kernel/kexec_file.c

   1 /*
   2  * kexec: kexec_file_load system call
   3  *
   4  * Copyright (C) 2014 Red Hat Inc.
   5  * Authors:
   6  *      Vivek Goyal <vgoyal@redhat.com>
   7  *
   8  * This source code is licensed under the GNU General Public License,
   9  * Version 2.  See the file COPYING for more details.
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/capability.h>
  15 #include <linux/mm.h>
  16 #include <linux/file.h>
  17 #include <linux/slab.h>
  18 #include <linux/kexec.h>
  19 #include <linux/mutex.h>
  20 #include <linux/list.h>
  21 #include <linux/fs.h>
  22 #include <linux/ima.h>
  23 #include <crypto/hash.h>
  24 #include <crypto/sha.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/vmalloc.h>
  27 #include "kexec_internal.h"
  28
  29 /*
  30  * Declare these symbols weak so that if architecture provides a purgatory,
  31  * these will be overridden.
  32  */
  33 char __weak kexec_purgatory[0];
  34 size_t __weak kexec_purgatory_size = 0;
  35
  36 static int kexec_calculate_store_digests(struct kimage *image);
  37
  38 /* Architectures can provide this probe function */
  39 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  40                                          unsigned long buf_len)
  41 {
  42         return -ENOEXEC;
  43 }
  44
  45 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
  46 {
  47         return ERR_PTR(-ENOEXEC);
  48 }
  49
  50 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
  51 {
  52         return -EINVAL;
  53 }
  54
  55 #ifdef CONFIG_KEXEC_VERIFY_SIG
  56 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
  57                                         unsigned long buf_len)
  58 {
  59         return -EKEYREJECTED;
  60 }
  61 #endif
  62
  63 /* Apply relocations of type RELA */
  64 int __weak
  65 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
  66                                  unsigned int relsec)
  67 {
  68         pr_err("RELA relocation unsupported.\n");
  69         return -ENOEXEC;
  70 }
  71
  72 /* Apply relocations of type REL */
  73 int __weak
  74 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
  75                              unsigned int relsec)
  76 {
  77         pr_err("REL relocation unsupported.\n");
  78         return -ENOEXEC;
  79 }
  80
  81 /*
  82  * Free up memory used by kernel, initrd, and command line. This is temporary
  83  * memory allocation which is not needed any more after these buffers have
  84  * been loaded into separate segments and have been copied elsewhere.
  85  */
  86 void kimage_file_post_load_cleanup(struct kimage *image)
  87 {
  88         struct purgatory_info *pi = &image->purgatory_info;
  89
  90         vfree(image->kernel_buf);
  91         image->kernel_buf = NULL;
  92
  93         vfree(image->initrd_buf);
  94         image->initrd_buf = NULL;
  95
  96         kfree(image->cmdline_buf);
  97         image->cmdline_buf = NULL;
  98
  99         vfree(pi->purgatory_buf);
 100         pi->purgatory_buf = NULL;
 101
 102         vfree(pi->sechdrs);
 103         pi->sechdrs = NULL;
 104
 105         /* See if architecture has anything to cleanup post load */
 106         arch_kimage_file_post_load_cleanup(image);
 107
 108         /*
 109          * Above call should have called into bootloader to free up
 110          * any data stored in kimage->image_loader_data. It should
 111          * be ok now to free it up.
 112          */
 113         kfree(image->image_loader_data);
 114         image->image_loader_data = NULL;
 115 }
 116
 117 /*
 118  * In file mode list of segments is prepared by kernel. Copy relevant
 119  * data from user space, do error checking, prepare segment list
 120  */
 121 static int
 122 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 123                              const char __user *cmdline_ptr,
 124                              unsigned long cmdline_len, unsigned flags)
 125 {
 126         int ret = 0;
 127         void *ldata;
 128         loff_t size;
 129
 130         ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
 131                                        &size, INT_MAX, READING_KEXEC_IMAGE);
 132         if (ret)
 133                 return ret;
 134         image->kernel_buf_len = size;
 135
 136         /* IMA needs to pass the measurement list to the next kernel. */
 137         ima_add_kexec_buffer(image);
 138
 139         /* Call arch image probe handlers */
 140         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 141                                             image->kernel_buf_len);
 142         if (ret)
 143                 goto out;
 144
 145 #ifdef CONFIG_KEXEC_VERIFY_SIG
 146         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 147                                            image->kernel_buf_len);
 148         if (ret) {
 149                 pr_debug("kernel signature verification failed.\n");
 150                 goto out;
 151         }
 152         pr_debug("kernel signature verification successful.\n");
 153 #endif
 154         /* It is possible that there no initramfs is being loaded */
 155         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 156                 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
 157                                                &size, INT_MAX,
 158                                                READING_KEXEC_INITRAMFS);
 159                 if (ret)
 160                         goto out;
 161                 image->initrd_buf_len = size;
 162         }
 163
 164         if (cmdline_len) {
 165                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 166                 if (IS_ERR(image->cmdline_buf)) {
 167                         ret = PTR_ERR(image->cmdline_buf);
 168                         image->cmdline_buf = NULL;
 169                         goto out;
 170                 }
 171
 172                 image->cmdline_buf_len = cmdline_len;
 173
 174                 /* command line should be a string with last byte null */
 175                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 176                         ret = -EINVAL;
 177                         goto out;
 178                 }
 179         }
 180
 181         /* Call arch image load handlers */
 182         ldata = arch_kexec_kernel_image_load(image);
 183
 184         if (IS_ERR(ldata)) {
 185                 ret = PTR_ERR(ldata);
 186                 goto out;
 187         }
 188
 189         image->image_loader_data = ldata;
 190 out:
 191         /* In case of error, free up all allocated memory in this function */
 192         if (ret)
 193                 kimage_file_post_load_cleanup(image);
 194         return ret;
 195 }
 196
 197 static int
 198 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 199                        int initrd_fd, const char __user *cmdline_ptr,
 200                        unsigned long cmdline_len, unsigned long flags)
 201 {
 202         int ret;
 203         struct kimage *image;
 204         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 205
 206         image = do_kimage_alloc_init();
 207         if (!image)
 208                 return -ENOMEM;
 209
 210         image->file_mode = 1;
 211
 212         if (kexec_on_panic) {
 213                 /* Enable special crash kernel control page alloc policy. */
 214                 image->control_page = crashk_res.start;
 215                 image->type = KEXEC_TYPE_CRASH;
 216         }
 217
 218         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 219                                            cmdline_ptr, cmdline_len, flags);
 220         if (ret)
 221                 goto out_free_image;
 222
 223         ret = sanity_check_segment_list(image);
 224         if (ret)
 225                 goto out_free_post_load_bufs;
 226
 227         ret = -ENOMEM;
 228         image->control_code_page = kimage_alloc_control_pages(image,
 229                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 230         if (!image->control_code_page) {
 231                 pr_err("Could not allocate control_code_buffer\n");
 232                 goto out_free_post_load_bufs;
 233         }
 234
 235         if (!kexec_on_panic) {
 236                 image->swap_page = kimage_alloc_control_pages(image, 0);
 237                 if (!image->swap_page) {
 238                         pr_err("Could not allocate swap buffer\n");
 239                         goto out_free_control_pages;
 240                 }
 241         }
 242
 243         *rimage = image;
 244         return 0;
 245 out_free_control_pages:
 246         kimage_free_page_list(&image->control_pages);
 247 out_free_post_load_bufs:
 248         kimage_file_post_load_cleanup(image);
 249 out_free_image:
 250         kfree(image);
 251         return ret;
 252 }
 253
 254 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 255                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 256                 unsigned long, flags)
 257 {
 258         int ret = 0, i;
 259         struct kimage **dest_image, *image;
 260
 261         /* We only trust the superuser with rebooting the system. */
 262         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 263                 return -EPERM;
 264
 265         /* Make sure we have a legal set of flags */
 266         if (flags != (flags & KEXEC_FILE_FLAGS))
 267                 return -EINVAL;
 268
 269         image = NULL;
 270
 271         if (!mutex_trylock(&kexec_mutex))
 272                 return -EBUSY;
 273
 274         dest_image = &kexec_image;
 275         if (flags & KEXEC_FILE_ON_CRASH) {
 276                 dest_image = &kexec_crash_image;
 277                 if (kexec_crash_image)
 278                         arch_kexec_unprotect_crashkres();
 279         }
 280
 281         if (flags & KEXEC_FILE_UNLOAD)
 282                 goto exchange;
 283
 284         /*
 285          * In case of crash, new kernel gets loaded in reserved region. It is
 286          * same memory where old crash kernel might be loaded. Free any
 287          * current crash dump kernel before we corrupt it.
 288          */
 289         if (flags & KEXEC_FILE_ON_CRASH)
 290                 kimage_free(xchg(&kexec_crash_image, NULL));
 291
 292         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 293                                      cmdline_len, flags);
 294         if (ret)
 295                 goto out;
 296
 297         ret = machine_kexec_prepare(image);
 298         if (ret)
 299                 goto out;
 300
 301         /*
 302          * Some architecture(like S390) may touch the crash memory before
 303          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 304          */
 305         ret = kimage_crash_copy_vmcoreinfo(image);
 306         if (ret)
 307                 goto out;
 308
 309         ret = kexec_calculate_store_digests(image);
 310         if (ret)
 311                 goto out;
 312
 313         for (i = 0; i < image->nr_segments; i++) {
 314                 struct kexec_segment *ksegment;
 315
 316                 ksegment = &image->segment[i];
 317                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 318                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 319                          ksegment->memsz);
 320
 321                 ret = kimage_load_segment(image, &image->segment[i]);
 322                 if (ret)
 323                         goto out;
 324         }
 325
 326         kimage_terminate(image);
 327
 328         /*
 329          * Free up any temporary buffers allocated which are not needed
 330          * after image has been loaded
 331          */
 332         kimage_file_post_load_cleanup(image);
 333 exchange:
 334         image = xchg(dest_image, image);
 335 out:
 336         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 337                 arch_kexec_protect_crashkres();
 338
 339         mutex_unlock(&kexec_mutex);
 340         kimage_free(image);
 341         return ret;
 342 }
 343
 344 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 345                                     struct kexec_buf *kbuf)
 346 {
 347         struct kimage *image = kbuf->image;
 348         unsigned long temp_start, temp_end;
 349
 350         temp_end = min(end, kbuf->buf_max);
 351         temp_start = temp_end - kbuf->memsz;
 352
 353         do {
 354                 /* align down start */
 355                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 356
 357                 if (temp_start < start || temp_start < kbuf->buf_min)
 358                         return 0;
 359
 360                 temp_end = temp_start + kbuf->memsz - 1;
 361
 362                 /*
 363                  * Make sure this does not conflict with any of existing
 364                  * segments
 365                  */
 366                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 367                         temp_start = temp_start - PAGE_SIZE;
 368                         continue;
 369                 }
 370
 371                 /* We found a suitable memory range */
 372                 break;
 373         } while (1);
 374
 375         /* If we are here, we found a suitable memory range */
 376         kbuf->mem = temp_start;
 377
 378         /* Success, stop navigating through remaining System RAM ranges */
 379         return 1;
 380 }
 381
 382 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 383                                      struct kexec_buf *kbuf)
 384 {
 385         struct kimage *image = kbuf->image;
 386         unsigned long temp_start, temp_end;
 387
 388         temp_start = max(start, kbuf->buf_min);
 389
 390         do {
 391                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 392                 temp_end = temp_start + kbuf->memsz - 1;
 393
 394                 if (temp_end > end || temp_end > kbuf->buf_max)
 395                         return 0;
 396                 /*
 397                  * Make sure this does not conflict with any of existing
 398                  * segments
 399                  */
 400                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 401                         temp_start = temp_start + PAGE_SIZE;
 402                         continue;
 403                 }
 404
 405                 /* We found a suitable memory range */
 406                 break;
 407         } while (1);
 408
 409         /* If we are here, we found a suitable memory range */
 410         kbuf->mem = temp_start;
 411
 412         /* Success, stop navigating through remaining System RAM ranges */
 413         return 1;
 414 }
 415
 416 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
 417 {
 418         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 419         unsigned long sz = end - start + 1;
 420
 421         /* Returning 0 will take to next memory range */
 422         if (sz < kbuf->memsz)
 423                 return 0;
 424
 425         if (end < kbuf->buf_min || start > kbuf->buf_max)
 426                 return 0;
 427
 428         /*
 429          * Allocate memory top down with-in ram range. Otherwise bottom up
 430          * allocation.
 431          */
 432         if (kbuf->top_down)
 433                 return locate_mem_hole_top_down(start, end, kbuf);
 434         return locate_mem_hole_bottom_up(start, end, kbuf);
 435 }
 436
 437 /**
 438  * arch_kexec_walk_mem - call func(data) on free memory regions
 439  * @kbuf:       Context info for the search. Also passed to @func.
 440  * @func:       Function to call for each memory region.
 441  *
 442  * Return: The memory walk will stop when func returns a non-zero value
 443  * and that value will be returned. If all free regions are visited without
 444  * func returning non-zero, then zero will be returned.
 445  */
 446 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
 447                                int (*func)(u64, u64, void *))
 448 {
 449         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 450                 return walk_iomem_res_desc(crashk_res.desc,
 451                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 452                                            crashk_res.start, crashk_res.end,
 453                                            kbuf, func);
 454         else
 455                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 456 }
 457
 458 /**
 459  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 460  * @kbuf:       Parameters for the memory search.
 461  *
 462  * On success, kbuf->mem will have the start address of the memory region found.
 463  *
 464  * Return: 0 on success, negative errno on error.
 465  */
 466 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 467 {
 468         int ret;
 469
 470         ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
 471
 472         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 473 }
 474
 475 /**
 476  * kexec_add_buffer - place a buffer in a kexec segment
 477  * @kbuf:       Buffer contents and memory parameters.
 478  *
 479  * This function assumes that kexec_mutex is held.
 480  * On successful return, @kbuf->mem will have the physical address of
 481  * the buffer in memory.
 482  *
 483  * Return: 0 on success, negative errno on error.
 484  */
 485 int kexec_add_buffer(struct kexec_buf *kbuf)
 486 {
 487
 488         struct kexec_segment *ksegment;
 489         int ret;
 490
 491         /* Currently adding segment this way is allowed only in file mode */
 492         if (!kbuf->image->file_mode)
 493                 return -EINVAL;
 494
 495         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 496                 return -EINVAL;
 497
 498         /*
 499          * Make sure we are not trying to add buffer after allocating
 500          * control pages. All segments need to be placed first before
 501          * any control pages are allocated. As control page allocation
 502          * logic goes through list of segments to make sure there are
 503          * no destination overlaps.
 504          */
 505         if (!list_empty(&kbuf->image->control_pages)) {
 506                 WARN_ON(1);
 507                 return -EINVAL;
 508         }
 509
 510         /* Ensure minimum alignment needed for segments. */
 511         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 512         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 513
 514         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 515         ret = kexec_locate_mem_hole(kbuf);
 516         if (ret)
 517                 return ret;
 518
 519         /* Found a suitable memory range */
 520         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 521         ksegment->kbuf = kbuf->buffer;
 522         ksegment->bufsz = kbuf->bufsz;
 523         ksegment->mem = kbuf->mem;
 524         ksegment->memsz = kbuf->memsz;
 525         kbuf->image->nr_segments++;
 526         return 0;
 527 }
 528
 529 /* Calculate and store the digest of segments */
 530 static int kexec_calculate_store_digests(struct kimage *image)
 531 {
 532         struct crypto_shash *tfm;
 533         struct shash_desc *desc;
 534         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 535         size_t desc_size, nullsz;
 536         char *digest;
 537         void *zero_buf;
 538         struct kexec_sha_region *sha_regions;
 539         struct purgatory_info *pi = &image->purgatory_info;
 540
 541         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 542         zero_buf_sz = PAGE_SIZE;
 543
 544         tfm = crypto_alloc_shash("sha256", 0, 0);
 545         if (IS_ERR(tfm)) {
 546                 ret = PTR_ERR(tfm);
 547                 goto out;
 548         }
 549
 550         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 551         desc = kzalloc(desc_size, GFP_KERNEL);
 552         if (!desc) {
 553                 ret = -ENOMEM;
 554                 goto out_free_tfm;
 555         }
 556
 557         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 558         sha_regions = vzalloc(sha_region_sz);
 559         if (!sha_regions)
 560                 goto out_free_desc;
 561
 562         desc->tfm   = tfm;
 563         desc->flags = 0;
 564
 565         ret = crypto_shash_init(desc);
 566         if (ret < 0)
 567                 goto out_free_sha_regions;
 568
 569         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 570         if (!digest) {
 571                 ret = -ENOMEM;
 572                 goto out_free_sha_regions;
 573         }
 574
 575         for (j = i = 0; i < image->nr_segments; i++) {
 576                 struct kexec_segment *ksegment;
 577
 578                 ksegment = &image->segment[i];
 579                 /*
 580                  * Skip purgatory as it will be modified once we put digest
 581                  * info in purgatory.
 582                  */
 583                 if (ksegment->kbuf == pi->purgatory_buf)
 584                         continue;
 585
 586                 ret = crypto_shash_update(desc, ksegment->kbuf,
 587                                           ksegment->bufsz);
 588                 if (ret)
 589                         break;
 590
 591                 /*
 592                  * Assume rest of the buffer is filled with zero and
 593                  * update digest accordingly.
 594                  */
 595                 nullsz = ksegment->memsz - ksegment->bufsz;
 596                 while (nullsz) {
 597                         unsigned long bytes = nullsz;
 598
 599                         if (bytes > zero_buf_sz)
 600                                 bytes = zero_buf_sz;
 601                         ret = crypto_shash_update(desc, zero_buf, bytes);
 602                         if (ret)
 603                                 break;
 604                         nullsz -= bytes;
 605                 }
 606
 607                 if (ret)
 608                         break;
 609
 610                 sha_regions[j].start = ksegment->mem;
 611                 sha_regions[j].len = ksegment->memsz;
 612                 j++;
 613         }
 614
 615         if (!ret) {
 616                 ret = crypto_shash_final(desc, digest);
 617                 if (ret)
 618                         goto out_free_digest;
 619                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 620                                                      sha_regions, sha_region_sz, 0);
 621                 if (ret)
 622                         goto out_free_digest;
 623
 624                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 625                                                      digest, SHA256_DIGEST_SIZE, 0);
 626                 if (ret)
 627                         goto out_free_digest;
 628         }
 629
 630 out_free_digest:
 631         kfree(digest);
 632 out_free_sha_regions:
 633         vfree(sha_regions);
 634 out_free_desc:
 635         kfree(desc);
 636 out_free_tfm:
 637         kfree(tfm);
 638 out:
 639         return ret;
 640 }
 641
 642 /* Actually load purgatory. Lot of code taken from kexec-tools */
 643 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
 644                                   unsigned long max, int top_down)
 645 {
 646         struct purgatory_info *pi = &image->purgatory_info;
 647         unsigned long align, bss_align, bss_sz, bss_pad;
 648         unsigned long entry, load_addr, curr_load_addr, bss_addr, offset;
 649         unsigned char *buf_addr, *src;
 650         int i, ret = 0, entry_sidx = -1;
 651         const Elf_Shdr *sechdrs_c;
 652         Elf_Shdr *sechdrs = NULL;
 653         struct kexec_buf kbuf = { .image = image, .bufsz = 0, .buf_align = 1,
 654                                   .buf_min = min, .buf_max = max,
 655                                   .top_down = top_down };
 656
 657         /*
 658          * sechdrs_c points to section headers in purgatory and are read
 659          * only. No modifications allowed.
 660          */
 661         sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
 662
 663         /*
 664          * We can not modify sechdrs_c[] and its fields. It is read only.
 665          * Copy it over to a local copy where one can store some temporary
 666          * data and free it at the end. We need to modify ->sh_addr and
 667          * ->sh_offset fields to keep track of permanent and temporary
 668          * locations of sections.
 669          */
 670         sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 671         if (!sechdrs)
 672                 return -ENOMEM;
 673
 674         memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 675
 676         /*
 677          * We seem to have multiple copies of sections. First copy is which
 678          * is embedded in kernel in read only section. Some of these sections
 679          * will be copied to a temporary buffer and relocated. And these
 680          * sections will finally be copied to their final destination at
 681          * segment load time.
 682          *
 683          * Use ->sh_offset to reflect section address in memory. It will
 684          * point to original read only copy if section is not allocatable.
 685          * Otherwise it will point to temporary copy which will be relocated.
 686          *
 687          * Use ->sh_addr to contain final address of the section where it
 688          * will go during execution time.
 689          */
 690         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 691                 if (sechdrs[i].sh_type == SHT_NOBITS)
 692                         continue;
 693
 694                 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
 695                                                 sechdrs[i].sh_offset;
 696         }
 697
 698         /*
 699          * Identify entry point section and make entry relative to section
 700          * start.
 701          */
 702         entry = pi->ehdr->e_entry;
 703         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 704                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 705                         continue;
 706
 707                 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
 708                         continue;
 709
 710                 /* Make entry section relative */
 711                 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
 712                     ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
 713                      pi->ehdr->e_entry)) {
 714                         entry_sidx = i;
 715                         entry -= sechdrs[i].sh_addr;
 716                         break;
 717                 }
 718         }
 719
 720         /* Determine how much memory is needed to load relocatable object. */
 721         bss_align = 1;
 722         bss_sz = 0;
 723
 724         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 725                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 726                         continue;
 727
 728                 align = sechdrs[i].sh_addralign;
 729                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 730                         if (kbuf.buf_align < align)
 731                                 kbuf.buf_align = align;
 732                         kbuf.bufsz = ALIGN(kbuf.bufsz, align);
 733                         kbuf.bufsz += sechdrs[i].sh_size;
 734                 } else {
 735                         /* bss section */
 736                         if (bss_align < align)
 737                                 bss_align = align;
 738                         bss_sz = ALIGN(bss_sz, align);
 739                         bss_sz += sechdrs[i].sh_size;
 740                 }
 741         }
 742
 743         /* Determine the bss padding required to align bss properly */
 744         bss_pad = 0;
 745         if (kbuf.bufsz & (bss_align - 1))
 746                 bss_pad = bss_align - (kbuf.bufsz & (bss_align - 1));
 747
 748         kbuf.memsz = kbuf.bufsz + bss_pad + bss_sz;
 749
 750         /* Allocate buffer for purgatory */
 751         kbuf.buffer = vzalloc(kbuf.bufsz);
 752         if (!kbuf.buffer) {
 753                 ret = -ENOMEM;
 754                 goto out;
 755         }
 756
 757         if (kbuf.buf_align < bss_align)
 758                 kbuf.buf_align = bss_align;
 759
 760         /* Add buffer to segment list */
 761         ret = kexec_add_buffer(&kbuf);
 762         if (ret)
 763                 goto out;
 764         pi->purgatory_load_addr = kbuf.mem;
 765
 766         /* Load SHF_ALLOC sections */
 767         buf_addr = kbuf.buffer;
 768         load_addr = curr_load_addr = pi->purgatory_load_addr;
 769         bss_addr = load_addr + kbuf.bufsz + bss_pad;
 770
 771         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 772                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 773                         continue;
 774
 775                 align = sechdrs[i].sh_addralign;
 776                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 777                         curr_load_addr = ALIGN(curr_load_addr, align);
 778                         offset = curr_load_addr - load_addr;
 779                         /* We already modifed ->sh_offset to keep src addr */
 780                         src = (char *) sechdrs[i].sh_offset;
 781                         memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
 782
 783                         /* Store load address and source address of section */
 784                         sechdrs[i].sh_addr = curr_load_addr;
 785
 786                         /*
 787                          * This section got copied to temporary buffer. Update
 788                          * ->sh_offset accordingly.
 789                          */
 790                         sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
 791
 792                         /* Advance to the next address */
 793                         curr_load_addr += sechdrs[i].sh_size;
 794                 } else {
 795                         bss_addr = ALIGN(bss_addr, align);
 796                         sechdrs[i].sh_addr = bss_addr;
 797                         bss_addr += sechdrs[i].sh_size;
 798                 }
 799         }
 800
 801         /* Update entry point based on load address of text section */
 802         if (entry_sidx >= 0)
 803                 entry += sechdrs[entry_sidx].sh_addr;
 804
 805         /* Make kernel jump to purgatory after shutdown */
 806         image->start = entry;
 807
 808         /* Used later to get/set symbol values */
 809         pi->sechdrs = sechdrs;
 810
 811         /*
 812          * Used later to identify which section is purgatory and skip it
 813          * from checksumming.
 814          */
 815         pi->purgatory_buf = kbuf.buffer;
 816         return ret;
 817 out:
 818         vfree(sechdrs);
 819         vfree(kbuf.buffer);
 820         return ret;
 821 }
 822
 823 static int kexec_apply_relocations(struct kimage *image)
 824 {
 825         int i, ret;
 826         struct purgatory_info *pi = &image->purgatory_info;
 827         Elf_Shdr *sechdrs = pi->sechdrs;
 828
 829         /* Apply relocations */
 830         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 831                 Elf_Shdr *section, *symtab;
 832
 833                 if (sechdrs[i].sh_type != SHT_RELA &&
 834                     sechdrs[i].sh_type != SHT_REL)
 835                         continue;
 836
 837                 /*
 838                  * For section of type SHT_RELA/SHT_REL,
 839                  * ->sh_link contains section header index of associated
 840                  * symbol table. And ->sh_info contains section header
 841                  * index of section to which relocations apply.
 842                  */
 843                 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
 844                     sechdrs[i].sh_link >= pi->ehdr->e_shnum)
 845                         return -ENOEXEC;
 846
 847                 section = &sechdrs[sechdrs[i].sh_info];
 848                 symtab = &sechdrs[sechdrs[i].sh_link];
 849
 850                 if (!(section->sh_flags & SHF_ALLOC))
 851                         continue;
 852
 853                 /*
 854                  * symtab->sh_link contain section header index of associated
 855                  * string table.
 856                  */
 857                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 858                         /* Invalid section number? */
 859                         continue;
 860
 861                 /*
 862                  * Respective architecture needs to provide support for applying
 863                  * relocations of type SHT_RELA/SHT_REL.
 864                  */
 865                 if (sechdrs[i].sh_type == SHT_RELA)
 866                         ret = arch_kexec_apply_relocations_add(pi->ehdr,
 867                                                                sechdrs, i);
 868                 else if (sechdrs[i].sh_type == SHT_REL)
 869                         ret = arch_kexec_apply_relocations(pi->ehdr,
 870                                                            sechdrs, i);
 871                 if (ret)
 872                         return ret;
 873         }
 874
 875         return 0;
 876 }
 877
 878 /* Load relocatable purgatory object and relocate it appropriately */
 879 int kexec_load_purgatory(struct kimage *image, unsigned long min,
 880                          unsigned long max, int top_down,
 881                          unsigned long *load_addr)
 882 {
 883         struct purgatory_info *pi = &image->purgatory_info;
 884         int ret;
 885
 886         if (kexec_purgatory_size <= 0)
 887                 return -EINVAL;
 888
 889         if (kexec_purgatory_size < sizeof(Elf_Ehdr))
 890                 return -ENOEXEC;
 891
 892         pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
 893
 894         if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
 895             || pi->ehdr->e_type != ET_REL
 896             || !elf_check_arch(pi->ehdr)
 897             || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
 898                 return -ENOEXEC;
 899
 900         if (pi->ehdr->e_shoff >= kexec_purgatory_size
 901             || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
 902             kexec_purgatory_size - pi->ehdr->e_shoff))
 903                 return -ENOEXEC;
 904
 905         ret = __kexec_load_purgatory(image, min, max, top_down);
 906         if (ret)
 907                 return ret;
 908
 909         ret = kexec_apply_relocations(image);
 910         if (ret)
 911                 goto out;
 912
 913         *load_addr = pi->purgatory_load_addr;
 914         return 0;
 915 out:
 916         vfree(pi->sechdrs);
 917         pi->sechdrs = NULL;
 918
 919         vfree(pi->purgatory_buf);
 920         pi->purgatory_buf = NULL;
 921         return ret;
 922 }
 923
 924 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 925                                             const char *name)
 926 {
 927         Elf_Sym *syms;
 928         Elf_Shdr *sechdrs;
 929         Elf_Ehdr *ehdr;
 930         int i, k;
 931         const char *strtab;
 932
 933         if (!pi->sechdrs || !pi->ehdr)
 934                 return NULL;
 935
 936         sechdrs = pi->sechdrs;
 937         ehdr = pi->ehdr;
 938
 939         for (i = 0; i < ehdr->e_shnum; i++) {
 940                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 941                         continue;
 942
 943                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 944                         /* Invalid strtab section number */
 945                         continue;
 946                 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
 947                 syms = (Elf_Sym *)sechdrs[i].sh_offset;
 948
 949                 /* Go through symbols for a match */
 950                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 951                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 952                                 continue;
 953
 954                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 955                                 continue;
 956
 957                         if (syms[k].st_shndx == SHN_UNDEF ||
 958                             syms[k].st_shndx >= ehdr->e_shnum) {
 959                                 pr_debug("Symbol: %s has bad section index %d.\n",
 960                                                 name, syms[k].st_shndx);
 961                                 return NULL;
 962                         }
 963
 964                         /* Found the symbol we are looking for */
 965                         return &syms[k];
 966                 }
 967         }
 968
 969         return NULL;
 970 }
 971
 972 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
 973 {
 974         struct purgatory_info *pi = &image->purgatory_info;
 975         Elf_Sym *sym;
 976         Elf_Shdr *sechdr;
 977
 978         sym = kexec_purgatory_find_symbol(pi, name);
 979         if (!sym)
 980                 return ERR_PTR(-EINVAL);
 981
 982         sechdr = &pi->sechdrs[sym->st_shndx];
 983
 984         /*
 985          * Returns the address where symbol will finally be loaded after
 986          * kexec_load_segment()
 987          */
 988         return (void *)(sechdr->sh_addr + sym->st_value);
 989 }
 990
 991 /*
 992  * Get or set value of a symbol. If "get_value" is true, symbol value is
 993  * returned in buf otherwise symbol value is set based on value in buf.
 994  */
 995 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
 996                                    void *buf, unsigned int size, bool get_value)
 997 {
 998         Elf_Sym *sym;
 999         Elf_Shdr *sechdrs;
1000         struct purgatory_info *pi = &image->purgatory_info;
1001         char *sym_buf;
1002
1003         sym = kexec_purgatory_find_symbol(pi, name);
1004         if (!sym)
1005                 return -EINVAL;
1006
1007         if (sym->st_size != size) {
1008                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1009                        name, (unsigned long)sym->st_size, size);
1010                 return -EINVAL;
1011         }
1012
1013         sechdrs = pi->sechdrs;
1014
1015         if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
1016                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1017                        get_value ? "get" : "set");
1018                 return -EINVAL;
1019         }
1020
1021         sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
1022                                         sym->st_value;
1023
1024         if (get_value)
1025                 memcpy((void *)buf, sym_buf, size);
1026         else
1027                 memcpy((void *)sym_buf, buf, size);
1028
1029         return 0;
1030 }