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