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         if (kexec_on_panic) {
 289                 /* Enable special crash kernel control page alloc policy. */
 290                 image->control_page = crashk_res.start;
 291                 image->type = KEXEC_TYPE_CRASH;
 292         }
 293
 294         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 295                                            cmdline_ptr, cmdline_len, flags);
 296         if (ret)
 297                 goto out_free_image;
 298
 299         ret = sanity_check_segment_list(image);
 300         if (ret)
 301                 goto out_free_post_load_bufs;
 302
 303         ret = -ENOMEM;
 304         image->control_code_page = kimage_alloc_control_pages(image,
 305                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 306         if (!image->control_code_page) {
 307                 pr_err("Could not allocate control_code_buffer\n");
 308                 goto out_free_post_load_bufs;
 309         }
 310
 311         if (!kexec_on_panic) {
 312                 image->swap_page = kimage_alloc_control_pages(image, 0);
 313                 if (!image->swap_page) {
 314                         pr_err("Could not allocate swap buffer\n");
 315                         goto out_free_control_pages;
 316                 }
 317         }
 318
 319         *rimage = image;
 320         return 0;
 321 out_free_control_pages:
 322         kimage_free_page_list(&image->control_pages);
 323 out_free_post_load_bufs:
 324         kimage_file_post_load_cleanup(image);
 325 out_free_image:
 326         kfree(image);
 327         return ret;
 328 }
 329
 330 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 331                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 332                 unsigned long, flags)
 333 {
 334         int image_type = (flags & KEXEC_FILE_ON_CRASH) ?
 335                          KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
 336         struct kimage **dest_image, *image;
 337         int ret = 0, i;
 338
 339         /* We only trust the superuser with rebooting the system. */
 340         if (!kexec_load_permitted(image_type))
 341                 return -EPERM;
 342
 343         /* Make sure we have a legal set of flags */
 344         if (flags != (flags & KEXEC_FILE_FLAGS))
 345                 return -EINVAL;
 346
 347         image = NULL;
 348
 349         if (!kexec_trylock())
 350                 return -EBUSY;
 351
 352         if (image_type == KEXEC_TYPE_CRASH) {
 353                 dest_image = &kexec_crash_image;
 354                 if (kexec_crash_image)
 355                         arch_kexec_unprotect_crashkres();
 356         } else {
 357                 dest_image = &kexec_image;
 358         }
 359
 360         if (flags & KEXEC_FILE_UNLOAD)
 361                 goto exchange;
 362
 363         /*
 364          * In case of crash, new kernel gets loaded in reserved region. It is
 365          * same memory where old crash kernel might be loaded. Free any
 366          * current crash dump kernel before we corrupt it.
 367          */
 368         if (flags & KEXEC_FILE_ON_CRASH)
 369                 kimage_free(xchg(&kexec_crash_image, NULL));
 370
 371         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 372                                      cmdline_len, flags);
 373         if (ret)
 374                 goto out;
 375
 376         ret = machine_kexec_prepare(image);
 377         if (ret)
 378                 goto out;
 379
 380         /*
 381          * Some architecture(like S390) may touch the crash memory before
 382          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 383          */
 384         ret = kimage_crash_copy_vmcoreinfo(image);
 385         if (ret)
 386                 goto out;
 387
 388         ret = kexec_calculate_store_digests(image);
 389         if (ret)
 390                 goto out;
 391
 392         kexec_dprintk("nr_segments = %lu\n", image->nr_segments);
 393         for (i = 0; i < image->nr_segments; i++) {
 394                 struct kexec_segment *ksegment;
 395
 396                 ksegment = &image->segment[i];
 397                 kexec_dprintk("segment[%d]: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 398                               i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 399                               ksegment->memsz);
 400
 401                 ret = kimage_load_segment(image, &image->segment[i]);
 402                 if (ret)
 403                         goto out;
 404         }
 405
 406         kimage_terminate(image);
 407
 408         ret = machine_kexec_post_load(image);
 409         if (ret)
 410                 goto out;
 411
 412         kexec_dprintk("kexec_file_load: type:%u, start:0x%lx head:0x%lx flags:0x%lx\n",
 413                       image->type, image->start, image->head, flags);
 414         /*
 415          * Free up any temporary buffers allocated which are not needed
 416          * after image has been loaded
 417          */
 418         kimage_file_post_load_cleanup(image);
 419 exchange:
 420         image = xchg(dest_image, image);
 421 out:
 422         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 423                 arch_kexec_protect_crashkres();
 424
 425         kexec_unlock();
 426         kimage_free(image);
 427         return ret;
 428 }
 429
 430 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 431                                     struct kexec_buf *kbuf)
 432 {
 433         struct kimage *image = kbuf->image;
 434         unsigned long temp_start, temp_end;
 435
 436         temp_end = min(end, kbuf->buf_max);
 437         temp_start = temp_end - kbuf->memsz + 1;
 438
 439         do {
