret) 256 kimage_file_post_load_cleanup(image); 257 return ret; 258 } 259 260 static int 261 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, 262 int initrd_fd, const char __user *cmdline_ptr, 263 unsigned long cmdline_len, unsigned long flags) 264 { 265 int ret; 266 struct kimage *image; 267 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; 268 269 image = do_kimage_alloc_init(); 270 if (!image) 271 return -ENOMEM; 272 273 image->file_mode = 1; 274 275 if (kexec_on_panic) { 276 /* Enable special crash kernel control page alloc policy. */ 277 image->control_page = crashk_res.start; 278 image->type = KEXEC_TYPE_CRASH; 279 } 280 281 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 282 cmdline_ptr, cmdline_len, flags); 283 if (ret) 284 goto out_free_image; 285 286 ret = sanity_check_segment_list(image); 287 if (ret) 288 goto out_free_post_load_bufs; 289 290 ret = -ENOMEM; 291 image->control_code_page = kimage_alloc_control_pages(image, 292 get_order(KEXEC_CONTROL_PAGE_SIZE)); 293 if (!image->control_code_page) { 294 pr_err("Could not allocate control_code_buffer\n"); 295 goto out_free_post_load_bufs; 296 } 297 298 if (!kexec_on_panic) { 299 image->swap_page = kimage_alloc_control_pages(image, 0); 300 if (!image->swap_page) { 301 pr_err("Could not allocate swap buffer\n"); 302 goto out_free_control_pages; 303 } 304 } 305 306 *rimage = image; 307 return 0; 308 out_free_control_pages: 309 kimage_free_page_list(&image->control_pages); 310 out_free_post_load_bufs: 311 kimage_file_post_load_cleanup(image); 312 out_free_image: 313 kfree(image); 314 return ret; 315 } 316 317 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 318 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 319 unsigned long, flags) 320 { 321 int ret = 0, i; 322 struct kimage **dest_image, *image; 323 324 /* We only trust the superuser with rebooting the system. */ 325 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) 326 return -EPERM; 327 328 /* Make sure we have a legal set of flags */ 329 if (flags != (flags & KEXEC_FILE_FLAGS)) 330 return -EINVAL; 331 332 image = NULL; 333 334 if (!mutex_trylock(&kexec_mutex)) 335 return -EBUSY; 336 337 dest_image = &kexec_image; 338 if (flags & KEXEC_FILE_ON_CRASH) { 339 dest_image = &kexec_crash_image; 340 if (kexec_crash_image) 341 arch_kexec_unprotect_crashkres(); 342 } 343 344 if (flags & KEXEC_FILE_UNLOAD) 345 goto exchange; 346 347 /* 348 * In case of crash, new kernel gets loaded in reserved region. It is 349 * same memory where old crash kernel might be loaded. Free any 350 * current crash dump kernel before we corrupt it. 351 */ 352 if (flags & KEXEC_FILE_ON_CRASH) 353 kimage_free(xchg(&kexec_crash_image, NULL)); 354 355 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 356 cmdline_len, flags); 357 if (ret) 358 goto out; 359 360 ret = machine_kexec_prepare(image); 361 if (ret) 362 goto out; 363 364 /* 365 * Some architecture(like S390) may touch the crash memory before 366 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 367 */ 368 ret = kimage_crash_copy_vmcoreinfo(image); 369 if (ret) 370 goto out; 371 372 ret = kexec_calculate_store_digests(image); 373 if (ret) 374 goto out; 375 376 for (i = 0; i < image->nr_segments; i++) { 377 struct kexec_segment *ksegment; 378 379 ksegment = &image->segment[i]; 380 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 381 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 382 ksegment->memsz); 383 384 ret = kimage_load_segment(image, &image->segment[i]); 385 if (ret) 386 goto out; 387 } 388 389 kimage_terminate(image); 390 391 /* 392 * Free up any temporary buffers allocated which are not needed 393 * after image has been loaded 394 */ 395 kimage_file_post_load_cleanup(image); 396 exchange: 397 image = xchg(dest_image, image); 398 out: 399 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 400 arch_kexec_protect_crashkres(); 401 402 mutex_unlock(&kexec_mutex); 403 kimage_free(image); 404 return ret; 405 } 406 407 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 408 struct kexec_buf *kbuf) 409 { 410 struct kimage *image = kbuf->image; 411 unsigned long temp_start, temp_end; 412 413 temp_end = min(end, kbuf->buf_max); 414 temp_start = temp_end - kbuf->memsz; 415 416 do { 417 /* align down start */ 418 temp_start = temp_start & (~(kbuf->buf_align - 1)); 419 420 if (temp_start < start || temp_start < kbuf->buf_min) 421 return 0; 422 423 temp_end = temp_start + kbuf->memsz - 1; 424 425 /* 426 * Make sure this does not conflict with any of existing 427 * segments 428 */ 429 if (kimage_is_destination_range(image, temp_start, temp_end)) { 430 temp_start = temp_start - PAGE_SIZE; 431 continue; 432 } 433 434 /* We found a suitable memory range */ 435 break; 436 } while (1); 437 438 /* If we are here, we found a suitable memory range */ 439 kbuf->mem = temp_start; 440 441 /* Success, stop navigating through remaining System RAM ranges */ 442 return 1; 443 } 444 445 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 446 struct kexec_buf *kbuf) 447 { 448 struct kimage *image = kbuf->image; 449 unsigned long temp_start, temp_end; 450 451 temp_start = max(start, kbuf->buf_min); 452 453 do { 454 temp_start = ALIGN(temp_start, kbuf->buf_align); 455 temp_end = temp_start + kbuf->memsz - 1; 456 457 if (temp_end > end || temp_end > kbuf->buf_max) 458 return 0; 459 /* 460 * Make sure this does not conflict with any of existing 461 * segments 462 */ 463 if (kimage_is_destination_range(image, temp_start, temp_end)) { 464 temp_start = temp_start + PAGE_SIZE; 465 continue; 466 } 467 468 /* We found a suitable memory range */ 469 break; 470 } while (1); 471 472 /* If we are here, we found a suitable memory range */ 473 kbuf->mem = temp_start; 474 475 /* Success, stop navigating through remaining System RAM ranges */ 476 return 1; 477 } 478 479 static int locate_mem_hole_callback(struct resource *res, void *arg) 480 { 481 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 482 u64 start = res->start, end = res->end; 483 unsigned long sz = end - start + 1; 484 485 /* Returning 0 will take to next memory range */ 486 if (sz < kbuf->memsz) 487 return 0; 488 489 if (end < kbuf->buf_min || start > kbuf->buf_max) 490 return 0; 491 492 /* 493 * Allocate memory top down with-in ram range. Otherwise bottom up 494 * allocation. 495 */ 496 if (kbuf->top_down) 497 return locate_mem_hole_top_down(start, end, kbuf); 498 return locate_mem_hole_bottom_up(start, end, kbuf); 499 } 500 501 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 502 static int kexec_walk_memblock(struct kexec_buf *kbuf, 503 int (*func)(struct resource *, void *)) 504 { 505 int ret = 0; 506 u64 i; 507 phys_addr_t mstart, mend; 508 struct resource res = { }; 509 510 if (kbuf->image->type == KEXEC_TYPE_CRASH) 511 return func(&crashk_res, kbuf); 512 513 if (kbuf->top_down) { 514 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 515 &mstart, &mend, NULL) { 516 /* 517 * In memblock, end points to the first byte after the 518 * range while in kexec, end points to the last byte 519 * in the range. 520 */ 521 res.start = mstart; 522 res.end = mend - 1; 523 ret = func(&res, kbuf); 524 if (ret) 525 break; 526 } 527 } else { 528 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 529 &mstart, &mend, NULL) { 530 /* 531 * In memblock, end points to the first byte after the 532 * range while in kexec, end points to the last byte 533 * in the range. 534 */ 535 res.start = mstart; 536 res.end = mend - 1; 537 ret = func(&res, kbuf); 538 if (ret) 539 break; 540 } 541 } 542 543 return ret; 544 } 545 #else 546 static int kexec_walk_memblock(struct kexec_buf *kbuf, 547 int (*func)(struct resource *, void *)) 548 { 549 return 0; 550 } 551 #endif 552 553 /** 554 * kexec_walk_resources - call func(data) on free memory regions 555 * @kbuf: Context info for the search. Also passed to @func. 556 * @func: Function to call for each memory region. 557 * 558 * Return: The memory walk will stop when func returns a non-zero value 559 * and that value will be returned. If all free regions are visited without 560 * func returning non-zero, then zero will be returned. 561 */ 562 static int kexec_walk_resources(struct kexec_buf *kbuf, 563 int (*func)(struct resource *, void *)) 564 { 565 if (kbuf->image->type == KEXEC_TYPE_CRASH) 566 return walk_iomem_res_desc(crashk_res.desc, 567 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 568 crashk_res.start, crashk_res.end, 569 kbuf, func); 570 else 571 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 572 } 573 574 /** 575 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 576 * @kbuf: Parameters for the memory search. 577 * 578 * On success, kbuf->mem will have the start address of the memory region found. 579 * 580 * Return: 0 on success, negative errno on error. 581 */ 582 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 583 { 584 int ret; 585 586 /* Arch knows where to place */ 587 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 588 return 0; 589 590 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 591 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 592 else 593 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 594 595 return ret == 1 ? 0 : -EADDRNOTAVAIL; 596 } 597 598 /** 599 * kexec_add_buffer - place a buffer in a kexec segment 600 * @kbuf: Buffer contents and memory parameters. 601 * 602 * This function assumes that kexec_mutex is held. 603 * On successful return, @kbuf->mem will have the physical address of 604 * the buffer in memory. 605 * 606 * Return: 0 on success, negative errno on error. 607 */ 608 int kexec_add_buffer(struct kexec_buf *kbuf) 609 { 610 611 struct kexec_segment *ksegment; 612 int ret; 613 614 /* Currently adding segment this way is allowed only in file mode */ 615 if (!kbuf->image->file_mode) 616 return -EINVAL; 617 618 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 619 return -EINVAL; 620 621 /* 622 * Make sure we are not trying to add buffer after allocating 623 * control pages. All segments need to be placed first before 624 * any control pages are allocated. As control page allocation 625 * logic goes through list of segments to make sure there are 626 * no destination overlaps. 627 */ 628 if (!list_empty(&kbuf->image->control_pages)) { 629 WARN_ON(1); 630 return -EINVAL; 631 } 632 633 /* Ensure minimum alignment needed for segments. */ 634 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 635 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 636 637 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 638 ret = kexec_locate_mem_hole(kbuf); 639 if (ret) 640 return ret; 641 642 /* Found a suitable memory range */ 643 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 644 ksegment->kbuf = kbuf->buffer; 645 ksegment->bufsz = kbuf->bufsz; 646 ksegment->mem = kbuf->mem; 647 ksegment->memsz = kbuf->memsz; 648 kbuf->image->nr_segments++; 649 return 0; 650 } 651 652 /* Calculate and store the digest of segments */ 653 static int kexec_calculate_store_digests(struct kimage *image) 654 { 655 struct crypto_shash *tfm; 656 struct shash_desc *desc; 657 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 658 size_t desc_size, nullsz; 659 char *digest; 660 void *zero_buf; 661 struct kexec_sha_region *sha_regions; 662 struct purgatory_info *pi = &image->purgatory_info; 663 664 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 665 return 0; 666 667 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 668 zero_buf_sz = PAGE_SIZE; 669 670 tfm = crypto_alloc_shash("sha256", 0, 0); 671 if (IS_ERR(tfm)) { 672 ret = PTR_ERR(tfm); 673 goto out; 674 } 675 676 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 677 desc = kzalloc(desc_size, GFP_KERNEL); 678 if (!desc) { 679 ret = -ENOMEM; 680 goto out_free_tfm; 681 } 682 683 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 684 sha_regions = vzalloc(sha_region_sz); 685 if (!sha_regions) 686 goto out_free_desc; 687 688 desc->tfm = tfm; 689 690 ret = crypto_shash_init(desc); 691 if (ret < 0) 692 goto out_free_sha_regions; 693 694 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 695 if (!digest) { 696 ret = -ENOMEM; 697 goto out_free_sha_regions; 698 } 699 700 for (j = i = 0; i < image->nr_segments; i++) { 701 struct kexec_segment *ksegment; 702 703 ksegment = &image->segment[i]; 704 /* 705 * Skip purgatory as it will be modified once we put digest 706 * info in purgatory. 