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