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 <crypto/hash.h>
  22 #include <crypto/sha.h>
  23 #include <linux/syscalls.h>
  24 #include <linux/vmalloc.h>
  25 #include "kexec_internal.h"
  26
  27 /*
  28  * Declare these symbols weak so that if architecture provides a purgatory,
  29  * these will be overridden.
  30  */
  31 char __weak kexec_purgatory[0];
  32 size_t __weak kexec_purgatory_size = 0;
  33
  34 static int kexec_calculate_store_digests(struct kimage *image);
  35
  36 static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len)
  37 {
  38         struct fd f = fdget(fd);
  39         int ret;
  40         struct kstat stat;
  41         loff_t pos;
  42         ssize_t bytes = 0;
  43
  44         if (!f.file)
  45                 return -EBADF;
  46
  47         ret = vfs_getattr(&f.file->f_path, &stat);
  48         if (ret)
  49                 goto out;
  50
  51         if (stat.size > INT_MAX) {
  52                 ret = -EFBIG;
  53                 goto out;
  54         }
  55
  56         /* Don't hand 0 to vmalloc, it whines. */
  57         if (stat.size == 0) {
  58                 ret = -EINVAL;
  59                 goto out;
  60         }
  61
  62         *buf = vmalloc(stat.size);
  63         if (!*buf) {
  64                 ret = -ENOMEM;
  65                 goto out;
  66         }
  67
  68         pos = 0;
  69         while (pos < stat.size) {
  70                 bytes = kernel_read(f.file, pos, (char *)(*buf) + pos,
  71                                     stat.size - pos);
  72                 if (bytes < 0) {
  73                         vfree(*buf);
  74                         ret = bytes;
  75                         goto out;
  76                 }
  77
  78                 if (bytes == 0)
  79                         break;
  80                 pos += bytes;
  81         }
  82
  83         if (pos != stat.size) {
  84                 ret = -EBADF;
  85                 vfree(*buf);
  86                 goto out;
  87         }
  88
  89         *buf_len = pos;
  90 out:
  91         fdput(f);
  92         return ret;
  93 }
  94
  95 /* Architectures can provide this probe function */
  96 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  97                                          unsigned long buf_len)
  98 {
  99         return -ENOEXEC;
 100 }
 101
 102 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
 103 {
 104         return ERR_PTR(-ENOEXEC);
 105 }
 106
 107 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
 108 {
 109         return -EINVAL;
 110 }
 111
 112 #ifdef CONFIG_KEXEC_VERIFY_SIG
 113 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
 114                                         unsigned long buf_len)
 115 {
 116         return -EKEYREJECTED;
 117 }
 118 #endif
 119
 120 /* Apply relocations of type RELA */
 121 int __weak
 122 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
 123                                  unsigned int relsec)
 124 {
 125         pr_err("RELA relocation unsupported.\n");
 126         return -ENOEXEC;
 127 }
 128
 129 /* Apply relocations of type REL */
 130 int __weak
 131 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
 132                              unsigned int relsec)
 133 {
 134         pr_err("REL relocation unsupported.\n");
 135         return -ENOEXEC;
 136 }
 137
 138 /*
 139  * Free up memory used by kernel, initrd, and command line. This is temporary
 140  * memory allocation which is not needed any more after these buffers have
 141  * been loaded into separate segments and have been copied elsewhere.
 142  */
 143 void kimage_file_post_load_cleanup(struct kimage *image)
 144 {
 145         struct purgatory_info *pi = &image->purgatory_info;
 146
 147         vfree(image->kernel_buf);
 148         image->kernel_buf = NULL;
 149
 150         vfree(image->initrd_buf);
 151         image->initrd_buf = NULL;
 152
 153         kfree(image->cmdline_buf);
 154         image->cmdline_buf = NULL;
 155
 156         vfree(pi->purgatory_buf);
 157         pi->purgatory_buf = NULL;
 158
 159         vfree(pi->sechdrs);
 160         pi->sechdrs = NULL;
 161
 162         /* See if architecture has anything to cleanup post load */
 163         arch_kimage_file_post_load_cleanup(image);
 164
 165         /*
 166          * Above call should have called into bootloader to free up
 167          * any data stored in kimage->image_loader_data. It should
 168          * be ok now to free it up.
