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 = kzalloc(cmdline_len, GFP_KERNEL);
 166                 if (!image->cmdline_buf) {
 167                         ret = -ENOMEM;
 168                         goto out;
 169                 }
 170
 171                 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
 172                                      cmdline_len);
 173                 if (ret) {
 174                         ret = -EFAULT;
 175                         goto out;
 176                 }
 177
 178                 image->cmdline_buf_len = cmdline_len;
 179
 180                 /* command line should be a string with last byte null */
 181                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 182                         ret = -EINVAL;
 183                         goto out;
 184                 }
 185         }
 186
 187         /* Call arch image load handlers */
 188         ldata = arch_kexec_kernel_image_load(image);
 189
 190         if (IS_ERR(ldata)) {
 191                 ret = PTR_ERR(ldata);
 192                 goto out;
 193         }
 194
 195         image->image_loader_data = ldata;
 196 out:
 197         /* In case of error, free up all allocated memory in this function */
 198         if (ret)
 199                 kimage_file_post_load_cleanup(image);
 200         return ret;
 201 }
 202
 203 static int
 204 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 205                        int initrd_fd, const char __user *cmdline_ptr,
 206                        unsigned long cmdline_len, unsigned long flags)
 207 {
 208         int ret;
 209         struct kimage *image;
 210         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 211
 212         image = do_kimage_alloc_init();
 213         if (!image)
 214                 return -ENOMEM;
 215
 216         image->file_mode = 1;
 217
 218         if (kexec_on_panic) {
 219                 /* Enable special crash kernel control page alloc policy. */
 220                 image->control_page = crashk_res.start;
 221                 image->type = KEXEC_TYPE_CRASH;
 222         }
 223
 224         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 225                                            cmdline_ptr, cmdline_len, flags);
 226         if (ret)
 227                 goto out_free_image;
 228
 229         ret = sanity_check_segment_list(image);
 230         if (ret)
 231                 goto out_free_post_load_bufs;
 232
 233         ret = -ENOMEM;
 234         image->control_code_page = kimage_alloc_control_pages(image,
 235                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 236         if (!image->control_code_page) {
 237                 pr_err("Could not allocate control_code_buffer\n");
 238                 goto out_free_post_load_bufs;
 239         }
 240
 241         if (!kexec_on_panic) {
 242                 image->swap_page = kimage_alloc_control_pages(image, 0);
 243                 if (!image->swap_page) {
 244                         pr_err("Could not allocate swap buffer\n");
 245                         goto out_free_control_pages;
 246                 }
 247         }
 248
 249         *rimage = image;
 250         return 0;
 251 out_free_control_pages:
 252         kimage_free_page_list(&image->control_pages);
 253 out_free_post_load_bufs:
 254         kimage_file_post_load_cleanup(image);
 255 out_free_image:
 256         kfree(image);
 257         return ret;
 258 }
 259
 260 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 261                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 262                 unsigned long, flags)
 263 {
 264         int ret = 0, i;
 265         struct kimage **dest_image, *image;
 266
 267         /* We only trust the superuser with rebooting the system. */
 268         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 269                 return -EPERM;
 270
 271         /* Make sure we have a legal set of flags */
 272         if (flags != (flags & KEXEC_FILE_FLAGS))
 273                 return -EINVAL;
 274
 275         image = NULL;
 276
 277         if (!mutex_trylock(&kexec_mutex))
 278                 return -EBUSY;
 279
 280         dest_image = &kexec_image;
 281         if (flags & KEXEC_FILE_ON_CRASH) {
 282                 dest_image = &kexec_crash_image;
 283                 if (kexec_crash_image)
 284                         arch_kexec_unprotect_crashkres();
 285         }
 286
 287         if (flags & KEXEC_FILE_UNLOAD)
 288                 goto exchange;
 289
 290         /*
 291          * In case of crash, new kernel gets loaded in reserved region. It is
 292          * same memory where old crash kernel might be loaded. Free any
 293          * current crash dump kernel before we corrupt it.