 440                 /* align down start */
 441                 temp_start = ALIGN_DOWN(temp_start, kbuf->buf_align);
 442
 443                 if (temp_start < start || temp_start < kbuf->buf_min)
 444                         return 0;
 445
 446                 temp_end = temp_start + kbuf->memsz - 1;
 447
 448                 /*
 449                  * Make sure this does not conflict with any of existing
 450                  * segments
 451                  */
 452                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 453                         temp_start = temp_start - PAGE_SIZE;
 454                         continue;
 455                 }
 456
 457                 /* We found a suitable memory range */
 458                 break;
 459         } while (1);
 460
 461         /* If we are here, we found a suitable memory range */
 462         kbuf->mem = temp_start;
 463
 464         /* Success, stop navigating through remaining System RAM ranges */
 465         return 1;
 466 }
 467
 468 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 469                                      struct kexec_buf *kbuf)
 470 {
 471         struct kimage *image = kbuf->image;
 472         unsigned long temp_start, temp_end;
 473
 474         temp_start = max(start, kbuf->buf_min);
 475
 476         do {
 477                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 478                 temp_end = temp_start + kbuf->memsz - 1;
 479
 480                 if (temp_end > end || temp_end > kbuf->buf_max)
 481                         return 0;
 482                 /*
 483                  * Make sure this does not conflict with any of existing
 484                  * segments
 485                  */
 486                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 487                         temp_start = temp_start + PAGE_SIZE;
 488                         continue;
 489                 }
 490
 491                 /* We found a suitable memory range */
 492                 break;
 493         } while (1);
 494
 495         /* If we are here, we found a suitable memory range */
 496         kbuf->mem = temp_start;
 497
 498         /* Success, stop navigating through remaining System RAM ranges */
 499         return 1;
 500 }
 501
 502 static int locate_mem_hole_callback(struct resource *res, void *arg)
 503 {
 504         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 505         u64 start = res->start, end = res->end;
 506         unsigned long sz = end - start + 1;
 507
 508         /* Returning 0 will take to next memory range */
 509
 510         /* Don't use memory that will be detected and handled by a driver. */
 511         if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
 512                 return 0;
 513
 514         if (sz < kbuf->memsz)
 515                 return 0;
 516
 517         if (end < kbuf->buf_min || start > kbuf->buf_max)
 518                 return 0;
 519
 520         /*
 521          * Allocate memory top down with-in ram range. Otherwise bottom up
 522          * allocation.
 523          */
 524         if (kbuf->top_down)
 525                 return locate_mem_hole_top_down(start, end, kbuf);
 526         return locate_mem_hole_bottom_up(start, end, kbuf);
 527 }
 528
 529 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
 530 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 531                                int (*func)(struct resource *, void *))
 532 {
 533         int ret = 0;
 534         u64 i;
 535         phys_addr_t mstart, mend;
 536         struct resource res = { };
 537
 538         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 539                 return func(&crashk_res, kbuf);
 540
 541         /*
 542          * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
 543          * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
 544          * locate_mem_hole_callback().
 545          */
 546         if (kbuf->top_down) {
 547                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 548                                                 &mstart, &mend, NULL) {
 549                         /*
 550                          * In memblock, end points to the first byte after the
 551                          * range while in kexec, end points to the last byte
 552                          * in the range.
 553                          */
 554                         res.start = mstart;
 555                         res.end = mend - 1;
 556                         ret = func(&res, kbuf);
 557                         if (ret)
 558                                 break;
 559                 }
 560         } else {
 561                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 562                                         &mstart, &mend, NULL) {
 563                         /*
 564                          * In memblock, end points to the first byte after the
 565                          * range while in kexec, end points to the last byte
 566                          * in the range.
 567                          */
 568                         res.start = mstart;
 569                         res.end = mend - 1;
 570                         ret = func(&res, kbuf);
 571                         if (ret)
 572                                 break;
 573                 }
 574         }
 575
 576         return ret;
 577 }
 578 #else
 579 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 580                                int (*func)(struct resource *, void *))
 581 {
 582         return 0;
 583 }
 584 #endif
 585
 586 /**
 587  * kexec_walk_resources - call func(data) on free memory regions
 588  * @kbuf:       Context info for the search. Also passed to @func.
