kernel/kexec_file.c

   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         ret = machine_kexec_prepare(image);
 380         if (ret)
 381                 goto out;
 382
 383         /*
 384          * Some architecture(like S390) may touch the crash memory before
 385          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 386          */
 387         ret = kimage_crash_copy_vmcoreinfo(image);
 388         if (ret)
 389                 goto out;
 390
 391         ret = kexec_calculate_store_digests(image);
 392         if (ret)
 393                 goto out;
 394
 395         kexec_dprintk("nr_segments = %lu\n", image->nr_segments);
 396         for (i = 0; i < image->nr_segments; i++) {
 397                 struct kexec_segment *ksegment;
 398
 399                 ksegment = &image->segment[i];
 400                 kexec_dprintk("segment[%d]: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 401                               i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 402                               ksegment->memsz);
 403
 404                 ret = kimage_load_segment(image, &image->segment[i]);
 405                 if (ret)
 406                         goto out;
 407         }
 408
 409         kimage_terminate(image);
 410
 411         ret = machine_kexec_post_load(image);
 412         if (ret)
 413                 goto out;
 414
 415         kexec_dprintk("kexec_file_load: type:%u, start:0x%lx head:0x%lx flags:0x%lx\n",
 416                       image->type, image->start, image->head, flags);
 417         /*
 418          * Free up any temporary buffers allocated which are not needed
 419          * after image has been loaded
 420          */
 421         kimage_file_post_load_cleanup(image);
 422 exchange:
 423         image = xchg(dest_image, image);
 424 out:
 425 #ifdef CONFIG_CRASH_DUMP
 426         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 427                 arch_kexec_protect_crashkres();
 428 #endif
 429
 430         kexec_unlock();
 431         kimage_free(image);
 432         return ret;
 433 }
 434
 435 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 436                                     struct kexec_buf *kbuf)
 437 {
 438         struct kimage *image = kbuf->image;
 439         unsigned long temp_start, temp_end;
 440
 441         temp_end = min(end, kbuf->buf_max);
 442         temp_start = temp_end - kbuf->memsz + 1;
 443
 444         do {
 445                 /* align down start */
 446                 temp_start = ALIGN_DOWN(temp_start, kbuf->buf_align);
 447
 448                 if (temp_start < start || temp_start < kbuf->buf_min)
 449                         return 0;
 450
 451                 temp_end = temp_start + kbuf->memsz - 1;
 452
 453                 /*
 454                  * Make sure this does not conflict with any of existing
 455                  * segments
 456                  */
 457                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 458                         temp_start = temp_start - PAGE_SIZE;
 459                         continue;
 460                 }
 461
 462                 /* We found a suitable memory range */
 463                 break;
 464         } while (1);
 465
 466         /* If we are here, we found a suitable memory range */
 467         kbuf->mem = temp_start;
 468
 469         /* Success, stop navigating through remaining System RAM ranges */
 470         return 1;
 471 }
 472
 473 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 474                                      struct kexec_buf *kbuf)
 475 {
 476         struct kimage *image = kbuf->image;
 477         unsigned long temp_start, temp_end;
 478
 479         temp_start = max(start, kbuf->buf_min);
 480
 481         do {
 482                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 483                 temp_end = temp_start + kbuf->memsz - 1;
 484
 485                 if (temp_end > end || temp_end > kbuf->buf_max)
 486                         return 0;
 487                 /*
 488                  * Make sure this does not conflict with any of existing
 489                  * segments
 490                  */
 491                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 492                         temp_start = temp_start + PAGE_SIZE;
 493                         continue;
 494                 }
 495
 496                 /* We found a suitable memory range */
 497                 break;
 498         } while (1);
 499
 500         /* If we are here, we found a suitable memory range */
 501         kbuf->mem = temp_start;
 502
 503         /* Success, stop navigating through remaining System RAM ranges */
 504         return 1;
 505 }
 506
 507 static int locate_mem_hole_callback(struct resource *res, void *arg)
 508 {
 509         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 510         u64 start = res->start, end = res->end;
 511         unsigned long sz = end - start + 1;
 512
 513         /* Returning 0 will take to next memory range */
 514
 515         /* Don't use memory that will be detected and handled by a driver. */
 516         if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
 517                 return 0;
 518
 519         if (sz < kbuf->memsz)
 520                 return 0;
 521
 522         if (end < kbuf->buf_min || start > kbuf->buf_max)
 523                 return 0;
 524
 525         /*
 526          * Allocate memory top down with-in ram range. Otherwise bottom up
 527          * allocation.
 528          */
 529         if (kbuf->top_down)
 530                 return locate_mem_hole_top_down(start, end, kbuf);
 531         return locate_mem_hole_bottom_up(start, end, kbuf);
 532 }
 533
 534 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
 535 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 536                                int (*func)(struct resource *, void *))
 537 {
 538         int ret = 0;
 539         u64 i;
 540         phys_addr_t mstart, mend;
 541         struct resource res = { };
 542
 543 #ifdef CONFIG_CRASH_DUMP
 544         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 545                 return func(&crashk_res, kbuf);
 546 #endif
 547
 548         /*
 549          * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
 550          * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
 551          * locate_mem_hole_callback().
 552          */
 553         if (kbuf->top_down) {
 554                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 555                                                 &mstart, &mend, NULL) {
 556                         /*
 557                          * In memblock, end points to the first byte after the
 558                          * range while in kexec, end points to the last byte
 559                          * in the range.
 560                          */
 561                         res.start = mstart;
 562                         res.end = mend - 1;
 563                         ret = func(&res, kbuf);
 564                         if (ret)
 565                                 break;
 566                 }
 567         } else {
 568                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 569                                         &mstart, &mend, NULL) {
 570                         /*
 571                          * In memblock, end points to the first byte after the
 572                          * range while in kexec, end points to the last byte
 573                          * in the range.
 574                          */
 575                         res.start = mstart;
 576                         res.end = mend - 1;
 577                         ret = func(&res, kbuf);
 578                         if (ret)
 579                                 break;
 580                 }
 581         }
 582
 583         return ret;
 584 }
 585 #else
 586 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 587                                int (*func)(struct resource *, void *))
 588 {
 589         return 0;
 590 }
 591 #endif
 592
 593 /**
 594  * kexec_walk_resources - call func(data) on free memory regions
 595  * @kbuf:       Context info for the search. Also passed to @func.
 596  * @func:       Function to call for each memory region.
 597  *
 598  * Return: The memory walk will stop when func returns a non-zero value
 599  * and that value will be returned. If all free regions are visited without
 600  * func returning non-zero, then zero will be returned.
 601  */
 602 static int kexec_walk_resources(struct kexec_buf *kbuf,
 603                                 int (*func)(struct resource *, void *))
 604 {
 605 #ifdef CONFIG_CRASH_DUMP
 606         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 607                 return walk_iomem_res_desc(crashk_res.desc,
 608                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 609                                            crashk_res.start, crashk_res.end,
 610                                            kbuf, func);
 611 #endif
 612         if (kbuf->top_down)
 613                 return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func);
 614         else
 615                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 616 }
 617
 618 /**
 619  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 620  * @kbuf:       Parameters for the memory search.
 621  *
 622  * On success, kbuf->mem will have the start address of the memory region found.
 623  *
 624  * Return: 0 on success, negative errno on error.
 625  */
 626 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 627 {
 628         int ret;
 629
 630         /* Arch knows where to place */
 631         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
 632                 return 0;
 633
 634         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 635                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
 636         else
 637                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
 638
 639         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 640 }
 641
 642 /**
 643  * kexec_add_buffer - place a buffer in a kexec segment
 644  * @kbuf:       Buffer contents and memory parameters.
 645  *
 646  * This function assumes that kexec_lock is held.
 647  * On successful return, @kbuf->mem will have the physical address of
 648  * the buffer in memory.
 649  *
 650  * Return: 0 on success, negative errno on error.
 651  */
 652 int kexec_add_buffer(struct kexec_buf *kbuf)
 653 {
 654         struct kexec_segment *ksegment;
 655         int ret;
 656
 657         /* Currently adding segment this way is allowed only in file mode */
 658         if (!kbuf->image->file_mode)
 659                 return -EINVAL;
 660
 661         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 662                 return -EINVAL;
 663
 664         /*
 665          * Make sure we are not trying to add buffer after allocating
 666          * control pages. All segments need to be placed first before
 667          * any control pages are allocated. As control page allocation
 668          * logic goes through list of segments to make sure there are
 669          * no destination overlaps.
 670          */
 671         if (!list_empty(&kbuf->image->control_pages)) {
 672                 WARN_ON(1);
 673                 return -EINVAL;
 674         }
 675
 676         /* Ensure minimum alignment needed for segments. */
 677         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 678         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 679
 680         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 681         ret = arch_kexec_locate_mem_hole(kbuf);
 682         if (ret)
 683                 return ret;
 684
 685         /* Found a suitable memory range */
 686         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 687         ksegment->kbuf = kbuf->buffer;
 688         ksegment->bufsz = kbuf->bufsz;
 689         ksegment->mem = kbuf->mem;
 690         ksegment->memsz = kbuf->memsz;
 691         kbuf->image->nr_segments++;
 692         return 0;
 693 }
 694
 695 /* Calculate and store the digest of segments */
 696 static int kexec_calculate_store_digests(struct kimage *image)
 697 {
 698         struct crypto_shash *tfm;
 699         struct shash_desc *desc;
 700         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 701         size_t desc_size, nullsz;
 702         char *digest;
 703         void *zero_buf;
 704         struct kexec_sha_region *sha_regions;
 705         struct purgatory_info *pi = &image->purgatory_info;
 706
 707         if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY))
 708                 return 0;
 709
 710         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 711         zero_buf_sz = PAGE_SIZE;
 712
 713         tfm = crypto_alloc_shash("sha256", 0, 0);
 714         if (IS_ERR(tfm)) {
 715                 ret = PTR_ERR(tfm);
 716                 goto out;
 717         }
 718
 719         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 720         desc = kzalloc(desc_size, GFP_KERNEL);
 721         if (!desc) {
 722                 ret = -ENOMEM;
 723                 goto out_free_tfm;
 724         }
 725
 726         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 727         sha_regions = vzalloc(sha_region_sz);
 728         if (!sha_regions) {
 729                 ret = -ENOMEM;
 730                 goto out_free_desc;
 731         }
 732
 733         desc->tfm   = tfm;
 734
 735         ret = crypto_shash_init(desc);
 736         if (ret < 0)
 737                 goto out_free_sha_regions;
 738
 739         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 740         if (!digest) {
 741                 ret = -ENOMEM;
 742                 goto out_free_sha_regions;
 743         }
 744
 745         for (j = i = 0; i < image->nr_segments; i++) {
 746                 struct kexec_segment *ksegment;
 747
 748 #ifdef CONFIG_CRASH_HOTPLUG
 749                 /* Exclude elfcorehdr segment to allow future changes via hotplug */
 750                 if (j == image->elfcorehdr_index)
 751                         continue;
 752 #endif
 753
 754                 ksegment = &image->segment[i];
 755                 /*
 756                  * Skip purgatory as it will be modified once we put digest
 757                  * info in purgatory.
 758                  */
 759                 if (ksegment->kbuf == pi->purgatory_buf)
 760                         continue;
 761
 762                 ret = crypto_shash_update(desc, ksegment->kbuf,
 763                                           ksegment->bufsz);
 764                 if (ret)
 765                         break;
 766
 767                 /*
 768                  * Assume rest of the buffer is filled with zero and
 769                  * update digest accordingly.
 770                  */
 771                 nullsz = ksegment->memsz - ksegment->bufsz;
 772                 while (nullsz) {
 773                         unsigned long bytes = nullsz;
 774
 775                         if (bytes > zero_buf_sz)
 776                                 bytes = zero_buf_sz;
 777                         ret = crypto_shash_update(desc, zero_buf, bytes);
 778                         if (ret)
 779                                 break;
 780                         nullsz -= bytes;
 781                 }
 782
 783                 if (ret)
 784                         break;
 785
 786                 sha_regions[j].start = ksegment->mem;
 787                 sha_regions[j].len = ksegment->memsz;
 788                 j++;
 789         }
 790
 791         if (!ret) {
 792                 ret = crypto_shash_final(desc, digest);
 793                 if (ret)
 794                         goto out_free_digest;
 795                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 796                                                      sha_regions, sha_region_sz, 0);
 797                 if (ret)
 798                         goto out_free_digest;
 799
 800                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 801                                                      digest, SHA256_DIGEST_SIZE, 0);
 802                 if (ret)
 803                         goto out_free_digest;
 804         }
 805
 806 out_free_digest:
 807         kfree(digest);
 808 out_free_sha_regions:
 809         vfree(sha_regions);
 810 out_free_desc:
 811         kfree(desc);
 812 out_free_tfm:
 813         kfree(tfm);
 814 out:
 815         return ret;
 816 }
 817
 818 #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY
 819 /*
 820  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 821  * @pi:         Purgatory to be loaded.
 822  * @kbuf:       Buffer to setup.
 823  *
 824  * Allocates the memory needed for the buffer. Caller is responsible to free
 825  * the memory after use.
 826  *
 827  * Return: 0 on success, negative errno on error.
 828  */
 829 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 830                                       struct kexec_buf *kbuf)
 831 {
 832         const Elf_Shdr *sechdrs;
 833         unsigned long bss_align;
 834         unsigned long bss_sz;
 835         unsigned long align;
 836         int i, ret;
 837
 838         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 839         kbuf->buf_align = bss_align = 1;
 840         kbuf->bufsz = bss_sz = 0;
 841
 842         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 843                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 844                         continue;
 845
 846                 align = sechdrs[i].sh_addralign;
 847                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 848                         if (kbuf->buf_align < align)
 849                                 kbuf->buf_align = align;
 850                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 851                         kbuf->bufsz += sechdrs[i].sh_size;
 852                 } else {
 853                         if (bss_align < align)
 854                                 bss_align = align;
 855                         bss_sz = ALIGN(bss_sz, align);
 856                         bss_sz += sechdrs[i].sh_size;
 857                 }
 858         }
 859         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 860         kbuf->memsz = kbuf->bufsz + bss_sz;
 861         if (kbuf->buf_align < bss_align)
 862                 kbuf->buf_align = bss_align;
 863
 864         kbuf->buffer = vzalloc(kbuf->bufsz);
 865         if (!kbuf->buffer)
 866                 return -ENOMEM;
 867         pi->purgatory_buf = kbuf->buffer;
 868
 869         ret = kexec_add_buffer(kbuf);
 870         if (ret)
 871                 goto out;
 872
 873         return 0;
 874 out:
 875         vfree(pi->purgatory_buf);
 876         pi->purgatory_buf = NULL;
 877         return ret;
 878 }
 879
 880 /*
 881  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 882  * @pi:         Purgatory to be loaded.
 883  * @kbuf:       Buffer prepared to store purgatory.
 884  *
 885  * Allocates the memory needed for the buffer. Caller is responsible to free
 886  * the memory after use.
 887  *
 888  * Return: 0 on success, negative errno on error.
 889  */
 890 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 891                                          struct kexec_buf *kbuf)
 892 {
 893         unsigned long bss_addr;
 894         unsigned long offset;
 895         size_t sechdrs_size;
 896         Elf_Shdr *sechdrs;
 897         int i;
 898
 899         /*
 900          * The section headers in kexec_purgatory are read-only. In order to
 901          * have them modifiable make a temporary copy.
 902          */
 903         sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum);
 904         sechdrs = vzalloc(sechdrs_size);
 905         if (!sechdrs)
 906                 return -ENOMEM;
 907         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size);
 908         pi->sechdrs = sechdrs;
 909
 910         offset = 0;
 911         bss_addr = kbuf->mem + kbuf->bufsz;
 912         kbuf->image->start = pi->ehdr->e_entry;
 913
 914         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 915                 unsigned long align;
 916                 void *src, *dst;
 917
 918                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 919                         continue;
 920
 921                 align = sechdrs[i].sh_addralign;
 922                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 923                         bss_addr = ALIGN(bss_addr, align);
 924                         sechdrs[i].sh_addr = bss_addr;
 925                         bss_addr += sechdrs[i].sh_size;
 926                         continue;
 927                 }
 928
 929                 offset = ALIGN(offset, align);
 930
 931                 /*
 932                  * Check if the segment contains the entry point, if so,
 933                  * calculate the value of image->start based on it.
 934                  * If the compiler has produced more than one .text section
 935                  * (Eg: .text.hot), they are generally after the main .text
 936                  * section, and they shall not be used to calculate
 937                  * image->start. So do not re-calculate image->start if it
 938                  * is not set to the initial value, and warn the user so they
 939                  * have a chance to fix their purgatory's linker script.
 940                  */
 941                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 942                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 943                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 944                                          + sechdrs[i].sh_size) &&
 945                     !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
 946                         kbuf->image->start -= sechdrs[i].sh_addr;
 947                         kbuf->image->start += kbuf->mem + offset;
 948                 }
 949
 950                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 951                 dst = pi->purgatory_buf + offset;
 952                 memcpy(dst, src, sechdrs[i].sh_size);
 953
 954                 sechdrs[i].sh_addr = kbuf->mem + offset;
 955                 sechdrs[i].sh_offset = offset;
 956                 offset += sechdrs[i].sh_size;
 957         }
 958
 959         return 0;
 960 }
 961
 962 static int kexec_apply_relocations(struct kimage *image)
 963 {
 964         int i, ret;
 965         struct purgatory_info *pi = &image->purgatory_info;
 966         const Elf_Shdr *sechdrs;
 967
 968         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 969
 970         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 971                 const Elf_Shdr *relsec;
 972                 const Elf_Shdr *symtab;
 973                 Elf_Shdr *section;
 974
 975                 relsec = sechdrs + i;
 976
 977                 if (relsec->sh_type != SHT_RELA &&
 978                     relsec->sh_type != SHT_REL)
 979                         continue;
 980
 981                 /*
 982                  * For section of type SHT_RELA/SHT_REL,
 983                  * ->sh_link contains section header index of associated
 984                  * symbol table. And ->sh_info contains section header
 985                  * index of section to which relocations apply.
 986                  */
 987                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 988                     relsec->sh_link >= pi->ehdr->e_shnum)
 989                         return -ENOEXEC;
 990
 991                 section = pi->sechdrs + relsec->sh_info;
 992                 symtab = sechdrs + relsec->sh_link;
 993
 994                 if (!(section->sh_flags & SHF_ALLOC))
 995                         continue;
 996
 997                 /*
 998                  * symtab->sh_link contain section header index of associated
 999                  * string table.
1000                  */
1001                 if (symtab->sh_link >= pi->ehdr->e_shnum)
1002                         /* Invalid section number? */
1003                         continue;
1004
1005                 /*
1006                  * Respective architecture needs to provide support for applying
1007                  * relocations of type SHT_RELA/SHT_REL.
1008                  */
1009                 if (relsec->sh_type == SHT_RELA)
1010                         ret = arch_kexec_apply_relocations_add(pi, section,
1011                                                                relsec, symtab);
1012                 else if (relsec->sh_type == SHT_REL)
1013                         ret = arch_kexec_apply_relocations(pi, section,
1014                                                            relsec, symtab);
1015                 if (ret)
1016                         return ret;
1017         }
1018
1019         return 0;
1020 }
1021
1022 /*
1023  * kexec_load_purgatory - Load and relocate the purgatory object.
1024  * @image:      Image to add the purgatory to.
1025  * @kbuf:       Memory parameters to use.
1026  *
1027  * Allocates the memory needed for image->purgatory_info.sechdrs and
1028  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1029  * to free the memory after use.
1030  *
1031  * Return: 0 on success, negative errno on error.
1032  */
1033 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1034 {
1035         struct purgatory_info *pi = &image->purgatory_info;
1036         int ret;
1037
1038         if (kexec_purgatory_size <= 0)
1039                 return -EINVAL;
1040
1041         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1042
1043         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1044         if (ret)
1045                 return ret;
1046
1047         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1048         if (ret)
1049                 goto out_free_kbuf;
1050
1051         ret = kexec_apply_relocations(image);
1052         if (ret)
1053                 goto out;
1054
1055         return 0;
1056 out:
1057         vfree(pi->sechdrs);
1058         pi->sechdrs = NULL;
1059 out_free_kbuf:
1060         vfree(pi->purgatory_buf);
1061         pi->purgatory_buf = NULL;
1062         return ret;
1063 }
1064
1065 /*
1066  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1067  * @pi:         Purgatory to search in.
1068  * @name:       Name of the symbol.
1069  *
1070  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1071  */
1072 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1073                                                   const char *name)
1074 {
1075         const Elf_Shdr *sechdrs;
1076         const Elf_Ehdr *ehdr;
1077         const Elf_Sym *syms;
1078         const char *strtab;
1079         int i, k;
1080
1081         if (!pi->ehdr)
1082                 return NULL;
1083
1084         ehdr = pi->ehdr;
1085         sechdrs = (void *)ehdr + ehdr->e_shoff;
1086
1087         for (i = 0; i < ehdr->e_shnum; i++) {
1088                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1089                         continue;
1090
1091                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1092                         /* Invalid strtab section number */
1093                         continue;
1094                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1095                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1096
1097                 /* Go through symbols for a match */
1098                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1099                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1100                                 continue;
1101
1102                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1103                                 continue;
1104
1105                         if (syms[k].st_shndx == SHN_UNDEF ||
1106                             syms[k].st_shndx >= ehdr->e_shnum) {
1107                                 pr_debug("Symbol: %s has bad section index %d.\n",
1108                                                 name, syms[k].st_shndx);
1109                                 return NULL;
1110                         }
1111
1112                         /* Found the symbol we are looking for */
1113                         return &syms[k];
1114                 }
1115         }
1116
1117         return NULL;
1118 }
1119
1120 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1121 {
1122         struct purgatory_info *pi = &image->purgatory_info;
1123         const Elf_Sym *sym;
1124         Elf_Shdr *sechdr;
1125
1126         sym = kexec_purgatory_find_symbol(pi, name);
1127         if (!sym)
1128                 return ERR_PTR(-EINVAL);
1129
1130         sechdr = &pi->sechdrs[sym->st_shndx];
1131
1132         /*
1133          * Returns the address where symbol will finally be loaded after
1134          * kexec_load_segment()
1135          */
1136         return (void *)(sechdr->sh_addr + sym->st_value);
1137 }
1138
1139 /*
1140  * Get or set value of a symbol. If "get_value" is true, symbol value is
1141  * returned in buf otherwise symbol value is set based on value in buf.
1142  */
1143 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1144                                    void *buf, unsigned int size, bool get_value)
1145 {
1146         struct purgatory_info *pi = &image->purgatory_info;
1147         const Elf_Sym *sym;
1148         Elf_Shdr *sec;
1149         char *sym_buf;
1150
1151         sym = kexec_purgatory_find_symbol(pi, name);
1152         if (!sym)
1153                 return -EINVAL;
1154
1155         if (sym->st_size != size) {
1156                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1157                        name, (unsigned long)sym->st_size, size);
1158                 return -EINVAL;
1159         }
1160
1161         sec = pi->sechdrs + sym->st_shndx;
1162
1163         if (sec->sh_type == SHT_NOBITS) {
1164                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1165                        get_value ? "get" : "set");
1166                 return -EINVAL;
1167         }
1168
1169         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1170
1171         if (get_value)
1172                 memcpy((void *)buf, sym_buf, size);
1173         else
1174                 memcpy((void *)sym_buf, buf, size);
1175
1176         return 0;
1177 }
1178 #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */