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
 208         /* Call arch image probe handlers */
 209         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 210                                             image->kernel_buf_len);
 211         if (ret)
 212                 goto out;
 213
 214 #ifdef CONFIG_KEXEC_SIG
 215         ret = kimage_validate_signature(image);
 216
 217         if (ret)
 218                 goto out;
 219 #endif
 220         /* It is possible that there no initramfs is being loaded */
 221         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 222                 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
 223                                                KEXEC_FILE_SIZE_MAX, NULL,
 224                                                READING_KEXEC_INITRAMFS);
 225                 if (ret < 0)
 226                         goto out;
 227                 image->initrd_buf_len = ret;
 228                 ret = 0;
 229         }
 230
 231         if (cmdline_len) {
 232                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 233                 if (IS_ERR(image->cmdline_buf)) {
 234                         ret = PTR_ERR(image->cmdline_buf);
 235                         image->cmdline_buf = NULL;
 236                         goto out;
 237                 }
 238
 239                 image->cmdline_buf_len = cmdline_len;
 240
 241                 /* command line should be a string with last byte null */
 242                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 243                         ret = -EINVAL;
 244                         goto out;
 245                 }
 246
 247                 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
 248                                   image->cmdline_buf_len - 1);
 249         }
 250
 251         /* IMA needs to pass the measurement list to the next kernel. */
 252         ima_add_kexec_buffer(image);
 253
 254         /* Call image load handler */
 255         ldata = kexec_image_load_default(image);
 256
 257         if (IS_ERR(ldata)) {
 258                 ret = PTR_ERR(ldata);
 259                 goto out;
 260         }
 261
 262         image->image_loader_data = ldata;
 263 out:
 264         /* In case of error, free up all allocated memory in this function */
 265         if (ret)
 266                 kimage_file_post_load_cleanup(image);
 267         return ret;
 268 }
 269
 270 static int
 271 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 272                        int initrd_fd, const char __user *cmdline_ptr,
 273                        unsigned long cmdline_len, unsigned long flags)
 274 {
 275         int ret;
 276         struct kimage *image;
 277         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 278
 279         image = do_kimage_alloc_init();
 280         if (!image)
 281                 return -ENOMEM;
 282
 283         kexec_file_dbg_print = !!(flags & KEXEC_FILE_DEBUG);
 284         image->file_mode = 1;
 285
 286         if (kexec_on_panic) {
 287                 /* Enable special crash kernel control page alloc policy. */
 288                 image->control_page = crashk_res.start;
 289                 image->type = KEXEC_TYPE_CRASH;
 290         }
 291
 292         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 293                                            cmdline_ptr, cmdline_len, flags);
 294         if (ret)
 295                 goto out_free_image;
 296
 297         ret = sanity_check_segment_list(image);
 298         if (ret)
 299                 goto out_free_post_load_bufs;
 300
 301         ret = -ENOMEM;
 302         image->control_code_page = kimage_alloc_control_pages(image,
 303                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 304         if (!image->control_code_page) {
 305                 pr_err("Could not allocate control_code_buffer\n");
 306                 goto out_free_post_load_bufs;
 307         }
 308
 309         if (!kexec_on_panic) {
 310                 image->swap_page = kimage_alloc_control_pages(image, 0);
 311                 if (!image->swap_page) {
 312                         pr_err("Could not allocate swap buffer\n");
 313                         goto out_free_control_pages;
 314                 }
 315         }
 316
 317         *rimage = image;
 318         return 0;
 319 out_free_control_pages:
 320         kimage_free_page_list(&image->control_pages);
 321 out_free_post_load_bufs:
 322         kimage_file_post_load_cleanup(image);
 323 out_free_image:
 324         kfree(image);
 325         return ret;
 326 }
 327
 328 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 329                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 330                 unsigned long, flags)
 331 {
 332         int image_type = (flags & KEXEC_FILE_ON_CRASH) ?