 589  * @func:       Function to call for each memory region.
 590  *
 591  * Return: The memory walk will stop when func returns a non-zero value
 592  * and that value will be returned. If all free regions are visited without
 593  * func returning non-zero, then zero will be returned.
 594  */
 595 static int kexec_walk_resources(struct kexec_buf *kbuf,
 596                                 int (*func)(struct resource *, void *))
 597 {
 598         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 599                 return walk_iomem_res_desc(crashk_res.desc,
 600                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 601                                            crashk_res.start, crashk_res.end,
 602                                            kbuf, func);
 603         else if (kbuf->top_down)
 604                 return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func);
 605         else
 606                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 607 }
 608
 609 /**
 610  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 611  * @kbuf:       Parameters for the memory search.
 612  *
 613  * On success, kbuf->mem will have the start address of the memory region found.
 614  *
 615  * Return: 0 on success, negative errno on error.
 616  */
 617 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 618 {
 619         int ret;
 620
 621         /* Arch knows where to place */
 622         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
 623                 return 0;
 624
 625         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 626                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
 627         else
 628                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
 629
 630         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 631 }
 632
 633 /**
 634  * kexec_add_buffer - place a buffer in a kexec segment
 635  * @kbuf:       Buffer contents and memory parameters.
 636  *
 637  * This function assumes that kexec_lock is held.
 638  * On successful return, @kbuf->mem will have the physical address of
 639  * the buffer in memory.
 640  *
 641  * Return: 0 on success, negative errno on error.
 642  */
 643 int kexec_add_buffer(struct kexec_buf *kbuf)
 644 {
 645         struct kexec_segment *ksegment;
 646         int ret;
 647
 648         /* Currently adding segment this way is allowed only in file mode */
 649         if (!kbuf->image->file_mode)
 650                 return -EINVAL;
 651
 652         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 653                 return -EINVAL;
 654
 655         /*
 656          * Make sure we are not trying to add buffer after allocating
 657          * control pages. All segments need to be placed first before
 658          * any control pages are allocated. As control page allocation
 659          * logic goes through list of segments to make sure there are
 660          * no destination overlaps.
 661          */
 662         if (!list_empty(&kbuf->image->control_pages)) {
 663                 WARN_ON(1);
 664                 return -EINVAL;
 665         }
 666
 667         /* Ensure minimum alignment needed for segments. */
 668         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 669         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 670
 671         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 672         ret = arch_kexec_locate_mem_hole(kbuf);
 673         if (ret)
 674                 return ret;
 675
 676         /* Found a suitable memory range */
 677         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 678         ksegment->kbuf = kbuf->buffer;
 679         ksegment->bufsz = kbuf->bufsz;
 680         ksegment->mem = kbuf->mem;
 681         ksegment->memsz = kbuf->memsz;
 682         kbuf->image->nr_segments++;
 683         return 0;
 684 }
 685
 686 /* Calculate and store the digest of segments */
 687 static int kexec_calculate_store_digests(struct kimage *image)
 688 {
 689         struct crypto_shash *tfm;
 690         struct shash_desc *desc;
 691         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 692         size_t desc_size, nullsz;
 693         char *digest;
 694         void *zero_buf;
 695         struct kexec_sha_region *sha_regions;
 696         struct purgatory_info *pi = &image->purgatory_info;
 697
 698         if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY))
 699                 return 0;
 700
 701         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 702         zero_buf_sz = PAGE_SIZE;
 703
 704         tfm = crypto_alloc_shash("sha256", 0, 0);
 705         if (IS_ERR(tfm)) {
 706                 ret = PTR_ERR(tfm);
 707                 goto out;
 708         }
 709
 710         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 711         desc = kzalloc(desc_size, GFP_KERNEL);
 712         if (!desc) {
 713                 ret = -ENOMEM;
 714                 goto out_free_tfm;
 715         }
 716
 717         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 718         sha_regions = vzalloc(sha_region_sz);
 719         if (!sha_regions) {
 720                 ret = -ENOMEM;
 721                 goto out_free_desc;
 722         }
 723
 724         desc->tfm   = tfm;
 725
 726         ret = crypto_shash_init(desc);
 727         if (ret < 0)
 728                 goto out_free_sha_regions;
 729
 730         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 731         if (!digest) {
 732                 ret = -ENOMEM;
 733                 goto out_free_sha_regions;
 734         }
 735
 736         for (j = i = 0; i < image->nr_segments; i++) {
 737                 struct kexec_segment *ksegment;
 738
 739 #ifdef CONFIG_CRASH_HOTPLUG
 740                 /* Exclude elfcorehdr segment to allow future changes via hotplug */
 741                 if (j == image->elfcorehdr_index)
 742                         continue;
 743 #endif
 744
 745                 ksegment = &image->segment[i];
 746                 /*
 747                  * Skip purgatory as it will be modified once we put digest
 748                  * info in purgatory.