707 */ 708 if (ksegment->kbuf == pi->purgatory_buf) 709 continue; 710 711 ret = crypto_shash_update(desc, ksegment->kbuf, 712 ksegment->bufsz); 713 if (ret) 714 break; 715 716 /* 717 * Assume rest of the buffer is filled with zero and 718 * update digest accordingly. 719 */ 720 nullsz = ksegment->memsz - ksegment->bufsz; 721 while (nullsz) { 722 unsigned long bytes = nullsz; 723 724 if (bytes > zero_buf_sz) 725 bytes = zero_buf_sz; 726 ret = crypto_shash_update(desc, zero_buf, bytes); 727 if (ret) 728 break; 729 nullsz -= bytes; 730 } 731 732 if (ret) 733 break; 734 735 sha_regions[j].start = ksegment->mem; 736 sha_regions[j].len = ksegment->memsz; 737 j++; 738 } 739 740 if (!ret) { 741 ret = crypto_shash_final(desc, digest); 742 if (ret) 743 goto out_free_digest; 744 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 745 sha_regions, sha_region_sz, 0); 746 if (ret) 747 goto out_free_digest; 748 749 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 750 digest, SHA256_DIGEST_SIZE, 0); 751 if (ret) 752 goto out_free_digest; 753 } 754 755 out_free_digest: 756 kfree(digest); 757 out_free_sha_regions: 758 vfree(sha_regions); 759 out_free_desc: 760 kfree(desc); 761 out_free_tfm: 762 kfree(tfm); 763 out: 764 return ret; 765 } 766 767 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 768 /* 769 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 770 * @pi: Purgatory to be loaded. 771 * @kbuf: Buffer to setup. 772 * 773 * Allocates the memory needed for the buffer. Caller is responsible to free 774 * the memory after use. 775 * 776 * Return: 0 on success, negative errno on error. 777 */ 778 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 779 struct kexec_buf *kbuf) 780 { 781 const Elf_Shdr *sechdrs; 782 unsigned long bss_align; 783 unsigned long bss_sz; 784 unsigned long align; 785 int i, ret; 786 787 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 788 kbuf->buf_align = bss_align = 1; 789 kbuf->bufsz = bss_sz = 0; 790 791 for (i = 0; i < pi->ehdr->e_shnum; i++) { 792 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 793 continue; 794 795 align = sechdrs[i].sh_addralign; 796 if (sechdrs[i].sh_type != SHT_NOBITS) { 797 if (kbuf->buf_align < align) 798 kbuf->buf_align = align; 799 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 800 kbuf->bufsz += sechdrs[i].sh_size; 801 } else { 802 if (bss_align < align) 803 bss_align = align; 804 bss_sz = ALIGN(bss_sz, align); 805 bss_sz += sechdrs[i].sh_size; 806 } 807 } 808 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 809 kbuf->memsz = kbuf->bufsz + bss_sz; 810 if (kbuf->buf_align < bss_align) 811 kbuf->buf_align = bss_align; 812 813 kbuf->buffer = vzalloc(kbuf->bufsz); 814 if (!kbuf->buffer) 815 return -ENOMEM; 816 pi->purgatory_buf = kbuf->buffer; 817 818 ret = kexec_add_buffer(kbuf); 819 if (ret) 820 goto out; 821 822 return 0; 823 out: 824 vfree(pi->purgatory_buf); 825 pi->purgatory_buf = NULL; 826 return ret; 827 } 828 829 /* 830 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 831 * @pi: Purgatory to be loaded. 832 * @kbuf: Buffer prepared to store purgatory. 833 * 834 * Allocates the memory needed for the buffer. Caller is responsible to free 835 * the memory after use. 836 * 837 * Return: 0 on success, negative errno on error. 