 294          */
 295         if (flags & KEXEC_FILE_ON_CRASH)
 296                 kimage_free(xchg(&kexec_crash_image, NULL));
 297
 298         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 299                                      cmdline_len, flags);
 300         if (ret)
 301                 goto out;
 302
 303         ret = machine_kexec_prepare(image);
 304         if (ret)
 305                 goto out;
 306
 307         ret = kexec_calculate_store_digests(image);
 308         if (ret)
 309                 goto out;
 310
 311         for (i = 0; i < image->nr_segments; i++) {
 312                 struct kexec_segment *ksegment;
 313
 314                 ksegment = &image->segment[i];
 315                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 316                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 317                          ksegment->memsz);
 318
 319                 ret = kimage_load_segment(image, &image->segment[i]);
 320                 if (ret)
 321                         goto out;
 322         }
 323
 324         kimage_terminate(image);
 325
 326         /*
 327          * Free up any temporary buffers allocated which are not needed
 328          * after image has been loaded
 329          */
 330         kimage_file_post_load_cleanup(image);
 331 exchange:
 332         image = xchg(dest_image, image);
 333 out:
 334         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 335                 arch_kexec_protect_crashkres();
 336
 337         mutex_unlock(&kexec_mutex);
 338         kimage_free(image);
 339         return ret;
 340 }
 341
 342 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 343                                     struct kexec_buf *kbuf)
 344 {
 345         struct kimage *image = kbuf->image;
 346         unsigned long temp_start, temp_end;
 347
 348         temp_end = min(end, kbuf->buf_max);
 349         temp_start = temp_end - kbuf->memsz;
 350
 351         do {
 352                 /* align down start */
 353                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 354
 355                 if (temp_start < start || temp_start < kbuf->buf_min)
 356                         return 0;
 357
 358                 temp_end = temp_start + kbuf->memsz - 1;
 359
 360                 /*
 361                  * Make sure this does not conflict with any of existing
 362                  * segments
 363                  */
 364                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 365                         temp_start = temp_start - PAGE_SIZE;
 366                         continue;
 367                 }
 368
 369                 /* We found a suitable memory range */
 370                 break;
 371         } while (1);
 372
 373         /* If we are here, we found a suitable memory range */
 374         kbuf->mem = temp_start;
 375
 376         /* Success, stop navigating through remaining System RAM ranges */
 377         return 1;
 378 }
 379
 380 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 381                                      struct kexec_buf *kbuf)
 382 {
 383         struct kimage *image = kbuf->image;
 384         unsigned long temp_start, temp_end;
 385
 386         temp_start = max(start, kbuf->buf_min);
 387
 388         do {
 389                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 390                 temp_end = temp_start + kbuf->memsz - 1;
 391
 392                 if (temp_end > end || temp_end > kbuf->buf_max)
 393                         return 0;
 394                 /*
 395                  * Make sure this does not conflict with any of existing
 396                  * segments
 397                  */
 398                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 399                         temp_start = temp_start + PAGE_SIZE;
 400                         continue;
 401                 }
 402
 403                 /* We found a suitable memory range */
 404                 break;
 405         } while (1);
 406
 407         /* If we are here, we found a suitable memory range */
 408         kbuf->mem = temp_start;
 409
 410         /* Success, stop navigating through remaining System RAM ranges */
 411         return 1;
 412 }
 413
 414 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
 415 {
 416         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 417         unsigned long sz = end - start + 1;
 418
 419         /* Returning 0 will take to next memory range */
 420         if (sz < kbuf->memsz)
 421                 return 0;
 422
 423         if (end < kbuf->buf_min || start > kbuf->buf_max)
 424                 return 0;
 425
 426         /*
 427          * Allocate memory top down with-in ram range. Otherwise bottom up
 428          * allocation.
