1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kexec: kexec_file_load system call 4 * 5 * Copyright (C) 2014 Red Hat Inc. 6 * Authors: 7 * Vivek Goyal <vgoyal@redhat.com> 8 */ 9 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11 12 #include <linux/capability.h> 13 #include <linux/mm.h> 14 #include <linux/file.h> 15 #include <linux/slab.h> 16 #include <linux/kexec.h> 17 #include <linux/memblock.h> 18 #include <linux/mutex.h> 19 #include <linux/list.h> 20 #include <linux/fs.h> 21 #include <linux/ima.h> 22 #include <crypto/hash.h> 23 #include <crypto/sha2.h> 24 #include <linux/elf.h> 25 #include <linux/elfcore.h> 26 #include <linux/kernel.h> 27 #include <linux/kernel_read_file.h> 28 #include <linux/syscalls.h> 29 #include <linux/vmalloc.h> 30 #include "kexec_internal.h" 31 32 #ifdef CONFIG_KEXEC_SIG 33 static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE); 34 35 void set_kexec_sig_enforced(void) 36 { 37 sig_enforce = true; 38 } 39 #endif 40 41 static int kexec_calculate_store_digests(struct kimage *image); 42 43 /* Maximum size in bytes for kernel/initrd files. */ 44 #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX) 45 46 /* 47 * Currently this is the only default function that is exported as some 48 * architectures need it to do additional handlings. 49 * In the future, other default functions may be exported too if required. 50 */ 51 int kexec_image_probe_default(struct kimage *image, void *buf, 52 unsigned long buf_len) 53 { 54 const struct kexec_file_ops * const *fops; 55 int ret = -ENOEXEC; 56 57 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 58 ret = (*fops)->probe(buf, buf_len); 59 if (!ret) { 60 image->fops = *fops; 61 return ret; 62 } 63 } 64 65 return ret; 66 } 67 68 static void *kexec_image_load_default(struct kimage *image) 69 { 70 if (!image->fops || !image->fops->load) 71 return ERR_PTR(-ENOEXEC); 72 73 return image->fops->load(image, image->kernel_buf, 74 image->kernel_buf_len, image->initrd_buf, 75 image->initrd_buf_len, image->cmdline_buf, 76 image->cmdline_buf_len); 77 } 78 79 int kexec_image_post_load_cleanup_default(struct kimage *image) 80 { 81 if (!image->fops || !image->fops->cleanup) 82 return 0; 83 84 return image->fops->cleanup(image->image_loader_data); 85 } 86 87 /* 88 * Free up memory used by kernel, initrd, and command line. This is temporary 89 * memory allocation which is not needed any more after these buffers have 90 * been loaded into separate segments and have been copied elsewhere. 91 */ 92 void kimage_file_post_load_cleanup(struct kimage *image) 93 { 94 struct purgatory_info *pi = &image->purgatory_info; 95 96 vfree(image->kernel_buf); 97 image->kernel_buf = NULL; 98 99 vfree(image->initrd_buf); 100 image->initrd_buf = NULL; 101 102 kfree(image->cmdline_buf); 103 image->cmdline_buf = NULL; 104 105 vfree(pi->purgatory_buf); 106 pi->purgatory_buf = NULL; 107 108 vfree(pi->sechdrs); 109 pi->sechdrs = NULL; 110 111 #ifdef CONFIG_IMA_KEXEC 112 vfree(image->ima_buffer); 113 image->ima_buffer = NULL; 114 #endif /* CONFIG_IMA_KEXEC */ 115 116 /* See if architecture has anything to cleanup post load */ 117 arch_kimage_file_post_load_cleanup(image); 118 119 /* 120 * Above call should have called into bootloader to free up 121 * any data stored in kimage->image_loader_data. It should 122 * be ok now to free it up. 123 */ 124 kfree(image->image_loader_data); 125 image->image_loader_data = NULL; 126 127 kexec_file_dbg_print = false; 128 } 129 130 #ifdef CONFIG_KEXEC_SIG 131 #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION 132 int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len) 133 { 134 int ret; 135 136 ret = verify_pefile_signature(kernel, kernel_len, 137 VERIFY_USE_SECONDARY_KEYRING, 138 VERIFYING_KEXEC_PE_SIGNATURE); 139 if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) { 140 ret = verify_pefile_signature(kernel, kernel_len, 141 VERIFY_USE_PLATFORM_KEYRING, 142 VERIFYING_KEXEC_PE_SIGNATURE); 143 } 144 return ret; 145 } 146 #endif 147 148 static int kexec_image_verify_sig(struct kimage *image, void *buf, 149 unsigned long buf_len) 150 { 151 if (!image->fops || !image->fops->verify_sig) { 152 pr_debug("kernel loader does not support signature verification.\n"); 153 return -EKEYREJECTED; 154 } 155 156 return image->fops->verify_sig(buf, buf_len); 157 } 158 159 static int 160 kimage_validate_signature(struct kimage *image) 161 { 162 int ret; 163 164 ret = kexec_image_verify_sig(image, image->kernel_buf, 165 image->kernel_buf_len); 166 if (ret) { 167 168 if (sig_enforce) { 169 pr_notice("Enforced kernel signature verification failed (%d).\n", ret); 170 return ret; 171 } 172 173 /* 174 * If IMA is guaranteed to appraise a signature on the kexec 175 * image, permit it even if the kernel is otherwise locked 176 * down. 177 */ 178 if (!ima_appraise_signature(READING_KEXEC_IMAGE) && 179 security_locked_down(LOCKDOWN_KEXEC)) 180 return -EPERM; 181 182 pr_debug("kernel signature verification failed (%d).\n", ret); 183 } 184 185 return 0; 186 } 187 #endif 188 189 /* 190 * In file mode list of segments is prepared by kernel. Copy relevant 191 * data from user space, do error checking, prepare segment list 192 */ 193 static int 194 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 195 const char __user *cmdline_ptr, 196 unsigned long cmdline_len, unsigned flags) 197 { 198 ssize_t ret; 199 void *ldata; 200 201 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf, 202 KEXEC_FILE_SIZE_MAX, NULL, 203 READING_KEXEC_IMAGE); 204 if (ret < 0) 205 return ret; 206 image->kernel_buf_len = ret; 207 kexec_dprintk("kernel: %p kernel_size: %#lx\n", 208 image->kernel_buf, image->kernel_buf_len); 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 KEXEC_FILE_SIZE_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 image load handler */ 257 ldata = kexec_image_load_default(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 kexec_file_dbg_print = !!(flags & KEXEC_FILE_DEBUG); 286 image->file_mode = 1; 287 288 #ifdef CONFIG_CRASH_DUMP 289 if (kexec_on_panic) { 290 /* Enable special crash kernel control page alloc policy. */ 291 image->control_page = crashk_res.start; 292 image->type = KEXEC_TYPE_CRASH; 293 } 294 #endif 295 296 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 297 cmdline_ptr, cmdline_len, flags); 298 if (ret) 299 goto out_free_image; 300 301 ret = sanity_check_segment_list(image); 302 if (ret) 303 goto out_free_post_load_bufs; 304 305 ret = -ENOMEM; 306 image->control_code_page = kimage_alloc_control_pages(image, 307 get_order(KEXEC_CONTROL_PAGE_SIZE)); 308 if (!image->control_code_page) { 309 pr_err("Could not allocate control_code_buffer\n"); 310 goto out_free_post_load_bufs; 311 } 312 313 if (!kexec_on_panic) { 314 image->swap_page = kimage_alloc_control_pages(image, 0); 315 if (!image->swap_page) { 316 pr_err("Could not allocate swap buffer\n"); 317 goto out_free_control_pages; 318 } 319 } 320 321 *rimage = image; 322 return 0; 323 out_free_control_pages: 324 kimage_free_page_list(&image->control_pages); 325 out_free_post_load_bufs: 326 kimage_file_post_load_cleanup(image); 327 out_free_image: 328 kfree(image); 329 return ret; 330 } 331 332 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 333 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 334 unsigned long, flags) 335 { 336 int image_type = (flags & KEXEC_FILE_ON_CRASH) ? 