 333                          KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
 334         struct kimage **dest_image, *image;
 335         int ret = 0, i;
 336
 337         /* We only trust the superuser with rebooting the system. */
 338         if (!kexec_load_permitted(image_type))
 339                 return -EPERM;
 340
 341         /* Make sure we have a legal set of flags */
 342         if (flags != (flags & KEXEC_FILE_FLAGS))
 343                 return -EINVAL;
 344
 345         image = NULL;
 346
 347         if (!kexec_trylock())
 348                 return -EBUSY;
 349
 350         if (image_type == KEXEC_TYPE_CRASH) {
 351                 dest_image = &kexec_crash_image;
 352                 if (kexec_crash_image)
 353                         arch_kexec_unprotect_crashkres();
 354         } else {
 355                 dest_image = &kexec_image;
 356         }
 357
 358         if (flags & KEXEC_FILE_UNLOAD)
 359                 goto exchange;
 360
 361         /*
 362          * In case of crash, new kernel gets loaded in reserved region. It is
 363          * same memory where old crash kernel might be loaded. Free any
 364          * current crash dump kernel before we corrupt it.
 365          */
 366         if (flags & KEXEC_FILE_ON_CRASH)
 367                 kimage_free(xchg(&kexec_crash_image, NULL));
 368
 369         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 370                                      cmdline_len, flags);
 371         if (ret)
 372                 goto out;
 373
 374         ret = machine_kexec_prepare(image);
 375         if (ret)
 376                 goto out;
 377
 378         /*
 379          * Some architecture(like S390) may touch the crash memory before
 380          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 381          */
 382         ret = kimage_crash_copy_vmcoreinfo(image);
 383         if (ret)
 384                 goto out;
 385
 386         ret = kexec_calculate_store_digests(image);
 387         if (ret)
 388                 goto out;
 389
 390         for (i = 0; i < image->nr_segments; i++) {
 391                 struct kexec_segment *ksegment;
 392
 393                 ksegment = &image->segment[i];
 394                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 395                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 396                          ksegment->memsz);
 397
 398                 ret = kimage_load_segment(image, &image->segment[i]);
 399                 if (ret)
 400                         goto out;
 401         }
 402
 403         kimage_terminate(image);
 404
 405         ret = machine_kexec_post_load(image);
 406         if (ret)
 407                 goto out;
 408
 409         /*
 410          * Free up any temporary buffers allocated which are not needed
 411          * after image has been loaded
 412          */
 413         kimage_file_post_load_cleanup(image);
 414 exchange:
 415         image = xchg(dest_image, image);
 416 out:
 417         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 418                 arch_kexec_protect_crashkres();
 419
 420         kexec_unlock();
 421         kimage_free(image);
 422         return ret;
 423 }
 424
 425 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 426                                     struct kexec_buf *kbuf)
 427 {
 428         struct kimage *image = kbuf->image;
 429         unsigned long temp_start, temp_end;
 430
 431         temp_end = min(end, kbuf->buf_max);
 432         temp_start = temp_end - kbuf->memsz;
 433
 434         do {
 435                 /* align down start */
 436                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 437
 438                 if (temp_start < start || temp_start < kbuf->buf_min)
 439                         return 0;
 440
 441                 temp_end = temp_start + kbuf->memsz - 1;
 442
 443                 /*
 444                  * Make sure this does not conflict with any of existing
 445                  * segments
 446                  */
 447                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 448                         temp_start = temp_start - PAGE_SIZE;
 449                         continue;
 450                 }
 451
 452                 /* We found a suitable memory range */
 453                 break;
 454         } while (1);
 455
 456         /* If we are here, we found a suitable memory range */
 457         kbuf->mem = temp_start;
 458
 459         /* Success, stop navigating through remaining System RAM ranges */
 460         return 1;
 461 }
 462
 463 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 464                                      struct kexec_buf *kbuf)
 465 {
 466         struct kimage *image = kbuf->image;
 467         unsigned long temp_start, temp_end;
 468
 469         temp_start = max(start, kbuf->buf_min);
 470
 471         do {
 472                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 473                 temp_end = temp_start + kbuf->memsz - 1;
 474
 475                 if (temp_end > end || temp_end > kbuf->buf_max)
 476                         return 0;
 477                 /*
 478                  * Make sure this does not conflict with any of existing
 479                  * segments
 480                  */
 481                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 482                         temp_start = temp_start + PAGE_SIZE;
 483                         continue;
 484                 }
 485
 486                 /* We found a suitable memory range */
 487                 break;
 488         } while (1);
 489
 490         /* If we are here, we found a suitable memory range */
 491         kbuf->mem = temp_start;
 492
 493         /* Success, stop navigating through remaining System RAM ranges */
 494         return 1;
 495 }
 496
 497 static int locate_mem_hole_callback(struct resource *res, void *arg)
 498 {
 499         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 500         u64 start = res->start, end = res->end;
 501         unsigned long sz = end - start + 1;
 502
 503         /* Returning 0 will take to next memory range */
 504
 505         /* Don't use memory that will be detected and handled by a driver. */
 506         if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
 507                 return 0;
 508
 509         if (sz < kbuf->memsz)
 510                 return 0;
 511
 512         if (end < kbuf->buf_min || start > kbuf->buf_max)
 513                 return 0;
 514
 515         /*
 516          * Allocate memory top down with-in ram range. Otherwise bottom up
 517          * allocation.
 518          */
 519         if (kbuf->top_down)
 520                 return locate_mem_hole_top_down(start, end, kbuf);
 521         return locate_mem_hole_bottom_up(start, end, kbuf);
 522 }
 523
 524 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
 525 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 526                                int (*func)(struct resource *, void *))
 527 {
 528         int ret = 0;
 529         u64 i;
 530         phys_addr_t mstart, mend;
 531         struct resource res = { };
 532
 533         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 534                 return func(&crashk_res, kbuf);
 535
 536         /*
 537          * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
 538          * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
 539          * locate_mem_hole_callback().
 540          */
 541         if (kbuf->top_down) {
 542                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 543                                                 &mstart, &mend, NULL) {
 544                         /*
 545                          * In memblock, end points to the first byte after the
 546                          * range while in kexec, end points to the last byte
 547                          * in the range.
 548                          */
 549                         res.start = mstart;
 550                         res.end = mend - 1;
 551                         ret = func(&res, kbuf);
 552                         if (ret)
 553                                 break;
 554                 }
 555         } else {
 556                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 557                                         &mstart, &mend, NULL) {
 558                         /*
 559                          * In memblock, end points to the first byte after the
 560                          * range while in kexec, end points to the last byte
 561                          * in the range.
 562                          */
 563                         res.start = mstart;
 564                         res.end = mend - 1;
 565                         ret = func(&res, kbuf);
 566                         if (ret)
 567                                 break;
 568                 }
 569         }
 570
 571         return ret;
 572 }
 573 #else
 574 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 575                                int (*func)(struct resource *, void *))
 576 {
 577         return 0;
 578 }
 579 #endif
 580
 581 /**
 582  * kexec_walk_resources - call func(data) on free memory regions
 583  * @kbuf:       Context info for the search. Also passed to @func.
 584  * @func:       Function to call for each memory region.
 585  *
 586  * Return: The memory walk will stop when func returns a non-zero value
 587  * and that value will be returned. If all free regions are visited without
 588  * func returning non-zero, then zero will be returned.
 589  */
 590 static int kexec_walk_resources(struct kexec_buf *kbuf,
 591                                 int (*func)(struct resource *, void *))
 592 {
 593         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 594                 return walk_iomem_res_desc(crashk_res.desc,
 595                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 596                                            crashk_res.start, crashk_res.end,
 597                                            kbuf, func);
 598         else if (kbuf->top_down)
 599                 return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func);
 600         else
 601                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 602 }
 603
 604 /**
 605  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 606  * @kbuf:       Parameters for the memory search.
 607  *
 608  * On success, kbuf->mem will have the start address of the memory region found.
 609  *
 610  * Return: 0 on success, negative errno on error.
 611  */
 612 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 613 {
 614         int ret;
 615
 616         /* Arch knows where to place */
 617         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
 618                 return 0;
 619
 620         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 621                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
 622         else
 623                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
 624
 625         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 626 }
 627
 628 /**
 629  * kexec_add_buffer - place a buffer in a kexec segment
 630  * @kbuf:       Buffer contents and memory parameters.
 631  *
 632  * This function assumes that kexec_lock is held.
 633  * On successful return, @kbuf->mem will have the physical address of
 634  * the buffer in memory.
 635  *
 636  * Return: 0 on success, negative errno on error.
 637  */
 638 int kexec_add_buffer(struct kexec_buf *kbuf)
 639 {
 640         struct kexec_segment *ksegment;
 641         int ret;
 642
 643         /* Currently adding segment this way is allowed only in file mode */
 644         if (!kbuf->image->file_mode)
 645                 return -EINVAL;
 646
 647         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 648                 return -EINVAL;
 649
 650         /*
 651          * Make sure we are not trying to add buffer after allocating
 652          * control pages. All segments need to be placed first before
 653          * any control pages are allocated. As control page allocation
 654          * logic goes through list of segments to make sure there are
 655          * no destination overlaps.
 656          */
 657         if (!list_empty(&kbuf->image->control_pages)) {
 658                 WARN_ON(1);
 659                 return -EINVAL;
 660         }
 661
 662         /* Ensure minimum alignment needed for segments. */
 663         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 664         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 665
 666         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 667         ret = arch_kexec_locate_mem_hole(kbuf);
 668         if (ret)
 669                 return ret;
 670
 671         /* Found a suitable memory range */
 672         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 673         ksegment->kbuf = kbuf->buffer;
 674         ksegment->bufsz = kbuf->bufsz;
 675         ksegment->mem = kbuf->mem;
 676         ksegment->memsz = kbuf->memsz;
 677         kbuf->image->nr_segments++;
 678         return 0;
 679 }
 680
 681 /* Calculate and store the digest of segments */
 682 static int kexec_calculate_store_digests(struct kimage *image)
 683 {
 684         struct crypto_shash *tfm;
 685         struct shash_desc *desc;
 686         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 687         size_t desc_size, nullsz;
 688         char *digest;
 689         void *zero_buf;
 690         struct kexec_sha_region *sha_regions;
 691         struct purgatory_info *pi = &image->purgatory_info;
 692
 693         if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY))
 694                 return 0;
 695
 696         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 697         zero_buf_sz = PAGE_SIZE;
 698
 699         tfm = crypto_alloc_shash("sha256", 0, 0);
 700         if (IS_ERR(tfm)) {
 701                 ret = PTR_ERR(tfm);
 702                 goto out;
 703         }
 704
 705         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 706         desc = kzalloc(desc_size, GFP_KERNEL);
 707         if (!desc) {
 708                 ret = -ENOMEM;
 709                 goto out_free_tfm;
 710         }
 711
 712         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 713         sha_regions = vzalloc(sha_region_sz);
 714         if (!sha_regions) {
 715                 ret = -ENOMEM;
 716                 goto out_free_desc;
 717         }
 718
 719         desc->tfm   = tfm;
 720
 721         ret = crypto_shash_init(desc);
 722         if (ret < 0)
 723                 goto out_free_sha_regions;
 724
 725         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 726         if (!digest) {
 727                 ret = -ENOMEM;
 728                 goto out_free_sha_regions;
 729         }
 730
 731         for (j = i = 0; i < image->nr_segments; i++) {
 732                 struct kexec_segment *ksegment;
 733
 734 #ifdef CONFIG_CRASH_HOTPLUG
 735                 /* Exclude elfcorehdr segment to allow future changes via hotplug */
 736                 if (j == image->elfcorehdr_index)
 737                         continue;
 738 #endif
 739
 740                 ksegment = &image->segment[i];
 741                 /*
 742                  * Skip purgatory as it will be modified once we put digest
 743                  * info in purgatory.
 744                  */
 745                 if (ksegment->kbuf == pi->purgatory_buf)
 746                         continue;
 747
 748                 ret = crypto_shash_update(desc, ksegment->kbuf,
 749                                           ksegment->bufsz);
 750                 if (ret)
 751                         break;
 752
 753                 /*
 754                  * Assume rest of the buffer is filled with zero and
 755                  * update digest accordingly.
 756                  */
 757                 nullsz = ksegment->memsz - ksegment->bufsz;
 758                 while (nullsz) {
 759                         unsigned long bytes = nullsz;
 760
 761                         if (bytes > zero_buf_sz)
 762                                 bytes = zero_buf_sz;
 763                         ret = crypto_shash_update(desc, zero_buf, bytes);
 764                         if (ret)
 765                                 break;
 766                         nullsz -= bytes;
 767                 }
 768
 769                 if (ret)
 770                         break;
 771
 772                 sha_regions[j].start = ksegment->mem;
 773                 sha_regions[j].len = ksegment->memsz;
 774                 j++;
 775         }
 776
 777         if (!ret) {
 778                 ret = crypto_shash_final(desc, digest);
 779                 if (ret)
 780                         goto out_free_digest;
 781                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 782                                                      sha_regions, sha_region_sz, 0);
 783                 if (ret)
 784                         goto out_free_digest;
 785
 786                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 787                                                      digest, SHA256_DIGEST_SIZE, 0);
 788                 if (ret)
 789                         goto out_free_digest;
 790         }
 791
 792 out_free_digest:
 793         kfree(digest);
 794 out_free_sha_regions:
 795         vfree(sha_regions);
 796 out_free_desc:
 797         kfree(desc);
 798 out_free_tfm:
 799         kfree(tfm);
 800 out:
 801         return ret;
 802 }
 803
 804 #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY
 805 /*
 806  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 807  * @pi:         Purgatory to be loaded.
 808  * @kbuf:       Buffer to setup.
 809  *
 810  * Allocates the memory needed for the buffer. Caller is responsible to free
 811  * the memory after use.
 812  *
 813  * Return: 0 on success, negative errno on error.
 814  */
 815 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 816                                       struct kexec_buf *kbuf)
 817 {
 818         const Elf_Shdr *sechdrs;
 819         unsigned long bss_align;
 820         unsigned long bss_sz;
 821         unsigned long align;
 822         int i, ret;
 823
 824         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 825         kbuf->buf_align = bss_align = 1;
 826         kbuf->bufsz = bss_sz = 0;
 827
 828         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 829                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 830                         continue;
 831
 832                 align = sechdrs[i].sh_addralign;
 833                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 834                         if (kbuf->buf_align < align)
 835                                 kbuf->buf_align = align;
 836                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 837                         kbuf->bufsz += sechdrs[i].sh_size;
 838                 } else {
 839                         if (bss_align < align)
 840                                 bss_align = align;
 841                         bss_sz = ALIGN(bss_sz, align);
 842                         bss_sz += sechdrs[i].sh_size;
 843                 }
 844         }
 845         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 846         kbuf->memsz = kbuf->bufsz + bss_sz;
 847         if (kbuf->buf_align < bss_align)
 848                 kbuf->buf_align = bss_align;
 849
 850         kbuf->buffer = vzalloc(kbuf->bufsz);
 851         if (!kbuf->buffer)
 852                 return -ENOMEM;
 853         pi->purgatory_buf = kbuf->buffer;
 854
 855         ret = kexec_add_buffer(kbuf);
 856         if (ret)
 857                 goto out;
 858
 859         return 0;
 860 out:
 861         vfree(pi->purgatory_buf);
 862         pi->purgatory_buf = NULL;
 863         return ret;
 864 }
 865
 866 /*
 867  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 868  * @pi:         Purgatory to be loaded.
 869  * @kbuf:       Buffer prepared to store purgatory.
 870  *
 871  * Allocates the memory needed for the buffer. Caller is responsible to free
 872  * the memory after use.
 873  *
 874  * Return: 0 on success, negative errno on error.
 875  */
 876 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 877                                          struct kexec_buf *kbuf)
 878 {
 879         unsigned long bss_addr;
 880         unsigned long offset;
 881         size_t sechdrs_size;
 882         Elf_Shdr *sechdrs;
 883         int i;
 884
 885         /*
 886          * The section headers in kexec_purgatory are read-only. In order to
 887          * have them modifiable make a temporary copy.