 429          */
 430         if (kbuf->top_down)
 431                 return locate_mem_hole_top_down(start, end, kbuf);
 432         return locate_mem_hole_bottom_up(start, end, kbuf);
 433 }
 434
 435 /**
 436  * arch_kexec_walk_mem - call func(data) on free memory regions
 437  * @kbuf:       Context info for the search. Also passed to @func.
 438  * @func:       Function to call for each memory region.
 439  *
 440  * Return: The memory walk will stop when func returns a non-zero value
 441  * and that value will be returned. If all free regions are visited without
 442  * func returning non-zero, then zero will be returned.
 443  */
 444 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
 445                                int (*func)(u64, u64, void *))
 446 {
 447         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 448                 return walk_iomem_res_desc(crashk_res.desc,
 449                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 450                                            crashk_res.start, crashk_res.end,
 451                                            kbuf, func);
 452         else
 453                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 454 }
 455
 456 /**
 457  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 458  * @kbuf:       Parameters for the memory search.
 459  *
 460  * On success, kbuf->mem will have the start address of the memory region found.
 461  *
 462  * Return: 0 on success, negative errno on error.
 463  */
 464 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 465 {
 466         int ret;
 467
 468         ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
 469
 470         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 471 }
 472
 473 /**
 474  * kexec_add_buffer - place a buffer in a kexec segment
 475  * @kbuf:       Buffer contents and memory parameters.
 476  *
 477  * This function assumes that kexec_mutex is held.
 478  * On successful return, @kbuf->mem will have the physical address of
 479  * the buffer in memory.
 480  *
 481  * Return: 0 on success, negative errno on error.
 482  */
 483 int kexec_add_buffer(struct kexec_buf *kbuf)
 484 {
 485
 486         struct kexec_segment *ksegment;
 487         int ret;
 488
 489         /* Currently adding segment this way is allowed only in file mode */
 490         if (!kbuf->image->file_mode)
 491                 return -EINVAL;
 492
 493         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 494                 return -EINVAL;
 495
 496         /*
 497          * Make sure we are not trying to add buffer after allocating
 498          * control pages. All segments need to be placed first before
 499          * any control pages are allocated. As control page allocation
 500          * logic goes through list of segments to make sure there are
 501          * no destination overlaps.
 502          */
 503         if (!list_empty(&kbuf->image->control_pages)) {
 504                 WARN_ON(1);
 505                 return -EINVAL;
 506         }
 507
 508         /* Ensure minimum alignment needed for segments. */
 509         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 510         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 511
 512         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 513         ret = kexec_locate_mem_hole(kbuf);
 514         if (ret)
 515                 return ret;
 516
 517         /* Found a suitable memory range */
 518         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 519         ksegment->kbuf = kbuf->buffer;
 520         ksegment->bufsz = kbuf->bufsz;
 521         ksegment->mem = kbuf->mem;
 522         ksegment->memsz = kbuf->memsz;
 523         kbuf->image->nr_segments++;
 524         return 0;
 525 }
 526
 527 /* Calculate and store the digest of segments */
 528 static int kexec_calculate_store_digests(struct kimage *image)
 529 {
 530         struct crypto_shash *tfm;
 531         struct shash_desc *desc;
 532         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 533         size_t desc_size, nullsz;
 534         char *digest;
 535         void *zero_buf;
 536         struct kexec_sha_region *sha_regions;
 537         struct purgatory_info *pi = &image->purgatory_info;
 538
 539         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 540         zero_buf_sz = PAGE_SIZE;
 541
 542         tfm = crypto_alloc_shash("sha256", 0, 0);
 543         if (IS_ERR(tfm)) {
 544                 ret = PTR_ERR(tfm);
 545                 goto out;
 546         }
 547
 548         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 549         desc = kzalloc(desc_size, GFP_KERNEL);
 550         if (!desc) {
 551                 ret = -ENOMEM;
 552                 goto out_free_tfm;
 553         }
 554
 555         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 556         sha_regions = vzalloc(sha_region_sz);
 557         if (!sha_regions)
 558                 goto out_free_desc;
 559
 560         desc->tfm   = tfm;
 561         desc->flags = 0;
 562
 563         ret = crypto_shash_init(desc);
 564         if (ret < 0)
 565                 goto out_free_sha_regions;
 566
 567         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 568         if (!digest) {
 569                 ret = -ENOMEM;
 570                 goto out_free_sha_regions;
 571         }
 572
 573         for (j = i = 0; i < image->nr_segments; i++) {
 574                 struct kexec_segment *ksegment;
 575
 576                 ksegment = &image->segment[i];
 577                 /*
 578                  * Skip purgatory as it will be modified once we put digest
 579                  * info in purgatory.