838 */ 839 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 840 struct kexec_buf *kbuf) 841 { 842 unsigned long bss_addr; 843 unsigned long offset; 844 Elf_Shdr *sechdrs; 845 int i; 846 847 /* 848 * The section headers in kexec_purgatory are read-only. In order to 849 * have them modifiable make a temporary copy. 850 */ 851 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 852 if (!sechdrs) 853 return -ENOMEM; 854 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 855 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 856 pi->sechdrs = sechdrs; 857 858 offset = 0; 859 bss_addr = kbuf->mem + kbuf->bufsz; 860 kbuf->image->start = pi->ehdr->e_entry; 861 862 for (i = 0; i < pi->ehdr->e_shnum; i++) { 863 unsigned long align; 864 void *src, *dst; 865 866 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 867 continue; 868 869 align = sechdrs[i].sh_addralign; 870 if (sechdrs[i].sh_type == SHT_NOBITS) { 871 bss_addr = ALIGN(bss_addr, align); 872 sechdrs[i].sh_addr = bss_addr; 873 bss_addr += sechdrs[i].sh_size; 874 continue; 875 } 876 877 offset = ALIGN(offset, align); 878 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 879 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 880 pi->ehdr->e_entry < (sechdrs[i].sh_addr 881 + sechdrs[i].sh_size)) { 882 kbuf->image->start -= sechdrs[i].sh_addr; 883 kbuf->image->start += kbuf->mem + offset; 884 } 885 886 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 887 dst = pi->purgatory_buf + offset; 888 memcpy(dst, src, sechdrs[i].sh_size); 889 890 sechdrs[i].sh_addr = kbuf->mem + offset; 891 sechdrs[i].sh_offset = offset; 892 offset += sechdrs[i].sh_size; 893 } 894 895 return 0; 896 } 897 898 static int kexec_apply_relocations(struct kimage *image) 899 { 900 int i, ret; 901 struct purgatory_info *pi = &image->purgatory_info; 902 const Elf_Shdr *sechdrs; 903 904 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 905 906 for (i = 0; i < pi->ehdr->e_shnum; i++) { 907 const Elf_Shdr *relsec; 908 const Elf_Shdr *symtab; 909 Elf_Shdr *section; 910 911 relsec = sechdrs + i; 912 913 if (relsec->sh_type != SHT_RELA && 914 relsec->sh_type != SHT_REL) 915 continue; 916 917 /* 918 * For section of type SHT_RELA/SHT_REL, 919 * ->sh_link contains section header index of associated 920 * symbol table. And ->sh_info contains section header 921 * index of section to which relocations apply. 922 */ 923 if (relsec->sh_info >= pi->ehdr->e_shnum || 924 relsec->sh_link >= pi->ehdr->e_shnum) 925 return -ENOEXEC; 926 927 section = pi->sechdrs + relsec->sh_info; 928 symtab = sechdrs + relsec->sh_link; 929 930 if (!(section->sh_flags & SHF_ALLOC)) 931 continue; 932 933 /* 934 * symtab->sh_link contain section header index of associated 935 * string table. 936 */ 937 if (symtab->sh_link >= pi->ehdr->e_shnum) 938 /* Invalid section number? */ 939 continue; 940 941 /* 942 * Respective architecture needs to provide support for applying 943 * relocations of type SHT_RELA/SHT_REL. 944 */ 945 if (relsec->sh_type == SHT_RELA) 946 ret = arch_kexec_apply_relocations_add(pi, section, 947 relsec, symtab); 948 else if (relsec->sh_type == SHT_REL) 949 ret = arch_kexec_apply_relocations(pi, section, 950 relsec, symtab); 951 if (ret) 952 return ret; 953 } 954 955 return 0; 956 } 957 958 /* 959 * kexec_load_purgatory - Load and relocate the purgatory object. 960 * @image: Image to add the purgatory to. 961 * @kbuf: Memory parameters to use. 962 * 963 * Allocates the memory needed for image->purgatory_info.sechdrs and 964 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 965 * to free the memory after use. 966 * 967 * Return: 0 on success, negative errno on error. 968 */ 969 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 970 { 971 struct purgatory_info *pi = &image->purgatory_info; 972 int ret; 973 974 if (kexec_purgatory_size <= 0) 975 return -EINVAL; 976 977 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 978 979 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 980 if (ret) 981 return ret; 982 983 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 984 if (ret) 985 goto out_free_kbuf; 986 987 ret = kexec_apply_relocations(image); 988 if (ret) 989 goto out; 990 991 return 0; 992 out: 993 vfree(pi->sechdrs); 994 pi->sechdrs = NULL; 995 out_free_kbuf: 996 vfree(pi->purgatory_buf); 997 pi->purgatory_buf = NULL; 998 return ret; 999 } 1000 1001 /* 1002 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1003 * @pi: Purgatory to search in. 1004 * @name: Name of the symbol. 