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