 580                  */
 581                 if (ksegment->kbuf == pi->purgatory_buf)
 582                         continue;
 583
 584                 ret = crypto_shash_update(desc, ksegment->kbuf,
 585                                           ksegment->bufsz);
 586                 if (ret)
 587                         break;
 588
 589                 /*
 590                  * Assume rest of the buffer is filled with zero and
 591                  * update digest accordingly.
 592                  */
 593                 nullsz = ksegment->memsz - ksegment->bufsz;
 594                 while (nullsz) {
 595                         unsigned long bytes = nullsz;
 596
 597                         if (bytes > zero_buf_sz)
 598                                 bytes = zero_buf_sz;
 599                         ret = crypto_shash_update(desc, zero_buf, bytes);
 600                         if (ret)
 601                                 break;
 602                         nullsz -= bytes;
 603                 }
 604
 605                 if (ret)
 606                         break;
 607
 608                 sha_regions[j].start = ksegment->mem;
 609                 sha_regions[j].len = ksegment->memsz;
 610                 j++;
 611         }
 612
 613         if (!ret) {
 614                 ret = crypto_shash_final(desc, digest);
 615                 if (ret)
 616                         goto out_free_digest;
 617                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 618                                                      sha_regions, sha_region_sz, 0);
 619                 if (ret)
 620                         goto out_free_digest;
 621
 622                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 623                                                      digest, SHA256_DIGEST_SIZE, 0);
 624                 if (ret)
 625                         goto out_free_digest;
 626         }
 627
 628 out_free_digest:
 629         kfree(digest);
 630 out_free_sha_regions:
 631         vfree(sha_regions);
 632 out_free_desc:
 633         kfree(desc);
 634 out_free_tfm:
 635         kfree(tfm);
 636 out:
 637         return ret;
 638 }
 639
 640 /* Actually load purgatory. Lot of code taken from kexec-tools */
 641 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
 642                                   unsigned long max, int top_down)
 643 {
 644         struct purgatory_info *pi = &image->purgatory_info;
 645         unsigned long align, bss_align, bss_sz, bss_pad;
 646         unsigned long entry, load_addr, curr_load_addr, bss_addr, offset;
 647         unsigned char *buf_addr, *src;
 648         int i, ret = 0, entry_sidx = -1;
 649         const Elf_Shdr *sechdrs_c;
 650         Elf_Shdr *sechdrs = NULL;
 651         struct kexec_buf kbuf = { .image = image, .bufsz = 0, .buf_align = 1,
 652                                   .buf_min = min, .buf_max = max,
 653                                   .top_down = top_down };
 654
 655         /*
 656          * sechdrs_c points to section headers in purgatory and are read
 657          * only. No modifications allowed.
 658          */
 659         sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
 660
 661         /*
 662          * We can not modify sechdrs_c[] and its fields. It is read only.
 663          * Copy it over to a local copy where one can store some temporary
 664          * data and free it at the end. We need to modify ->sh_addr and
 665          * ->sh_offset fields to keep track of permanent and temporary
 666          * locations of sections.
 667          */
 668         sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 669         if (!sechdrs)
 670                 return -ENOMEM;
 671
 672         memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 673
 674         /*
 675          * We seem to have multiple copies of sections. First copy is which
 676          * is embedded in kernel in read only section. Some of these sections
 677          * will be copied to a temporary buffer and relocated. And these
 678          * sections will finally be copied to their final destination at
 679          * segment load time.
 680          *
 681          * Use ->sh_offset to reflect section address in memory. It will
 682          * point to original read only copy if section is not allocatable.
 683          * Otherwise it will point to temporary copy which will be relocated.
 684          *
 685          * Use ->sh_addr to contain final address of the section where it
 686          * will go during execution time.
 687          */
 688         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 689                 if (sechdrs[i].sh_type == SHT_NOBITS)
 690                         continue;
 691
 692                 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
 693                                                 sechdrs[i].sh_offset;
 694         }
 695
 696         /*
 697          * Identify entry point section and make entry relative to section
 698          * start.
 699          */
 700         entry = pi->ehdr->e_entry;
 701         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 702                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 703                         continue;
 704
 705                 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
 706                         continue;
 707
 708                 /* Make entry section relative */
 709                 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
 710                     ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
 711                      pi->ehdr->e_entry)) {
 712                         entry_sidx = i;
 713                         entry -= sechdrs[i].sh_addr;
 714                         break;
 715                 }
 716         }
 717
 718         /* Determine how much memory is needed to load relocatable object. */
 719         bss_align = 1;
 720         bss_sz = 0;
 721
 722         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 723                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 724                         continue;
 725
 726                 align = sechdrs[i].sh_addralign;
 727                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 728                         if (kbuf.buf_align < align)
 729                                 kbuf.buf_align = align;
 730                         kbuf.bufsz = ALIGN(kbuf.bufsz, align);
 731                         kbuf.bufsz += sechdrs[i].sh_size;
 732                 } else {
 733                         /* bss section */
 734                         if (bss_align < align)
 735                                 bss_align = align;
 736                         bss_sz = ALIGN(bss_sz, align);
 737                         bss_sz += sechdrs[i].sh_size;
 738                 }
 739         }
 740
 741         /* Determine the bss padding required to align bss properly */
 742         bss_pad = 0;
 743         if (kbuf.bufsz & (bss_align - 1))
 744                 bss_pad = bss_align - (kbuf.bufsz & (bss_align - 1));
 745
 746         kbuf.memsz = kbuf.bufsz + bss_pad + bss_sz;
 747
 748         /* Allocate buffer for purgatory */
 749         kbuf.buffer = vzalloc(kbuf.bufsz);
 750         if (!kbuf.buffer) {
 751                 ret = -ENOMEM;
 752                 goto out;
 753         }
 754
 755         if (kbuf.buf_align < bss_align)
 756                 kbuf.buf_align = bss_align;
 757
 758         /* Add buffer to segment list */
 759         ret = kexec_add_buffer(&kbuf);
 760         if (ret)
 761                 goto out;
 762         pi->purgatory_load_addr = kbuf.mem;
 763
 764         /* Load SHF_ALLOC sections */
 765         buf_addr = kbuf.buffer;
 766         load_addr = curr_load_addr = pi->purgatory_load_addr;
 767         bss_addr = load_addr + kbuf.bufsz + bss_pad;
 768
 769         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 770                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 771                         continue;
 772
 773                 align = sechdrs[i].sh_addralign;
 774                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 775                         curr_load_addr = ALIGN(curr_load_addr, align);
 776                         offset = curr_load_addr - load_addr;
 777                         /* We already modifed ->sh_offset to keep src addr */
 778                         src = (char *) sechdrs[i].sh_offset;
 779                         memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
 780
 781                         /* Store load address and source address of section */
 782                         sechdrs[i].sh_addr = curr_load_addr;
 783
 784                         /*
 785                          * This section got copied to temporary buffer. Update
 786                          * ->sh_offset accordingly.
 787                          */
 788                         sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
 789
 790                         /* Advance to the next address */
 791                         curr_load_addr += sechdrs[i].sh_size;
 792                 } else {
 793                         bss_addr = ALIGN(bss_addr, align);
 794                         sechdrs[i].sh_addr = bss_addr;
 795                         bss_addr += sechdrs[i].sh_size;
 796                 }
 797         }
 798
 799         /* Update entry point based on load address of text section */
 800         if (entry_sidx >= 0)
 801                 entry += sechdrs[entry_sidx].sh_addr;
 802
 803         /* Make kernel jump to purgatory after shutdown */
 804         image->start = entry;
 805
 806         /* Used later to get/set symbol values */
 807         pi->sechdrs = sechdrs;
 808
 809         /*
 810          * Used later to identify which section is purgatory and skip it
 811          * from checksumming.
 812          */
 813         pi->purgatory_buf = kbuf.buffer;
 814         return ret;
 815 out:
 816         vfree(sechdrs);
 817         vfree(kbuf.buffer);
 818         return ret;
 819 }
 820
 821 static int kexec_apply_relocations(struct kimage *image)
 822 {
 823         int i, ret;
 824         struct purgatory_info *pi = &image->purgatory_info;
 825         Elf_Shdr *sechdrs = pi->sechdrs;
 826
 827         /* Apply relocations */
 828         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 829                 Elf_Shdr *section, *symtab;
 830
 831                 if (sechdrs[i].sh_type != SHT_RELA &&
 832                     sechdrs[i].sh_type != SHT_REL)
 833                         continue;
 834
 835                 /*
 836                  * For section of type SHT_RELA/SHT_REL,
 837                  * ->sh_link contains section header index of associated
 838                  * symbol table. And ->sh_info contains section header
 839                  * index of section to which relocations apply.
 840                  */
 841                 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
 842                     sechdrs[i].sh_link >= pi->ehdr->e_shnum)
 843                         return -ENOEXEC;
 844
 845                 section = &sechdrs[sechdrs[i].sh_info];
 846                 symtab = &sechdrs[sechdrs[i].sh_link];
 847
 848                 if (!(section->sh_flags & SHF_ALLOC))
 849                         continue;
 850
 851                 /*
 852                  * symtab->sh_link contain section header index of associated
 853                  * string table.
 854                  */
 855                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 856                         /* Invalid section number? */
 857                         continue;
 858
 859                 /*
 860                  * Respective architecture needs to provide support for applying
 861                  * relocations of type SHT_RELA/SHT_REL.
 862                  */
 863                 if (sechdrs[i].sh_type == SHT_RELA)
 864                         ret = arch_kexec_apply_relocations_add(pi->ehdr,
 865                                                                sechdrs, i);
 866                 else if (sechdrs[i].sh_type == SHT_REL)
 867                         ret = arch_kexec_apply_relocations(pi->ehdr,
 868                                                            sechdrs, i);
 869                 if (ret)
 870                         return ret;
 871         }
 872
 873         return 0;
 874 }
 875
 876 /* Load relocatable purgatory object and relocate it appropriately */
 877 int kexec_load_purgatory(struct kimage *image, unsigned long min,
 878                          unsigned long max, int top_down,
 879                          unsigned long *load_addr)
 880 {
 881         struct purgatory_info *pi = &image->purgatory_info;
 882         int ret;
 883
 884         if (kexec_purgatory_size <= 0)
 885                 return -EINVAL;
 886
 887         if (kexec_purgatory_size < sizeof(Elf_Ehdr))
 888                 return -ENOEXEC;
 889
 890         pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
 891
 892         if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
 893             || pi->ehdr->e_type != ET_REL
 894             || !elf_check_arch(pi->ehdr)
 895             || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
 896                 return -ENOEXEC;
 897
 898         if (pi->ehdr->e_shoff >= kexec_purgatory_size
 899             || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
 900             kexec_purgatory_size - pi->ehdr->e_shoff))
 901                 return -ENOEXEC;
 902
 903         ret = __kexec_load_purgatory(image, min, max, top_down);
 904         if (ret)
 905                 return ret;
 906
 907         ret = kexec_apply_relocations(image);
 908         if (ret)
 909                 goto out;
 910
 911         *load_addr = pi->purgatory_load_addr;
 912         return 0;
 913 out:
 914         vfree(pi->sechdrs);
 915         pi->sechdrs = NULL;
 916
 917         vfree(pi->purgatory_buf);
 918         pi->purgatory_buf = NULL;
 919         return ret;
 920 }
 921
 922 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 923                                             const char *name)
 924 {
 925         Elf_Sym *syms;
 926         Elf_Shdr *sechdrs;
 927         Elf_Ehdr *ehdr;
 928         int i, k;
 929         const char *strtab;
 930
 931         if (!pi->sechdrs || !pi->ehdr)
 932                 return NULL;
 933
 934         sechdrs = pi->sechdrs;
 935         ehdr = pi->ehdr;
 936
 937         for (i = 0; i < ehdr->e_shnum; i++) {
 938                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 939                         continue;
 940
 941                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 942                         /* Invalid strtab section number */
 943                         continue;
 944                 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
 945                 syms = (Elf_Sym *)sechdrs[i].sh_offset;
 946
 947                 /* Go through symbols for a match */
 948                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 949                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 950                                 continue;
 951
 952                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 953                                 continue;
 954
 955                         if (syms[k].st_shndx == SHN_UNDEF ||
 956                             syms[k].st_shndx >= ehdr->e_shnum) {
 957                                 pr_debug("Symbol: %s has bad section index %d.\n",
 958                                                 name, syms[k].st_shndx);
 959                                 return NULL;
 960                         }
 961
 962                         /* Found the symbol we are looking for */
 963                         return &syms[k];
 964                 }
 965         }
 966
 967         return NULL;
 968 }
 969
 970 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
 971 {
 972         struct purgatory_info *pi = &image->purgatory_info;
 973         Elf_Sym *sym;
 974         Elf_Shdr *sechdr;
 975
 976         sym = kexec_purgatory_find_symbol(pi, name);
 977         if (!sym)
 978                 return ERR_PTR(-EINVAL);
 979
 980         sechdr = &pi->sechdrs[sym->st_shndx];
 981
 982         /*
 983          * Returns the address where symbol will finally be loaded after
 984          * kexec_load_segment()
 985          */
 986         return (void *)(sechdr->sh_addr + sym->st_value);
 987 }
 988
 989 /*
 990  * Get or set value of a symbol. If "get_value" is true, symbol value is
 991  * returned in buf otherwise symbol value is set based on value in buf.
 992  */
 993 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
 994                                    void *buf, unsigned int size, bool get_value)
 995 {
 996         Elf_Sym *sym;
 997         Elf_Shdr *sechdrs;
 998         struct purgatory_info *pi = &image->purgatory_info;
 999         char *sym_buf;
1000
1001         sym = kexec_purgatory_find_symbol(pi, name);
1002         if (!sym)
1003                 return -EINVAL;
1004
1005         if (sym->st_size != size) {
1006                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1007                        name, (unsigned long)sym->st_size, size);
1008                 return -EINVAL;
1009         }
1010
1011         sechdrs = pi->sechdrs;
1012
1013         if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
1014                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1015                        get_value ? "get" : "set");
1016                 return -EINVAL;
1017         }
1018
1019         sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
1020                                         sym->st_value;
1021
1022         if (get_value)
1023                 memcpy((void *)buf, sym_buf, size);
1024         else
1025                 memcpy((void *)sym_buf, buf, size);
1026
1027         return 0;
1028 }