 749                  */
 750                 if (ksegment->kbuf == pi->purgatory_buf)
 751                         continue;
 752
 753                 ret = crypto_shash_update(desc, ksegment->kbuf,
 754                                           ksegment->bufsz);
 755                 if (ret)
 756                         break;
 757
 758                 /*
 759                  * Assume rest of the buffer is filled with zero and
 760                  * update digest accordingly.
 761                  */
 762                 nullsz = ksegment->memsz - ksegment->bufsz;
 763                 while (nullsz) {
 764                         unsigned long bytes = nullsz;
 765
 766                         if (bytes > zero_buf_sz)
 767                                 bytes = zero_buf_sz;
 768                         ret = crypto_shash_update(desc, zero_buf, bytes);
 769                         if (ret)
 770                                 break;
 771                         nullsz -= bytes;
 772                 }
 773
 774                 if (ret)
 775                         break;
 776
 777                 sha_regions[j].start = ksegment->mem;
 778                 sha_regions[j].len = ksegment->memsz;
 779                 j++;
 780         }
 781
 782         if (!ret) {
 783                 ret = crypto_shash_final(desc, digest);
 784                 if (ret)
 785                         goto out_free_digest;
 786                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 787                                                      sha_regions, sha_region_sz, 0);
 788                 if (ret)
 789                         goto out_free_digest;
 790
 791                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 792                                                      digest, SHA256_DIGEST_SIZE, 0);
 793                 if (ret)
 794                         goto out_free_digest;
 795         }
 796
 797 out_free_digest:
 798         kfree(digest);
 799 out_free_sha_regions:
 800         vfree(sha_regions);
 801 out_free_desc:
 802         kfree(desc);
 803 out_free_tfm:
 804         kfree(tfm);
 805 out:
 806         return ret;
 807 }
 808
 809 #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY
 810 /*
 811  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 812  * @pi:         Purgatory to be loaded.
 813  * @kbuf:       Buffer to setup.
 814  *
 815  * Allocates the memory needed for the buffer. Caller is responsible to free
 816  * the memory after use.
 817  *
 818  * Return: 0 on success, negative errno on error.
 819  */
 820 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 821                                       struct kexec_buf *kbuf)
 822 {
 823         const Elf_Shdr *sechdrs;
 824         unsigned long bss_align;
 825         unsigned long bss_sz;
 826         unsigned long align;
 827         int i, ret;
 828
 829         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 830         kbuf->buf_align = bss_align = 1;
 831         kbuf->bufsz = bss_sz = 0;
 832
 833         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 834                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 835                         continue;
 836
 837                 align = sechdrs[i].sh_addralign;
 838                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 839                         if (kbuf->buf_align < align)
 840                                 kbuf->buf_align = align;
 841                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 842                         kbuf->bufsz += sechdrs[i].sh_size;
 843                 } else {
 844                         if (bss_align < align)
 845                                 bss_align = align;
 846                         bss_sz = ALIGN(bss_sz, align);
 847                         bss_sz += sechdrs[i].sh_size;
 848                 }
 849         }
 850         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 851         kbuf->memsz = kbuf->bufsz + bss_sz;
 852         if (kbuf->buf_align < bss_align)
 853                 kbuf->buf_align = bss_align;
 854
 855         kbuf->buffer = vzalloc(kbuf->bufsz);
 856         if (!kbuf->buffer)
 857                 return -ENOMEM;
 858         pi->purgatory_buf = kbuf->buffer;
 859
 860         ret = kexec_add_buffer(kbuf);
 861         if (ret)
 862                 goto out;
 863
 864         return 0;
 865 out:
 866         vfree(pi->purgatory_buf);
 867         pi->purgatory_buf = NULL;
 868         return ret;
 869 }
 870
 871 /*
 872  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 873  * @pi:         Purgatory to be loaded.
 874  * @kbuf:       Buffer prepared to store purgatory.
