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/memblock.h>
  20 #include <linux/mutex.h>
  21 #include <linux/list.h>
  22 #include <linux/fs.h>
  23 #include <linux/ima.h>
  24 #include <crypto/hash.h>
  25 #include <crypto/sha.h>
  26 #include <linux/elf.h>
  27 #include <linux/elfcore.h>
  28 #include <linux/kernel.h>
  29 #include <linux/syscalls.h>
  30 #include <linux/vmalloc.h>
  31 #include "kexec_internal.h"
  32
  33 static int kexec_calculate_store_digests(struct kimage *image);
  34
  35 /*
  36  * Currently this is the only default function that is exported as some
  37  * architectures need it to do additional handlings.
  38  * In the future, other default functions may be exported too if required.
  39  */
  40 int kexec_image_probe_default(struct kimage *image, void *buf,
  41                               unsigned long buf_len)
  42 {
  43         const struct kexec_file_ops * const *fops;
  44         int ret = -ENOEXEC;
  45
  46         for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
  47                 ret = (*fops)->probe(buf, buf_len);
  48                 if (!ret) {
  49                         image->fops = *fops;
  50                         return ret;
  51                 }
  52         }
  53
  54         return ret;
  55 }
  56
  57 /* Architectures can provide this probe function */
  58 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  59                                          unsigned long buf_len)
  60 {
  61         return kexec_image_probe_default(image, buf, buf_len);
  62 }
  63
  64 static void *kexec_image_load_default(struct kimage *image)
  65 {
  66         if (!image->fops || !image->fops->load)
  67                 return ERR_PTR(-ENOEXEC);
  68
  69         return image->fops->load(image, image->kernel_buf,
  70                                  image->kernel_buf_len, image->initrd_buf,
  71                                  image->initrd_buf_len, image->cmdline_buf,
  72                                  image->cmdline_buf_len);
  73 }
  74
  75 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
  76 {
  77         return kexec_image_load_default(image);
  78 }
  79
  80 int kexec_image_post_load_cleanup_default(struct kimage *image)
  81 {
  82         if (!image->fops || !image->fops->cleanup)
  83                 return 0;
  84
  85         return image->fops->cleanup(image->image_loader_data);
  86 }
  87
  88 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
  89 {
  90         return kexec_image_post_load_cleanup_default(image);
  91 }
  92
  93 #ifdef CONFIG_KEXEC_VERIFY_SIG
  94 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
  95                                           unsigned long buf_len)
  96 {
  97         if (!image->fops || !image->fops->verify_sig) {
  98                 pr_debug("kernel loader does not support signature verification.\n");
  99                 return -EKEYREJECTED;
 100         }
 101
 102         return image->fops->verify_sig(buf, buf_len);
 103 }
 104
 105 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
 106                                         unsigned long buf_len)
 107 {
 108         return kexec_image_verify_sig_default(image, buf, buf_len);
 109 }
 110 #endif
 111
 112 /*
 113  * arch_kexec_apply_relocations_add - apply relocations of type RELA
 114  * @pi:         Purgatory to be relocated.
 115  * @section:    Section relocations applying to.
 116  * @relsec:     Section containing RELAs.
 117  * @symtab:     Corresponding symtab.
 118  *
 119  * Return: 0 on success, negative errno on error.
 120  */
 121 int __weak
 122 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
 123                                  const Elf_Shdr *relsec, const Elf_Shdr *symtab)
 124 {
 125         pr_err("RELA relocation unsupported.\n");
 126         return -ENOEXEC;
 127 }
 128
 129 /*
 130  * arch_kexec_apply_relocations - apply relocations of type REL
 131  * @pi:         Purgatory to be relocated.
 132  * @section:    Section relocations applying to.
 133  * @relsec:     Section containing RELs.
 134  * @symtab:     Corresponding symtab.
 135  *
 136  * Return: 0 on success, negative errno on error.
 137  */
 138 int __weak
 139 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
 140                              const Elf_Shdr *relsec, const Elf_Shdr *symtab)
 141 {
 142         pr_err("REL relocation unsupported.\n");
 143         return -ENOEXEC;
 144 }
 145
 146 /*
 147  * Free up memory used by kernel, initrd, and command line. This is temporary
 148  * memory allocation which is not needed any more after these buffers have
 149  * been loaded into separate segments and have been copied elsewhere.
 150  */
 151 void kimage_file_post_load_cleanup(struct kimage *image)
 152 {
 153         struct purgatory_info *pi = &image->purgatory_info;
 154
 155         vfree(image->kernel_buf);
 156         image->kernel_buf = NULL;
 157
 158         vfree(image->initrd_buf);
 159         image->initrd_buf = NULL;
 160
 161         kfree(image->cmdline_buf);
 162         image->cmdline_buf = NULL;
 163
 164         vfree(pi->purgatory_buf);
 165         pi->purgatory_buf = NULL;
 166
 167         vfree(pi->sechdrs);
 168         pi->sechdrs = NULL;
 169
 170         /* See if architecture has anything to cleanup post load */
 171         arch_kimage_file_post_load_cleanup(image);
 172
 173         /*
 174          * Above call should have called into bootloader to free up
 175          * any data stored in kimage->image_loader_data. It should
 176          * be ok now to free it up.
 177          */
 178         kfree(image->image_loader_data);
 179         image->image_loader_data = NULL;
 180 }
 181
 182 /*
 183  * In file mode list of segments is prepared by kernel. Copy relevant
 184  * data from user space, do error checking, prepare segment list
 185  */
 186 static int
 187 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 188                              const char __user *cmdline_ptr,
 189                              unsigned long cmdline_len, unsigned flags)
 190 {
 191         int ret = 0;
 192         void *ldata;
 193         loff_t size;
 194
 195         ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
 196                                        &size, INT_MAX, READING_KEXEC_IMAGE);
 197         if (ret)
 198                 return ret;
 199         image->kernel_buf_len = size;
 200
 201         /* IMA needs to pass the measurement list to the next kernel. */
 202         ima_add_kexec_buffer(image);
 203
 204         /* Call arch image probe handlers */
 205         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 206                                             image->kernel_buf_len);
 207         if (ret)
 208                 goto out;
 209
 210 #ifdef CONFIG_KEXEC_VERIFY_SIG
 211         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 212                                            image->kernel_buf_len);
 213         if (ret) {
 214                 pr_debug("kernel signature verification failed.\n");
 215                 goto out;
 216         }
 217         pr_debug("kernel signature verification successful.\n");
 218 #endif
 219         /* It is possible that there no initramfs is being loaded */
 220         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 221                 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
 222                                                &size, INT_MAX,
 223                                                READING_KEXEC_INITRAMFS);
 224                 if (ret)
 225                         goto out;
 226                 image->initrd_buf_len = size;
 227         }
 228
 229         if (cmdline_len) {
 230                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 231                 if (IS_ERR(image->cmdline_buf)) {
 232                         ret = PTR_ERR(image->cmdline_buf);
 233                         image->cmdline_buf = NULL;
 234                         goto out;
 235                 }
 236
 237                 image->cmdline_buf_len = cmdline_len;
 238
 239                 /* command line should be a string with last byte null */
 240                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 241                         ret = -EINVAL;
 242                         goto out;
 243                 }
 244         }
 245
 246         /* Call arch image load handlers */
 247         ldata = arch_kexec_kernel_image_load(image);
 248
 249         if (IS_ERR(ldata)) {
 250                 ret = PTR_ERR(ldata);
 251                 goto out;
 252         }
 253
 254         image->image_loader_data = ldata;
 255 out:
 256         /* In case of error, free up all allocated memory in this function */
 257         if (ret)
 258                 kimage_file_post_load_cleanup(image);
 259         return ret;
 260 }
 261
 262 static int
 263 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 264                        int initrd_fd, const char __user *cmdline_ptr,
 265                        unsigned long cmdline_len, unsigned long flags)
 266 {
 267         int ret;
 268         struct kimage *image;
 269         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 270
 271         image = do_kimage_alloc_init();
 272         if (!image)
 273                 return -ENOMEM;
 274
 275         image->file_mode = 1;
 276
 277         if (kexec_on_panic) {
 278                 /* Enable special crash kernel control page alloc policy. */
 279                 image->control_page = crashk_res.start;
 280                 image->type = KEXEC_TYPE_CRASH;
 281         }
 282
 283         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 284                                            cmdline_ptr, cmdline_len, flags);
 285         if (ret)
 286                 goto out_free_image;
 287
 288         ret = sanity_check_segment_list(image);
 289         if (ret)
 290                 goto out_free_post_load_bufs;
 291
 292         ret = -ENOMEM;
 293         image->control_code_page = kimage_alloc_control_pages(image,
 294                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 295         if (!image->control_code_page) {
 296                 pr_err("Could not allocate control_code_buffer\n");
 297                 goto out_free_post_load_bufs;
 298         }
 299
 300         if (!kexec_on_panic) {
 301                 image->swap_page = kimage_alloc_control_pages(image, 0);
 302                 if (!image->swap_page) {
 303                         pr_err("Could not allocate swap buffer\n");
 304                         goto out_free_control_pages;
 305                 }
 306         }
 307
 308         *rimage = image;
 309         return 0;
 310 out_free_control_pages:
 311         kimage_free_page_list(&image->control_pages);
 312 out_free_post_load_bufs:
 313         kimage_file_post_load_cleanup(image);
 314 out_free_image:
 315         kfree(image);
 316         return ret;
 317 }
 318
 319 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 320                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 321                 unsigned long, flags)
 322 {
 323         int ret = 0, i;
 324         struct kimage **dest_image, *image;
 325
 326         /* We only trust the superuser with rebooting the system. */
 327         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 328                 return -EPERM;
 329
 330         /* Make sure we have a legal set of flags */
 331         if (flags != (flags & KEXEC_FILE_FLAGS))
 332                 return -EINVAL;
 333
 334         image = NULL;
 335
 336         if (!mutex_trylock(&kexec_mutex))
 337                 return -EBUSY;
 338
 339         dest_image = &kexec_image;
 340         if (flags & KEXEC_FILE_ON_CRASH) {
 341                 dest_image = &kexec_crash_image;
 342                 if (kexec_crash_image)
 343                         arch_kexec_unprotect_crashkres();
 344         }
 345
 346         if (flags & KEXEC_FILE_UNLOAD)
 347                 goto exchange;
 348
 349         /*
 350          * In case of crash, new kernel gets loaded in reserved region. It is
 351          * same memory where old crash kernel might be loaded. Free any
 352          * current crash dump kernel before we corrupt it.
 353          */
 354         if (flags & KEXEC_FILE_ON_CRASH)
 355                 kimage_free(xchg(&kexec_crash_image, NULL));
 356
 357         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 358                                      cmdline_len, flags);
 359         if (ret)
 360                 goto out;
 361
 362         ret = machine_kexec_prepare(image);
 363         if (ret)
 364                 goto out;
 365
 366         /*
 367          * Some architecture(like S390) may touch the crash memory before
 368          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 369          */
 370         ret = kimage_crash_copy_vmcoreinfo(image);
 371         if (ret)
 372                 goto out;
 373
 374         ret = kexec_calculate_store_digests(image);
 375         if (ret)
 376                 goto out;
 377
 378         for (i = 0; i < image->nr_segments; i++) {
 379                 struct kexec_segment *ksegment;
 380
 381                 ksegment = &image->segment[i];
 382                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 383                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 384                          ksegment->memsz);
 385
 386                 ret = kimage_load_segment(image, &image->segment[i]);
 387                 if (ret)
 388                         goto out;
 389         }
 390
 391         kimage_terminate(image);
 392
 393         /*
 394          * Free up any temporary buffers allocated which are not needed
 395          * after image has been loaded
 396          */
 397         kimage_file_post_load_cleanup(image);
 398 exchange:
 399         image = xchg(dest_image, image);
 400 out:
 401         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 402                 arch_kexec_protect_crashkres();
 403
 404         mutex_unlock(&kexec_mutex);
 405         kimage_free(image);
 406         return ret;
 407 }
 408
 409 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 410                                     struct kexec_buf *kbuf)
 411 {
 412         struct kimage *image = kbuf->image;
 413         unsigned long temp_start, temp_end;
 414
 415         temp_end = min(end, kbuf->buf_max);
 416         temp_start = temp_end - kbuf->memsz;
 417
 418         do {
 419                 /* align down start */
 420                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 421
 422                 if (temp_start < start || temp_start < kbuf->buf_min)
 423                         return 0;
 424
 425                 temp_end = temp_start + kbuf->memsz - 1;
 426
 427                 /*
 428                  * Make sure this does not conflict with any of existing
 429                  * segments
 430                  */
 431                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 432                         temp_start = temp_start - PAGE_SIZE;
 433                         continue;
 434                 }
 435
 436                 /* We found a suitable memory range */
 437                 break;
 438         } while (1);
 439
 440         /* If we are here, we found a suitable memory range */
 441         kbuf->mem = temp_start;
 442
 443         /* Success, stop navigating through remaining System RAM ranges */
 444         return 1;
 445 }
 446
 447 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 448                                      struct kexec_buf *kbuf)
 449 {
 450         struct kimage *image = kbuf->image;
 451         unsigned long temp_start, temp_end;
 452
 453         temp_start = max(start, kbuf->buf_min);
 454
 455         do {
 456                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 457                 temp_end = temp_start + kbuf->memsz - 1;
 458
 459                 if (temp_end > end || temp_end > kbuf->buf_max)
 460                         return 0;
 461                 /*
 462                  * Make sure this does not conflict with any of existing
 463                  * segments
 464                  */
 465                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 466                         temp_start = temp_start + PAGE_SIZE;
 467                         continue;
 468                 }
 469
 470                 /* We found a suitable memory range */
 471                 break;
 472         } while (1);
 473
 474         /* If we are here, we found a suitable memory range */
 475         kbuf->mem = temp_start;
 476
 477         /* Success, stop navigating through remaining System RAM ranges */
 478         return 1;
 479 }
 480
 481 static int locate_mem_hole_callback(struct resource *res, void *arg)
 482 {
 483         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 484         u64 start = res->start, end = res->end;
 485         unsigned long sz = end - start + 1;
 486
 487         /* Returning 0 will take to next memory range */
 488         if (sz < kbuf->memsz)
 489                 return 0;
 490
 491         if (end < kbuf->buf_min || start > kbuf->buf_max)
 492                 return 0;
 493
 494         /*
 495          * Allocate memory top down with-in ram range. Otherwise bottom up
 496          * allocation.
 497          */
 498         if (kbuf->top_down)
 499                 return locate_mem_hole_top_down(start, end, kbuf);
 500         return locate_mem_hole_bottom_up(start, end, kbuf);
 501 }
 502
 503 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
 504 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 505                                int (*func)(struct resource *, void *))
 506 {
 507         return 0;
 508 }
 509 #else
 510 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 511                                int (*func)(struct resource *, void *))
 512 {
 513         int ret = 0;
 514         u64 i;
 515         phys_addr_t mstart, mend;
 516         struct resource res = { };
 517
 518         if (kbuf->top_down) {
 519                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, 0,
 520                                                 &mstart, &mend, NULL) {
 521                         /*
 522                          * In memblock, end points to the first byte after the
 523                          * range while in kexec, end points to the last byte
 524                          * in the range.
 525                          */
 526                         res.start = mstart;
 527                         res.end = mend - 1;
 528                         ret = func(&res, kbuf);
 529                         if (ret)
 530                                 break;
 531                 }
 532         } else {
 533                 for_each_free_mem_range(i, NUMA_NO_NODE, 0, &mstart, &mend,
 534                                         NULL) {
 535                         /*
 536                          * In memblock, end points to the first byte after the
 537                          * range while in kexec, end points to the last byte
 538                          * in the range.
 539                          */
 540                         res.start = mstart;
 541                         res.end = mend - 1;
 542                         ret = func(&res, kbuf);
 543                         if (ret)
 544                                 break;
 545                 }
 546         }
 547
 548         return ret;
 549 }
 550 #endif
 551
 552 /**
 553  * kexec_walk_resources - call func(data) on free memory regions
 554  * @kbuf:       Context info for the search. Also passed to @func.
 555  * @func:       Function to call for each memory region.
 556  *
 557  * Return: The memory walk will stop when func returns a non-zero value
 558  * and that value will be returned. If all free regions are visited without
 559  * func returning non-zero, then zero will be returned.
 560  */
 561 static int kexec_walk_resources(struct kexec_buf *kbuf,
 562                                 int (*func)(struct resource *, void *))
 563 {
 564         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 565                 return walk_iomem_res_desc(crashk_res.desc,
 566                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 567                                            crashk_res.start, crashk_res.end,
 568                                            kbuf, func);
 569         else
 570                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 571 }
 572
 573 /**
 574  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 575  * @kbuf:       Parameters for the memory search.
 576  *
 577  * On success, kbuf->mem will have the start address of the memory region found.
 578  *
 579  * Return: 0 on success, negative errno on error.
 580  */
 581 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 582 {
 583         int ret;
 584
 585         /* Arch knows where to place */
 586         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
 587                 return 0;
 588
 589         if (IS_ENABLED(CONFIG_ARCH_DISCARD_MEMBLOCK))
 590                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
 591         else
 592                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
 593
 594         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 595 }
 596
 597 /**
 598  * kexec_add_buffer - place a buffer in a kexec segment
 599  * @kbuf:       Buffer contents and memory parameters.
 600  *
 601  * This function assumes that kexec_mutex is held.
 602  * On successful return, @kbuf->mem will have the physical address of
 603  * the buffer in memory.
 604  *
 605  * Return: 0 on success, negative errno on error.
 606  */
 607 int kexec_add_buffer(struct kexec_buf *kbuf)
 608 {
 609
 610         struct kexec_segment *ksegment;
 611         int ret;
 612
 613         /* Currently adding segment this way is allowed only in file mode */
 614         if (!kbuf->image->file_mode)
 615                 return -EINVAL;
 616
 617         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 618                 return -EINVAL;
 619
 620         /*
 621          * Make sure we are not trying to add buffer after allocating
 622          * control pages. All segments need to be placed first before
 623          * any control pages are allocated. As control page allocation
 624          * logic goes through list of segments to make sure there are
 625          * no destination overlaps.
 626          */
 627         if (!list_empty(&kbuf->image->control_pages)) {
 628                 WARN_ON(1);
 629                 return -EINVAL;
 630         }
 631
 632         /* Ensure minimum alignment needed for segments. */
 633         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 634         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 635
 636         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 637         ret = kexec_locate_mem_hole(kbuf);
 638         if (ret)
 639                 return ret;
 640
 641         /* Found a suitable memory range */
 642         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 643         ksegment->kbuf = kbuf->buffer;
 644         ksegment->bufsz = kbuf->bufsz;
 645         ksegment->mem = kbuf->mem;
 646         ksegment->memsz = kbuf->memsz;
 647         kbuf->image->nr_segments++;
 648         return 0;
 649 }
 650
 651 /* Calculate and store the digest of segments */
 652 static int kexec_calculate_store_digests(struct kimage *image)
 653 {
 654         struct crypto_shash *tfm;
 655         struct shash_desc *desc;
 656         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 657         size_t desc_size, nullsz;
 658         char *digest;
 659         void *zero_buf;
 660         struct kexec_sha_region *sha_regions;
 661         struct purgatory_info *pi = &image->purgatory_info;
 662
 663         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
 664                 return 0;
 665
 666         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 667         zero_buf_sz = PAGE_SIZE;
 668
 669         tfm = crypto_alloc_shash("sha256", 0, 0);
 670         if (IS_ERR(tfm)) {
 671                 ret = PTR_ERR(tfm);
 672                 goto out;
 673         }
 674
 675         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 676         desc = kzalloc(desc_size, GFP_KERNEL);
 677         if (!desc) {
 678                 ret = -ENOMEM;
 679                 goto out_free_tfm;
 680         }
 681
 682         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 683         sha_regions = vzalloc(sha_region_sz);
 684         if (!sha_regions)
 685                 goto out_free_desc;
 686
 687         desc->tfm   = tfm;
 688         desc->flags = 0;
 689
 690         ret = crypto_shash_init(desc);
 691         if (ret < 0)
 692                 goto out_free_sha_regions;
 693
 694         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 695         if (!digest) {
 696                 ret = -ENOMEM;
 697                 goto out_free_sha_regions;
 698         }
 699
 700         for (j = i = 0; i < image->nr_segments; i++) {
 701                 struct kexec_segment *ksegment;
 702
 703                 ksegment = &image->segment[i];
 704                 /*
 705                  * Skip purgatory as it will be modified once we put digest
 706                  * info in purgatory.
 707                  */
 708                 if (ksegment->kbuf == pi->purgatory_buf)
 709                         continue;
 710
 711                 ret = crypto_shash_update(desc, ksegment->kbuf,
 712                                           ksegment->bufsz);
 713                 if (ret)
 714                         break;
 715
 716                 /*
 717                  * Assume rest of the buffer is filled with zero and
 718                  * update digest accordingly.
 719                  */
 720                 nullsz = ksegment->memsz - ksegment->bufsz;
 721                 while (nullsz) {
 722                         unsigned long bytes = nullsz;
 723
 724                         if (bytes > zero_buf_sz)
 725                                 bytes = zero_buf_sz;
 726                         ret = crypto_shash_update(desc, zero_buf, bytes);
 727                         if (ret)
 728                                 break;
 729                         nullsz -= bytes;
 730                 }
 731
 732                 if (ret)
 733                         break;
 734
 735                 sha_regions[j].start = ksegment->mem;
 736                 sha_regions[j].len = ksegment->memsz;
 737                 j++;
 738         }
 739
 740         if (!ret) {
 741                 ret = crypto_shash_final(desc, digest);
 742                 if (ret)
 743                         goto out_free_digest;
 744                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 745                                                      sha_regions, sha_region_sz, 0);
 746                 if (ret)
 747                         goto out_free_digest;
 748
 749                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 750                                                      digest, SHA256_DIGEST_SIZE, 0);
 751                 if (ret)
 752                         goto out_free_digest;
 753         }
 754
 755 out_free_digest:
 756         kfree(digest);
 757 out_free_sha_regions:
 758         vfree(sha_regions);
 759 out_free_desc:
 760         kfree(desc);
 761 out_free_tfm:
 762         kfree(tfm);
 763 out:
 764         return ret;
 765 }
 766
 767 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
 768 /*
 769  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 770  * @pi:         Purgatory to be loaded.
 771  * @kbuf:       Buffer to setup.
 772  *
 773  * Allocates the memory needed for the buffer. Caller is responsible to free
 774  * the memory after use.
 775  *
 776  * Return: 0 on success, negative errno on error.
 777  */
 778 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 779                                       struct kexec_buf *kbuf)
 780 {
 781         const Elf_Shdr *sechdrs;
 782         unsigned long bss_align;
 783         unsigned long bss_sz;
 784         unsigned long align;
 785         int i, ret;
 786
 787         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 788         kbuf->buf_align = bss_align = 1;
 789         kbuf->bufsz = bss_sz = 0;
 790
 791         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 792                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 793                         continue;
 794
 795                 align = sechdrs[i].sh_addralign;
 796                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 797                         if (kbuf->buf_align < align)
 798                                 kbuf->buf_align = align;
 799                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 800                         kbuf->bufsz += sechdrs[i].sh_size;
 801                 } else {
 802                         if (bss_align < align)
 803                                 bss_align = align;
 804                         bss_sz = ALIGN(bss_sz, align);
 805                         bss_sz += sechdrs[i].sh_size;
 806                 }
 807         }
 808         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 809         kbuf->memsz = kbuf->bufsz + bss_sz;
 810         if (kbuf->buf_align < bss_align)
 811                 kbuf->buf_align = bss_align;
 812
 813         kbuf->buffer = vzalloc(kbuf->bufsz);
 814         if (!kbuf->buffer)
 815                 return -ENOMEM;
 816         pi->purgatory_buf = kbuf->buffer;
 817
 818         ret = kexec_add_buffer(kbuf);
 819         if (ret)
 820                 goto out;
 821
 822         return 0;
 823 out:
 824         vfree(pi->purgatory_buf);
 825         pi->purgatory_buf = NULL;
 826         return ret;
 827 }
 828
 829 /*
 830  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 831  * @pi:         Purgatory to be loaded.
 832  * @kbuf:       Buffer prepared to store purgatory.
 833  *
 834  * Allocates the memory needed for the buffer. Caller is responsible to free
 835  * the memory after use.
 836  *
 837  * Return: 0 on success, negative errno on error.
 838  */
 839 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 840                                          struct kexec_buf *kbuf)
 841 {
 842         unsigned long bss_addr;
 843         unsigned long offset;
 844         Elf_Shdr *sechdrs;
 845         int i;
 846
 847         /*
 848          * The section headers in kexec_purgatory are read-only. In order to
 849          * have them modifiable make a temporary copy.
 850          */
 851         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
 852         if (!sechdrs)
 853                 return -ENOMEM;
 854         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
 855                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 856         pi->sechdrs = sechdrs;
 857
 858         offset = 0;
 859         bss_addr = kbuf->mem + kbuf->bufsz;
 860         kbuf->image->start = pi->ehdr->e_entry;
 861
 862         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 863                 unsigned long align;
 864                 void *src, *dst;
 865
 866                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 867                         continue;
 868
 869                 align = sechdrs[i].sh_addralign;
 870                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 871                         bss_addr = ALIGN(bss_addr, align);
 872                         sechdrs[i].sh_addr = bss_addr;
 873                         bss_addr += sechdrs[i].sh_size;
 874                         continue;
 875                 }
 876
 877                 offset = ALIGN(offset, align);
 878                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 879                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 880                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 881                                          + sechdrs[i].sh_size)) {
 882                         kbuf->image->start -= sechdrs[i].sh_addr;
 883                         kbuf->image->start += kbuf->mem + offset;
 884                 }
 885
 886                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 887                 dst = pi->purgatory_buf + offset;
 888                 memcpy(dst, src, sechdrs[i].sh_size);
 889
 890                 sechdrs[i].sh_addr = kbuf->mem + offset;
 891                 sechdrs[i].sh_offset = offset;
 892                 offset += sechdrs[i].sh_size;
 893         }
 894
 895         return 0;
 896 }
 897
 898 static int kexec_apply_relocations(struct kimage *image)
 899 {
 900         int i, ret;
 901         struct purgatory_info *pi = &image->purgatory_info;
 902         const Elf_Shdr *sechdrs;
 903
 904         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 905
 906         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 907                 const Elf_Shdr *relsec;
 908                 const Elf_Shdr *symtab;
 909                 Elf_Shdr *section;
 910
 911                 relsec = sechdrs + i;
 912
 913                 if (relsec->sh_type != SHT_RELA &&
 914                     relsec->sh_type != SHT_REL)
 915                         continue;
 916
 917                 /*
 918                  * For section of type SHT_RELA/SHT_REL,
 919                  * ->sh_link contains section header index of associated
 920                  * symbol table. And ->sh_info contains section header
 921                  * index of section to which relocations apply.
 922                  */
 923                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 924                     relsec->sh_link >= pi->ehdr->e_shnum)
 925                         return -ENOEXEC;
 926
 927                 section = pi->sechdrs + relsec->sh_info;
 928                 symtab = sechdrs + relsec->sh_link;
 929
 930                 if (!(section->sh_flags & SHF_ALLOC))
 931                         continue;
 932
 933                 /*
 934                  * symtab->sh_link contain section header index of associated
 935                  * string table.
 936                  */
 937                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 938                         /* Invalid section number? */
 939                         continue;
 940
 941                 /*
 942                  * Respective architecture needs to provide support for applying
 943                  * relocations of type SHT_RELA/SHT_REL.
 944                  */
 945                 if (relsec->sh_type == SHT_RELA)
 946                         ret = arch_kexec_apply_relocations_add(pi, section,
 947                                                                relsec, symtab);
 948                 else if (relsec->sh_type == SHT_REL)
 949                         ret = arch_kexec_apply_relocations(pi, section,
 950                                                            relsec, symtab);
 951                 if (ret)
 952                         return ret;
 953         }
 954
 955         return 0;
 956 }
 957
 958 /*
 959  * kexec_load_purgatory - Load and relocate the purgatory object.
 960  * @image:      Image to add the purgatory to.
 961  * @kbuf:       Memory parameters to use.
 962  *
 963  * Allocates the memory needed for image->purgatory_info.sechdrs and
 964  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
 965  * to free the memory after use.
 966  *
 967  * Return: 0 on success, negative errno on error.
 968  */
 969 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
 970 {
 971         struct purgatory_info *pi = &image->purgatory_info;
 972         int ret;
 973
 974         if (kexec_purgatory_size <= 0)
 975                 return -EINVAL;
 976
 977         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
 978
 979         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
 980         if (ret)
 981                 return ret;
 982
 983         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
 984         if (ret)
 985                 goto out_free_kbuf;
 986
 987         ret = kexec_apply_relocations(image);
 988         if (ret)
 989                 goto out;
 990
 991         return 0;
 992 out:
 993         vfree(pi->sechdrs);
 994         pi->sechdrs = NULL;
 995 out_free_kbuf:
 996         vfree(pi->purgatory_buf);
 997         pi->purgatory_buf = NULL;
 998         return ret;
 999 }
1000
1001 /*
1002  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1003  * @pi:         Purgatory to search in.
1004  * @name:       Name of the symbol.
1005  *
1006  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1007  */
1008 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1009                                                   const char *name)
1010 {
1011         const Elf_Shdr *sechdrs;
1012         const Elf_Ehdr *ehdr;
1013         const Elf_Sym *syms;
1014         const char *strtab;
1015         int i, k;
1016
1017         if (!pi->ehdr)
1018                 return NULL;
1019
1020         ehdr = pi->ehdr;
1021         sechdrs = (void *)ehdr + ehdr->e_shoff;
1022
1023         for (i = 0; i < ehdr->e_shnum; i++) {
1024                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1025                         continue;
1026
1027                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1028                         /* Invalid strtab section number */
1029                         continue;
1030                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1031                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1032
1033                 /* Go through symbols for a match */
1034                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1035                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1036                                 continue;
1037
1038                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1039                                 continue;
1040
1041                         if (syms[k].st_shndx == SHN_UNDEF ||
1042                             syms[k].st_shndx >= ehdr->e_shnum) {
1043                                 pr_debug("Symbol: %s has bad section index %d.\n",
1044                                                 name, syms[k].st_shndx);
1045                                 return NULL;
1046                         }
1047
1048                         /* Found the symbol we are looking for */
1049                         return &syms[k];
1050                 }
1051         }
1052
1053         return NULL;
1054 }
1055
1056 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1057 {
1058         struct purgatory_info *pi = &image->purgatory_info;
1059         const Elf_Sym *sym;
1060         Elf_Shdr *sechdr;
1061
1062         sym = kexec_purgatory_find_symbol(pi, name);
1063         if (!sym)
1064                 return ERR_PTR(-EINVAL);
1065
1066         sechdr = &pi->sechdrs[sym->st_shndx];
1067
1068         /*
1069          * Returns the address where symbol will finally be loaded after
1070          * kexec_load_segment()
1071          */
1072         return (void *)(sechdr->sh_addr + sym->st_value);
1073 }
1074
1075 /*
1076  * Get or set value of a symbol. If "get_value" is true, symbol value is
1077  * returned in buf otherwise symbol value is set based on value in buf.
1078  */
1079 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1080                                    void *buf, unsigned int size, bool get_value)
1081 {
1082         struct purgatory_info *pi = &image->purgatory_info;
1083         const Elf_Sym *sym;
1084         Elf_Shdr *sec;
1085         char *sym_buf;
1086
1087         sym = kexec_purgatory_find_symbol(pi, name);
1088         if (!sym)
1089                 return -EINVAL;
1090
1091         if (sym->st_size != size) {
1092                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1093                        name, (unsigned long)sym->st_size, size);
1094                 return -EINVAL;
1095         }
1096
1097         sec = pi->sechdrs + sym->st_shndx;
1098
1099         if (sec->sh_type == SHT_NOBITS) {
1100                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1101                        get_value ? "get" : "set");
1102                 return -EINVAL;
1103         }
1104
1105         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1106
1107         if (get_value)
1108                 memcpy((void *)buf, sym_buf, size);
1109         else
1110                 memcpy((void *)sym_buf, buf, size);
1111
1112         return 0;
1113 }
1114 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1115
1116 int crash_exclude_mem_range(struct crash_mem *mem,
1117                             unsigned long long mstart, unsigned long long mend)
1118 {
1119         int i, j;
1120         unsigned long long start, end;
1121         struct crash_mem_range temp_range = {0, 0};
1122
1123         for (i = 0; i < mem->nr_ranges; i++) {
1124                 start = mem->ranges[i].start;
1125                 end = mem->ranges[i].end;
1126
1127                 if (mstart > end || mend < start)
1128                         continue;
1129
1130                 /* Truncate any area outside of range */
1131                 if (mstart < start)
1132                         mstart = start;
1133                 if (mend > end)
1134                         mend = end;
1135
1136                 /* Found completely overlapping range */
1137                 if (mstart == start && mend == end) {
1138                         mem->ranges[i].start = 0;
1139                         mem->ranges[i].end = 0;
1140                         if (i < mem->nr_ranges - 1) {
1141                                 /* Shift rest of the ranges to left */
1142                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1143                                         mem->ranges[j].start =
1144                                                 mem->ranges[j+1].start;
1145                                         mem->ranges[j].end =
1146                                                         mem->ranges[j+1].end;
1147                                 }
1148                         }
1149                         mem->nr_ranges--;
1150                         return 0;
1151                 }
1152
1153                 if (mstart > start && mend < end) {
1154                         /* Split original range */
1155                         mem->ranges[i].end = mstart - 1;
1156                         temp_range.start = mend + 1;
1157                         temp_range.end = end;
1158                 } else if (mstart != start)
1159                         mem->ranges[i].end = mstart - 1;
1160                 else
1161                         mem->ranges[i].start = mend + 1;
1162                 break;
1163         }
1164
1165         /* If a split happened, add the split to array */
1166         if (!temp_range.end)
1167                 return 0;
1168
1169         /* Split happened */
1170         if (i == mem->max_nr_ranges - 1)
1171                 return -ENOMEM;
1172
1173         /* Location where new range should go */
1174         j = i + 1;
1175         if (j < mem->nr_ranges) {
1176                 /* Move over all ranges one slot towards the end */
1177                 for (i = mem->nr_ranges - 1; i >= j; i--)
1178                         mem->ranges[i + 1] = mem->ranges[i];
1179         }
1180
1181         mem->ranges[j].start = temp_range.start;
1182         mem->ranges[j].end = temp_range.end;
1183         mem->nr_ranges++;
1184         return 0;
1185 }
1186
1187 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1188                           void **addr, unsigned long *sz)
1189 {
1190         Elf64_Ehdr *ehdr;
1191         Elf64_Phdr *phdr;
1192         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1193         unsigned char *buf;
1194         unsigned int cpu, i;
1195         unsigned long long notes_addr;
1196         unsigned long mstart, mend;
1197
1198         /* extra phdr for vmcoreinfo elf note */
1199         nr_phdr = nr_cpus + 1;
1200         nr_phdr += mem->nr_ranges;
1201
1202         /*
1203          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1204          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1205          * I think this is required by tools like gdb. So same physical
1206          * memory will be mapped in two elf headers. One will contain kernel
1207          * text virtual addresses and other will have __va(physical) addresses.
1208          */
1209
1210         nr_phdr++;
1211         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1212         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1213
1214         buf = vzalloc(elf_sz);
1215         if (!buf)
1216                 return -ENOMEM;
1217
1218         ehdr = (Elf64_Ehdr *)buf;
1219         phdr = (Elf64_Phdr *)(ehdr + 1);
1220         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1221         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1222         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1223         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1224         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1225         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1226         ehdr->e_type = ET_CORE;
1227         ehdr->e_machine = ELF_ARCH;
1228         ehdr->e_version = EV_CURRENT;
1229         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1230         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1231         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1232
1233         /* Prepare one phdr of type PT_NOTE for each present cpu */
1234         for_each_present_cpu(cpu) {
1235                 phdr->p_type = PT_NOTE;
1236                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1237                 phdr->p_offset = phdr->p_paddr = notes_addr;
1238                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1239                 (ehdr->e_phnum)++;
1240                 phdr++;
1241         }
1242
1243         /* Prepare one PT_NOTE header for vmcoreinfo */
1244         phdr->p_type = PT_NOTE;
1245         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1246         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1247         (ehdr->e_phnum)++;
1248         phdr++;
1249
1250         /* Prepare PT_LOAD type program header for kernel text region */
1251         if (kernel_map) {
1252                 phdr->p_type = PT_LOAD;
1253                 phdr->p_flags = PF_R|PF_W|PF_X;
1254                 phdr->p_vaddr = (Elf64_Addr)_text;
1255                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1256                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1257                 ehdr->e_phnum++;
1258                 phdr++;
1259         }
1260
1261         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1262         for (i = 0; i < mem->nr_ranges; i++) {
1263                 mstart = mem->ranges[i].start;
1264                 mend = mem->ranges[i].end;
1265
1266                 phdr->p_type = PT_LOAD;
1267                 phdr->p_flags = PF_R|PF_W|PF_X;
1268                 phdr->p_offset  = mstart;
1269
1270                 phdr->p_paddr = mstart;
1271                 phdr->p_vaddr = (unsigned long long) __va(mstart);
1272                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1273                 phdr->p_align = 0;
1274                 ehdr->e_phnum++;
1275                 phdr++;
1276                 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",
1277                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1278                         ehdr->e_phnum, phdr->p_offset);
1279         }
1280
1281         *addr = buf;
1282         *sz = elf_sz;
1283         return 0;
1284 }