 888          */
 889         sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum);
 890         sechdrs = vzalloc(sechdrs_size);
 891         if (!sechdrs)
 892                 return -ENOMEM;
 893         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size);
 894         pi->sechdrs = sechdrs;
 895
 896         offset = 0;
 897         bss_addr = kbuf->mem + kbuf->bufsz;
 898         kbuf->image->start = pi->ehdr->e_entry;
 899
 900         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 901                 unsigned long align;
 902                 void *src, *dst;
 903
 904                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 905                         continue;
 906
 907                 align = sechdrs[i].sh_addralign;
 908                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 909                         bss_addr = ALIGN(bss_addr, align);
 910                         sechdrs[i].sh_addr = bss_addr;
 911                         bss_addr += sechdrs[i].sh_size;
 912                         continue;
 913                 }
 914
 915                 offset = ALIGN(offset, align);
 916
 917                 /*
 918                  * Check if the segment contains the entry point, if so,
 919                  * calculate the value of image->start based on it.
 920                  * If the compiler has produced more than one .text section
 921                  * (Eg: .text.hot), they are generally after the main .text
 922                  * section, and they shall not be used to calculate
 923                  * image->start. So do not re-calculate image->start if it
 924                  * is not set to the initial value, and warn the user so they
 925                  * have a chance to fix their purgatory's linker script.
 926                  */
 927                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 928                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 929                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 930                                          + sechdrs[i].sh_size) &&
 931                     !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
 932                         kbuf->image->start -= sechdrs[i].sh_addr;
 933                         kbuf->image->start += kbuf->mem + offset;
 934                 }
 935
 936                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 937                 dst = pi->purgatory_buf + offset;
 938                 memcpy(dst, src, sechdrs[i].sh_size);
 939
 940                 sechdrs[i].sh_addr = kbuf->mem + offset;
 941                 sechdrs[i].sh_offset = offset;
 942                 offset += sechdrs[i].sh_size;
 943         }
 944
 945         return 0;
 946 }
 947
 948 static int kexec_apply_relocations(struct kimage *image)
 949 {
 950         int i, ret;
 951         struct purgatory_info *pi = &image->purgatory_info;
 952         const Elf_Shdr *sechdrs;
 953
 954         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 955
 956         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 957                 const Elf_Shdr *relsec;
 958                 const Elf_Shdr *symtab;
 959                 Elf_Shdr *section;
 960
 961                 relsec = sechdrs + i;
 962
 963                 if (relsec->sh_type != SHT_RELA &&
 964                     relsec->sh_type != SHT_REL)
 965                         continue;
 966
 967                 /*
 968                  * For section of type SHT_RELA/SHT_REL,
 969                  * ->sh_link contains section header index of associated
 970                  * symbol table. And ->sh_info contains section header
 971                  * index of section to which relocations apply.
 972                  */
 973                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 974                     relsec->sh_link >= pi->ehdr->e_shnum)
 975                         return -ENOEXEC;
 976
 977                 section = pi->sechdrs + relsec->sh_info;
 978                 symtab = sechdrs + relsec->sh_link;
 979
 980                 if (!(section->sh_flags & SHF_ALLOC))
 981                         continue;
 982
 983                 /*
 984                  * symtab->sh_link contain section header index of associated
 985                  * string table.
 986                  */
 987                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 988                         /* Invalid section number? */
 989                         continue;
 990
 991                 /*
 992                  * Respective architecture needs to provide support for applying
 993                  * relocations of type SHT_RELA/SHT_REL.
 994                  */
 995                 if (relsec->sh_type == SHT_RELA)
 996                         ret = arch_kexec_apply_relocations_add(pi, section,
 997                                                                relsec, symtab);
 998                 else if (relsec->sh_type == SHT_REL)
 999                         ret = arch_kexec_apply_relocations(pi, section,
1000                                                            relsec, symtab);
1001                 if (ret)
1002                         return ret;
1003         }
1004
1005         return 0;
1006 }
1007
1008 /*
1009  * kexec_load_purgatory - Load and relocate the purgatory object.
1010  * @image:      Image to add the purgatory to.
1011  * @kbuf:       Memory parameters to use.
1012  *
1013  * Allocates the memory needed for image->purgatory_info.sechdrs and
1014  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1015  * to free the memory after use.
