1 /* 2 * kexec: kexec_file_load system call 3 * 4 * Copyright (C) 2014 Red Hat Inc. 5 * Authors: 6 * Vivek Goyal <vgoyal@redhat.com> 7 * 8 * This source code is licensed under the GNU General Public License, 9 * Version 2. See the file COPYING for more details. 10 */ 11 12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 13 14 #include <linux/capability.h> 15 #include <linux/mm.h> 16 #include <linux/file.h> 17 #include <linux/slab.h> 18 #include <linux/kexec.h> 19 #include <linux/mutex.h> 20 #include <linux/list.h> 21 #include <linux/fs.h> 22 #include <linux/ima.h> 23 #include <crypto/hash.h> 24 #include <crypto/sha.h> 25 #include <linux/elf.h> 26 #include <linux/elfcore.h> 27 #include <linux/kernel.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 /* 35 * Currently this is the only default function that is exported as some 36 * architectures need it to do additional handlings. 37 * In the future, other default functions may be exported too if required. 38 */ 39 int kexec_image_probe_default(struct kimage *image, void *buf, 40 unsigned long buf_len) 41 { 42 const struct kexec_file_ops * const *fops; 43 int ret = -ENOEXEC; 44 45 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { 46 ret = (*fops)->probe(buf, buf_len); 47 if (!ret) { 48 image->fops = *fops; 49 return ret; 50 } 51 } 52 53 return ret; 54 } 55 56 /* Architectures can provide this probe function */ 57 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, 58 unsigned long buf_len) 59 { 60 return kexec_image_probe_default(image, buf, buf_len); 61 } 62 63 static void *kexec_image_load_default(struct kimage *image) 64 { 65 if (!image->fops || !image->fops->load) 66 return ERR_PTR(-ENOEXEC); 67 68 return image->fops->load(image, image->kernel_buf, 69 image->kernel_buf_len, image->initrd_buf, 70 image->initrd_buf_len, image->cmdline_buf, 71 image->cmdline_buf_len); 72 } 73 74 void * __weak arch_kexec_kernel_image_load(struct kimage *image) 75 { 76 return kexec_image_load_default(image); 77 } 78 79 static 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 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) 88 { 89 return kexec_image_post_load_cleanup_default(image); 90 } 91 92 #ifdef CONFIG_KEXEC_VERIFY_SIG 93 static int kexec_image_verify_sig_default(struct kimage *image, void *buf, 94 unsigned long buf_len) 95 { 96 if (!image->fops || !image->fops->verify_sig) { 97 pr_debug("kernel loader does not support signature verification.\n"); 98 return -EKEYREJECTED; 99 } 100 101 return image->fops->verify_sig(buf, buf_len); 102 } 103 104 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, 105 unsigned long buf_len) 106 { 107 return kexec_image_verify_sig_default(image, buf, buf_len); 108 } 109 #endif 110 111 /* 112 * arch_kexec_apply_relocations_add - apply relocations of type RELA 113 * @pi: Purgatory to be relocated. 114 * @section: Section relocations applying to. 115 * @relsec: Section containing RELAs. 116 * @symtab: Corresponding symtab. 117 * 118 * Return: 0 on success, negative errno on error. 119 */ 120 int __weak 121 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section, 122 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 123 { 124 pr_err("RELA relocation unsupported.\n"); 125 return -ENOEXEC; 126 } 127 128 /* 129 * arch_kexec_apply_relocations - apply relocations of type REL 130 * @pi: Purgatory to be relocated. 131 * @section: Section relocations applying to. 132 * @relsec: Section containing RELs. 133 * @symtab: Corresponding symtab. 134 * 135 * Return: 0 on success, negative errno on error. 136 */ 137 int __weak 138 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section, 139 const Elf_Shdr *relsec, const Elf_Shdr *symtab) 140 { 141 pr_err("REL relocation unsupported.\n"); 142 return -ENOEXEC; 143 } 144 145 /* 146 * Free up memory used by kernel, initrd, and command line. This is temporary 147 * memory allocation which is not needed any more after these buffers have 148 * been loaded into separate segments and have been copied elsewhere. 149 */ 150 void kimage_file_post_load_cleanup(struct kimage *image) 151 { 152 struct purgatory_info *pi = &image->purgatory_info; 153 154 vfree(image->kernel_buf); 155 image->kernel_buf = NULL; 156 157 vfree(image->initrd_buf); 158 image->initrd_buf = NULL; 159 160 kfree(image->cmdline_buf); 161 image->cmdline_buf = NULL; 162 163 vfree(pi->purgatory_buf); 164 pi->purgatory_buf = NULL; 165 166 vfree(pi->sechdrs); 167 pi->sechdrs = NULL; 168 169 /* See if architecture has anything to cleanup post load */ 170 arch_kimage_file_post_load_cleanup(image); 171 172 /* 173 * Above call should have called into bootloader to free up 174 * any data stored in kimage->image_loader_data. It should 175 * be ok now to free it up. 176 */ 177 kfree(image->image_loader_data); 178 image->image_loader_data = NULL; 179 } 180 181 /* 182 * In file mode list of segments is prepared by kernel. Copy relevant 183 * data from user space, do error checking, prepare segment list 184 */ 185 static int 186 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, 187 const char __user *cmdline_ptr, 188 unsigned long cmdline_len, unsigned flags) 189 { 190 int ret = 0; 191 void *ldata; 192 loff_t size; 193 194 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf, 195 &size, INT_MAX, READING_KEXEC_IMAGE); 196 if (ret) 197 return ret; 198 image->kernel_buf_len = size; 199 200 /* IMA needs to pass the measurement list to the next kernel. */ 201 ima_add_kexec_buffer(image); 202 203 /* Call arch image probe handlers */ 204 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, 205 image->kernel_buf_len); 206 if (ret) 207 goto out; 208 209 #ifdef CONFIG_KEXEC_VERIFY_SIG 210 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, 211 image->kernel_buf_len); 212 if (ret) { 213 pr_debug("kernel signature verification failed.\n"); 214 goto out; 215 } 216 pr_debug("kernel signature verification successful.\n"); 217 #endif 218 /* It is possible that there no initramfs is being loaded */ 219 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { 220 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf, 221 &size, INT_MAX, 222 READING_KEXEC_INITRAMFS); 223 if (ret) 224 goto out; 225 image->initrd_buf_len = size; 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 } 244 245 /* Call arch image load handlers */ 246 ldata = arch_kexec_kernel_image_load(image); 247 248 if (IS_ERR(ldata)) { 249 ret = PTR_ERR(ldata); 250 goto out; 251 } 252 253 image->image_loader_data = ldata; 254 out: 255 /* In case of error, free up all allocated memory in this function */ 256 if (ret) 257 kimage_file_post_load_cleanup(image); 258 return ret; 259 } 260 261 static int 262 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, 263 int initrd_fd, const char __user *cmdline_ptr, 264 unsigned long cmdline_len, unsigned long flags) 265 { 266 int ret; 267 struct kimage *image; 268 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; 269 270 image = do_kimage_alloc_init(); 271 if (!image) 272 return -ENOMEM; 273 274 image->file_mode = 1; 275 276 if (kexec_on_panic) { 277 /* Enable special crash kernel control page alloc policy. */ 278 image->control_page = crashk_res.start; 279 image->type = KEXEC_TYPE_CRASH; 280 } 281 282 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, 283 cmdline_ptr, cmdline_len, flags); 284 if (ret) 285 goto out_free_image; 286 287 ret = sanity_check_segment_list(image); 288 if (ret) 289 goto out_free_post_load_bufs; 290 291 ret = -ENOMEM; 292 image->control_code_page = kimage_alloc_control_pages(image, 293 get_order(KEXEC_CONTROL_PAGE_SIZE)); 294 if (!image->control_code_page) { 295 pr_err("Could not allocate control_code_buffer\n"); 296 goto out_free_post_load_bufs; 297 } 298 299 if (!kexec_on_panic) { 300 image->swap_page = kimage_alloc_control_pages(image, 0); 301 if (!image->swap_page) { 302 pr_err("Could not allocate swap buffer\n"); 303 goto out_free_control_pages; 304 } 305 } 306 307 *rimage = image; 308 return 0; 309 out_free_control_pages: 310 kimage_free_page_list(&image->control_pages); 311 out_free_post_load_bufs: 312 kimage_file_post_load_cleanup(image); 313 out_free_image: 314 kfree(image); 315 return ret; 316 } 317 318 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 319 unsigned long, cmdline_len, const char __user *, cmdline_ptr, 320 unsigned long, flags) 321 { 322 int ret = 0, i; 323 struct kimage **dest_image, *image; 324 325 /* We only trust the superuser with rebooting the system. */ 326 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) 327 return -EPERM; 328 329 /* Make sure we have a legal set of flags */ 330 if (flags != (flags & KEXEC_FILE_FLAGS)) 331 return -EINVAL; 332 333 image = NULL; 334 335 if (!mutex_trylock(&kexec_mutex)) 336 return -EBUSY; 337 338 dest_image = &kexec_image; 339 if (flags & KEXEC_FILE_ON_CRASH) { 340 dest_image = &kexec_crash_image; 341 if (kexec_crash_image) 342 arch_kexec_unprotect_crashkres(); 343 } 344 345 if (flags & KEXEC_FILE_UNLOAD) 346 goto exchange; 347 348 /* 349 * In case of crash, new kernel gets loaded in reserved region. It is 350 * same memory where old crash kernel might be loaded. Free any 351 * current crash dump kernel before we corrupt it. 352 */ 353 if (flags & KEXEC_FILE_ON_CRASH) 354 kimage_free(xchg(&kexec_crash_image, NULL)); 355 356 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, 357 cmdline_len, flags); 358 if (ret) 359 goto out; 360 361 ret = machine_kexec_prepare(image); 362 if (ret) 363 goto out; 364 365 /* 366 * Some architecture(like S390) may touch the crash memory before 367 * machine_kexec_prepare(), we must copy vmcoreinfo data after it. 368 */ 369 ret = kimage_crash_copy_vmcoreinfo(image); 370 if (ret) 371 goto out; 372 373 ret = kexec_calculate_store_digests(image); 374 if (ret) 375 goto out; 376 377 for (i = 0; i < image->nr_segments; i++) { 378 struct kexec_segment *ksegment; 379 380 ksegment = &image->segment[i]; 381 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", 382 i, ksegment->buf, ksegment->bufsz, ksegment->mem, 383 ksegment->memsz); 384 385 ret = kimage_load_segment(image, &image->segment[i]); 386 if (ret) 387 goto out; 388 } 389 390 kimage_terminate(image); 391 392 /* 393 * Free up any temporary buffers allocated which are not needed 394 * after image has been loaded 395 */ 396 kimage_file_post_load_cleanup(image); 397 exchange: 398 image = xchg(dest_image, image); 399 out: 400 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) 401 arch_kexec_protect_crashkres(); 402 403 mutex_unlock(&kexec_mutex); 404 kimage_free(image); 405 return ret; 406 } 407 408 static int locate_mem_hole_top_down(unsigned long start, unsigned long end, 409 struct kexec_buf *kbuf) 410 { 411 struct kimage *image = kbuf->image; 412 unsigned long temp_start, temp_end; 413 414 temp_end = min(end, kbuf->buf_max); 415 temp_start = temp_end - kbuf->memsz; 416 417 do { 418 /* align down start */ 419 temp_start = temp_start & (~(kbuf->buf_align - 1)); 420 421 if (temp_start < start || temp_start < kbuf->buf_min) 422 return 0; 423 424 temp_end = temp_start + kbuf->memsz - 1; 425 426 /* 427 * Make sure this does not conflict with any of existing 428 * segments 429 */ 430 if (kimage_is_destination_range(image, temp_start, temp_end)) { 431 temp_start = temp_start - PAGE_SIZE; 432 continue; 433 } 434 435 /* We found a suitable memory range */ 436 break; 437 } while (1); 438 439 /* If we are here, we found a suitable memory range */ 440 kbuf->mem = temp_start; 441 442 /* Success, stop navigating through remaining System RAM ranges */ 443 return 1; 444 } 445 446 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, 447 struct kexec_buf *kbuf) 448 { 449 struct kimage *image = kbuf->image; 450 unsigned long temp_start, temp_end; 451 452 temp_start = max(start, kbuf->buf_min); 453 454 do { 455 temp_start = ALIGN(temp_start, kbuf->buf_align); 456 temp_end = temp_start + kbuf->memsz - 1; 457 458 if (temp_end > end || temp_end > kbuf->buf_max) 459 return 0; 460 /* 461 * Make sure this does not conflict with any of existing 462 * segments 463 */ 464 if (kimage_is_destination_range(image, temp_start, temp_end)) { 465 temp_start = temp_start + PAGE_SIZE; 466 continue; 467 } 468 469 /* We found a suitable memory range */ 470 break; 471 } while (1); 472 473 /* If we are here, we found a suitable memory range */ 474 kbuf->mem = temp_start; 475 476 /* Success, stop navigating through remaining System RAM ranges */ 477 return 1; 478 } 479 480 static int locate_mem_hole_callback(struct resource *res, void *arg) 481 { 482 struct kexec_buf *kbuf = (struct kexec_buf *)arg; 483 u64 start = res->start, end = res->end; 484 unsigned long sz = end - start + 1; 485 486 /* Returning 0 will take to next memory range */ 487 if (sz < kbuf->memsz) 488 return 0; 489 490 if (end < kbuf->buf_min || start > kbuf->buf_max) 491 return 0; 492 493 /* 494 * Allocate memory top down with-in ram range. Otherwise bottom up 495 * allocation. 496 */ 497 if (kbuf->top_down) 498 return locate_mem_hole_top_down(start, end, kbuf); 499 return locate_mem_hole_bottom_up(start, end, kbuf); 500 } 501 502 /** 503 * arch_kexec_walk_mem - call func(data) on free memory regions 504 * @kbuf: Context info for the search. Also passed to @func. 505 * @func: Function to call for each memory region. 506 * 507 * Return: The memory walk will stop when func returns a non-zero value 508 * and that value will be returned. If all free regions are visited without 509 * func returning non-zero, then zero will be returned. 510 */ 511 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf, 512 int (*func)(struct resource *, void *)) 513 { 514 if (kbuf->image->type == KEXEC_TYPE_CRASH) 515 return walk_iomem_res_desc(crashk_res.desc, 516 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, 517 crashk_res.start, crashk_res.end, 518 kbuf, func); 519 else 520 return walk_system_ram_res(0, ULONG_MAX, kbuf, func); 521 } 522 523 /** 524 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel 525 * @kbuf: Parameters for the memory search. 526 * 527 * On success, kbuf->mem will have the start address of the memory region found. 528 * 529 * Return: 0 on success, negative errno on error. 530 */ 531 int kexec_locate_mem_hole(struct kexec_buf *kbuf) 532 { 533 int ret; 534 535 ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback); 536 537 return ret == 1 ? 0 : -EADDRNOTAVAIL; 538 } 539 540 /** 541 * kexec_add_buffer - place a buffer in a kexec segment 542 * @kbuf: Buffer contents and memory parameters. 543 * 544 * This function assumes that kexec_mutex is held. 545 * On successful return, @kbuf->mem will have the physical address of 546 * the buffer in memory. 547 * 548 * Return: 0 on success, negative errno on error. 549 */ 550 int kexec_add_buffer(struct kexec_buf *kbuf) 551 { 552 553 struct kexec_segment *ksegment; 554 int ret; 555 556 /* Currently adding segment this way is allowed only in file mode */ 557 if (!kbuf->image->file_mode) 558 return -EINVAL; 559 560 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) 561 return -EINVAL; 562 563 /* 564 * Make sure we are not trying to add buffer after allocating 565 * control pages. All segments need to be placed first before 566 * any control pages are allocated. As control page allocation 567 * logic goes through list of segments to make sure there are 568 * no destination overlaps. 569 */ 570 if (!list_empty(&kbuf->image->control_pages)) { 571 WARN_ON(1); 572 return -EINVAL; 573 } 574 575 /* Ensure minimum alignment needed for segments. */ 576 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); 577 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); 578 579 /* Walk the RAM ranges and allocate a suitable range for the buffer */ 580 ret = kexec_locate_mem_hole(kbuf); 581 if (ret) 582 return ret; 583 584 /* Found a suitable memory range */ 585 ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; 586 ksegment->kbuf = kbuf->buffer; 587 ksegment->bufsz = kbuf->bufsz; 588 ksegment->mem = kbuf->mem; 589 ksegment->memsz = kbuf->memsz; 590 kbuf->image->nr_segments++; 591 return 0; 592 } 593 594 /* Calculate and store the digest of segments */ 595 static int kexec_calculate_store_digests(struct kimage *image) 596 { 597 struct crypto_shash *tfm; 598 struct shash_desc *desc; 599 int ret = 0, i, j, zero_buf_sz, sha_region_sz; 600 size_t desc_size, nullsz; 601 char *digest; 602 void *zero_buf; 603 struct kexec_sha_region *sha_regions; 604 struct purgatory_info *pi = &image->purgatory_info; 605 606 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY)) 607 return 0; 608 609 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); 610 zero_buf_sz = PAGE_SIZE; 611 612 tfm = crypto_alloc_shash("sha256", 0, 0); 613 if (IS_ERR(tfm)) { 614 ret = PTR_ERR(tfm); 615 goto out; 616 } 617 618 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); 619 desc = kzalloc(desc_size, GFP_KERNEL); 620 if (!desc) { 621 ret = -ENOMEM; 622 goto out_free_tfm; 623 } 624 625 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); 626 sha_regions = vzalloc(sha_region_sz); 627 if (!sha_regions) 628 goto out_free_desc; 629 630 desc->tfm = tfm; 631 desc->flags = 0; 632 633 ret = crypto_shash_init(desc); 634 if (ret < 0) 635 goto out_free_sha_regions; 636 637 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); 638 if (!digest) { 639 ret = -ENOMEM; 640 goto out_free_sha_regions; 641 } 642 643 for (j = i = 0; i < image->nr_segments; i++) { 644 struct kexec_segment *ksegment; 645 646 ksegment = &image->segment[i]; 647 /* 648 * Skip purgatory as it will be modified once we put digest 649 * info in purgatory. 650 */ 651 if (ksegment->kbuf == pi->purgatory_buf) 652 continue; 653 654 ret = crypto_shash_update(desc, ksegment->kbuf, 655 ksegment->bufsz); 656 if (ret) 657 break; 658 659 /* 660 * Assume rest of the buffer is filled with zero and 661 * update digest accordingly. 662 */ 663 nullsz = ksegment->memsz - ksegment->bufsz; 664 while (nullsz) { 665 unsigned long bytes = nullsz; 666 667 if (bytes > zero_buf_sz) 668 bytes = zero_buf_sz; 669 ret = crypto_shash_update(desc, zero_buf, bytes); 670 if (ret) 671 break; 672 nullsz -= bytes; 673 } 674 675 if (ret) 676 break; 677 678 sha_regions[j].start = ksegment->mem; 679 sha_regions[j].len = ksegment->memsz; 680 j++; 681 } 682 683 if (!ret) { 684 ret = crypto_shash_final(desc, digest); 685 if (ret) 686 goto out_free_digest; 687 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", 688 sha_regions, sha_region_sz, 0); 689 if (ret) 690 goto out_free_digest; 691 692 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", 693 digest, SHA256_DIGEST_SIZE, 0); 694 if (ret) 695 goto out_free_digest; 696 } 697 698 out_free_digest: 699 kfree(digest); 700 out_free_sha_regions: 701 vfree(sha_regions); 702 out_free_desc: 703 kfree(desc); 704 out_free_tfm: 705 kfree(tfm); 706 out: 707 return ret; 708 } 709 710 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY 711 /* 712 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. 713 * @pi: Purgatory to be loaded. 714 * @kbuf: Buffer to setup. 715 * 716 * Allocates the memory needed for the buffer. Caller is responsible to free 717 * the memory after use. 718 * 719 * Return: 0 on success, negative errno on error. 720 */ 721 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, 722 struct kexec_buf *kbuf) 723 { 724 const Elf_Shdr *sechdrs; 725 unsigned long bss_align; 726 unsigned long bss_sz; 727 unsigned long align; 728 int i, ret; 729 730 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 731 kbuf->buf_align = bss_align = 1; 732 kbuf->bufsz = bss_sz = 0; 733 734 for (i = 0; i < pi->ehdr->e_shnum; i++) { 735 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 736 continue; 737 738 align = sechdrs[i].sh_addralign; 739 if (sechdrs[i].sh_type != SHT_NOBITS) { 740 if (kbuf->buf_align < align) 741 kbuf->buf_align = align; 742 kbuf->bufsz = ALIGN(kbuf->bufsz, align); 743 kbuf->bufsz += sechdrs[i].sh_size; 744 } else { 745 if (bss_align < align) 746 bss_align = align; 747 bss_sz = ALIGN(bss_sz, align); 748 bss_sz += sechdrs[i].sh_size; 749 } 750 } 751 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); 752 kbuf->memsz = kbuf->bufsz + bss_sz; 753 if (kbuf->buf_align < bss_align) 754 kbuf->buf_align = bss_align; 755 756 kbuf->buffer = vzalloc(kbuf->bufsz); 757 if (!kbuf->buffer) 758 return -ENOMEM; 759 pi->purgatory_buf = kbuf->buffer; 760 761 ret = kexec_add_buffer(kbuf); 762 if (ret) 763 goto out; 764 765 return 0; 766 out: 767 vfree(pi->purgatory_buf); 768 pi->purgatory_buf = NULL; 769 return ret; 770 } 771 772 /* 773 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. 774 * @pi: Purgatory to be loaded. 775 * @kbuf: Buffer prepared to store purgatory. 776 * 777 * Allocates the memory needed for the buffer. Caller is responsible to free 778 * the memory after use. 779 * 780 * Return: 0 on success, negative errno on error. 781 */ 782 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, 783 struct kexec_buf *kbuf) 784 { 785 unsigned long bss_addr; 786 unsigned long offset; 787 Elf_Shdr *sechdrs; 788 int i; 789 790 /* 791 * The section headers in kexec_purgatory are read-only. In order to 792 * have them modifiable make a temporary copy. 793 */ 794 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum)); 795 if (!sechdrs) 796 return -ENOMEM; 797 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, 798 pi->ehdr->e_shnum * sizeof(Elf_Shdr)); 799 pi->sechdrs = sechdrs; 800 801 offset = 0; 802 bss_addr = kbuf->mem + kbuf->bufsz; 803 kbuf->image->start = pi->ehdr->e_entry; 804 805 for (i = 0; i < pi->ehdr->e_shnum; i++) { 806 unsigned long align; 807 void *src, *dst; 808 809 if (!(sechdrs[i].sh_flags & SHF_ALLOC)) 810 continue; 811 812 align = sechdrs[i].sh_addralign; 813 if (sechdrs[i].sh_type == SHT_NOBITS) { 814 bss_addr = ALIGN(bss_addr, align); 815 sechdrs[i].sh_addr = bss_addr; 816 bss_addr += sechdrs[i].sh_size; 817 continue; 818 } 819 820 offset = ALIGN(offset, align); 821 if (sechdrs[i].sh_flags & SHF_EXECINSTR && 822 pi->ehdr->e_entry >= sechdrs[i].sh_addr && 823 pi->ehdr->e_entry < (sechdrs[i].sh_addr 824 + sechdrs[i].sh_size)) { 825 kbuf->image->start -= sechdrs[i].sh_addr; 826 kbuf->image->start += kbuf->mem + offset; 827 } 828 829 src = (void *)pi->ehdr + sechdrs[i].sh_offset; 830 dst = pi->purgatory_buf + offset; 831 memcpy(dst, src, sechdrs[i].sh_size); 832 833 sechdrs[i].sh_addr = kbuf->mem + offset; 834 sechdrs[i].sh_offset = offset; 835 offset += sechdrs[i].sh_size; 836 } 837 838 return 0; 839 } 840 841 static int kexec_apply_relocations(struct kimage *image) 842 { 843 int i, ret; 844 struct purgatory_info *pi = &image->purgatory_info; 845 const Elf_Shdr *sechdrs; 846 847 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 848 849 for (i = 0; i < pi->ehdr->e_shnum; i++) { 850 const Elf_Shdr *relsec; 851 const Elf_Shdr *symtab; 852 Elf_Shdr *section; 853 854 relsec = sechdrs + i; 855 856 if (relsec->sh_type != SHT_RELA && 857 relsec->sh_type != SHT_REL) 858 continue; 859 860 /* 861 * For section of type SHT_RELA/SHT_REL, 862 * ->sh_link contains section header index of associated 863 * symbol table. And ->sh_info contains section header 864 * index of section to which relocations apply. 865 */ 866 if (relsec->sh_info >= pi->ehdr->e_shnum || 867 relsec->sh_link >= pi->ehdr->e_shnum) 868 return -ENOEXEC; 869 870 section = pi->sechdrs + relsec->sh_info; 871 symtab = sechdrs + relsec->sh_link; 872 873 if (!(section->sh_flags & SHF_ALLOC)) 874 continue; 875 876 /* 877 * symtab->sh_link contain section header index of associated 878 * string table. 879 */ 880 if (symtab->sh_link >= pi->ehdr->e_shnum) 881 /* Invalid section number? */ 882 continue; 883 884 /* 885 * Respective architecture needs to provide support for applying 886 * relocations of type SHT_RELA/SHT_REL. 887 */ 888 if (relsec->sh_type == SHT_RELA) 889 ret = arch_kexec_apply_relocations_add(pi, section, 890 relsec, symtab); 891 else if (relsec->sh_type == SHT_REL) 892 ret = arch_kexec_apply_relocations(pi, section, 893 relsec, symtab); 894 if (ret) 895 return ret; 896 } 897 898 return 0; 899 } 900 901 /* 902 * kexec_load_purgatory - Load and relocate the purgatory object. 903 * @image: Image to add the purgatory to. 904 * @kbuf: Memory parameters to use. 905 * 906 * Allocates the memory needed for image->purgatory_info.sechdrs and 907 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible 908 * to free the memory after use. 909 * 910 * Return: 0 on success, negative errno on error. 911 */ 912 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) 913 { 914 struct purgatory_info *pi = &image->purgatory_info; 915 int ret; 916 917 if (kexec_purgatory_size <= 0) 918 return -EINVAL; 919 920 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; 921 922 ret = kexec_purgatory_setup_kbuf(pi, kbuf); 923 if (ret) 924 return ret; 925 926 ret = kexec_purgatory_setup_sechdrs(pi, kbuf); 927 if (ret) 928 goto out_free_kbuf; 929 930 ret = kexec_apply_relocations(image); 931 if (ret) 932 goto out; 933 934 return 0; 935 out: 936 vfree(pi->sechdrs); 937 pi->sechdrs = NULL; 938 out_free_kbuf: 939 vfree(pi->purgatory_buf); 940 pi->purgatory_buf = NULL; 941 return ret; 942 } 943 944 /* 945 * kexec_purgatory_find_symbol - find a symbol in the purgatory 946 * @pi: Purgatory to search in. 947 * @name: Name of the symbol. 948 * 949 * Return: pointer to symbol in read-only symtab on success, NULL on error. 950 */ 951 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, 952 const char *name) 953 { 954 const Elf_Shdr *sechdrs; 955 const Elf_Ehdr *ehdr; 956 const Elf_Sym *syms; 957 const char *strtab; 958 int i, k; 959 960 if (!pi->ehdr) 961 return NULL; 962 963 ehdr = pi->ehdr; 964 sechdrs = (void *)ehdr + ehdr->e_shoff; 965 966 for (i = 0; i < ehdr->e_shnum; i++) { 967 if (sechdrs[i].sh_type != SHT_SYMTAB) 968 continue; 969 970 if (sechdrs[i].sh_link >= ehdr->e_shnum) 971 /* Invalid strtab section number */ 972 continue; 973 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; 974 syms = (void *)ehdr + sechdrs[i].sh_offset; 975 976 /* Go through symbols for a match */ 977 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { 978 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) 979 continue; 980 981 if (strcmp(strtab + syms[k].st_name, name) != 0) 982 continue; 983 984 if (syms[k].st_shndx == SHN_UNDEF || 985 syms[k].st_shndx >= ehdr->e_shnum) { 986 pr_debug("Symbol: %s has bad section index %d.\n", 987 name, syms[k].st_shndx); 988 return NULL; 989 } 990 991 /* Found the symbol we are looking for */ 992 return &syms[k]; 993 } 994 } 995 996 return NULL; 997 } 998 999 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) 1000 { 1001 struct purgatory_info *pi = &image->purgatory_info; 1002 const Elf_Sym *sym; 1003 Elf_Shdr *sechdr; 1004 1005 sym = kexec_purgatory_find_symbol(pi, name); 1006 if (!sym) 1007 return ERR_PTR(-EINVAL); 1008 1009 sechdr = &pi->sechdrs[sym->st_shndx]; 1010 1011 /* 1012 * Returns the address where symbol will finally be loaded after 1013 * kexec_load_segment() 1014 */ 1015 return (void *)(sechdr->sh_addr + sym->st_value); 1016 } 1017 1018 /* 1019 * Get or set value of a symbol. If "get_value" is true, symbol value is 1020 * returned in buf otherwise symbol value is set based on value in buf. 1021 */ 1022 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, 1023 void *buf, unsigned int size, bool get_value) 1024 { 1025 struct purgatory_info *pi = &image->purgatory_info; 1026 const Elf_Sym *sym; 1027 Elf_Shdr *sec; 1028 char *sym_buf; 1029 1030 sym = kexec_purgatory_find_symbol(pi, name); 1031 if (!sym) 1032 return -EINVAL; 1033 1034 if (sym->st_size != size) { 1035 pr_err("symbol %s size mismatch: expected %lu actual %u\n", 1036 name, (unsigned long)sym->st_size, size); 1037 return -EINVAL; 1038 } 1039 1040 sec = pi->sechdrs + sym->st_shndx; 1041 1042 if (sec->sh_type == SHT_NOBITS) { 1043 pr_err("symbol %s is in a bss section. Cannot %s\n", name, 1044 get_value ? "get" : "set"); 1045 return -EINVAL; 1046 } 1047 1048 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; 1049 1050 if (get_value) 1051 memcpy((void *)buf, sym_buf, size); 1052 else 1053 memcpy((void *)sym_buf, buf, size); 1054 1055 return 0; 1056 } 1057 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */ 1058 1059 int crash_exclude_mem_range(struct crash_mem *mem, 1060 unsigned long long mstart, unsigned long long mend) 1061 { 1062 int i, j; 1063 unsigned long long start, end; 1064 struct crash_mem_range temp_range = {0, 0}; 1065 1066 for (i = 0; i < mem->nr_ranges; i++) { 1067 start = mem->ranges[i].start; 1068 end = mem->ranges[i].end; 1069 1070 if (mstart > end || mend < start) 1071 continue; 1072 1073 /* Truncate any area outside of range */ 1074 if (mstart < start) 1075 mstart = start; 1076 if (mend > end) 1077 mend = end; 1078 1079 /* Found completely overlapping range */ 1080 if (mstart == start && mend == end) { 1081 mem->ranges[i].start = 0; 1082 mem->ranges[i].end = 0; 1083 if (i < mem->nr_ranges - 1) { 1084 /* Shift rest of the ranges to left */ 1085 for (j = i; j < mem->nr_ranges - 1; j++) { 1086 mem->ranges[j].start = 1087 mem->ranges[j+1].start; 1088 mem->ranges[j].end = 1089 mem->ranges[j+1].end; 1090 } 1091 } 1092 mem->nr_ranges--; 1093 return 0; 1094 } 1095 1096 if (mstart > start && mend < end) { 1097 /* Split original range */ 1098 mem->ranges[i].end = mstart - 1; 1099 temp_range.start = mend + 1; 1100 temp_range.end = end; 1101 } else if (mstart != start) 1102 mem->ranges[i].end = mstart - 1; 1103 else 1104 mem->ranges[i].start = mend + 1; 1105 break; 1106 } 1107 1108 /* If a split happened, add the split to array */ 1109 if (!temp_range.end) 1110 return 0; 1111 1112 /* Split happened */ 1113 if (i == mem->max_nr_ranges - 1) 1114 return -ENOMEM; 1115 1116 /* Location where new range should go */ 1117 j = i + 1; 1118 if (j < mem->nr_ranges) { 1119 /* Move over all ranges one slot towards the end */ 1120 for (i = mem->nr_ranges - 1; i >= j; i--) 1121 mem->ranges[i + 1] = mem->ranges[i]; 1122 } 1123 1124 mem->ranges[j].start = temp_range.start; 1125 mem->ranges[j].end = temp_range.end; 1126 mem->nr_ranges++; 1127 return 0; 1128 } 1129 1130 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, 1131 void **addr, unsigned long *sz) 1132 { 1133 Elf64_Ehdr *ehdr; 1134 Elf64_Phdr *phdr; 1135 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; 1136 unsigned char *buf; 1137 unsigned int cpu, i; 1138 unsigned long long notes_addr; 1139 unsigned long mstart, mend; 1140 1141 /* extra phdr for vmcoreinfo elf note */ 1142 nr_phdr = nr_cpus + 1; 1143 nr_phdr += mem->nr_ranges; 1144 1145 /* 1146 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping 1147 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). 1148 * I think this is required by tools like gdb. So same physical 1149 * memory will be mapped in two elf headers. One will contain kernel 1150 * text virtual addresses and other will have __va(physical) addresses. 1151 */ 1152 1153 nr_phdr++; 1154 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); 1155 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); 1156 1157 buf = vzalloc(elf_sz); 1158 if (!buf) 1159 return -ENOMEM; 1160 1161 ehdr = (Elf64_Ehdr *)buf; 1162 phdr = (Elf64_Phdr *)(ehdr + 1); 1163 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 1164 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 1165 ehdr->e_ident[EI_DATA] = ELFDATA2LSB; 1166 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1167 ehdr->e_ident[EI_OSABI] = ELF_OSABI; 1168 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 1169 ehdr->e_type = ET_CORE; 1170 ehdr->e_machine = ELF_ARCH; 1171 ehdr->e_version = EV_CURRENT; 1172 ehdr->e_phoff = sizeof(Elf64_Ehdr); 1173 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 1174 ehdr->e_phentsize = sizeof(Elf64_Phdr); 1175 1176 /* Prepare one phdr of type PT_NOTE for each present cpu */ 1177 for_each_present_cpu(cpu) { 1178 phdr->p_type = PT_NOTE; 1179 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); 1180 phdr->p_offset = phdr->p_paddr = notes_addr; 1181 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); 1182 (ehdr->e_phnum)++; 1183 phdr++; 1184 } 1185 1186 /* Prepare one PT_NOTE header for vmcoreinfo */ 1187 phdr->p_type = PT_NOTE; 1188 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); 1189 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; 1190 (ehdr->e_phnum)++; 1191 phdr++; 1192 1193 /* Prepare PT_LOAD type program header for kernel text region */ 1194 if (kernel_map) { 1195 phdr->p_type = PT_LOAD; 1196 phdr->p_flags = PF_R|PF_W|PF_X; 1197 phdr->p_vaddr = (Elf64_Addr)_text; 1198 phdr->p_filesz = phdr->p_memsz = _end - _text; 1199 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); 1200 ehdr->e_phnum++; 1201 phdr++; 1202 } 1203 1204 /* Go through all the ranges in mem->ranges[] and prepare phdr */ 1205 for (i = 0; i < mem->nr_ranges; i++) { 1206 mstart = mem->ranges[i].start; 1207 mend = mem->ranges[i].end; 1208 1209 phdr->p_type = PT_LOAD; 1210 phdr->p_flags = PF_R|PF_W|PF_X; 1211 phdr->p_offset = mstart; 1212 1213 phdr->p_paddr = mstart; 1214 phdr->p_vaddr = (unsigned long long) __va(mstart); 1215 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; 1216 phdr->p_align = 0; 1217 ehdr->e_phnum++; 1218 phdr++; 1219 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", 1220 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, 1221 ehdr->e_phnum, phdr->p_offset); 1222 } 1223 1224 *addr = buf; 1225 *sz = elf_sz; 1226 return 0; 1227 } 1228