1005 * 1006 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1007 */ 1008 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1009 const char *name) 1010 { 1011 const Elf_Shdr *sechdrs; 1012 const Elf_Ehdr *ehdr; 1013 const Elf_Sym *syms; 1014 const char *strtab; 1015 int i, k; 1016 1017 if (!pi->ehdr) 1018 return NULL; 1019 1020 ehdr = pi->ehdr; 1021 sechdrs = (void *)ehdr + ehdr->e_shoff; 1022 1023 for (i = 0; i < ehdr->e_shnum; i++) { 1024 if (sechdrs[i].sh_type != SHT_SYMTAB) 1025 continue; 1026 1027 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1028 /* Invalid strtab section number */ 1029 continue; 1030 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1031 syms = (void *)ehdr + sechdrs[i].sh_offset; 1032 1033 /* Go through symbols for a match */ 1034 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1035 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1036 continue; 1037 1038 if (strcmp(strtab + syms[k].st_name, name) != 0) 1039 continue; 1040 1041 if (syms[k].st_shndx == SHN_UNDEF || 1042 syms[k].st_shndx >= ehdr->e_shnum) { 1043 pr_debug("Symbol: %s has bad section index %d.\n", 1044 name, syms[k].st_shndx); 1045 return NULL; 1046 } 1047 1048 /* Found the symbol we are looking for */ 1049 return &syms[k]; 1050 } 1051 } 1052 1053 return NULL; 1054 } 1055 1056 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1057 { 1058 struct purgatory_info *pi = &image->purgatory_info; 1059 const Elf_Sym *sym; 1060 Elf_Shdr *sechdr; 1061 1062 sym = kexec_purgatory_find_symbol(pi, name); 1063 if (!sym) 1064 return ERR_PTR(-EINVAL); 1065 1066 sechdr = &pi->sechdrs[sym->st_shndx]; 1067 1068 /* 1069 * Returns the address where symbol will finally be loaded after 1070 * kexec_load_segment() 1071 */ 1072 return (void *)(sechdr->sh_addr + sym->st_value); 1073 } 1074 1075 /* 1076 * Get or set value of a symbol. If "get_value" is true, symbol value is 1077 * returned in buf otherwise symbol value is set based on value in buf. 1078 */ 1079 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1080 void *buf, unsigned int size, bool get_value) 1081 { 1082 struct purgatory_info *pi = &image->purgatory_info; 1083 const Elf_Sym *sym; 1084 Elf_Shdr *sec; 1085 char *sym_buf; 1086 1087 sym = kexec_purgatory_find_symbol(pi, name); 1088 if (!sym) 1089 return -EINVAL; 1090 1091 if (sym->st_size != size) { 1092 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1093 name, (unsigned long)sym->st_size, size); 1094 return -EINVAL; 1095 } 1096 1097 sec = pi->sechdrs + sym->st_shndx; 1098 1099 if (sec->sh_type == SHT_NOBITS) { 1100 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1101 get_value ? "get" : "set"); 1102 return -EINVAL; 1103 } 1104 1105 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1106 1107 if (get_value) 1108 memcpy((void *)buf, sym_buf, size); 1109 else 1110 memcpy((void *)sym_buf, buf, size); 1111 1112 return 0; 1113 } 1114 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1115 1116 int crash_exclude_mem_range(struct crash_mem *mem, 1117 unsigned long long mstart, unsigned long long mend) 1118 { 1119 int i, j; 1120 unsigned long long start, end; 1121 struct crash_mem_range temp_range = {0, 0}; 1122 1123 for (i = 0; i < mem->nr_ranges; i++) { 1124 start = mem->ranges[i].start; 1125 end = mem->ranges[i].end; 1126 1127 if (mstart > end || mend < start) 1128 continue; 1129 1130 /* Truncate any area outside of range */ 1131 if (mstart < start) 1132 mstart = start; 1133 if (mend > end) 1134 mend = end; 1135 1136 /* Found completely overlapping range */ 1137 if (mstart == start && mend == end) { 1138 mem->ranges[i].start = 0; 1139 mem->ranges[i].end = 0; 1140 if (i < mem->nr_ranges - 1) { 1141 /* Shift rest of the ranges to left */ 1142 for (j = i; j < mem->nr_ranges - 1; j++) { 1143 mem->ranges[j].start = 1144 mem->ranges[j+1].start; 1145 mem->ranges[j].end = 1146 mem->ranges[j+1].end; 1147 } 1148 } 1149 mem->nr_ranges--; 1150 return 0; 1151 } 1152 1153 if (mstart > start && mend < end) { 1154 /* Split original range */ 1155 mem->ranges[i].end = mstart - 1; 1156 temp_range.start = mend + 1; 1157 temp_range.end = end; 1158 } else if (mstart != start) 1159 mem->ranges[i].end = mstart - 1; 1160 else 1161 mem->ranges[i].start = mend + 1; 1162 break; 1163 } 1164 1165 /* If a split happened, add the split to array */ 1166 if (!temp_range.end) 1167 return 0; 1168 1169 /* Split happened */ 1170 if (i == mem->max_nr_ranges - 1) 1171 return -ENOMEM; 1172 1173 /* Location where new range should go */ 1174 j = i + 1; 1175 if (j < mem->nr_ranges) { 1176 /* Move over all ranges one slot towards the end */ 1177 for (i = mem->nr_ranges - 1; i >= j; i--) 1178 mem->ranges[i + 1] = mem->ranges[i]; 1179 } 1180 1181 mem->ranges[j].start = temp_range.start; 1182 mem->ranges[j].end = temp_range.end; 1183 mem->nr_ranges++; 1184 return 0; 1185 } 1186 1187 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, 1188 void **addr, unsigned long *sz) 1189 { 1190 Elf64_Ehdr *ehdr; 1191 Elf64_Phdr *phdr; 1192 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1193 unsigned char *buf; 1194 unsigned int cpu, i; 1195 unsigned long long notes_addr; 1196 unsigned long mstart, mend; 1197 1198 /* extra phdr for vmcoreinfo elf note */ 1199 nr_phdr = nr_cpus + 1; 1200 nr_phdr += mem->nr_ranges; 1201 1202 /* 1203 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1204 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1205 * I think this is required by tools like gdb. So same physical 1206 * memory will be mapped in two elf headers. One will contain kernel 1207 * text virtual addresses and other will have __va(physical) addresses. 1208 */ 1209 1210 nr_phdr++; 1211 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1212 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1213 1214 buf = vzalloc(elf_sz); 1215 if (!buf) 1216 return -ENOMEM; 1217 1218 ehdr = (Elf64_Ehdr *)buf; 1219 phdr = (Elf64_Phdr *)(ehdr + 1); 1220 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1221 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1222 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1223 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1224 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1225 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1226 ehdr->e_type = ET_CORE; 1227 ehdr->e_machine = ELF_ARCH; 1228 ehdr->e_version = EV_CURRENT; 1229 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1230 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1231 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1232 1233 /* Prepare one phdr of type PT_NOTE for each present cpu */ 1234 for_each_present_cpu(cpu) { 1235 phdr->p_type = PT_NOTE; 1236 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1237 phdr->p_offset = phdr->p_paddr = notes_addr; 1238 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1239 (ehdr->e_phnum)++; 1240 phdr++; 1241 } 1242 1243 /* Prepare one PT_NOTE header for vmcoreinfo */ 1244 phdr->p_type = PT_NOTE; 1245 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1246 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1247 (ehdr->e_phnum)++; 1248 phdr++; 1249 1250 /* Prepare PT_LOAD type program header for kernel text region */ 1251 if (kernel_map) { 1252 phdr->p_type = PT_LOAD; 1253 phdr->p_flags = PF_R|PF_W|PF_X; 1254 phdr->p_vaddr = (Elf64_Addr)_text; 1255 phdr->p_filesz = phdr->p_memsz = _end - _text; 1256 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1257 ehdr->e_phnum++; 1258 phdr++; 1259 } 1260 1261 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1262 for (i = 0; i < mem->nr_ranges; i++) { 1263 mstart = mem->ranges[i].start; 1264 mend = mem->ranges[i].end; 1265 1266 phdr->p_type = PT_LOAD; 1267 phdr->p_flags = PF_R|PF_W|PF_X; 1268 phdr->p_offset = mstart; 1269 1270 phdr->p_paddr = mstart; 1271 phdr->p_vaddr = (unsigned long long) __va(mstart); 1272 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1273 phdr->p_align = 0; 1274 ehdr->e_phnum++; 1275 phdr++; 1276 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", 1277 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1278 ehdr->e_phnum, phdr->p_offset); 1279 } 1280 1281 *addr = buf; 1282 *sz = elf_sz; 1283 return 0; 1284 } 1285