 875  *
 876  * Allocates the memory needed for the buffer. Caller is responsible to free
 877  * the memory after use.
 878  *
 879  * Return: 0 on success, negative errno on error.
 880  */
 881 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 882                                          struct kexec_buf *kbuf)
 883 {
 884         unsigned long bss_addr;
 885         unsigned long offset;
 886         size_t sechdrs_size;
 887         Elf_Shdr *sechdrs;
 888         int i;
 889
 890         /*
 891          * The section headers in kexec_purgatory are read-only. In order to
 892          * have them modifiable make a temporary copy.
 893          */
 894         sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum);
 895         sechdrs = vzalloc(sechdrs_size);
 896         if (!sechdrs)
 897                 return -ENOMEM;
 898         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size);
 899         pi->sechdrs = sechdrs;
 900
 901         offset = 0;
 902         bss_addr = kbuf->mem + kbuf->bufsz;
 903         kbuf->image->start = pi->ehdr->e_entry;
 904
 905         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 906                 unsigned long align;
 907                 void *src, *dst;
 908
 909                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 910                         continue;
 911
 912                 align = sechdrs[i].sh_addralign;
 913                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 914                         bss_addr = ALIGN(bss_addr, align);
 915                         sechdrs[i].sh_addr = bss_addr;
 916                         bss_addr += sechdrs[i].sh_size;
 917                         continue;
 918                 }
 919
 920                 offset = ALIGN(offset, align);
 921
 922                 /*
 923                  * Check if the segment contains the entry point, if so,
 924                  * calculate the value of image->start based on it.
 925                  * If the compiler has produced more than one .text section
 926                  * (Eg: .text.hot), they are generally after the main .text
 927                  * section, and they shall not be used to calculate
 928                  * image->start. So do not re-calculate image->start if it
 929                  * is not set to the initial value, and warn the user so they
 930                  * have a chance to fix their purgatory's linker script.
 931                  */
 932                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 933                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 934                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 935                                          + sechdrs[i].sh_size) &&
 936                     !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
 937                         kbuf->image->start -= sechdrs[i].sh_addr;
 938                         kbuf->image->start += kbuf->mem + offset;
 939                 }
 940
 941                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 942                 dst = pi->purgatory_buf + offset;
 943                 memcpy(dst, src, sechdrs[i].sh_size);
 944
 945                 sechdrs[i].sh_addr = kbuf->mem + offset;
 946                 sechdrs[i].sh_offset = offset;
 947                 offset += sechdrs[i].sh_size;
 948         }
 949
 950         return 0;
 951 }
 952
 953 static int kexec_apply_relocations(struct kimage *image)
 954 {
 955         int i, ret;
 956         struct purgatory_info *pi = &image->purgatory_info;
 957         const Elf_Shdr *sechdrs;
 958
 959         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 960
 961         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 962                 const Elf_Shdr *relsec;
 963                 const Elf_Shdr *symtab;
 964                 Elf_Shdr *section;
 965
 966                 relsec = sechdrs + i;
 967
 968                 if (relsec->sh_type != SHT_RELA &&
 969                     relsec->sh_type != SHT_REL)
 970                         continue;
 971
 972                 /*
 973                  * For section of type SHT_RELA/SHT_REL,
 974                  * ->sh_link contains section header index of associated
 975                  * symbol table. And ->sh_info contains section header
 976                  * index of section to which relocations apply.
 977                  */
 978                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 979                     relsec->sh_link >= pi->ehdr->e_shnum)
 980                         return -ENOEXEC;
 981
 982                 section = pi->sechdrs + relsec->sh_info;
 983                 symtab = sechdrs + relsec->sh_link;
 984
 985                 if (!(section->sh_flags & SHF_ALLOC))
 986                         continue;
 987
 988                 /*
 989                  * symtab->sh_link contain section header index of associated
 990                  * string table.
 991                  */
 992                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 993                         /* Invalid section number? */
 994                         continue;
 995
 996                 /*
 997                  * Respective architecture needs to provide support for applying
 998                  * relocations of type SHT_RELA/SHT_REL.
 999                  */
1000                 if (relsec->sh_type == SHT_RELA)
1001                         ret = arch_kexec_apply_relocations_add(pi, section,
1002                                                                relsec, symtab);
1003                 else if (relsec->sh_type == SHT_REL)
1004                         ret = arch_kexec_apply_relocations(pi, section,
1005                                                            relsec, symtab);
1006                 if (ret)
1007                         return ret;
1008         }
1009
1010         return 0;
1011 }
1012
1013 /*
1014  * kexec_load_purgatory - Load and relocate the purgatory object.