1016  *
1017  * Return: 0 on success, negative errno on error.
1018  */
1019 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1020 {
1021         struct purgatory_info *pi = &image->purgatory_info;
1022         int ret;
1023
1024         if (kexec_purgatory_size <= 0)
1025                 return -EINVAL;
1026
1027         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1028
1029         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1030         if (ret)
1031                 return ret;
1032
1033         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1034         if (ret)
1035                 goto out_free_kbuf;
1036
1037         ret = kexec_apply_relocations(image);
1038         if (ret)
1039                 goto out;
1040
1041         return 0;
1042 out:
1043         vfree(pi->sechdrs);
1044         pi->sechdrs = NULL;
1045 out_free_kbuf:
1046         vfree(pi->purgatory_buf);
1047         pi->purgatory_buf = NULL;
1048         return ret;
1049 }
1050
1051 /*
1052  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1053  * @pi:         Purgatory to search in.
1054  * @name:       Name of the symbol.
1055  *
1056  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1057  */
1058 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1059                                                   const char *name)
1060 {
1061         const Elf_Shdr *sechdrs;
1062         const Elf_Ehdr *ehdr;
1063         const Elf_Sym *syms;
1064         const char *strtab;
1065         int i, k;
1066
1067         if (!pi->ehdr)
1068                 return NULL;
1069
1070         ehdr = pi->ehdr;
1071         sechdrs = (void *)ehdr + ehdr->e_shoff;
1072
1073         for (i = 0; i < ehdr->e_shnum; i++) {
1074                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1075                         continue;
1076
1077                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1078                         /* Invalid strtab section number */
1079                         continue;
1080                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1081                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1082
1083                 /* Go through symbols for a match */
1084                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1085                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1086                                 continue;
1087
1088                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1089                                 continue;
1090
1091                         if (syms[k].st_shndx == SHN_UNDEF ||
1092                             syms[k].st_shndx >= ehdr->e_shnum) {
1093                                 pr_debug("Symbol: %s has bad section index %d.\n",
1094                                                 name, syms[k].st_shndx);
1095                                 return NULL;
1096                         }
1097
1098                         /* Found the symbol we are looking for */
1099                         return &syms[k];
1100                 }
1101         }
1102
1103         return NULL;
1104 }
1105
1106 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1107 {
1108         struct purgatory_info *pi = &image->purgatory_info;
1109         const Elf_Sym *sym;
1110         Elf_Shdr *sechdr;
1111
1112         sym = kexec_purgatory_find_symbol(pi, name);
1113         if (!sym)
1114                 return ERR_PTR(-EINVAL);
1115
1116         sechdr = &pi->sechdrs[sym->st_shndx];
1117
1118         /*
1119          * Returns the address where symbol will finally be loaded after
1120          * kexec_load_segment()
1121          */
1122         return (void *)(sechdr->sh_addr + sym->st_value);
1123 }
1124
1125 /*
1126  * Get or set value of a symbol. If "get_value" is true, symbol value is
1127  * returned in buf otherwise symbol value is set based on value in buf.
1128  */
1129 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1130                                    void *buf, unsigned int size, bool get_value)
1131 {
1132         struct purgatory_info *pi = &image->purgatory_info;
1133         const Elf_Sym *sym;
1134         Elf_Shdr *sec;
1135         char *sym_buf;
1136
1137         sym = kexec_purgatory_find_symbol(pi, name);
1138         if (!sym)
1139                 return -EINVAL;
1140
1141         if (sym->st_size != size) {
1142                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1143                        name, (unsigned long)sym->st_size, size);
1144                 return -EINVAL;
1145         }
1146
1147         sec = pi->sechdrs + sym->st_shndx;
1148
1149         if (sec->sh_type == SHT_NOBITS) {
1150                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1151                        get_value ? "get" : "set");
1152                 return -EINVAL;
1153         }
1154
1155         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1156
1157         if (get_value)
1158                 memcpy((void *)buf, sym_buf, size);
1159         else
1160                 memcpy((void *)sym_buf, buf, size);
1161
1162         return 0;
1163 }
1164 #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */