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