kernel/kexec_file.c

   1 /*
   2  * kexec: kexec_file_load system call
   3  *
   4  * Copyright (C) 2014 Red Hat Inc.
   5  * Authors:
   6  *      Vivek Goyal <vgoyal@redhat.com>
   7  *
   8  * This source code is licensed under the GNU General Public License,
   9  * Version 2.  See the file COPYING for more details.
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/capability.h>
  15 #include <linux/mm.h>
  16 #include <linux/file.h>
  17 #include <linux/slab.h>
  18 #include <linux/kexec.h>
  19 #include <linux/mutex.h>
  20 #include <linux/list.h>
  21 #include <linux/fs.h>
  22 #include <linux/ima.h>
  23 #include <crypto/hash.h>
  24 #include <crypto/sha.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/vmalloc.h>
  27 #include "kexec_internal.h"
  28
  29 /*
  30  * Declare these symbols weak so that if architecture provides a purgatory,
  31  * these will be overridden.
  32  */
  33 char __weak kexec_purgatory[0];
  34 size_t __weak kexec_purgatory_size = 0;
  35
  36 static int kexec_calculate_store_digests(struct kimage *image);
  37
  38 /* Architectures can provide this probe function */
  39 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  40                                          unsigned long buf_len)
  41 {
  42         return -ENOEXEC;
  43 }
  44
  45 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
  46 {
  47         return ERR_PTR(-ENOEXEC);
  48 }
  49
  50 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
  51 {
  52         return -EINVAL;
  53 }
  54
  55 #ifdef CONFIG_KEXEC_VERIFY_SIG
  56 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
  57                                         unsigned long buf_len)
  58 {
  59         return -EKEYREJECTED;
  60 }
  61 #endif
  62
  63 /* Apply relocations of type RELA */
  64 int __weak
  65 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
  66                                  unsigned int relsec)
  67 {
  68         pr_err("RELA relocation unsupported.\n");
  69         return -ENOEXEC;
  70 }
  71
  72 /* Apply relocations of type REL */
  73 int __weak
  74 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
  75                              unsigned int relsec)
  76 {
  77         pr_err("REL relocation unsupported.\n");
  78         return -ENOEXEC;
  79 }
  80
  81 /*
  82  * Free up memory used by kernel, initrd, and command line. This is temporary
  83  * memory allocation which is not needed any more after these buffers have
  84  * been loaded into separate segments and have been copied elsewhere.
  85  */
  86 void kimage_file_post_load_cleanup(struct kimage *image)
  87 {
  88         struct purgatory_info *pi = &image->purgatory_info;
  89
  90         vfree(image->kernel_buf);
  91         image->kernel_buf = NULL;
  92
  93         vfree(image->initrd_buf);
  94         image->initrd_buf = NULL;
  95
  96         kfree(image->cmdline_buf);
  97         image->cmdline_buf = NULL;
  98
  99         vfree(pi->purgatory_buf);
 100         pi->purgatory_buf = NULL;
 101
 102         vfree(pi->sechdrs);
 103         pi->sechdrs = NULL;
 104
 105         /* See if architecture has anything to cleanup post load */
 106         arch_kimage_file_post_load_cleanup(image);
 107
 108         /*
 109          * Above call should have called into bootloader to free up
 110          * any data stored in kimage->image_loader_data. It should
 111          * be ok now to free it up.
 112          */
 113         kfree(image->image_loader_data);
 114         image->image_loader_data = NULL;
 115 }
 116
 117 /*
 118  * In file mode list of segments is prepared by kernel. Copy relevant
 119  * data from user space, do error checking, prepare segment list
 120  */
 121 static int
 122 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 123                              const char __user *cmdline_ptr,
 124                              unsigned long cmdline_len, unsigned flags)
 125 {
 126         int ret = 0;
 127         void *ldata;
 128         loff_t size;
 129
 130         ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
 131                                        &size, INT_MAX, READING_KEXEC_IMAGE);
 132         if (ret)
 133                 return ret;
 134         image->kernel_buf_len = size;
 135
 136         /* IMA needs to pass the measurement list to the next kernel. */
 137         ima_add_kexec_buffer(image);
 138
 139         /* Call arch image probe handlers */
 140         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 141                                             image->kernel_buf_len);
 142         if (ret)
 143                 goto out;
 144
 145 #ifdef CONFIG_KEXEC_VERIFY_SIG
 146         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 147                                            image->kernel_buf_len);
 148         if (ret) {
 149                 pr_debug("kernel signature verification failed.\n");
 150                 goto out;
 151         }
 152         pr_debug("kernel signature verification successful.\n");
 153 #endif
 154         /* It is possible that there no initramfs is being loaded */
 155         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 156                 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
 157                                                &size, INT_MAX,
 158                                                READING_KEXEC_INITRAMFS);
 159                 if (ret)
 160                         goto out;
 161                 image->initrd_buf_len = size;
 162         }
 163
 164         if (cmdline_len) {
 165                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 166                 if (IS_ERR(image->cmdline_buf)) {
 167                         ret = PTR_ERR(image->cmdline_buf);
 168                         image->cmdline_buf = NULL;
 169                         goto out;
 170                 }
 171
 172                 image->cmdline_buf_len = cmdline_len;
 173
 174                 /* command line should be a string with last byte null */
 175                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 176                         ret = -EINVAL;
 177                         goto out;
 178                 }
 179         }
 180
 181         /* Call arch image load handlers */
 182         ldata = arch_kexec_kernel_image_load(image);
 183
 184         if (IS_ERR(ldata)) {
 185                 ret = PTR_ERR(ldata);
 186                 goto out;
 187         }
 188
 189         image->image_loader_data = ldata;
 190 out:
 191         /* In case of error, free up all allocated memory in this function */
 192         if (ret)
 193                 kimage_file_post_load_cleanup(image);
 194         return ret;
 195 }
 196
 197 static int
 198 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 199                        int initrd_fd, const char __user *cmdline_ptr,
 200                        unsigned long cmdline_len, unsigned long flags)
 201 {
 202         int ret;
 203         struct kimage *image;
 204         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 205
 206         image = do_kimage_alloc_init();
 207         if (!image)
 208                 return -ENOMEM;
 209
 210         image->file_mode = 1;
 211
 212         if (kexec_on_panic) {
 213                 /* Enable special crash kernel control page alloc policy. */
 214                 image->control_page = crashk_res.start;
 215                 image->type = KEXEC_TYPE_CRASH;
 216         }
 217
 218         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 219                                            cmdline_ptr, cmdline_len, flags);
 220         if (ret)
 221                 goto out_free_image;
 222
 223         ret = sanity_check_segment_list(image);
 224         if (ret)
 225                 goto out_free_post_load_bufs;
 226
 227         ret = -ENOMEM;
 228         image->control_code_page = kimage_alloc_control_pages(image,
 229                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 230         if (!image->control_code_page) {
 231                 pr_err("Could not allocate control_code_buffer\n");
 232                 goto out_free_post_load_bufs;
 233         }
 234
 235         if (!kexec_on_panic) {
 236                 image->swap_page = kimage_alloc_control_pages(image, 0);
 237                 if (!image->swap_page) {
 238                         pr_err("Could not allocate swap buffer\n");
 239                         goto out_free_control_pages;
 240                 }
 241         }
 242
 243         *rimage = image;
 244         return 0;
 245 out_free_control_pages:
 246         kimage_free_page_list(&image->control_pages);
 247 out_free_post_load_bufs:
 248         kimage_file_post_load_cleanup(image);
 249 out_free_image:
 250         kfree(image);
 251         return ret;
 252 }
 253
 254 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 255                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 256                 unsigned long, flags)
 257 {
 258         int ret = 0, i;
 259         struct kimage **dest_image, *image;
 260
 261         /* We only trust the superuser with rebooting the system. */
 262         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 263                 return -EPERM;
 264
 265         /* Make sure we have a legal set of flags */
 266         if (flags != (flags & KEXEC_FILE_FLAGS))
 267                 return -EINVAL;
 268
 269         image = NULL;
 270
 271         if (!mutex_trylock(&kexec_mutex))
 272                 return -EBUSY;
 273
 274         dest_image = &kexec_image;
 275         if (flags & KEXEC_FILE_ON_CRASH) {
 276                 dest_image = &kexec_crash_image;
 277                 if (kexec_crash_image)
 278                         arch_kexec_unprotect_crashkres();
 279         }
 280
 281         if (flags & KEXEC_FILE_UNLOAD)
 282                 goto exchange;
 283
 284         /*
 285          * In case of crash, new kernel gets loaded in reserved region. It is
 286          * same memory where old crash kernel might be loaded. Free any
 287          * current crash dump kernel before we corrupt it.
 288          */
 289         if (flags & KEXEC_FILE_ON_CRASH)
 290                 kimage_free(xchg(&kexec_crash_image, NULL));
 291
 292         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 293                                      cmdline_len, flags);
 294         if (ret)
 295                 goto out;
 296
 297         ret = machine_kexec_prepare(image);
 298         if (ret)
 299                 goto out;
 300
 301         ret = kexec_calculate_store_digests(image);
 302         if (ret)
 303                 goto out;
 304
 305         for (i = 0; i < image->nr_segments; i++) {
 306                 struct kexec_segment *ksegment;
 307
 308                 ksegment = &image->segment[i];
 309                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 310                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 311                          ksegment->memsz);
 312
 313                 ret = kimage_load_segment(image, &image->segment[i]);
 314                 if (ret)
 315                         goto out;
 316         }
 317
 318         kimage_terminate(image);
 319
 320         /*
 321          * Free up any temporary buffers allocated which are not needed
 322          * after image has been loaded
 323          */
 324         kimage_file_post_load_cleanup(image);
 325 exchange:
 326         image = xchg(dest_image, image);
 327 out:
 328         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 329                 arch_kexec_protect_crashkres();
 330
 331         mutex_unlock(&kexec_mutex);
 332         kimage_free(image);
 333         return ret;
 334 }
 335
 336 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 337                                     struct kexec_buf *kbuf)
 338 {
 339         struct kimage *image = kbuf->image;
 340         unsigned long temp_start, temp_end;
 341
 342         temp_end = min(end, kbuf->buf_max);
 343         temp_start = temp_end - kbuf->memsz;
 344
 345         do {
 346                 /* align down start */
 347                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 348
 349                 if (temp_start < start || temp_start < kbuf->buf_min)
 350                         return 0;
 351
 352                 temp_end = temp_start + kbuf->memsz - 1;
 353
 354                 /*
 355                  * Make sure this does not conflict with any of existing
 356                  * segments
 357                  */
 358                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 359                         temp_start = temp_start - PAGE_SIZE;
 360                         continue;
 361                 }
 362
 363                 /* We found a suitable memory range */
 364                 break;
 365         } while (1);
 366
 367         /* If we are here, we found a suitable memory range */
 368         kbuf->mem = temp_start;
 369
 370         /* Success, stop navigating through remaining System RAM ranges */
 371         return 1;
 372 }
 373
 374 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 375                                      struct kexec_buf *kbuf)
 376 {
 377         struct kimage *image = kbuf->image;
 378         unsigned long temp_start, temp_end;
 379
 380         temp_start = max(start, kbuf->buf_min);
 381
 382         do {
 383                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 384                 temp_end = temp_start + kbuf->memsz - 1;
 385
 386                 if (temp_end > end || temp_end > kbuf->buf_max)
 387                         return 0;
 388                 /*
 389                  * Make sure this does not conflict with any of existing
 390                  * segments
 391                  */
 392                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 393                         temp_start = temp_start + PAGE_SIZE;
 394                         continue;
 395                 }
 396
 397                 /* We found a suitable memory range */
 398                 break;
 399         } while (1);
 400
 401         /* If we are here, we found a suitable memory range */
 402         kbuf->mem = temp_start;
 403
 404         /* Success, stop navigating through remaining System RAM ranges */
 405         return 1;
 406 }
 407
 408 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
 409 {
 410         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 411         unsigned long sz = end - start + 1;
 412
 413         /* Returning 0 will take to next memory range */
 414         if (sz < kbuf->memsz)
 415                 return 0;
 416
 417         if (end < kbuf->buf_min || start > kbuf->buf_max)
 418                 return 0;
 419
 420         /*
 421          * Allocate memory top down with-in ram range. Otherwise bottom up
 422          * allocation.
 423          */
 424         if (kbuf->top_down)
 425                 return locate_mem_hole_top_down(start, end, kbuf);
 426         return locate_mem_hole_bottom_up(start, end, kbuf);
 427 }
 428
 429 /**
 430  * arch_kexec_walk_mem - call func(data) on free memory regions
 431  * @kbuf:       Context info for the search. Also passed to @func.
 432  * @func:       Function to call for each memory region.
 433  *
 434  * Return: The memory walk will stop when func returns a non-zero value
 435  * and that value will be returned. If all free regions are visited without
 436  * func returning non-zero, then zero will be returned.
 437  */
 438 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
 439                                int (*func)(u64, u64, void *))
 440 {
 441         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 442                 return walk_iomem_res_desc(crashk_res.desc,
 443                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 444                                            crashk_res.start, crashk_res.end,
 445                                            kbuf, func);
 446         else
 447                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 448 }
 449
 450 /**
 451  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 452  * @kbuf:       Parameters for the memory search.
 453  *
 454  * On success, kbuf->mem will have the start address of the memory region found.
 455  *
 456  * Return: 0 on success, negative errno on error.
 457  */
 458 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 459 {
 460         int ret;
 461
 462         ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
 463
 464         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 465 }
 466
 467 /**
 468  * kexec_add_buffer - place a buffer in a kexec segment
 469  * @kbuf:       Buffer contents and memory parameters.
 470  *
 471  * This function assumes that kexec_mutex is held.
 472  * On successful return, @kbuf->mem will have the physical address of
 473  * the buffer in memory.
 474  *
 475  * Return: 0 on success, negative errno on error.
 476  */
 477 int kexec_add_buffer(struct kexec_buf *kbuf)
 478 {
 479
 480         struct kexec_segment *ksegment;
 481         int ret;
 482
 483         /* Currently adding segment this way is allowed only in file mode */
 484         if (!kbuf->image->file_mode)
 485                 return -EINVAL;
 486
 487         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 488                 return -EINVAL;
 489
 490         /*
 491          * Make sure we are not trying to add buffer after allocating
 492          * control pages. All segments need to be placed first before
 493          * any control pages are allocated. As control page allocation
 494          * logic goes through list of segments to make sure there are
 495          * no destination overlaps.
 496          */
 497         if (!list_empty(&kbuf->image->control_pages)) {
 498                 WARN_ON(1);
 499                 return -EINVAL;
 500         }
 501
 502         /* Ensure minimum alignment needed for segments. */
 503         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 504         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 505
 506         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 507         ret = kexec_locate_mem_hole(kbuf);
 508         if (ret)
 509                 return ret;
 510
 511         /* Found a suitable memory range */
 512         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 513         ksegment->kbuf = kbuf->buffer;
 514         ksegment->bufsz = kbuf->bufsz;
 515         ksegment->mem = kbuf->mem;
 516         ksegment->memsz = kbuf->memsz;
 517         kbuf->image->nr_segments++;
 518         return 0;
 519 }
 520
 521 /* Calculate and store the digest of segments */
 522 static int kexec_calculate_store_digests(struct kimage *image)
 523 {
 524         struct crypto_shash *tfm;
 525         struct shash_desc *desc;
 526         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 527         size_t desc_size, nullsz;
 528         char *digest;
 529         void *zero_buf;
 530         struct kexec_sha_region *sha_regions;
 531         struct purgatory_info *pi = &image->purgatory_info;
 532
 533         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 534         zero_buf_sz = PAGE_SIZE;
 535
 536         tfm = crypto_alloc_shash("sha256", 0, 0);
 537         if (IS_ERR(tfm)) {
 538                 ret = PTR_ERR(tfm);
 539                 goto out;
 540         }
 541
 542         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 543         desc = kzalloc(desc_size, GFP_KERNEL);
 544         if (!desc) {
 545                 ret = -ENOMEM;
 546                 goto out_free_tfm;
 547         }
 548
 549         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 550         sha_regions = vzalloc(sha_region_sz);
 551         if (!sha_regions)
 552                 goto out_free_desc;
 553
 554         desc->tfm   = tfm;
 555         desc->flags = 0;
 556
 557         ret = crypto_shash_init(desc);
 558         if (ret < 0)
 559                 goto out_free_sha_regions;
 560
 561         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 562         if (!digest) {
 563                 ret = -ENOMEM;
 564                 goto out_free_sha_regions;
 565         }
 566
 567         for (j = i = 0; i < image->nr_segments; i++) {
 568                 struct kexec_segment *ksegment;
 569
 570                 ksegment = &image->segment[i];
 571                 /*
 572                  * Skip purgatory as it will be modified once we put digest
 573                  * info in purgatory.
 574                  */
 575                 if (ksegment->kbuf == pi->purgatory_buf)
 576                         continue;
 577
 578                 ret = crypto_shash_update(desc, ksegment->kbuf,
 579                                           ksegment->bufsz);
 580                 if (ret)
 581                         break;
 582
 583                 /*
 584                  * Assume rest of the buffer is filled with zero and
 585                  * update digest accordingly.
 586                  */
 587                 nullsz = ksegment->memsz - ksegment->bufsz;
 588                 while (nullsz) {
 589                         unsigned long bytes = nullsz;
 590
 591                         if (bytes > zero_buf_sz)
 592                                 bytes = zero_buf_sz;
 593                         ret = crypto_shash_update(desc, zero_buf, bytes);
 594                         if (ret)
 595                                 break;
 596                         nullsz -= bytes;
 597                 }
 598
 599                 if (ret)
 600                         break;
 601
 602                 sha_regions[j].start = ksegment->mem;
 603                 sha_regions[j].len = ksegment->memsz;
 604                 j++;
 605         }
 606
 607         if (!ret) {
 608                 ret = crypto_shash_final(desc, digest);
 609                 if (ret)
 610                         goto out_free_digest;
 611                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 612                                                      sha_regions, sha_region_sz, 0);
 613                 if (ret)
 614                         goto out_free_digest;
 615
 616                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 617                                                      digest, SHA256_DIGEST_SIZE, 0);
 618                 if (ret)
 619                         goto out_free_digest;
 620         }
 621
 622 out_free_digest:
 623         kfree(digest);
 624 out_free_sha_regions:
 625         vfree(sha_regions);
 626 out_free_desc:
 627         kfree(desc);
 628 out_free_tfm:
 629         kfree(tfm);
 630 out:
 631         return ret;
 632 }
 633
 634 /* Actually load purgatory. Lot of code taken from kexec-tools */
 635 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
 636                                   unsigned long max, int top_down)
 637 {
 638         struct purgatory_info *pi = &image->purgatory_info;
 639         unsigned long align, bss_align, bss_sz, bss_pad;
 640         unsigned long entry, load_addr, curr_load_addr, bss_addr, offset;
 641         unsigned char *buf_addr, *src;
 642         int i, ret = 0, entry_sidx = -1;
 643         const Elf_Shdr *sechdrs_c;
 644         Elf_Shdr *sechdrs = NULL;
 645         struct kexec_buf kbuf = { .image = image, .bufsz = 0, .buf_align = 1,
 646                                   .buf_min = min, .buf_max = max,
 647                                   .top_down = top_down };
 648
 649         /*
 650          * sechdrs_c points to section headers in purgatory and are read
 651          * only. No modifications allowed.
 652          */
 653         sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
 654
 655         /*
 656          * We can not modify sechdrs_c[] and its fields. It is read only.
 657          * Copy it over to a local copy where one can store some temporary
 658          * data and free it at the end. We need to modify ->sh_addr and
 659          * ->sh_offset fields to keep track of permanent and temporary
 660          * locations of sections.
 661          */
 662         sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 663         if (!sechdrs)
 664                 return -ENOMEM;
 665
 666         memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 667
 668         /*
 669          * We seem to have multiple copies of sections. First copy is which
 670          * is embedded in kernel in read only section. Some of these sections
 671          * will be copied to a temporary buffer and relocated. And these
 672          * sections will finally be copied to their final destination at
 673          * segment load time.
 674          *
 675          * Use ->sh_offset to reflect section address in memory. It will
 676          * point to original read only copy if section is not allocatable.
 677          * Otherwise it will point to temporary copy which will be relocated.
 678          *
 679          * Use ->sh_addr to contain final address of the section where it
 680          * will go during execution time.
 681          */
 682         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 683                 if (sechdrs[i].sh_type == SHT_NOBITS)
 684                         continue;
 685
 686                 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
 687                                                 sechdrs[i].sh_offset;
 688         }
 689
 690         /*
 691          * Identify entry point section and make entry relative to section
 692          * start.
 693          */
 694         entry = pi->ehdr->e_entry;
 695         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 696                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 697                         continue;
 698
 699                 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
 700                         continue;
 701
 702                 /* Make entry section relative */
 703                 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
 704                     ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
 705                      pi->ehdr->e_entry)) {
 706                         entry_sidx = i;
 707                         entry -= sechdrs[i].sh_addr;
 708                         break;
 709                 }
 710         }
 711
 712         /* Determine how much memory is needed to load relocatable object. */
 713         bss_align = 1;
 714         bss_sz = 0;
 715
 716         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 717                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 718                         continue;
 719
 720                 align = sechdrs[i].sh_addralign;
 721                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 722                         if (kbuf.buf_align < align)
 723                                 kbuf.buf_align = align;
 724                         kbuf.bufsz = ALIGN(kbuf.bufsz, align);
 725                         kbuf.bufsz += sechdrs[i].sh_size;
 726                 } else {
 727                         /* bss section */
 728                         if (bss_align < align)
 729                                 bss_align = align;
 730                         bss_sz = ALIGN(bss_sz, align);
 731                         bss_sz += sechdrs[i].sh_size;
 732                 }
 733         }
 734
 735         /* Determine the bss padding required to align bss properly */
 736         bss_pad = 0;
 737         if (kbuf.bufsz & (bss_align - 1))
 738                 bss_pad = bss_align - (kbuf.bufsz & (bss_align - 1));
 739
 740         kbuf.memsz = kbuf.bufsz + bss_pad + bss_sz;
 741
 742         /* Allocate buffer for purgatory */
 743         kbuf.buffer = vzalloc(kbuf.bufsz);
 744         if (!kbuf.buffer) {
 745                 ret = -ENOMEM;
 746                 goto out;
 747         }
 748
 749         if (kbuf.buf_align < bss_align)
 750                 kbuf.buf_align = bss_align;
 751
 752         /* Add buffer to segment list */
 753         ret = kexec_add_buffer(&kbuf);
 754         if (ret)
 755                 goto out;
 756         pi->purgatory_load_addr = kbuf.mem;
 757
 758         /* Load SHF_ALLOC sections */
 759         buf_addr = kbuf.buffer;
 760         load_addr = curr_load_addr = pi->purgatory_load_addr;
 761         bss_addr = load_addr + kbuf.bufsz + bss_pad;
 762
 763         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 764                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 765                         continue;
 766
 767                 align = sechdrs[i].sh_addralign;
 768                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 769                         curr_load_addr = ALIGN(curr_load_addr, align);
 770                         offset = curr_load_addr - load_addr;
 771                         /* We already modifed ->sh_offset to keep src addr */
 772                         src = (char *) sechdrs[i].sh_offset;
 773                         memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
 774
 775                         /* Store load address and source address of section */
 776                         sechdrs[i].sh_addr = curr_load_addr;
 777
 778                         /*
 779                          * This section got copied to temporary buffer. Update
 780                          * ->sh_offset accordingly.
 781                          */
 782                         sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
 783
 784                         /* Advance to the next address */
 785                         curr_load_addr += sechdrs[i].sh_size;
 786                 } else {
 787                         bss_addr = ALIGN(bss_addr, align);
 788                         sechdrs[i].sh_addr = bss_addr;
 789                         bss_addr += sechdrs[i].sh_size;
 790                 }
 791         }
 792
 793         /* Update entry point based on load address of text section */
 794         if (entry_sidx >= 0)
 795                 entry += sechdrs[entry_sidx].sh_addr;
 796
 797         /* Make kernel jump to purgatory after shutdown */
 798         image->start = entry;
 799
 800         /* Used later to get/set symbol values */
 801         pi->sechdrs = sechdrs;
 802
 803         /*
 804          * Used later to identify which section is purgatory and skip it
 805          * from checksumming.
 806          */
 807         pi->purgatory_buf = kbuf.buffer;
 808         return ret;
 809 out:
 810         vfree(sechdrs);
 811         vfree(kbuf.buffer);
 812         return ret;
 813 }
 814
 815 static int kexec_apply_relocations(struct kimage *image)
 816 {
 817         int i, ret;
 818         struct purgatory_info *pi = &image->purgatory_info;
 819         Elf_Shdr *sechdrs = pi->sechdrs;
 820
 821         /* Apply relocations */
 822         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 823                 Elf_Shdr *section, *symtab;
 824
 825                 if (sechdrs[i].sh_type != SHT_RELA &&
 826                     sechdrs[i].sh_type != SHT_REL)
 827                         continue;
 828
 829                 /*
 830                  * For section of type SHT_RELA/SHT_REL,
 831                  * ->sh_link contains section header index of associated
 832                  * symbol table. And ->sh_info contains section header
 833                  * index of section to which relocations apply.
 834                  */
 835                 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
 836                     sechdrs[i].sh_link >= pi->ehdr->e_shnum)
 837                         return -ENOEXEC;
 838
 839                 section = &sechdrs[sechdrs[i].sh_info];
 840                 symtab = &sechdrs[sechdrs[i].sh_link];
 841
 842                 if (!(section->sh_flags & SHF_ALLOC))
 843                         continue;
 844
 845                 /*
 846                  * symtab->sh_link contain section header index of associated
 847                  * string table.
 848                  */
 849                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 850                         /* Invalid section number? */
 851                         continue;
 852
 853                 /*
 854                  * Respective architecture needs to provide support for applying
 855                  * relocations of type SHT_RELA/SHT_REL.
 856                  */
 857                 if (sechdrs[i].sh_type == SHT_RELA)
 858                         ret = arch_kexec_apply_relocations_add(pi->ehdr,
 859                                                                sechdrs, i);
 860                 else if (sechdrs[i].sh_type == SHT_REL)
 861                         ret = arch_kexec_apply_relocations(pi->ehdr,
 862                                                            sechdrs, i);
 863                 if (ret)
 864                         return ret;
 865         }
 866
 867         return 0;
 868 }
 869
 870 /* Load relocatable purgatory object and relocate it appropriately */
 871 int kexec_load_purgatory(struct kimage *image, unsigned long min,
 872                          unsigned long max, int top_down,
 873                          unsigned long *load_addr)
 874 {
 875         struct purgatory_info *pi = &image->purgatory_info;
 876         int ret;
 877
 878         if (kexec_purgatory_size <= 0)
 879                 return -EINVAL;
 880
 881         if (kexec_purgatory_size < sizeof(Elf_Ehdr))
 882                 return -ENOEXEC;
 883
 884         pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
 885
 886         if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
 887             || pi->ehdr->e_type != ET_REL
 888             || !elf_check_arch(pi->ehdr)
 889             || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
 890                 return -ENOEXEC;
 891
 892         if (pi->ehdr->e_shoff >= kexec_purgatory_size
 893             || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
 894             kexec_purgatory_size - pi->ehdr->e_shoff))
 895                 return -ENOEXEC;
 896
 897         ret = __kexec_load_purgatory(image, min, max, top_down);
 898         if (ret)
 899                 return ret;
 900
 901         ret = kexec_apply_relocations(image);
 902         if (ret)
 903                 goto out;
 904
 905         *load_addr = pi->purgatory_load_addr;
 906         return 0;
 907 out:
 908         vfree(pi->sechdrs);
 909         pi->sechdrs = NULL;
 910
 911         vfree(pi->purgatory_buf);
 912         pi->purgatory_buf = NULL;
 913         return ret;
 914 }
 915
 916 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 917                                             const char *name)
 918 {
 919         Elf_Sym *syms;
 920         Elf_Shdr *sechdrs;
 921         Elf_Ehdr *ehdr;
 922         int i, k;
 923         const char *strtab;
 924
 925         if (!pi->sechdrs || !pi->ehdr)
 926                 return NULL;
 927
 928         sechdrs = pi->sechdrs;
 929         ehdr = pi->ehdr;
 930
 931         for (i = 0; i < ehdr->e_shnum; i++) {
 932                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 933                         continue;
 934
 935                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 936                         /* Invalid strtab section number */
 937                         continue;
 938                 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
 939                 syms = (Elf_Sym *)sechdrs[i].sh_offset;
 940
 941                 /* Go through symbols for a match */
 942                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 943                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 944                                 continue;
 945
 946                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 947                                 continue;
 948
 949                         if (syms[k].st_shndx == SHN_UNDEF ||
 950                             syms[k].st_shndx >= ehdr->e_shnum) {
 951                                 pr_debug("Symbol: %s has bad section index %d.\n",
 952                                                 name, syms[k].st_shndx);
 953                                 return NULL;
 954                         }
 955
 956                         /* Found the symbol we are looking for */
 957                         return &syms[k];
 958                 }
 959         }
 960
 961         return NULL;
 962 }
 963
 964 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
 965 {
 966         struct purgatory_info *pi = &image->purgatory_info;
 967         Elf_Sym *sym;
 968         Elf_Shdr *sechdr;
 969
 970         sym = kexec_purgatory_find_symbol(pi, name);
 971         if (!sym)
 972                 return ERR_PTR(-EINVAL);
 973
 974         sechdr = &pi->sechdrs[sym->st_shndx];
 975
 976         /*
 977          * Returns the address where symbol will finally be loaded after
 978          * kexec_load_segment()
 979          */
 980         return (void *)(sechdr->sh_addr + sym->st_value);
 981 }
 982
 983 /*
 984  * Get or set value of a symbol. If "get_value" is true, symbol value is
 985  * returned in buf otherwise symbol value is set based on value in buf.
 986  */
 987 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
 988                                    void *buf, unsigned int size, bool get_value)
 989 {
 990         Elf_Sym *sym;
 991         Elf_Shdr *sechdrs;
 992         struct purgatory_info *pi = &image->purgatory_info;
 993         char *sym_buf;
 994
 995         sym = kexec_purgatory_find_symbol(pi, name);
 996         if (!sym)
 997                 return -EINVAL;
 998
 999         if (sym->st_size != size) {
1000                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1001                        name, (unsigned long)sym->st_size, size);
1002                 return -EINVAL;
1003         }
1004
1005         sechdrs = pi->sechdrs;
1006
1007         if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
1008                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1009                        get_value ? "get" : "set");
1010                 return -EINVAL;
1011         }
1012
1013         sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
1014                                         sym->st_value;
1015
1016         if (get_value)
1017                 memcpy((void *)buf, sym_buf, size);
1018         else
1019                 memcpy((void *)sym_buf, buf, size);
1020
1021         return 0;
1022 }