 169          */
 170         kfree(image->image_loader_data);
 171         image->image_loader_data = NULL;
 172 }
 173
 174 /*
 175  * In file mode list of segments is prepared by kernel. Copy relevant
 176  * data from user space, do error checking, prepare segment list
 177  */
 178 static int
 179 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 180                              const char __user *cmdline_ptr,
 181                              unsigned long cmdline_len, unsigned flags)
 182 {
 183         int ret = 0;
 184         void *ldata;
 185
 186         ret = copy_file_from_fd(kernel_fd, &image->kernel_buf,
 187                                 &image->kernel_buf_len);
 188         if (ret)
 189                 return ret;
 190
 191         /* Call arch image probe handlers */
 192         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 193                                             image->kernel_buf_len);
 194
 195         if (ret)
 196                 goto out;
 197
 198 #ifdef CONFIG_KEXEC_VERIFY_SIG
 199         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 200                                            image->kernel_buf_len);
 201         if (ret) {
 202                 pr_debug("kernel signature verification failed.\n");
 203                 goto out;
 204         }
 205         pr_debug("kernel signature verification successful.\n");
 206 #endif
 207         /* It is possible that there no initramfs is being loaded */
 208         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 209                 ret = copy_file_from_fd(initrd_fd, &image->initrd_buf,
 210                                         &image->initrd_buf_len);
 211                 if (ret)
 212                         goto out;
 213         }
 214
 215         if (cmdline_len) {
 216                 image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
 217                 if (!image->cmdline_buf) {
 218                         ret = -ENOMEM;
 219                         goto out;
 220                 }
 221
 222                 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
 223                                      cmdline_len);
 224                 if (ret) {
 225                         ret = -EFAULT;
 226                         goto out;
 227                 }
 228
 229                 image->cmdline_buf_len = cmdline_len;
 230
 231                 /* command line should be a string with last byte null */
 232                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 233                         ret = -EINVAL;
 234                         goto out;
 235                 }
 236         }
 237
 238         /* Call arch image load handlers */
 239         ldata = arch_kexec_kernel_image_load(image);
 240
 241         if (IS_ERR(ldata)) {
 242                 ret = PTR_ERR(ldata);
 243                 goto out;
 244         }
 245
 246         image->image_loader_data = ldata;
 247 out:
 248         /* In case of error, free up all allocated memory in this function */
 249         if (ret)
 250                 kimage_file_post_load_cleanup(image);
 251         return ret;
 252 }
 253
 254 static int
 255 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 256                        int initrd_fd, const char __user *cmdline_ptr,
 257                        unsigned long cmdline_len, unsigned long flags)
 258 {
 259         int ret;
 260         struct kimage *image;
 261         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 262
 263         image = do_kimage_alloc_init();
 264         if (!image)
 265                 return -ENOMEM;
 266
 267         image->file_mode = 1;
 268
 269         if (kexec_on_panic) {
 270                 /* Enable special crash kernel control page alloc policy. */
 271                 image->control_page = crashk_res.start;
 272                 image->type = KEXEC_TYPE_CRASH;
 273         }
 274
 275         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 276                                            cmdline_ptr, cmdline_len, flags);
 277         if (ret)
 278                 goto out_free_image;
 279
 280         ret = sanity_check_segment_list(image);
 281         if (ret)
 282                 goto out_free_post_load_bufs;
 283
 284         ret = -ENOMEM;
 285         image->control_code_page = kimage_alloc_control_pages(image,
 286                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 287         if (!image->control_code_page) {
 288                 pr_err("Could not allocate control_code_buffer\n");
 289                 goto out_free_post_load_bufs;
 290         }
 291
 292         if (!kexec_on_panic) {
 293                 image->swap_page = kimage_alloc_control_pages(image, 0);
 294                 if (!image->swap_page) {
 295                         pr_err("Could not allocate swap buffer\n");
 296                         goto out_free_control_pages;
 297                 }
 298         }
 299
 300         *rimage = image;
 301         return 0;
 302 out_free_control_pages:
 303         kimage_free_page_list(&image->control_pages);
 304 out_free_post_load_bufs:
 305         kimage_file_post_load_cleanup(image);
 306 out_free_image:
 307         kfree(image);
 308         return ret;
 309 }
 310
 311 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 312                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 313                 unsigned long, flags)
 314 {
 315         int ret = 0, i;
 316         struct kimage **dest_image, *image;
 317
 318         /* We only trust the superuser with rebooting the system. */
 319         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 320                 return -EPERM;
 321
 322         /* Make sure we have a legal set of flags */
 323         if (flags != (flags & KEXEC_FILE_FLAGS))
 324                 return -EINVAL;
 325
 326         image = NULL;
 327
 328         if (!mutex_trylock(&kexec_mutex))
 329                 return -EBUSY;
 330
 331         dest_image = &kexec_image;
 332         if (flags & KEXEC_FILE_ON_CRASH) {
 333                 dest_image = &kexec_crash_image;
 334                 if (kexec_crash_image)
 335                         arch_kexec_unprotect_crashkres();
 336         }
 337
 338         if (flags & KEXEC_FILE_UNLOAD)
 339                 goto exchange;
 340
 341         /*
 342          * In case of crash, new kernel gets loaded in reserved region. It is
 343          * same memory where old crash kernel might be loaded. Free any
 344          * current crash dump kernel before we corrupt it.
 345          */
 346         if (flags & KEXEC_FILE_ON_CRASH)
 347                 kimage_free(xchg(&kexec_crash_image, NULL));
 348
 349         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 350                                      cmdline_len, flags);
 351         if (ret)
 352                 goto out;
 353
 354         ret = machine_kexec_prepare(image);
 355         if (ret)
 356                 goto out;
 357
 358         ret = kexec_calculate_store_digests(image);
 359         if (ret)
 360                 goto out;
 361
 362         for (i = 0; i < image->nr_segments; i++) {
 363                 struct kexec_segment *ksegment;
 364
 365                 ksegment = &image->segment[i];
 366                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 367                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 368                          ksegment->memsz);
 369
 370                 ret = kimage_load_segment(image, &image->segment[i]);
 371                 if (ret)
 372                         goto out;
 373         }
 374
 375         kimage_terminate(image);
 376
 377         /*
 378          * Free up any temporary buffers allocated which are not needed
 379          * after image has been loaded
 380          */
 381         kimage_file_post_load_cleanup(image);
 382 exchange:
 383         image = xchg(dest_image, image);
 384 out:
 385         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 386                 arch_kexec_protect_crashkres();
 387
 388         mutex_unlock(&kexec_mutex);
 389         kimage_free(image);
 390         return ret;
 391 }
 392
 393 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 394                                     struct kexec_buf *kbuf)
 395 {
 396         struct kimage *image = kbuf->image;
 397         unsigned long temp_start, temp_end;
 398
 399         temp_end = min(end, kbuf->buf_max);
 400         temp_start = temp_end - kbuf->memsz;
 401
 402         do {
 403                 /* align down start */
 404                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 405
 406                 if (temp_start < start || temp_start < kbuf->buf_min)
 407                         return 0;
 408
 409                 temp_end = temp_start + kbuf->memsz - 1;
 410
 411                 /*
 412                  * Make sure this does not conflict with any of existing
 413                  * segments
 414                  */
 415                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 416                         temp_start = temp_start - PAGE_SIZE;
 417                         continue;
 418                 }
 419
 420                 /* We found a suitable memory range */
 421                 break;
 422         } while (1);
 423
 424         /* If we are here, we found a suitable memory range */
 425         kbuf->mem = temp_start;
 426
 427         /* Success, stop navigating through remaining System RAM ranges */
 428         return 1;
 429 }
 430
 431 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 432                                      struct kexec_buf *kbuf)
 433 {
 434         struct kimage *image = kbuf->image;
 435         unsigned long temp_start, temp_end;
 436
 437         temp_start = max(start, kbuf->buf_min);
 438
 439         do {
 440                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 441                 temp_end = temp_start + kbuf->memsz - 1;
 442
 443                 if (temp_end > end || temp_end > kbuf->buf_max)
 444                         return 0;
 445                 /*
 446                  * Make sure this does not conflict with any of existing
 447                  * segments
 448                  */
 449                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 450                         temp_start = temp_start + PAGE_SIZE;
 451                         continue;
 452                 }
 453
 454                 /* We found a suitable memory range */
 455                 break;
 456         } while (1);
 457
 458         /* If we are here, we found a suitable memory range */
 459         kbuf->mem = temp_start;
 460
 461         /* Success, stop navigating through remaining System RAM ranges */
 462         return 1;
 463 }
 464
 465 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
 466 {
 467         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 468         unsigned long sz = end - start + 1;
 469
 470         /* Returning 0 will take to next memory range */
 471         if (sz < kbuf->memsz)
 472                 return 0;
 473
 474         if (end < kbuf->buf_min || start > kbuf->buf_max)
 475                 return 0;
 476
 477         /*
 478          * Allocate memory top down with-in ram range. Otherwise bottom up
 479          * allocation.
 480          */
 481         if (kbuf->top_down)
 482                 return locate_mem_hole_top_down(start, end, kbuf);
 483         return locate_mem_hole_bottom_up(start, end, kbuf);
 484 }
 485
 486 /*
 487  * Helper function for placing a buffer in a kexec segment. This assumes
 488  * that kexec_mutex is held.
 489  */
 490 int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
 491                      unsigned long memsz, unsigned long buf_align,
 492                      unsigned long buf_min, unsigned long buf_max,
 493                      bool top_down, unsigned long *load_addr)
 494 {
 495
 496         struct kexec_segment *ksegment;
 497         struct kexec_buf buf, *kbuf;
 498         int ret;
 499
 500         /* Currently adding segment this way is allowed only in file mode */
 501         if (!image->file_mode)
 502                 return -EINVAL;
 503
 504         if (image->nr_segments >= KEXEC_SEGMENT_MAX)
 505                 return -EINVAL;
 506
 507         /*
 508          * Make sure we are not trying to add buffer after allocating
 509          * control pages. All segments need to be placed first before
 510          * any control pages are allocated. As control page allocation
 511          * logic goes through list of segments to make sure there are
 512          * no destination overlaps.
 513          */
 514         if (!list_empty(&image->control_pages)) {
 515                 WARN_ON(1);
 516                 return -EINVAL;
 517         }
 518
 519         memset(&buf, 0, sizeof(struct kexec_buf));
 520         kbuf = &buf;
 521         kbuf->image = image;
 522         kbuf->buffer = buffer;
 523         kbuf->bufsz = bufsz;
 524
 525         kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
 526         kbuf->buf_align = max(buf_align, PAGE_SIZE);
 527         kbuf->buf_min = buf_min;
 528         kbuf->buf_max = buf_max;
 529         kbuf->top_down = top_down;
 530
 531         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 532         if (image->type == KEXEC_TYPE_CRASH)
 533                 ret = walk_iomem_res_desc(crashk_res.desc,
 534                                 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 535                                 crashk_res.start, crashk_res.end, kbuf,
 536                                 locate_mem_hole_callback);
 537         else
 538                 ret = walk_system_ram_res(0, -1, kbuf,
 539                                           locate_mem_hole_callback);
 540         if (ret != 1) {
 541                 /* A suitable memory range could not be found for buffer */
 542                 return -EADDRNOTAVAIL;
 543         }
 544
 545         /* Found a suitable memory range */
 546         ksegment = &image->segment[image->nr_segments];
 547         ksegment->kbuf = kbuf->buffer;
 548         ksegment->bufsz = kbuf->bufsz;
 549         ksegment->mem = kbuf->mem;
 550         ksegment->memsz = kbuf->memsz;
 551         image->nr_segments++;
 552         *load_addr = ksegment->mem;
 553         return 0;
 554 }
 555
 556 /* Calculate and store the digest of segments */
 557 static int kexec_calculate_store_digests(struct kimage *image)
 558 {
 559         struct crypto_shash *tfm;
 560         struct shash_desc *desc;
 561         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 562         size_t desc_size, nullsz;
 563         char *digest;
 564         void *zero_buf;
 565         struct kexec_sha_region *sha_regions;
 566         struct purgatory_info *pi = &image->purgatory_info;
 567
 568         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 569         zero_buf_sz = PAGE_SIZE;
 570
 571         tfm = crypto_alloc_shash("sha256", 0, 0);
 572         if (IS_ERR(tfm)) {
 573                 ret = PTR_ERR(tfm);
 574                 goto out;
 575         }
 576
 577         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 578         desc = kzalloc(desc_size, GFP_KERNEL);
 579         if (!desc) {
 580                 ret = -ENOMEM;
 581                 goto out_free_tfm;
 582         }
 583
 584         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 585         sha_regions = vzalloc(sha_region_sz);
 586         if (!sha_regions)
 587                 goto out_free_desc;
 588
 589         desc->tfm   = tfm;
 590         desc->flags = 0;
 591
 592         ret = crypto_shash_init(desc);
 593         if (ret < 0)
 594                 goto out_free_sha_regions;
 595
 596         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 597         if (!digest) {
 598                 ret = -ENOMEM;
 599                 goto out_free_sha_regions;
 600         }
 601
 602         for (j = i = 0; i < image->nr_segments; i++) {
 603                 struct kexec_segment *ksegment;
 604
 605                 ksegment = &image->segment[i];
 606                 /*
 607                  * Skip purgatory as it will be modified once we put digest
 608                  * info in purgatory.
 609                  */
 610                 if (ksegment->kbuf == pi->purgatory_buf)
 611                         continue;
 612
 613                 ret = crypto_shash_update(desc, ksegment->kbuf,
 614                                           ksegment->bufsz);
 615                 if (ret)
 616                         break;
 617
 618                 /*
 619                  * Assume rest of the buffer is filled with zero and
 620                  * update digest accordingly.
 621                  */
 622                 nullsz = ksegment->memsz - ksegment->bufsz;
 623                 while (nullsz) {
 624                         unsigned long bytes = nullsz;
 625
 626                         if (bytes > zero_buf_sz)
 627                                 bytes = zero_buf_sz;
 628                         ret = crypto_shash_update(desc, zero_buf, bytes);
 629                         if (ret)
 630                                 break;
 631                         nullsz -= bytes;
 632                 }
 633
 634                 if (ret)
 635                         break;
 636
 637                 sha_regions[j].start = ksegment->mem;
 638                 sha_regions[j].len = ksegment->memsz;
 639                 j++;
 640         }
 641
 642         if (!ret) {
 643                 ret = crypto_shash_final(desc, digest);
 644                 if (ret)
 645                         goto out_free_digest;
 646                 ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
 647                                                 sha_regions, sha_region_sz, 0);
 648                 if (ret)
 649                         goto out_free_digest;
 650
 651                 ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
 652                                                 digest, SHA256_DIGEST_SIZE, 0);
 653                 if (ret)
 654                         goto out_free_digest;
 655         }
 656
 657 out_free_digest:
 658         kfree(digest);
 659 out_free_sha_regions:
 660         vfree(sha_regions);
 661 out_free_desc:
 662         kfree(desc);
 663 out_free_tfm:
 664         kfree(tfm);
 665 out:
 666         return ret;
 667 }
 668
 669 /* Actually load purgatory. Lot of code taken from kexec-tools */
 670 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
 671                                   unsigned long max, int top_down)
 672 {
 673         struct purgatory_info *pi = &image->purgatory_info;
 674         unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
 675         unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
 676         unsigned char *buf_addr, *src;
 677         int i, ret = 0, entry_sidx = -1;
 678         const Elf_Shdr *sechdrs_c;
 679         Elf_Shdr *sechdrs = NULL;
 680         void *purgatory_buf = NULL;
 681
 682         /*
 683          * sechdrs_c points to section headers in purgatory and are read
 684          * only. No modifications allowed.
 685          */
 686         sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
 687
 688         /*
 689          * We can not modify sechdrs_c[] and its fields. It is read only.
 690          * Copy it over to a local copy where one can store some temporary
 691          * data and free it at the end. We need to modify ->sh_addr and
 692          * ->sh_offset fields to keep track of permanent and temporary
 693          * locations of sections.
 694          */
 695         sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 696         if (!sechdrs)
 697                 return -ENOMEM;
 698
 699         memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 700
 701         /*
 702          * We seem to have multiple copies of sections. First copy is which
 703          * is embedded in kernel in read only section. Some of these sections
 704          * will be copied to a temporary buffer and relocated. And these
 705          * sections will finally be copied to their final destination at
 706          * segment load time.
 707          *
 708          * Use ->sh_offset to reflect section address in memory. It will
 709          * point to original read only copy if section is not allocatable.
 710          * Otherwise it will point to temporary copy which will be relocated.
 711          *
 712          * Use ->sh_addr to contain final address of the section where it
 713          * will go during execution time.
 714          */
 715         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 716                 if (sechdrs[i].sh_type == SHT_NOBITS)
 717                         continue;
 718
 719                 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
 720                                                 sechdrs[i].sh_offset;
 721         }
 722
 723         /*
 724          * Identify entry point section and make entry relative to section
 725          * start.
 726          */
 727         entry = pi->ehdr->e_entry;
 728         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 729                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 730                         continue;
 731
 732                 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
 733                         continue;
 734
 735                 /* Make entry section relative */
 736                 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
 737                     ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
 738                      pi->ehdr->e_entry)) {
 739                         entry_sidx = i;
 740                         entry -= sechdrs[i].sh_addr;
 741                         break;
 742                 }
 743         }
 744
 745         /* Determine how much memory is needed to load relocatable object. */
 746         buf_align = 1;
 747         bss_align = 1;
 748         buf_sz = 0;
 749         bss_sz = 0;
 750
 751         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 752                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 753                         continue;
 754
 755                 align = sechdrs[i].sh_addralign;
 756                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 757                         if (buf_align < align)
 758                                 buf_align = align;
 759                         buf_sz = ALIGN(buf_sz, align);
 760                         buf_sz += sechdrs[i].sh_size;
 761                 } else {
 762                         /* bss section */
 763                         if (bss_align < align)
 764                                 bss_align = align;
 765                         bss_sz = ALIGN(bss_sz, align);
 766                         bss_sz += sechdrs[i].sh_size;
 767                 }
 768         }
 769
 770         /* Determine the bss padding required to align bss properly */
 771         bss_pad = 0;
 772         if (buf_sz & (bss_align - 1))
 773                 bss_pad = bss_align - (buf_sz & (bss_align - 1));
 774
 775         memsz = buf_sz + bss_pad + bss_sz;
 776
 777         /* Allocate buffer for purgatory */
 778         purgatory_buf = vzalloc(buf_sz);
 779         if (!purgatory_buf) {
 780                 ret = -ENOMEM;
 781                 goto out;
 782         }
 783
 784         if (buf_align < bss_align)
 785                 buf_align = bss_align;
 786
 787         /* Add buffer to segment list */
 788         ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
 789                                 buf_align, min, max, top_down,
 790                                 &pi->purgatory_load_addr);
 791         if (ret)
 792                 goto out;
 793
 794         /* Load SHF_ALLOC sections */
 795         buf_addr = purgatory_buf;
 796         load_addr = curr_load_addr = pi->purgatory_load_addr;
 797         bss_addr = load_addr + buf_sz + bss_pad;
 798
 799         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 800                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 801                         continue;
 802
 803                 align = sechdrs[i].sh_addralign;
 804                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 805                         curr_load_addr = ALIGN(curr_load_addr, align);
 806                         offset = curr_load_addr - load_addr;
 807                         /* We already modifed ->sh_offset to keep src addr */
 808                         src = (char *) sechdrs[i].sh_offset;
 809                         memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
 810
 811                         /* Store load address and source address of section */
 812                         sechdrs[i].sh_addr = curr_load_addr;
 813
 814                         /*
 815                          * This section got copied to temporary buffer. Update
 816                          * ->sh_offset accordingly.
 817                          */
 818                         sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
 819
 820                         /* Advance to the next address */
 821                         curr_load_addr += sechdrs[i].sh_size;
 822                 } else {
 823                         bss_addr = ALIGN(bss_addr, align);
 824                         sechdrs[i].sh_addr = bss_addr;
 825                         bss_addr += sechdrs[i].sh_size;
 826                 }
 827         }
 828
 829         /* Update entry point based on load address of text section */
 830         if (entry_sidx >= 0)
 831                 entry += sechdrs[entry_sidx].sh_addr;
 832
 833         /* Make kernel jump to purgatory after shutdown */
 834         image->start = entry;
 835
 836         /* Used later to get/set symbol values */
 837         pi->sechdrs = sechdrs;
 838
 839         /*
 840          * Used later to identify which section is purgatory and skip it
 841          * from checksumming.
 842          */
 843         pi->purgatory_buf = purgatory_buf;
 844         return ret;
 845 out:
 846         vfree(sechdrs);
 847         vfree(purgatory_buf);
 848         return ret;
 849 }
 850
 851 static int kexec_apply_relocations(struct kimage *image)
 852 {
 853         int i, ret;
 854         struct purgatory_info *pi = &image->purgatory_info;
 855         Elf_Shdr *sechdrs = pi->sechdrs;
 856
 857         /* Apply relocations */
 858         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 859                 Elf_Shdr *section, *symtab;
 860
 861                 if (sechdrs[i].sh_type != SHT_RELA &&
 862                     sechdrs[i].sh_type != SHT_REL)
 863                         continue;
 864
 865                 /*
 866                  * For section of type SHT_RELA/SHT_REL,
 867                  * ->sh_link contains section header index of associated
 868                  * symbol table. And ->sh_info contains section header
 869                  * index of section to which relocations apply.
 870                  */
 871                 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
 872                     sechdrs[i].sh_link >= pi->ehdr->e_shnum)
 873                         return -ENOEXEC;
 874
 875                 section = &sechdrs[sechdrs[i].sh_info];
 876                 symtab = &sechdrs[sechdrs[i].sh_link];
 877
 878                 if (!(section->sh_flags & SHF_ALLOC))
 879                         continue;
 880
 881                 /*
 882                  * symtab->sh_link contain section header index of associated
 883                  * string table.
 884                  */
 885                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 886                         /* Invalid section number? */
 887                         continue;
 888
 889                 /*
 890                  * Respective architecture needs to provide support for applying
 891                  * relocations of type SHT_RELA/SHT_REL.
 892                  */
 893                 if (sechdrs[i].sh_type == SHT_RELA)
 894                         ret = arch_kexec_apply_relocations_add(pi->ehdr,
 895                                                                sechdrs, i);
 896                 else if (sechdrs[i].sh_type == SHT_REL)
 897                         ret = arch_kexec_apply_relocations(pi->ehdr,
 898                                                            sechdrs, i);
 899                 if (ret)
 900                         return ret;
 901         }
 902
 903         return 0;
 904 }
 905
 906 /* Load relocatable purgatory object and relocate it appropriately */
 907 int kexec_load_purgatory(struct kimage *image, unsigned long min,
 908                          unsigned long max, int top_down,
 909                          unsigned long *load_addr)
 910 {
 911         struct purgatory_info *pi = &image->purgatory_info;
 912         int ret;
 913
 914         if (kexec_purgatory_size <= 0)
 915                 return -EINVAL;
 916
 917         if (kexec_purgatory_size < sizeof(Elf_Ehdr))
 918                 return -ENOEXEC;
 919
 920         pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
 921
 922         if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
 923             || pi->ehdr->e_type != ET_REL
 924             || !elf_check_arch(pi->ehdr)
 925             || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
 926                 return -ENOEXEC;
 927
 928         if (pi->ehdr->e_shoff >= kexec_purgatory_size
 929             || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
 930             kexec_purgatory_size - pi->ehdr->e_shoff))
 931                 return -ENOEXEC;
 932
 933         ret = __kexec_load_purgatory(image, min, max, top_down);
 934         if (ret)
 935                 return ret;
 936
 937         ret = kexec_apply_relocations(image);
 938         if (ret)
 939                 goto out;
 940
 941         *load_addr = pi->purgatory_load_addr;
 942         return 0;
 943 out:
 944         vfree(pi->sechdrs);
 945         vfree(pi->purgatory_buf);
 946         return ret;
 947 }
 948
 949 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 950                                             const char *name)
 951 {
 952         Elf_Sym *syms;
 953         Elf_Shdr *sechdrs;
 954         Elf_Ehdr *ehdr;
 955         int i, k;
 956         const char *strtab;
 957
 958         if (!pi->sechdrs || !pi->ehdr)
 959                 return NULL;
 960
 961         sechdrs = pi->sechdrs;
 962         ehdr = pi->ehdr;
 963
 964         for (i = 0; i < ehdr->e_shnum; i++) {
 965                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 966                         continue;
 967
 968                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 969                         /* Invalid strtab section number */
 970                         continue;
 971                 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
 972                 syms = (Elf_Sym *)sechdrs[i].sh_offset;
 973
 974                 /* Go through symbols for a match */
 975                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 976                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 977                                 continue;
 978
 979                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 980                                 continue;
 981
 982                         if (syms[k].st_shndx == SHN_UNDEF ||
 983                             syms[k].st_shndx >= ehdr->e_shnum) {
 984                                 pr_debug("Symbol: %s has bad section index %d.\n",
 985                                                 name, syms[k].st_shndx);
 986                                 return NULL;
 987                         }
 988
 989                         /* Found the symbol we are looking for */
 990                         return &syms[k];
 991                 }
 992         }
 993
 994         return NULL;
 995 }
 996
 997 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
 998 {
 999         struct purgatory_info *pi = &image->purgatory_info;
1000         Elf_Sym *sym;
1001         Elf_Shdr *sechdr;
1002
1003         sym = kexec_purgatory_find_symbol(pi, name);
1004         if (!sym)
1005                 return ERR_PTR(-EINVAL);
1006
1007         sechdr = &pi->sechdrs[sym->st_shndx];
1008
1009         /*
1010          * Returns the address where symbol will finally be loaded after
1011          * kexec_load_segment()
1012          */
1013         return (void *)(sechdr->sh_addr + sym->st_value);
1014 }
1015
1016 /*
1017  * Get or set value of a symbol. If "get_value" is true, symbol value is
1018  * returned in buf otherwise symbol value is set based on value in buf.
1019  */
1020 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1021                                    void *buf, unsigned int size, bool get_value)
1022 {
1023         Elf_Sym *sym;
1024         Elf_Shdr *sechdrs;
1025         struct purgatory_info *pi = &image->purgatory_info;
1026         char *sym_buf;
1027
1028         sym = kexec_purgatory_find_symbol(pi, name);
1029         if (!sym)
1030                 return -EINVAL;
1031
1032         if (sym->st_size != size) {
1033                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1034                        name, (unsigned long)sym->st_size, size);
1035                 return -EINVAL;
1036         }
1037
1038         sechdrs = pi->sechdrs;
1039
1040         if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
1041                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1042                        get_value ? "get" : "set");
1043                 return -EINVAL;
1044         }
1045
1046         sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
1047                                         sym->st_value;
1048
1049         if (get_value)
1050                 memcpy((void *)buf, sym_buf, size);
1051         else
1052                 memcpy((void *)sym_buf, buf, size);
1053
1054         return 0;
1055 }