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/elf.h>
  26 #include <linux/elfcore.h>
  27 #include <linux/kernel.h>
  28 #include <linux/kexec.h>
  29 #include <linux/slab.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/vmalloc.h>
  32 #include "kexec_internal.h"
  33
  34 static int kexec_calculate_store_digests(struct kimage *image);
  35
  36 /*
  37  * Currently this is the only default function that is exported as some
  38  * architectures need it to do additional handlings.
  39  * In the future, other default functions may be exported too if required.
  40  */
  41 int kexec_image_probe_default(struct kimage *image, void *buf,
  42                               unsigned long buf_len)
  43 {
  44         const struct kexec_file_ops * const *fops;
  45         int ret = -ENOEXEC;
  46
  47         for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
  48                 ret = (*fops)->probe(buf, buf_len);
  49                 if (!ret) {
  50                         image->fops = *fops;
  51                         return ret;
  52                 }
  53         }
  54
  55         return ret;
  56 }
  57
  58 /* Architectures can provide this probe function */
  59 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  60                                          unsigned long buf_len)
  61 {
  62         return kexec_image_probe_default(image, buf, buf_len);
  63 }
  64
  65 static void *kexec_image_load_default(struct kimage *image)
  66 {
  67         if (!image->fops || !image->fops->load)
  68                 return ERR_PTR(-ENOEXEC);
  69
  70         return image->fops->load(image, image->kernel_buf,
  71                                  image->kernel_buf_len, image->initrd_buf,
  72                                  image->initrd_buf_len, image->cmdline_buf,
  73                                  image->cmdline_buf_len);
  74 }
  75
  76 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
  77 {
  78         return kexec_image_load_default(image);
  79 }
  80
  81 static int kexec_image_post_load_cleanup_default(struct kimage *image)
  82 {
  83         if (!image->fops || !image->fops->cleanup)
  84                 return 0;
  85
  86         return image->fops->cleanup(image->image_loader_data);
  87 }
  88
  89 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
  90 {
  91         return kexec_image_post_load_cleanup_default(image);
  92 }
  93
  94 #ifdef CONFIG_KEXEC_VERIFY_SIG
  95 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
  96                                           unsigned long buf_len)
  97 {
  98         if (!image->fops || !image->fops->verify_sig) {
  99                 pr_debug("kernel loader does not support signature verification.\n");
 100                 return -EKEYREJECTED;
 101         }
 102
 103         return image->fops->verify_sig(buf, buf_len);
 104 }
 105
 106 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
 107                                         unsigned long buf_len)
 108 {
 109         return kexec_image_verify_sig_default(image, buf, buf_len);
 110 }
 111 #endif
 112
 113 /*
 114  * arch_kexec_apply_relocations_add - apply relocations of type RELA
 115  * @pi:         Purgatory to be relocated.
 116  * @section:    Section relocations applying to.
 117  * @relsec:     Section containing RELAs.
 118  * @symtab:     Corresponding symtab.
 119  *
 120  * Return: 0 on success, negative errno on error.
 121  */
 122 int __weak
 123 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
 124                                  const Elf_Shdr *relsec, const Elf_Shdr *symtab)
 125 {
 126         pr_err("RELA relocation unsupported.\n");
 127         return -ENOEXEC;
 128 }
 129
 130 /*
 131  * arch_kexec_apply_relocations - apply relocations of type REL
 132  * @pi:         Purgatory to be relocated.
 133  * @section:    Section relocations applying to.
 134  * @relsec:     Section containing RELs.
 135  * @symtab:     Corresponding symtab.
 136  *
 137  * Return: 0 on success, negative errno on error.
 138  */
 139 int __weak
 140 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
 141                              const Elf_Shdr *relsec, const Elf_Shdr *symtab)
 142 {
 143         pr_err("REL relocation unsupported.\n");
 144         return -ENOEXEC;
 145 }
 146
 147 /*
 148  * Free up memory used by kernel, initrd, and command line. This is temporary
 149  * memory allocation which is not needed any more after these buffers have
 150  * been loaded into separate segments and have been copied elsewhere.
 151  */
 152 void kimage_file_post_load_cleanup(struct kimage *image)
 153 {
 154         struct purgatory_info *pi = &image->purgatory_info;
 155
 156         vfree(image->kernel_buf);
 157         image->kernel_buf = NULL;
 158
 159         vfree(image->initrd_buf);
 160         image->initrd_buf = NULL;
 161
 162         kfree(image->cmdline_buf);
 163         image->cmdline_buf = NULL;
 164
 165         vfree(pi->purgatory_buf);
 166         pi->purgatory_buf = NULL;
 167
 168         vfree(pi->sechdrs);
 169         pi->sechdrs = NULL;
 170
 171 #ifdef CONFIG_IMA_KEXEC
 172         vfree(image->ima_buffer);
 173         image->ima_buffer = NULL;
 174 #endif /* CONFIG_IMA_KEXEC */
 175
 176         /* See if architecture has anything to cleanup post load */
 177         arch_kimage_file_post_load_cleanup(image);
 178
 179         /*
 180          * Above call should have called into bootloader to free up
 181          * any data stored in kimage->image_loader_data. It should
 182          * be ok now to free it up.
 183          */
 184         kfree(image->image_loader_data);
 185         image->image_loader_data = NULL;
 186 }
 187
 188 /*
 189  * In file mode list of segments is prepared by kernel. Copy relevant
 190  * data from user space, do error checking, prepare segment list
 191  */
 192 static int
 193 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 194                              const char __user *cmdline_ptr,
 195                              unsigned long cmdline_len, unsigned flags)
 196 {
 197         int ret = 0;
 198         void *ldata;
 199         loff_t size;
 200
 201         ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
 202                                        &size, INT_MAX, READING_KEXEC_IMAGE);
 203         if (ret)
 204                 return ret;
 205         image->kernel_buf_len = size;
 206
 207         /* IMA needs to pass the measurement list to the next kernel. */
 208         ima_add_kexec_buffer(image);
 209
 210         /* Call arch image probe handlers */
 211         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 212                                             image->kernel_buf_len);
 213         if (ret)
 214                 goto out;
 215
 216 #ifdef CONFIG_KEXEC_VERIFY_SIG
 217         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 218                                            image->kernel_buf_len);
 219         if (ret) {
 220                 pr_debug("kernel signature verification failed.\n");
 221                 goto out;
 222         }
 223         pr_debug("kernel signature verification successful.\n");
 224 #endif
 225         /* It is possible that there no initramfs is being loaded */
 226         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 227                 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
 228                                                &size, INT_MAX,
 229                                                READING_KEXEC_INITRAMFS);
 230                 if (ret)
 231                         goto out;
 232                 image->initrd_buf_len = size;
 233         }
 234
 235         if (cmdline_len) {
 236                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 237                 if (IS_ERR(image->cmdline_buf)) {
 238                         ret = PTR_ERR(image->cmdline_buf);
 239                         image->cmdline_buf = NULL;
 240                         goto out;
 241                 }
 242
 243                 image->cmdline_buf_len = cmdline_len;
 244
 245                 /* command line should be a string with last byte null */
 246                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 247                         ret = -EINVAL;
 248                         goto out;
 249                 }
 250         }
 251
 252         /* Call arch image load handlers */
 253         ldata = arch_kexec_kernel_image_load(image);
 254
 255         if (IS_ERR(ldata)) {
 256                 ret = PTR_ERR(ldata);
 257                 goto out;
 258         }
 259
 260         image->image_loader_data = ldata;
 261 out:
 262         /* In case of error, free up all allocated memory in this function */
 263         if (ret)
 264                 kimage_file_post_load_cleanup(image);
 265         return ret;
 266 }
 267
 268 static int
 269 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 270                        int initrd_fd, const char __user *cmdline_ptr,
 271                        unsigned long cmdline_len, unsigned long flags)
 272 {
 273         int ret;
 274         struct kimage *image;
 275         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 276
 277         image = do_kimage_alloc_init();
 278         if (!image)
 279                 return -ENOMEM;
 280
 281         image->file_mode = 1;
 282
 283         if (kexec_on_panic) {
 284                 /* Enable special crash kernel control page alloc policy. */
 285                 image->control_page = crashk_res.start;
 286                 image->type = KEXEC_TYPE_CRASH;
 287         }
 288
 289         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 290                                            cmdline_ptr, cmdline_len, flags);
 291         if (ret)
 292                 goto out_free_image;
 293
 294         ret = sanity_check_segment_list(image);
 295         if (ret)
 296                 goto out_free_post_load_bufs;
 297
 298         ret = -ENOMEM;
 299         image->control_code_page = kimage_alloc_control_pages(image,
 300                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 301         if (!image->control_code_page) {
 302                 pr_err("Could not allocate control_code_buffer\n");
 303                 goto out_free_post_load_bufs;
 304         }
 305
 306         if (!kexec_on_panic) {
 307                 image->swap_page = kimage_alloc_control_pages(image, 0);
 308                 if (!image->swap_page) {
 309                         pr_err("Could not allocate swap buffer\n");
 310                         goto out_free_control_pages;
 311                 }
 312         }
 313
 314         *rimage = image;
 315         return 0;
 316 out_free_control_pages:
 317         kimage_free_page_list(&image->control_pages);
 318 out_free_post_load_bufs:
 319         kimage_file_post_load_cleanup(image);
 320 out_free_image:
 321         kfree(image);
 322         return ret;
 323 }
 324
 325 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 326                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 327                 unsigned long, flags)
 328 {
 329         int ret = 0, i;
 330         struct kimage **dest_image, *image;
 331
 332         /* We only trust the superuser with rebooting the system. */
 333         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 334                 return -EPERM;
 335
 336         /* Make sure we have a legal set of flags */
 337         if (flags != (flags & KEXEC_FILE_FLAGS))
 338                 return -EINVAL;
 339
 340         image = NULL;
 341
 342         if (!mutex_trylock(&kexec_mutex))
 343                 return -EBUSY;
 344
 345         dest_image = &kexec_image;
 346         if (flags & KEXEC_FILE_ON_CRASH) {
 347                 dest_image = &kexec_crash_image;
 348                 if (kexec_crash_image)
 349                         arch_kexec_unprotect_crashkres();
 350         }
 351
 352         if (flags & KEXEC_FILE_UNLOAD)
 353                 goto exchange;
 354
 355         /*
 356          * In case of crash, new kernel gets loaded in reserved region. It is
 357          * same memory where old crash kernel might be loaded. Free any
 358          * current crash dump kernel before we corrupt it.
 359          */
 360         if (flags & KEXEC_FILE_ON_CRASH)
 361                 kimage_free(xchg(&kexec_crash_image, NULL));
 362
 363         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 364                                      cmdline_len, flags);
 365         if (ret)
 366                 goto out;
 367
 368         ret = machine_kexec_prepare(image);
 369         if (ret)
 370                 goto out;
 371
 372         /*
 373          * Some architecture(like S390) may touch the crash memory before
 374          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 375          */
 376         ret = kimage_crash_copy_vmcoreinfo(image);
 377         if (ret)
 378                 goto out;
 379
 380         ret = kexec_calculate_store_digests(image);
 381         if (ret)
 382                 goto out;
 383
 384         for (i = 0; i < image->nr_segments; i++) {
 385                 struct kexec_segment *ksegment;
 386
 387                 ksegment = &image->segment[i];
 388                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 389                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 390                          ksegment->memsz);
 391
 392                 ret = kimage_load_segment(image, &image->segment[i]);
 393                 if (ret)
 394                         goto out;
 395         }
 396
 397         kimage_terminate(image);
 398
 399         /*
 400          * Free up any temporary buffers allocated which are not needed
 401          * after image has been loaded
 402          */
 403         kimage_file_post_load_cleanup(image);
 404 exchange:
 405         image = xchg(dest_image, image);
 406 out:
 407         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 408                 arch_kexec_protect_crashkres();
 409
 410         mutex_unlock(&kexec_mutex);
 411         kimage_free(image);
 412         return ret;
 413 }
 414
 415 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 416                                     struct kexec_buf *kbuf)
 417 {
 418         struct kimage *image = kbuf->image;
 419         unsigned long temp_start, temp_end;
 420
 421         temp_end = min(end, kbuf->buf_max);
 422         temp_start = temp_end - kbuf->memsz;
 423
 424         do {
 425                 /* align down start */
 426                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 427
 428                 if (temp_start < start || temp_start < kbuf->buf_min)
 429                         return 0;
 430
 431                 temp_end = temp_start + kbuf->memsz - 1;
 432
 433                 /*
 434                  * Make sure this does not conflict with any of existing
 435                  * segments
 436                  */
 437                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 438                         temp_start = temp_start - PAGE_SIZE;
 439                         continue;
 440                 }
 441
 442                 /* We found a suitable memory range */
 443                 break;
 444         } while (1);
 445
 446         /* If we are here, we found a suitable memory range */
 447         kbuf->mem = temp_start;
 448
 449         /* Success, stop navigating through remaining System RAM ranges */
 450         return 1;
 451 }
 452
 453 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 454                                      struct kexec_buf *kbuf)
 455 {
 456         struct kimage *image = kbuf->image;
 457         unsigned long temp_start, temp_end;
 458
 459         temp_start = max(start, kbuf->buf_min);
 460
 461         do {
 462                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 463                 temp_end = temp_start + kbuf->memsz - 1;
 464
 465                 if (temp_end > end || temp_end > kbuf->buf_max)
 466                         return 0;
 467                 /*
 468                  * Make sure this does not conflict with any of existing
 469                  * segments
 470                  */
 471                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 472                         temp_start = temp_start + PAGE_SIZE;
 473                         continue;
 474                 }
 475
 476                 /* We found a suitable memory range */
 477                 break;
 478         } while (1);
 479
 480         /* If we are here, we found a suitable memory range */
 481         kbuf->mem = temp_start;
 482
 483         /* Success, stop navigating through remaining System RAM ranges */
 484         return 1;
 485 }
 486
 487 static int locate_mem_hole_callback(struct resource *res, void *arg)
 488 {
 489         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 490         u64 start = res->start, end = res->end;
 491         unsigned long sz = end - start + 1;
 492
 493         /* Returning 0 will take to next memory range */
 494         if (sz < kbuf->memsz)
 495                 return 0;
 496
 497         if (end < kbuf->buf_min || start > kbuf->buf_max)
 498                 return 0;
 499
 500         /*
 501          * Allocate memory top down with-in ram range. Otherwise bottom up
 502          * allocation.
 503          */
 504         if (kbuf->top_down)
 505                 return locate_mem_hole_top_down(start, end, kbuf);
 506         return locate_mem_hole_bottom_up(start, end, kbuf);
 507 }
 508
 509 /**
 510  * arch_kexec_walk_mem - call func(data) on free memory regions
 511  * @kbuf:       Context info for the search. Also passed to @func.
 512  * @func:       Function to call for each memory region.
 513  *
 514  * Return: The memory walk will stop when func returns a non-zero value
 515  * and that value will be returned. If all free regions are visited without
 516  * func returning non-zero, then zero will be returned.
 517  */
 518 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
 519                                int (*func)(struct resource *, void *))
 520 {
 521         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 522                 return walk_iomem_res_desc(crashk_res.desc,
 523                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 524                                            crashk_res.start, crashk_res.end,
 525                                            kbuf, func);
 526         else
 527                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 528 }
 529
 530 /**
 531  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 532  * @kbuf:       Parameters for the memory search.
 533  *
 534  * On success, kbuf->mem will have the start address of the memory region found.
 535  *
 536  * Return: 0 on success, negative errno on error.
 537  */
 538 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 539 {
 540         int ret;
 541
 542         ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
 543
 544         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 545 }
 546
 547 /**
 548  * kexec_add_buffer - place a buffer in a kexec segment
 549  * @kbuf:       Buffer contents and memory parameters.
 550  *
 551  * This function assumes that kexec_mutex is held.
 552  * On successful return, @kbuf->mem will have the physical address of
 553  * the buffer in memory.
 554  *
 555  * Return: 0 on success, negative errno on error.
 556  */
 557 int kexec_add_buffer(struct kexec_buf *kbuf)
 558 {
 559
 560         struct kexec_segment *ksegment;
 561         int ret;
 562
 563         /* Currently adding segment this way is allowed only in file mode */
 564         if (!kbuf->image->file_mode)
 565                 return -EINVAL;
 566
 567         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 568                 return -EINVAL;
 569
 570         /*
 571          * Make sure we are not trying to add buffer after allocating
 572          * control pages. All segments need to be placed first before
 573          * any control pages are allocated. As control page allocation
 574          * logic goes through list of segments to make sure there are
 575          * no destination overlaps.
 576          */
 577         if (!list_empty(&kbuf->image->control_pages)) {
 578                 WARN_ON(1);
 579                 return -EINVAL;
 580         }
 581
 582         /* Ensure minimum alignment needed for segments. */
 583         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 584         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 585
 586         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 587         ret = kexec_locate_mem_hole(kbuf);
 588         if (ret)
 589                 return ret;
 590
 591         /* Found a suitable memory range */
 592         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 593         ksegment->kbuf = kbuf->buffer;
 594         ksegment->bufsz = kbuf->bufsz;
 595         ksegment->mem = kbuf->mem;
 596         ksegment->memsz = kbuf->memsz;
 597         kbuf->image->nr_segments++;
 598         return 0;
 599 }
 600
 601 /* Calculate and store the digest of segments */
 602 static int kexec_calculate_store_digests(struct kimage *image)
 603 {
 604         struct crypto_shash *tfm;
 605         struct shash_desc *desc;
 606         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 607         size_t desc_size, nullsz;
 608         char *digest;
 609         void *zero_buf;
 610         struct kexec_sha_region *sha_regions;
 611         struct purgatory_info *pi = &image->purgatory_info;
 612
 613         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
 614                 return 0;
 615
 616         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 617         zero_buf_sz = PAGE_SIZE;
 618
 619         tfm = crypto_alloc_shash("sha256", 0, 0);
 620         if (IS_ERR(tfm)) {
 621                 ret = PTR_ERR(tfm);
 622                 goto out;
 623         }
 624
 625         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 626         desc = kzalloc(desc_size, GFP_KERNEL);
 627         if (!desc) {
 628                 ret = -ENOMEM;
 629                 goto out_free_tfm;
 630         }
 631
 632         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 633         sha_regions = vzalloc(sha_region_sz);
 634         if (!sha_regions) {
 635                 ret = -ENOMEM;
 636                 goto out_free_desc;
 637         }
 638
 639         desc->tfm   = tfm;
 640         desc->flags = 0;
 641
 642         ret = crypto_shash_init(desc);
 643         if (ret < 0)
 644                 goto out_free_sha_regions;
 645
 646         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 647         if (!digest) {
 648                 ret = -ENOMEM;
 649                 goto out_free_sha_regions;
 650         }
 651
 652         for (j = i = 0; i < image->nr_segments; i++) {
 653                 struct kexec_segment *ksegment;
 654
 655                 ksegment = &image->segment[i];
 656                 /*
 657                  * Skip purgatory as it will be modified once we put digest
 658                  * info in purgatory.
 659                  */
 660                 if (ksegment->kbuf == pi->purgatory_buf)
 661                         continue;
 662
 663                 ret = crypto_shash_update(desc, ksegment->kbuf,
 664                                           ksegment->bufsz);
 665                 if (ret)
 666                         break;
 667
 668                 /*
 669                  * Assume rest of the buffer is filled with zero and
 670                  * update digest accordingly.
 671                  */
 672                 nullsz = ksegment->memsz - ksegment->bufsz;
 673                 while (nullsz) {
 674                         unsigned long bytes = nullsz;
 675
 676                         if (bytes > zero_buf_sz)
 677                                 bytes = zero_buf_sz;
 678                         ret = crypto_shash_update(desc, zero_buf, bytes);
 679                         if (ret)
 680                                 break;
 681                         nullsz -= bytes;
 682                 }
 683
 684                 if (ret)
 685                         break;
 686
 687                 sha_regions[j].start = ksegment->mem;
 688                 sha_regions[j].len = ksegment->memsz;
 689                 j++;
 690         }
 691
 692         if (!ret) {
 693                 ret = crypto_shash_final(desc, digest);
 694                 if (ret)
 695                         goto out_free_digest;
 696                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 697                                                      sha_regions, sha_region_sz, 0);
 698                 if (ret)
 699                         goto out_free_digest;
 700
 701                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 702                                                      digest, SHA256_DIGEST_SIZE, 0);
 703                 if (ret)
 704                         goto out_free_digest;
 705         }
 706
 707 out_free_digest:
 708         kfree(digest);
 709 out_free_sha_regions:
 710         vfree(sha_regions);
 711 out_free_desc:
 712         kfree(desc);
 713 out_free_tfm:
 714         kfree(tfm);
 715 out:
 716         return ret;
 717 }
 718
 719 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
 720 /*
 721  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 722  * @pi:         Purgatory to be loaded.
 723  * @kbuf:       Buffer to setup.
 724  *
 725  * Allocates the memory needed for the buffer. Caller is responsible to free
 726  * the memory after use.
 727  *
 728  * Return: 0 on success, negative errno on error.
 729  */
 730 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 731                                       struct kexec_buf *kbuf)
 732 {
 733         const Elf_Shdr *sechdrs;
 734         unsigned long bss_align;
 735         unsigned long bss_sz;
 736         unsigned long align;
 737         int i, ret;
 738
 739         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 740         kbuf->buf_align = bss_align = 1;
 741         kbuf->bufsz = bss_sz = 0;
 742
 743         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 744                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 745                         continue;
 746
 747                 align = sechdrs[i].sh_addralign;
 748                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 749                         if (kbuf->buf_align < align)
 750                                 kbuf->buf_align = align;
 751                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 752                         kbuf->bufsz += sechdrs[i].sh_size;
 753                 } else {
 754                         if (bss_align < align)
 755                                 bss_align = align;
 756                         bss_sz = ALIGN(bss_sz, align);
 757                         bss_sz += sechdrs[i].sh_size;
 758                 }
 759         }
 760         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 761         kbuf->memsz = kbuf->bufsz + bss_sz;
 762         if (kbuf->buf_align < bss_align)
 763                 kbuf->buf_align = bss_align;
 764
 765         kbuf->buffer = vzalloc(kbuf->bufsz);
 766         if (!kbuf->buffer)
 767                 return -ENOMEM;
 768         pi->purgatory_buf = kbuf->buffer;
 769
 770         ret = kexec_add_buffer(kbuf);
 771         if (ret)
 772                 goto out;
 773
 774         return 0;
 775 out:
 776         vfree(pi->purgatory_buf);
 777         pi->purgatory_buf = NULL;
 778         return ret;
 779 }
 780
 781 /*
 782  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 783  * @pi:         Purgatory to be loaded.
 784  * @kbuf:       Buffer prepared to store purgatory.
 785  *
 786  * Allocates the memory needed for the buffer. Caller is responsible to free
 787  * the memory after use.
 788  *
 789  * Return: 0 on success, negative errno on error.
 790  */
 791 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 792                                          struct kexec_buf *kbuf)
 793 {
 794         unsigned long bss_addr;
 795         unsigned long offset;
 796         Elf_Shdr *sechdrs;
 797         int i;
 798
 799         /*
 800          * The section headers in kexec_purgatory are read-only. In order to
 801          * have them modifiable make a temporary copy.
 802          */
 803         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
 804         if (!sechdrs)
 805                 return -ENOMEM;
 806         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
 807                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 808         pi->sechdrs = sechdrs;
 809
 810         offset = 0;
 811         bss_addr = kbuf->mem + kbuf->bufsz;
 812         kbuf->image->start = pi->ehdr->e_entry;
 813
 814         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 815                 unsigned long align;
 816                 void *src, *dst;
 817
 818                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 819                         continue;
 820
 821                 align = sechdrs[i].sh_addralign;
 822                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 823                         bss_addr = ALIGN(bss_addr, align);
 824                         sechdrs[i].sh_addr = bss_addr;
 825                         bss_addr += sechdrs[i].sh_size;
 826                         continue;
 827                 }
 828
 829                 offset = ALIGN(offset, align);
 830                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 831                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 832                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 833                                          + sechdrs[i].sh_size)) {
 834                         kbuf->image->start -= sechdrs[i].sh_addr;
 835                         kbuf->image->start += kbuf->mem + offset;
 836                 }
 837
 838                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 839                 dst = pi->purgatory_buf + offset;
 840                 memcpy(dst, src, sechdrs[i].sh_size);
 841
 842                 sechdrs[i].sh_addr = kbuf->mem + offset;
 843                 sechdrs[i].sh_offset = offset;
 844                 offset += sechdrs[i].sh_size;
 845         }
 846
 847         return 0;
 848 }
 849
 850 static int kexec_apply_relocations(struct kimage *image)
 851 {
 852         int i, ret;
 853         struct purgatory_info *pi = &image->purgatory_info;
 854         const Elf_Shdr *sechdrs;
 855
 856         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 857
 858         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 859                 const Elf_Shdr *relsec;
 860                 const Elf_Shdr *symtab;
 861                 Elf_Shdr *section;
 862
 863                 relsec = sechdrs + i;
 864
 865                 if (relsec->sh_type != SHT_RELA &&
 866                     relsec->sh_type != SHT_REL)
 867                         continue;
 868
 869                 /*
 870                  * For section of type SHT_RELA/SHT_REL,
 871                  * ->sh_link contains section header index of associated
 872                  * symbol table. And ->sh_info contains section header
 873                  * index of section to which relocations apply.
 874                  */
 875                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 876                     relsec->sh_link >= pi->ehdr->e_shnum)
 877                         return -ENOEXEC;
 878
 879                 section = pi->sechdrs + relsec->sh_info;
 880                 symtab = sechdrs + relsec->sh_link;
 881
 882                 if (!(section->sh_flags & SHF_ALLOC))
 883                         continue;
 884
 885                 /*
 886                  * symtab->sh_link contain section header index of associated
 887                  * string table.
 888                  */
 889                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 890                         /* Invalid section number? */
 891                         continue;
 892
 893                 /*
 894                  * Respective architecture needs to provide support for applying
 895                  * relocations of type SHT_RELA/SHT_REL.
 896                  */
 897                 if (relsec->sh_type == SHT_RELA)
 898                         ret = arch_kexec_apply_relocations_add(pi, section,
 899                                                                relsec, symtab);
 900                 else if (relsec->sh_type == SHT_REL)
 901                         ret = arch_kexec_apply_relocations(pi, section,
 902                                                            relsec, symtab);
 903                 if (ret)
 904                         return ret;
 905         }
 906
 907         return 0;
 908 }
 909
 910 /*
 911  * kexec_load_purgatory - Load and relocate the purgatory object.
 912  * @image:      Image to add the purgatory to.
 913  * @kbuf:       Memory parameters to use.
 914  *
 915  * Allocates the memory needed for image->purgatory_info.sechdrs and
 916  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
 917  * to free the memory after use.
 918  *
 919  * Return: 0 on success, negative errno on error.
 920  */
 921 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
 922 {
 923         struct purgatory_info *pi = &image->purgatory_info;
 924         int ret;
 925
 926         if (kexec_purgatory_size <= 0)
 927                 return -EINVAL;
 928
 929         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
 930
 931         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
 932         if (ret)
 933                 return ret;
 934
 935         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
 936         if (ret)
 937                 goto out_free_kbuf;
 938
 939         ret = kexec_apply_relocations(image);
 940         if (ret)
 941                 goto out;
 942
 943         return 0;
 944 out:
 945         vfree(pi->sechdrs);
 946         pi->sechdrs = NULL;
 947 out_free_kbuf:
 948         vfree(pi->purgatory_buf);
 949         pi->purgatory_buf = NULL;
 950         return ret;
 951 }
 952
 953 /*
 954  * kexec_purgatory_find_symbol - find a symbol in the purgatory
 955  * @pi:         Purgatory to search in.
 956  * @name:       Name of the symbol.
 957  *
 958  * Return: pointer to symbol in read-only symtab on success, NULL on error.
 959  */
 960 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 961                                                   const char *name)
 962 {
 963         const Elf_Shdr *sechdrs;
 964         const Elf_Ehdr *ehdr;
 965         const Elf_Sym *syms;
 966         const char *strtab;
 967         int i, k;
 968
 969         if (!pi->ehdr)
 970                 return NULL;
 971
 972         ehdr = pi->ehdr;
 973         sechdrs = (void *)ehdr + ehdr->e_shoff;
 974
 975         for (i = 0; i < ehdr->e_shnum; i++) {
 976                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 977                         continue;
 978
 979                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 980                         /* Invalid strtab section number */
 981                         continue;
 982                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
 983                 syms = (void *)ehdr + sechdrs[i].sh_offset;
 984
 985                 /* Go through symbols for a match */
 986                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 987                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 988                                 continue;
 989
 990                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 991                                 continue;
 992
 993                         if (syms[k].st_shndx == SHN_UNDEF ||
 994                             syms[k].st_shndx >= ehdr->e_shnum) {
 995                                 pr_debug("Symbol: %s has bad section index %d.\n",
 996                                                 name, syms[k].st_shndx);
 997                                 return NULL;
 998                         }
 999
1000                         /* Found the symbol we are looking for */
1001                         return &syms[k];
1002                 }
1003         }
1004
1005         return NULL;
1006 }
1007
1008 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1009 {
1010         struct purgatory_info *pi = &image->purgatory_info;
1011         const Elf_Sym *sym;
1012         Elf_Shdr *sechdr;
1013
1014         sym = kexec_purgatory_find_symbol(pi, name);
1015         if (!sym)
1016                 return ERR_PTR(-EINVAL);
1017
1018         sechdr = &pi->sechdrs[sym->st_shndx];
1019
1020         /*
1021          * Returns the address where symbol will finally be loaded after
1022          * kexec_load_segment()
1023          */
1024         return (void *)(sechdr->sh_addr + sym->st_value);
1025 }
1026
1027 /*
1028  * Get or set value of a symbol. If "get_value" is true, symbol value is
1029  * returned in buf otherwise symbol value is set based on value in buf.
1030  */
1031 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1032                                    void *buf, unsigned int size, bool get_value)
1033 {
1034         struct purgatory_info *pi = &image->purgatory_info;
1035         const Elf_Sym *sym;
1036         Elf_Shdr *sec;
1037         char *sym_buf;
1038
1039         sym = kexec_purgatory_find_symbol(pi, name);
1040         if (!sym)
1041                 return -EINVAL;
1042
1043         if (sym->st_size != size) {
1044                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1045                        name, (unsigned long)sym->st_size, size);
1046                 return -EINVAL;
1047         }
1048
1049         sec = pi->sechdrs + sym->st_shndx;
1050
1051         if (sec->sh_type == SHT_NOBITS) {
1052                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1053                        get_value ? "get" : "set");
1054                 return -EINVAL;
1055         }
1056
1057         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1058
1059         if (get_value)
1060                 memcpy((void *)buf, sym_buf, size);
1061         else
1062                 memcpy((void *)sym_buf, buf, size);
1063
1064         return 0;
1065 }
1066 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1067
1068 int crash_exclude_mem_range(struct crash_mem *mem,
1069                             unsigned long long mstart, unsigned long long mend)
1070 {
1071         int i, j;
1072         unsigned long long start, end;
1073         struct crash_mem_range temp_range = {0, 0};
1074
1075         for (i = 0; i < mem->nr_ranges; i++) {
1076                 start = mem->ranges[i].start;
1077                 end = mem->ranges[i].end;
1078
1079                 if (mstart > end || mend < start)
1080                         continue;
1081
1082                 /* Truncate any area outside of range */
1083                 if (mstart < start)
1084                         mstart = start;
1085                 if (mend > end)
1086                         mend = end;
1087
1088                 /* Found completely overlapping range */
1089                 if (mstart == start && mend == end) {
1090                         mem->ranges[i].start = 0;
1091                         mem->ranges[i].end = 0;
1092                         if (i < mem->nr_ranges - 1) {
1093                                 /* Shift rest of the ranges to left */
1094                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1095                                         mem->ranges[j].start =
1096                                                 mem->ranges[j+1].start;
1097                                         mem->ranges[j].end =
1098                                                         mem->ranges[j+1].end;
1099                                 }
1100                         }
1101                         mem->nr_ranges--;
1102                         return 0;
1103                 }
1104
1105                 if (mstart > start && mend < end) {
1106                         /* Split original range */
1107                         mem->ranges[i].end = mstart - 1;
1108                         temp_range.start = mend + 1;
1109                         temp_range.end = end;
1110                 } else if (mstart != start)
1111                         mem->ranges[i].end = mstart - 1;
1112                 else
1113                         mem->ranges[i].start = mend + 1;
1114                 break;
1115         }
1116
1117         /* If a split happened, add the split to array */
1118         if (!temp_range.end)
1119                 return 0;
1120
1121         /* Split happened */
1122         if (i == mem->max_nr_ranges - 1)
1123                 return -ENOMEM;
1124
1125         /* Location where new range should go */
1126         j = i + 1;
1127         if (j < mem->nr_ranges) {
1128                 /* Move over all ranges one slot towards the end */
1129                 for (i = mem->nr_ranges - 1; i >= j; i--)
1130                         mem->ranges[i + 1] = mem->ranges[i];
1131         }
1132
1133         mem->ranges[j].start = temp_range.start;
1134         mem->ranges[j].end = temp_range.end;
1135         mem->nr_ranges++;
1136         return 0;
1137 }
1138
1139 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1140                           void **addr, unsigned long *sz)
1141 {
1142         Elf64_Ehdr *ehdr;
1143         Elf64_Phdr *phdr;
1144         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1145         unsigned char *buf;
1146         unsigned int cpu, i;
1147         unsigned long long notes_addr;
1148         unsigned long mstart, mend;
1149
1150         /* extra phdr for vmcoreinfo elf note */
1151         nr_phdr = nr_cpus + 1;
1152         nr_phdr += mem->nr_ranges;
1153
1154         /*
1155          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1156          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1157          * I think this is required by tools like gdb. So same physical
1158          * memory will be mapped in two elf headers. One will contain kernel
1159          * text virtual addresses and other will have __va(physical) addresses.
1160          */
1161
1162         nr_phdr++;
1163         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1164         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1165
1166         buf = vzalloc(elf_sz);
1167         if (!buf)
1168                 return -ENOMEM;
1169
1170         ehdr = (Elf64_Ehdr *)buf;
1171         phdr = (Elf64_Phdr *)(ehdr + 1);
1172         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1173         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1174         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1175         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1176         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1177         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1178         ehdr->e_type = ET_CORE;
1179         ehdr->e_machine = ELF_ARCH;
1180         ehdr->e_version = EV_CURRENT;
1181         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1182         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1183         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1184
1185         /* Prepare one phdr of type PT_NOTE for each present cpu */
1186         for_each_present_cpu(cpu) {
1187                 phdr->p_type = PT_NOTE;
1188                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1189                 phdr->p_offset = phdr->p_paddr = notes_addr;
1190                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1191                 (ehdr->e_phnum)++;
1192                 phdr++;
1193         }
1194
1195         /* Prepare one PT_NOTE header for vmcoreinfo */
1196         phdr->p_type = PT_NOTE;
1197         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1198         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1199         (ehdr->e_phnum)++;
1200         phdr++;
1201
1202         /* Prepare PT_LOAD type program header for kernel text region */
1203         if (kernel_map) {
1204                 phdr->p_type = PT_LOAD;
1205                 phdr->p_flags = PF_R|PF_W|PF_X;
1206                 phdr->p_vaddr = (Elf64_Addr)_text;
1207                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1208                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1209                 ehdr->e_phnum++;
1210                 phdr++;
1211         }
1212
1213         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1214         for (i = 0; i < mem->nr_ranges; i++) {
1215                 mstart = mem->ranges[i].start;
1216                 mend = mem->ranges[i].end;
1217
1218                 phdr->p_type = PT_LOAD;
1219                 phdr->p_flags = PF_R|PF_W|PF_X;
1220                 phdr->p_offset  = mstart;
1221
1222                 phdr->p_paddr = mstart;
1223                 phdr->p_vaddr = (unsigned long long) __va(mstart);
1224                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1225                 phdr->p_align = 0;
1226                 ehdr->e_phnum++;
1227                 phdr++;
1228                 pr_debug("Crash PT_LOAD elf header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
1229                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1230                         ehdr->e_phnum, phdr->p_offset);
1231         }
1232
1233         *addr = buf;
1234         *sz = elf_sz;
1235         return 0;
1236 }