337 KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT; 338 struct kimage **dest_image, *image; 339 int ret = 0, i; 340 341 /* We only trust the superuser with rebooting the system. */ 342 if (!kexec_load_permitted(image_type)) 343 return -EPERM; 344 345 /* Make sure we have a legal set of flags */ 346 if (flags != (flags & KEXEC_FILE_FLAGS)) 347 return -EINVAL; 348 349 image = NULL; 350 351 if (!kexec_trylock()) 352 return -EBUSY; 353 354 #ifdef CONFIG_CRASH_DUMP 355 if (image_type == KEXEC_TYPE_CRASH) { 356 dest_image = &kexec_crash_image; 357 if (kexec_crash_image) 358 arch_kexec_unprotect_crashkres(); 359 } else 360 #endif 361 dest_image = &kexec_image; 362 363 if (flags & KEXEC_FILE_UNLOAD) 364 goto exchange; 365 366 /* 367 * In case of crash, new kernel gets loaded in reserved region. It is 368 * same memory where old crash kernel might be loaded. Free any 369 * current crash dump kernel before we corrupt it. 370 */ 371 if (flags & KEXEC_FILE_ON_CRASH) 372 kimage_free(xchg(&kexec_crash_image, NULL)); 373 374 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 375 cmdline_len, flags); 376 if (ret) 377 goto out; 378 379 #ifdef CONFIG_CRASH_HOTPLUG 380 if ((flags & KEXEC_FILE_ON_CRASH) && arch_crash_hotplug_support(image, flags)) 381 image->hotplug_support = 1; 382 #endif 383 384 ret = machine_kexec_prepare(image); 385 if (ret) 386 goto out; 387 388 /* 389 * Some architecture(like S390) may touch the crash memory before 390 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 391 */ 392 ret = kimage_crash_copy_vmcoreinfo(image); 393 if (ret) 394 goto out; 395 396 ret = kexec_calculate_store_digests(image); 397 if (ret) 398 goto out; 399 400 kexec_dprintk("nr_segments = %lu\n", image->nr_segments); 401 for (i = 0; i < image->nr_segments; i++) { 402 struct kexec_segment *ksegment; 403 404 ksegment = &image->segment[i]; 405 kexec_dprintk("segment[%d]: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 406 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 407 ksegment->memsz); 408 409 ret = kimage_load_segment(image, &image->segment[i]); 410 if (ret) 411 goto out; 412 } 413 414 kimage_terminate(image); 415 416 ret = machine_kexec_post_load(image); 417 if (ret) 418 goto out; 419 420 kexec_dprintk("kexec_file_load: type:%u, start:0x%lx head:0x%lx flags:0x%lx\n", 421 image->type, image->start, image->head, flags); 422 /* 423 * Free up any temporary buffers allocated which are not needed 424 * after image has been loaded 425 */ 426 kimage_file_post_load_cleanup(image); 427 exchange: 428 image = xchg(dest_image, image); 429 out: 430 #ifdef CONFIG_CRASH_DUMP 431 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 432 arch_kexec_protect_crashkres(); 433 #endif 434 435 kexec_unlock(); 436 kimage_free(image); 437 return ret; 438 } 439 440 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 441 struct kexec_buf *kbuf) 442 { 443 struct kimage *image = kbuf->image; 444 unsigned long temp_start, temp_end; 445 446 temp_end = min(end, kbuf->buf_max); 447 temp_start = temp_end - kbuf->memsz + 1; 448 449 do { 450 /* align down start */ 451 temp_start = ALIGN_DOWN(temp_start, kbuf->buf_align); 452 453 if (temp_start < start || temp_start < kbuf->buf_min) 454 return 0; 455 456 temp_end = temp_start + kbuf->memsz - 1; 457 458 /* 459 * Make sure this does not conflict with any of existing 460 * segments 461 */ 462 if (kimage_is_destination_range(image, temp_start, temp_end)) { 463 temp_start = temp_start - PAGE_SIZE; 464 continue; 465 } 466 467 /* Make sure this does not conflict with exclude range */ 468 if (arch_check_excluded_range(image, temp_start, temp_end)) { 469 temp_start = temp_start - PAGE_SIZE; 470 continue; 471 } 472 473 /* We found a suitable memory range */ 474 break; 475 } while (1); 476 477 /* If we are here, we found a suitable memory range */ 478 kbuf->mem = temp_start; 479 480 /* Success, stop navigating through remaining System RAM ranges */ 481 return 1; 482 } 483 484 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 485 struct kexec_buf *kbuf) 486 { 487 struct kimage *image = kbuf->image; 488 unsigned long temp_start, temp_end; 489 490 temp_start = max(start, kbuf->buf_min); 491 492 do { 493 temp_start = ALIGN(temp_start, kbuf->buf_align); 494 temp_end = temp_start + kbuf->memsz - 1; 495 496 if (temp_end > end || temp_end > kbuf->buf_max) 497 return 0; 498 /* 499 * Make sure this does not conflict with any of existing 500 * segments 501 */ 502 if (kimage_is_destination_range(image, temp_start, temp_end)) { 503 temp_start = temp_start + PAGE_SIZE; 504 continue; 505 } 506 507 /* Make sure this does not conflict with exclude range */ 508 if (arch_check_excluded_range(image, temp_start, temp_end)) { 509 temp_start = temp_start + PAGE_SIZE; 510 continue; 511 } 512 513 /* We found a suitable memory range */ 514 break; 515 } while (1); 516 517 /* If we are here, we found a suitable memory range */ 518 kbuf->mem = temp_start; 519 520 /* Success, stop navigating through remaining System RAM ranges */ 521 return 1; 522 } 523 524 static int locate_mem_hole_callback(struct resource *res, void *arg) 525 { 526 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 527 u64 start = res->start, end = res->end; 528 unsigned long sz = end - start + 1; 529 530 /* Returning 0 will take to next memory range */ 531 532 /* Don't use memory that will be detected and handled by a driver. */ 533 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) 534 return 0; 535 536 if (sz < kbuf->memsz) 537 return 0; 538 539 if (end < kbuf->buf_min || start > kbuf->buf_max) 540 return 0; 541 542 /* 543 * Allocate memory top down with-in ram range. Otherwise bottom up 544 * allocation. 545 */ 546 if (kbuf->top_down) 547 return locate_mem_hole_top_down(start, end, kbuf); 548 return locate_mem_hole_bottom_up(start, end, kbuf); 549 } 550 551 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 552 static int kexec_walk_memblock(struct kexec_buf *kbuf, 553 int (*func)(struct resource *, void *)) 554 { 555 int ret = 0; 556 u64 i; 557 phys_addr_t mstart, mend; 558 struct resource res = { }; 559 560 #ifdef CONFIG_CRASH_DUMP 561 if (kbuf->image->type == KEXEC_TYPE_CRASH) 562 return func(&crashk_res, kbuf); 563 #endif 564 565 /* 566 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See 567 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in 568 * locate_mem_hole_callback(). 569 */ 570 if (kbuf->top_down) { 571 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, 572 &mstart, &mend, NULL) { 573 /* 574 * In memblock, end points to the first byte after the 575 * range while in kexec, end points to the last byte 576 * in the range. 577 */ 578 res.start = mstart; 579 res.end = mend - 1; 580 ret = func(&res, kbuf); 581 if (ret) 582 break; 583 } 584 } else { 585 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, 586 &mstart, &mend, NULL) { 587 /* 588 * In memblock, end points to the first byte after the 589 * range while in kexec, end points to the last byte 590 * in the range. 591 */ 592 res.start = mstart; 593 res.end = mend - 1; 594 ret = func(&res, kbuf); 595 if (ret) 596 break; 597 } 598 } 599 600 return ret; 601 } 602 #else 603 static int kexec_walk_memblock(struct kexec_buf *kbuf, 604 int (*func)(struct resource *, void *)) 605 { 606 return 0; 607 } 608 #endif 609 610 /** 611 * kexec_walk_resources - call func(data) on free memory regions 612 * @kbuf: Context info for the search. Also passed to @func. 613 * @func: Function to call for each memory region. 614 * 615 * Return: The memory walk will stop when func returns a non-zero value 616 * and that value will be returned. If all free regions are visited without 617 * func returning non-zero, then zero will be returned. 618 */ 619 static int kexec_walk_resources(struct kexec_buf *kbuf, 620 int (*func)(struct resource *, void *)) 621 { 622 #ifdef CONFIG_CRASH_DUMP 623 if (kbuf->image->type == KEXEC_TYPE_CRASH) 624 return walk_iomem_res_desc(crashk_res.desc, 625 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 626 crashk_res.start, crashk_res.end, 627 kbuf, func); 628 #endif 629 if (kbuf->top_down) 630 return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func); 631 else 632 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 633 } 634 635 /** 636 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 637 * @kbuf: Parameters for the memory search. 638 * 639 * On success, kbuf->mem will have the start address of the memory region found. 640 * 641 * Return: 0 on success, negative errno on error. 642 */ 643 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 644 { 645 int ret; 646 647 /* Arch knows where to place */ 648 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) 649 return 0; 650 651 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 652 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback); 653 else 654 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback); 655 656 return ret == 1 ? 0 : -EADDRNOTAVAIL; 657 } 658 659 /** 660 * kexec_add_buffer - place a buffer in a kexec segment 661 * @kbuf: Buffer contents and memory parameters. 662 * 663 * This function assumes that kexec_lock is held. 664 * On successful return, @kbuf->mem will have the physical address of 665 * the buffer in memory. 666 * 667 * Return: 0 on success, negative errno on error. 668 */ 669 int kexec_add_buffer(struct kexec_buf *kbuf) 670 { 671 struct kexec_segment *ksegment; 672 int ret; 673 674 /* Currently adding segment this way is allowed only in file mode */ 675 if (!kbuf->image->file_mode) 676 return -EINVAL; 677 678 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 679 return -EINVAL; 680 681 /* 682 * Make sure we are not trying to add buffer after allocating 683 * control pages. All segments need to be placed first before 684 * any control pages are allocated. As control page allocation 685 * logic goes through list of segments to make sure there are 686 * no destination overlaps. 687 */ 688 if (!list_empty(&kbuf->image->control_pages)) { 689 WARN_ON(1); 690 return -EINVAL; 691 } 692 693 /* Ensure minimum alignment needed for segments. */ 694 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 695 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 696 697 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 698 ret = arch_kexec_locate_mem_hole(kbuf); 699 if (ret) 700 return ret; 701 702 /* Found a suitable memory range */ 703 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 704 ksegment->kbuf = kbuf->buffer; 705 ksegment->bufsz = kbuf->bufsz; 706 ksegment->mem = kbuf->mem; 707 ksegment->memsz = kbuf->memsz; 708 kbuf->image->nr_segments++; 709 return 0; 710 } 711 712 /* Calculate and store the digest of segments */ 713 static int kexec_calculate_store_digests(struct kimage *image) 714 { 715 struct crypto_shash *tfm; 716 struct shash_desc *desc; 717 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 718 size_t desc_size, nullsz; 719 char *digest; 720 void *zero_buf; 721 struct kexec_sha_region *sha_regions; 722 struct purgatory_info *pi = &image->purgatory_info; 723 724 if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY)) 725 return 0; 726 727 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 728 zero_buf_sz = PAGE_SIZE; 729 730 tfm = crypto_alloc_shash("sha256", 0, 0); 731 if (IS_ERR(tfm)) { 732 ret = PTR_ERR(tfm); 733 goto out; 734 } 735 736 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 737 desc = kzalloc(desc_size, GFP_KERNEL); 738 if (!desc) { 739 ret = -ENOMEM; 740 goto out_free_tfm; 741 } 742 743 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 744 sha_regions = vzalloc(sha_region_sz); 745 if (!sha_regions) { 746 ret = -ENOMEM; 747 goto out_free_desc; 748 } 749 750 desc->tfm = tfm; 751 752 ret = crypto_shash_init(desc); 753 if (ret < 0) 754 goto out_free_sha_regions; 755 756 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 757 if (!digest) { 758 ret = -ENOMEM; 759 goto out_free_sha_regions; 760 } 761 762 for (j = i = 0; i < image->nr_segments; i++) { 763 struct kexec_segment *ksegment; 764 765 #ifdef CONFIG_CRASH_HOTPLUG 766 /* Exclude elfcorehdr segment to allow future changes via hotplug */ 767 if (i == image->elfcorehdr_index) 768 continue; 769 #endif 770 771 ksegment = &image->segment[i]; 772 /* 773 * Skip purgatory as it will be modified once we put digest 774 * info in purgatory. 775 */ 776 if (ksegment->kbuf == pi->purgatory_buf) 777 continue; 778 779 ret = crypto_shash_update(desc, ksegment->kbuf, 780 ksegment->bufsz); 781 if (ret) 782 break; 783 784 /* 785 * Assume rest of the buffer is filled with zero and 786 * update digest accordingly. 787 */ 788 nullsz = ksegment->memsz - ksegment->bufsz; 789 while (nullsz) { 790 unsigned long bytes = nullsz; 791 792 if (bytes > zero_buf_sz) 793 bytes = zero_buf_sz; 794 ret = crypto_shash_update(desc, zero_buf, bytes); 795 if (ret) 796 break; 797 nullsz -= bytes; 798 } 799 800 if (ret) 801 break; 802 803 sha_regions[j].start = ksegment->mem; 804 sha_regions[j].len = ksegment->memsz; 805 j++; 806 } 807 808 if (!ret) { 809 ret = crypto_shash_final(desc, digest); 810 if (ret) 811 goto out_free_digest; 812 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 813 sha_regions, sha_region_sz, 0); 814 if (ret) 815 goto out_free_digest; 816 817 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 818 digest, SHA256_DIGEST_SIZE, 0); 819 if (ret) 820 goto out_free_digest; 821 } 822 823 out_free_digest: 824 kfree(digest); 825 out_free_sha_regions: 826 vfree(sha_regions); 827 out_free_desc: 828 kfree(desc); 829 out_free_tfm: 830 kfree(tfm); 831 out: 832 return ret; 833 } 834 835 #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY 836 /* 837 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 838 * @pi: Purgatory to be loaded. 839 * @kbuf: Buffer to setup. 840 * 841 * Allocates the memory needed for the buffer. Caller is responsible to free 842 * the memory after use. 843 * 844 * Return: 0 on success, negative errno on error. 845 */ 846 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 847 struct kexec_buf *kbuf) 848 { 849 const Elf_Shdr *sechdrs; 850 unsigned long bss_align; 851 unsigned long bss_sz; 852 unsigned long align; 853 int i, ret; 854 855 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 856 kbuf->buf_align = bss_align = 1; 857 kbuf->bufsz = bss_sz = 0; 858 859 for (i = 0; i < pi->ehdr->e_shnum; i++) { 860 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 861 continue; 862 863 align = sechdrs[i].sh_addralign; 864 if (sechdrs[i].sh_type != SHT_NOBITS) { 865 if (kbuf->buf_align < align) 866 kbuf->buf_align = align; 867 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 868 kbuf->bufsz += sechdrs[i].sh_size; 869 } else { 870 if (bss_align < align) 871 bss_align = align; 872 bss_sz = ALIGN(bss_sz, align); 873 bss_sz += sechdrs[i].sh_size; 874 } 875 } 876 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 877 kbuf->memsz = kbuf->bufsz + bss_sz; 878 if (kbuf->buf_align < bss_align) 879 kbuf->buf_align = bss_align; 880 881 kbuf->buffer = vzalloc(kbuf->bufsz); 882 if (!kbuf->buffer) 883 return -ENOMEM; 884 pi->purgatory_buf = kbuf->buffer; 885 886 ret = kexec_add_buffer(kbuf); 887 if (ret) 888 goto out; 889 890 return 0; 891 out: 892 vfree(pi->purgatory_buf); 893 pi->purgatory_buf = NULL; 894 return ret; 895 } 896 897 /* 898 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 899 * @pi: Purgatory to be loaded. 900 * @kbuf: Buffer prepared to store purgatory. 901 * 902 * Allocates the memory needed for the buffer. Caller is responsible to free 903 * the memory after use. 904 * 905 * Return: 0 on success, negative errno on error. 906 */ 907 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 908 struct kexec_buf *kbuf) 909 { 910 unsigned long bss_addr; 911 unsigned long offset; 912 size_t sechdrs_size; 913 Elf_Shdr *sechdrs; 914 int i; 915 916 /* 917 * The section headers in kexec_purgatory are read-only. In order to 918 * have them modifiable make a temporary copy. 919 */ 920 sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum); 921 sechdrs = vzalloc(sechdrs_size); 922 if (!sechdrs) 923 return -ENOMEM; 924 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size); 925 pi->sechdrs = sechdrs; 926 927 offset = 0; 928 bss_addr = kbuf->mem + kbuf->bufsz; 929 kbuf->image->start = pi->ehdr->e_entry; 930 931 for (i = 0; i < pi->ehdr->e_shnum; i++) { 932 unsigned long align; 933 void *src, *dst; 934 935 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 936 continue; 937 938 align = sechdrs[i].sh_addralign; 939 if (sechdrs[i].sh_type == SHT_NOBITS) { 940 bss_addr = ALIGN(bss_addr, align); 941 sechdrs[i].sh_addr = bss_addr; 942 bss_addr += sechdrs[i].sh_size; 943 continue; 944 } 945 946 offset = ALIGN(offset, align); 947 948 /* 949 * Check if the segment contains the entry point, if so, 950 * calculate the value of image->start based on it. 951 * If the compiler has produced more than one .text section 952 * (Eg: .text.hot), they are generally after the main .text 953 * section, and they shall not be used to calculate 954 * image->start. So do not re-calculate image->start if it 955 * is not set to the initial value, and warn the user so they 956 * have a chance to fix their purgatory's linker script. 957 */ 958 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 959 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 960 pi->ehdr->e_entry < (sechdrs[i].sh_addr 961 + sechdrs[i].sh_size) && 962 !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) { 963 kbuf->image->start -= sechdrs[i].sh_addr; 964 kbuf->image->start += kbuf->mem + offset; 965 } 966 967 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 968 dst = pi->purgatory_buf + offset; 969 memcpy(dst, src, sechdrs[i].sh_size); 970 971 sechdrs[i].sh_addr = kbuf->mem + offset; 972 sechdrs[i].sh_offset = offset; 973 offset += sechdrs[i].sh_size; 974 } 975 976 return 0; 977 } 978 979 static int kexec_apply_relocations(struct kimage *image) 980 { 981 int i, ret; 982 struct purgatory_info *pi = &image->purgatory_info; 983 const Elf_Shdr *sechdrs; 984 985 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 986 987 for (i = 0; i < pi->ehdr->e_shnum; i++) { 988 const Elf_Shdr *relsec; 989 const Elf_Shdr *symtab; 990 Elf_Shdr *section; 991 992 relsec = sechdrs + i; 993 994 if (relsec->sh_type != SHT_RELA && 995 relsec->sh_type != SHT_REL) 996 continue; 997 998 /* 999 * For section of type SHT_RELA/SHT_REL, 1000 * ->sh_link contains section header index of associated 1001 * symbol table. And ->sh_info contains section header 1002 * index of section to which relocations apply. 1003 */ 1004 if (relsec->sh_info >= pi->ehdr->e_shnum || 1005 relsec->sh_link >= pi->ehdr->e_shnum) 1006 return -ENOEXEC; 1007 1008 section = pi->sechdrs + relsec->sh_info; 1009 symtab = sechdrs + relsec->sh_link; 1010 1011 if (!(section->sh_flags & SHF_ALLOC)) 1012 continue; 1013 1014 /* 1015 * symtab->sh_link contain section header index of associated 1016 * string table. 1017 */ 1018 if (symtab->sh_link >= pi->ehdr->e_shnum) 1019 /* Invalid section number? */ 1020 continue; 1021 1022 /* 1023 * Respective architecture needs to provide support for applying 1024 * relocations of type SHT_RELA/SHT_REL. 1025 */ 1026 if (relsec->sh_type == SHT_RELA) 1027 ret = arch_kexec_apply_relocations_add(pi, section, 1028 relsec, symtab); 1029 else if (relsec->sh_type == SHT_REL) 1030 ret = arch_kexec_apply_relocations(pi, section, 1031 relsec, symtab); 1032 if (ret) 1033 return ret; 1034 } 1035 1036 return 0; 1037 } 1038 1039 /* 1040 * kexec_load_purgatory - Load and relocate the purgatory object. 1041 * @image: Image to add the purgatory to. 1042 * @kbuf: Memory parameters to use. 1043 * 1044 * Allocates the memory needed for image->purgatory_info.sechdrs and 1045 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 1046 * to free the memory after use. 1047 * 1048 * Return: 0 on success, negative errno on error. 1049 */ 1050 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 1051 { 1052 struct purgatory_info *pi = &image->purgatory_info; 1053 int ret; 1054 1055 if (kexec_purgatory_size <= 0) 1056 return -EINVAL; 1057 1058 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 1059 1060 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 1061 if (ret) 1062 return ret; 1063 1064 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 1065 if (ret) 1066 goto out_free_kbuf; 1067 1068 ret = kexec_apply_relocations(image); 1069 if (ret) 1070 goto out; 1071 1072 return 0; 1073 out: 1074 vfree(pi->sechdrs); 1075 pi->sechdrs = NULL; 1076 out_free_kbuf: 1077 vfree(pi->purgatory_buf); 1078 pi->purgatory_buf = NULL; 1079 return ret; 1080 } 1081 1082 /* 1083 * kexec_purgatory_find_symbol - find a symbol in the purgatory 1084 * @pi: Purgatory to search in. 1085 * @name: Name of the symbol. 1086 * 1087 * Return: pointer to symbol in read-only symtab on success, NULL on error. 1088 */ 1089 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 1090 const char *name) 1091 { 1092 const Elf_Shdr *sechdrs; 1093 const Elf_Ehdr *ehdr; 1094 const Elf_Sym *syms; 1095 const char *strtab; 1096 int i, k; 1097 1098 if (!pi->ehdr) 1099 return NULL; 1100 1101 ehdr = pi->ehdr; 1102 sechdrs = (void *)ehdr + ehdr->e_shoff; 1103 1104 for (i = 0; i < ehdr->e_shnum; i++) { 1105 if (sechdrs[i].sh_type != SHT_SYMTAB) 1106 continue; 1107 1108 if (sechdrs[i].sh_link >= ehdr->e_shnum) 1109 /* Invalid strtab section number */ 1110 continue; 1111 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 1112 syms = (void *)ehdr + sechdrs[i].sh_offset; 1113 1114 /* Go through symbols for a match */ 1115 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 1116 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 1117 continue; 1118 1119 if (strcmp(strtab + syms[k].st_name, name) != 0) 1120 continue; 1121 1122 if (syms[k].st_shndx == SHN_UNDEF || 1123 syms[k].st_shndx >= ehdr->e_shnum) { 1124 pr_debug("Symbol: %s has bad section index %d.\n", 1125 name, syms[k].st_shndx); 1126 return NULL; 1127 } 1128 1129 /* Found the symbol we are looking for */ 1130 return &syms[k]; 1131 } 1132 } 1133 1134 return NULL; 1135 } 1136 1137 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1138 { 1139 struct purgatory_info *pi = &image->purgatory_info; 1140 const Elf_Sym *sym; 1141 Elf_Shdr *sechdr; 1142 1143 sym = kexec_purgatory_find_symbol(pi, name); 1144 if (!sym) 1145 return ERR_PTR(-EINVAL); 1146 1147 sechdr = &pi->sechdrs[sym->st_shndx]; 1148 1149 /* 1150 * Returns the address where symbol will finally be loaded after 1151 * kexec_load_segment() 1152 */ 1153 return (void *)(sechdr->sh_addr + sym->st_value); 1154 } 1155 1156 /* 1157 * Get or set value of a symbol. If "get_value" is true, symbol value is 1158 * returned in buf otherwise symbol value is set based on value in buf. 1159 */ 1160 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1161 void *buf, unsigned int size, bool get_value) 1162 { 1163 struct purgatory_info *pi = &image->purgatory_info; 1164 const Elf_Sym *sym; 1165 Elf_Shdr *sec; 1166 char *sym_buf; 1167 1168 sym = kexec_purgatory_find_symbol(pi, name); 1169 if (!sym) 1170 return -EINVAL; 1171 1172 if (sym->st_size != size) { 1173 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1174 name, (unsigned long)sym->st_size, size); 1175 return -EINVAL; 1176 } 1177 1178 sec = pi->sechdrs + sym->st_shndx; 1179 1180 if (sec->sh_type == SHT_NOBITS) { 1181 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1182 get_value ? "get" : "set"); 1183 return -EINVAL; 1184 } 1185 1186 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1187 1188 if (get_value) 1189 memcpy((void *)buf, sym_buf, size); 1190 else 1191 memcpy((void *)sym_buf, buf, size); 1192 1193 return 0; 1194 } 1195 #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */ 1196