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