1015  * @image:      Image to add the purgatory to.
1016  * @kbuf:       Memory parameters to use.
1017  *
1018  * Allocates the memory needed for image->purgatory_info.sechdrs and
1019  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1020  * to free the memory after use.
1021  *
1022  * Return: 0 on success, negative errno on error.
1023  */
1024 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1025 {
1026         struct purgatory_info *pi = &image->purgatory_info;
1027         int ret;
1028
1029         if (kexec_purgatory_size <= 0)
1030                 return -EINVAL;
1031
1032         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1033
1034         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1035         if (ret)
1036                 return ret;
1037
1038         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1039         if (ret)
1040                 goto out_free_kbuf;
1041
1042         ret = kexec_apply_relocations(image);
1043         if (ret)
1044                 goto out;
1045
1046         return 0;
1047 out:
1048         vfree(pi->sechdrs);
1049         pi->sechdrs = NULL;
1050 out_free_kbuf:
1051         vfree(pi->purgatory_buf);
1052         pi->purgatory_buf = NULL;
1053         return ret;
1054 }
1055
1056 /*
1057  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1058  * @pi:         Purgatory to search in.
1059  * @name:       Name of the symbol.
1060  *
1061  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1062  */
1063 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1064                                                   const char *name)
1065 {
1066         const Elf_Shdr *sechdrs;
1067         const Elf_Ehdr *ehdr;
1068         const Elf_Sym *syms;
1069         const char *strtab;
1070         int i, k;
1071
1072         if (!pi->ehdr)
1073                 return NULL;
1074
1075         ehdr = pi->ehdr;
1076         sechdrs = (void *)ehdr + ehdr->e_shoff;
1077
1078         for (i = 0; i < ehdr->e_shnum; i++) {
1079                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1080                         continue;
1081
1082                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1083                         /* Invalid strtab section number */
1084                         continue;
1085                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1086                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1087
1088                 /* Go through symbols for a match */
1089                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1090                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1091                                 continue;
1092
1093                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1094                                 continue;
1095
1096                         if (syms[k].st_shndx == SHN_UNDEF ||
1097                             syms[k].st_shndx >= ehdr->e_shnum) {
1098                                 pr_debug("Symbol: %s has bad section index %d.\n",
1099                                                 name, syms[k].st_shndx);
1100                                 return NULL;
1101                         }
1102
1103                         /* Found the symbol we are looking for */
1104                         return &syms[k];
1105                 }
1106         }
1107
1108         return NULL;
1109 }
1110
1111 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1112 {
1113         struct purgatory_info *pi = &image->purgatory_info;
1114         const Elf_Sym *sym;
1115         Elf_Shdr *sechdr;
1116
1117         sym = kexec_purgatory_find_symbol(pi, name);
1118         if (!sym)
1119                 return ERR_PTR(-EINVAL);
1120
1121         sechdr = &pi->sechdrs[sym->st_shndx];
1122
1123         /*
1124          * Returns the address where symbol will finally be loaded after
1125          * kexec_load_segment()
1126          */
1127         return (void *)(sechdr->sh_addr + sym->st_value);
1128 }
1129
1130 /*
1131  * Get or set value of a symbol. If "get_value" is true, symbol value is
1132  * returned in buf otherwise symbol value is set based on value in buf.
1133  */
1134 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1135                                    void *buf, unsigned int size, bool get_value)
1136 {
1137         struct purgatory_info *pi = &image->purgatory_info;
1138         const Elf_Sym *sym;
1139         Elf_Shdr *sec;
1140         char *sym_buf;
1141
1142         sym = kexec_purgatory_find_symbol(pi, name);
1143         if (!sym)
1144                 return -EINVAL;
1145
1146         if (sym->st_size != size) {
1147                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1148                        name, (unsigned long)sym->st_size, size);
1149                 return -EINVAL;
1150         }
1151
1152         sec = pi->sechdrs + sym->st_shndx;
1153
1154         if (sec->sh_type == SHT_NOBITS) {
1155                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1156                        get_value ? "get" : "set");
1157                 return -EINVAL;
1158         }
1159
1160         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1161
1162         if (get_value)
1163                 memcpy((void *)buf, sym_buf, size);
1164         else
1165                 memcpy((void *)sym_buf, buf, size);
1166
1167         return 0;
1168 }
1169 #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */