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_KEEP_MEMBLOCK
 504 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 505                                int (*func)(struct resource *, void *))
 506 {
 507         int ret = 0;
 508         u64 i;
 509         phys_addr_t mstart, mend;
 510         struct resource res = { };
 511
 512         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 513                 return func(&crashk_res, kbuf);
 514
 515         if (kbuf->top_down) {
 516                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 517                                                 &mstart, &mend, NULL) {
 518                         /*
 519                          * In memblock, end points to the first byte after the
 520                          * range while in kexec, end points to the last byte
 521                          * in the range.
 522                          */
 523                         res.start = mstart;
 524                         res.end = mend - 1;
 525                         ret = func(&res, kbuf);
 526                         if (ret)
 527                                 break;
 528                 }
 529         } else {
 530                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 531                                         &mstart, &mend, NULL) {
 532                         /*
 533                          * In memblock, end points to the first byte after the
 534                          * range while in kexec, end points to the last byte
 535                          * in the range.
 536                          */
 537                         res.start = mstart;
 538                         res.end = mend - 1;
 539                         ret = func(&res, kbuf);
 540                         if (ret)
 541                                 break;
 542                 }
 543         }
 544
 545         return ret;
 546 }
 547 #else
 548 static int kexec_walk_memblock(struct kexec_buf *kbuf,
 549                                int (*func)(struct resource *, void *))
 550 {
 551         return 0;
 552 }
 553 #endif
 554
 555 /**
 556  * kexec_walk_resources - call func(data) on free memory regions
 557  * @kbuf:       Context info for the search. Also passed to @func.
 558  * @func:       Function to call for each memory region.
 559  *
 560  * Return: The memory walk will stop when func returns a non-zero value
 561  * and that value will be returned. If all free regions are visited without
 562  * func returning non-zero, then zero will be returned.
 563  */
 564 static int kexec_walk_resources(struct kexec_buf *kbuf,
 565                                 int (*func)(struct resource *, void *))
 566 {
 567         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 568                 return walk_iomem_res_desc(crashk_res.desc,
 569                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 570                                            crashk_res.start, crashk_res.end,
 571                                            kbuf, func);
 572         else
 573                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 574 }
 575
 576 /**
 577  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 578  * @kbuf:       Parameters for the memory search.
 579  *
 580  * On success, kbuf->mem will have the start address of the memory region found.
 581  *
 582  * Return: 0 on success, negative errno on error.
 583  */
 584 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 585 {
 586         int ret;
 587
 588         /* Arch knows where to place */
 589         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
 590                 return 0;
 591
 592         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 593                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
 594         else
 595                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
 596
 597         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 598 }
 599
 600 /**
 601  * kexec_add_buffer - place a buffer in a kexec segment
 602  * @kbuf:       Buffer contents and memory parameters.
 603  *
 604  * This function assumes that kexec_mutex is held.
 605  * On successful return, @kbuf->mem will have the physical address of
 606  * the buffer in memory.
 607  *
 608  * Return: 0 on success, negative errno on error.
 609  */
 610 int kexec_add_buffer(struct kexec_buf *kbuf)
 611 {
 612
 613         struct kexec_segment *ksegment;
 614         int ret;
 615
 616         /* Currently adding segment this way is allowed only in file mode */
 617         if (!kbuf->image->file_mode)
 618                 return -EINVAL;
 619
 620         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 621                 return -EINVAL;
 622
 623         /*
 624          * Make sure we are not trying to add buffer after allocating
 625          * control pages. All segments need to be placed first before
 626          * any control pages are allocated. As control page allocation
 627          * logic goes through list of segments to make sure there are
 628          * no destination overlaps.
 629          */
 630         if (!list_empty(&kbuf->image->control_pages)) {
 631                 WARN_ON(1);
 632                 return -EINVAL;
 633         }
 634
 635         /* Ensure minimum alignment needed for segments. */
 636         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 637         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 638
 639         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 640         ret = kexec_locate_mem_hole(kbuf);
 641         if (ret)
 642                 return ret;
 643
 644         /* Found a suitable memory range */
 645         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 646         ksegment->kbuf = kbuf->buffer;
 647         ksegment->bufsz = kbuf->bufsz;
 648         ksegment->mem = kbuf->mem;
 649         ksegment->memsz = kbuf->memsz;
 650         kbuf->image->nr_segments++;
 651         return 0;
 652 }
 653
 654 /* Calculate and store the digest of segments */
 655 static int kexec_calculate_store_digests(struct kimage *image)
 656 {
 657         struct crypto_shash *tfm;
 658         struct shash_desc *desc;
 659         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 660         size_t desc_size, nullsz;
 661         char *digest;
 662         void *zero_buf;
 663         struct kexec_sha_region *sha_regions;
 664         struct purgatory_info *pi = &image->purgatory_info;
 665
 666         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
 667                 return 0;
 668
 669         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 670         zero_buf_sz = PAGE_SIZE;
 671
 672         tfm = crypto_alloc_shash("sha256", 0, 0);
 673         if (IS_ERR(tfm)) {
 674                 ret = PTR_ERR(tfm);
 675                 goto out;
 676         }
 677
 678         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 679         desc = kzalloc(desc_size, GFP_KERNEL);
 680         if (!desc) {
 681                 ret = -ENOMEM;
 682                 goto out_free_tfm;
 683         }
 684
 685         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 686         sha_regions = vzalloc(sha_region_sz);
 687         if (!sha_regions)
 688                 goto out_free_desc;
 689
 690         desc->tfm   = tfm;
 691
 692         ret = crypto_shash_init(desc);
 693         if (ret < 0)
 694                 goto out_free_sha_regions;
 695
 696         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 697         if (!digest) {
 698                 ret = -ENOMEM;
 699                 goto out_free_sha_regions;
 700         }
 701
 702         for (j = i = 0; i < image->nr_segments; i++) {
 703                 struct kexec_segment *ksegment;
 704
 705                 ksegment = &image->segment[i];
 706                 /*
 707                  * Skip purgatory as it will be modified once we put digest
 708                  * info in purgatory.
 709                  */
 710                 if (ksegment->kbuf == pi->purgatory_buf)
 711                         continue;
 712
 713                 ret = crypto_shash_update(desc, ksegment->kbuf,
 714                                           ksegment->bufsz);
 715                 if (ret)
 716                         break;
 717
 718                 /*
 719                  * Assume rest of the buffer is filled with zero and
 720                  * update digest accordingly.
 721                  */
 722                 nullsz = ksegment->memsz - ksegment->bufsz;
 723                 while (nullsz) {
 724                         unsigned long bytes = nullsz;
 725
 726                         if (bytes > zero_buf_sz)
 727                                 bytes = zero_buf_sz;
 728                         ret = crypto_shash_update(desc, zero_buf, bytes);
 729                         if (ret)
 730                                 break;
 731                         nullsz -= bytes;
 732                 }
 733
 734                 if (ret)
 735                         break;
 736
 737                 sha_regions[j].start = ksegment->mem;
 738                 sha_regions[j].len = ksegment->memsz;
 739                 j++;
 740         }
 741
 742         if (!ret) {
 743                 ret = crypto_shash_final(desc, digest);
 744                 if (ret)
 745                         goto out_free_digest;
 746                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 747                                                      sha_regions, sha_region_sz, 0);
 748                 if (ret)
 749                         goto out_free_digest;
 750
 751                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 752                                                      digest, SHA256_DIGEST_SIZE, 0);
 753                 if (ret)
 754                         goto out_free_digest;
 755         }
 756
 757 out_free_digest:
 758         kfree(digest);
 759 out_free_sha_regions:
 760         vfree(sha_regions);
 761 out_free_desc:
 762         kfree(desc);
 763 out_free_tfm:
 764         kfree(tfm);
 765 out:
 766         return ret;
 767 }
 768
 769 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
 770 /*
 771  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 772  * @pi:         Purgatory to be loaded.
 773  * @kbuf:       Buffer to setup.
 774  *
 775  * Allocates the memory needed for the buffer. Caller is responsible to free
 776  * the memory after use.
 777  *
 778  * Return: 0 on success, negative errno on error.
 779  */
 780 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 781                                       struct kexec_buf *kbuf)
 782 {
 783         const Elf_Shdr *sechdrs;
 784         unsigned long bss_align;
 785         unsigned long bss_sz;
 786         unsigned long align;
 787         int i, ret;
 788
 789         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 790         kbuf->buf_align = bss_align = 1;
 791         kbuf->bufsz = bss_sz = 0;
 792
 793         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 794                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 795                         continue;
 796
 797                 align = sechdrs[i].sh_addralign;
 798                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 799                         if (kbuf->buf_align < align)
 800                                 kbuf->buf_align = align;
 801                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 802                         kbuf->bufsz += sechdrs[i].sh_size;
 803                 } else {
 804                         if (bss_align < align)
 805                                 bss_align = align;
 806                         bss_sz = ALIGN(bss_sz, align);
 807                         bss_sz += sechdrs[i].sh_size;
 808                 }
 809         }
 810         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 811         kbuf->memsz = kbuf->bufsz + bss_sz;
 812         if (kbuf->buf_align < bss_align)
 813                 kbuf->buf_align = bss_align;
 814
 815         kbuf->buffer = vzalloc(kbuf->bufsz);
 816         if (!kbuf->buffer)
 817                 return -ENOMEM;
 818         pi->purgatory_buf = kbuf->buffer;
 819
 820         ret = kexec_add_buffer(kbuf);
 821         if (ret)
 822                 goto out;
 823
 824         return 0;
 825 out:
 826         vfree(pi->purgatory_buf);
 827         pi->purgatory_buf = NULL;
 828         return ret;
 829 }
 830
 831 /*
 832  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 833  * @pi:         Purgatory to be loaded.
 834  * @kbuf:       Buffer prepared to store purgatory.
 835  *
 836  * Allocates the memory needed for the buffer. Caller is responsible to free
 837  * the memory after use.
 838  *
 839  * Return: 0 on success, negative errno on error.
 840  */
 841 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 842                                          struct kexec_buf *kbuf)
 843 {
 844         unsigned long bss_addr;
 845         unsigned long offset;
 846         Elf_Shdr *sechdrs;
 847         int i;
 848
 849         /*
 850          * The section headers in kexec_purgatory are read-only. In order to
 851          * have them modifiable make a temporary copy.
 852          */
 853         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
 854         if (!sechdrs)
 855                 return -ENOMEM;
 856         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
 857                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 858         pi->sechdrs = sechdrs;
 859
 860         offset = 0;
 861         bss_addr = kbuf->mem + kbuf->bufsz;
 862         kbuf->image->start = pi->ehdr->e_entry;
 863
 864         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 865                 unsigned long align;
 866                 void *src, *dst;
 867
 868                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 869                         continue;
 870
 871                 align = sechdrs[i].sh_addralign;
 872                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 873                         bss_addr = ALIGN(bss_addr, align);
 874                         sechdrs[i].sh_addr = bss_addr;
 875                         bss_addr += sechdrs[i].sh_size;
 876                         continue;
 877                 }
 878
 879                 offset = ALIGN(offset, align);
 880                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 881                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 882                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 883                                          + sechdrs[i].sh_size)) {
 884                         kbuf->image->start -= sechdrs[i].sh_addr;
 885                         kbuf->image->start += kbuf->mem + offset;
 886                 }
 887
 888                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 889                 dst = pi->purgatory_buf + offset;
 890                 memcpy(dst, src, sechdrs[i].sh_size);
 891
 892                 sechdrs[i].sh_addr = kbuf->mem + offset;
 893                 sechdrs[i].sh_offset = offset;
 894                 offset += sechdrs[i].sh_size;
 895         }
 896
 897         return 0;
 898 }
 899
 900 static int kexec_apply_relocations(struct kimage *image)
 901 {
 902         int i, ret;
 903         struct purgatory_info *pi = &image->purgatory_info;
 904         const Elf_Shdr *sechdrs;
 905
 906         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 907
 908         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 909                 const Elf_Shdr *relsec;
 910                 const Elf_Shdr *symtab;
 911                 Elf_Shdr *section;
 912
 913                 relsec = sechdrs + i;
 914
 915                 if (relsec->sh_type != SHT_RELA &&
 916                     relsec->sh_type != SHT_REL)
 917                         continue;
 918
 919                 /*
 920                  * For section of type SHT_RELA/SHT_REL,
 921                  * ->sh_link contains section header index of associated
 922                  * symbol table. And ->sh_info contains section header
 923                  * index of section to which relocations apply.
 924                  */
 925                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 926                     relsec->sh_link >= pi->ehdr->e_shnum)
 927                         return -ENOEXEC;
 928
 929                 section = pi->sechdrs + relsec->sh_info;
 930                 symtab = sechdrs + relsec->sh_link;
 931
 932                 if (!(section->sh_flags & SHF_ALLOC))
 933                         continue;
 934
 935                 /*
 936                  * symtab->sh_link contain section header index of associated
 937                  * string table.
 938                  */
 939                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 940                         /* Invalid section number? */
 941                         continue;
 942
 943                 /*
 944                  * Respective architecture needs to provide support for applying
 945                  * relocations of type SHT_RELA/SHT_REL.
 946                  */
 947                 if (relsec->sh_type == SHT_RELA)
 948                         ret = arch_kexec_apply_relocations_add(pi, section,
 949                                                                relsec, symtab);
 950                 else if (relsec->sh_type == SHT_REL)
 951                         ret = arch_kexec_apply_relocations(pi, section,
 952                                                            relsec, symtab);
 953                 if (ret)
 954                         return ret;
 955         }
 956
 957         return 0;
 958 }
 959
 960 /*
 961  * kexec_load_purgatory - Load and relocate the purgatory object.
 962  * @image:      Image to add the purgatory to.
 963  * @kbuf:       Memory parameters to use.
 964  *
 965  * Allocates the memory needed for image->purgatory_info.sechdrs and
 966  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
 967  * to free the memory after use.
 968  *
 969  * Return: 0 on success, negative errno on error.
 970  */
 971 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
 972 {
 973         struct purgatory_info *pi = &image->purgatory_info;
 974         int ret;
 975
 976         if (kexec_purgatory_size <= 0)
 977                 return -EINVAL;
 978
 979         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
 980
 981         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
 982         if (ret)
 983                 return ret;
 984
 985         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
 986         if (ret)
 987                 goto out_free_kbuf;
 988
 989         ret = kexec_apply_relocations(image);
 990         if (ret)
 991                 goto out;
 992
 993         return 0;
 994 out:
 995         vfree(pi->sechdrs);
 996         pi->sechdrs = NULL;
 997 out_free_kbuf:
 998         vfree(pi->purgatory_buf);
 999         pi->purgatory_buf = NULL;
1000         return ret;
1001 }
1002
1003 /*
1004  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1005  * @pi:         Purgatory to search in.
1006  * @name:       Name of the symbol.
1007  *
1008  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1009  */
1010 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1011                                                   const char *name)
1012 {
1013         const Elf_Shdr *sechdrs;
1014         const Elf_Ehdr *ehdr;
1015         const Elf_Sym *syms;
1016         const char *strtab;
1017         int i, k;
1018
1019         if (!pi->ehdr)
1020                 return NULL;
1021
1022         ehdr = pi->ehdr;
1023         sechdrs = (void *)ehdr + ehdr->e_shoff;
1024
1025         for (i = 0; i < ehdr->e_shnum; i++) {
1026                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1027                         continue;
1028
1029                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1030                         /* Invalid strtab section number */
1031                         continue;
1032                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1033                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1034
1035                 /* Go through symbols for a match */
1036                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1037                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1038                                 continue;
1039
1040                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1041                                 continue;
1042
1043                         if (syms[k].st_shndx == SHN_UNDEF ||
1044                             syms[k].st_shndx >= ehdr->e_shnum) {
1045                                 pr_debug("Symbol: %s has bad section index %d.\n",
1046                                                 name, syms[k].st_shndx);
1047                                 return NULL;
1048                         }
1049
1050                         /* Found the symbol we are looking for */
1051                         return &syms[k];
1052                 }
1053         }
1054
1055         return NULL;
1056 }
1057
1058 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1059 {
1060         struct purgatory_info *pi = &image->purgatory_info;
1061         const Elf_Sym *sym;
1062         Elf_Shdr *sechdr;
1063
1064         sym = kexec_purgatory_find_symbol(pi, name);
1065         if (!sym)
1066                 return ERR_PTR(-EINVAL);
1067
1068         sechdr = &pi->sechdrs[sym->st_shndx];
1069
1070         /*
1071          * Returns the address where symbol will finally be loaded after
1072          * kexec_load_segment()
1073          */
1074         return (void *)(sechdr->sh_addr + sym->st_value);
1075 }
1076
1077 /*
1078  * Get or set value of a symbol. If "get_value" is true, symbol value is
1079  * returned in buf otherwise symbol value is set based on value in buf.
1080  */
1081 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1082                                    void *buf, unsigned int size, bool get_value)
1083 {
1084         struct purgatory_info *pi = &image->purgatory_info;
1085         const Elf_Sym *sym;
1086         Elf_Shdr *sec;
1087         char *sym_buf;
1088
1089         sym = kexec_purgatory_find_symbol(pi, name);
1090         if (!sym)
1091                 return -EINVAL;
1092
1093         if (sym->st_size != size) {
1094                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1095                        name, (unsigned long)sym->st_size, size);
1096                 return -EINVAL;
1097         }
1098
1099         sec = pi->sechdrs + sym->st_shndx;
1100
1101         if (sec->sh_type == SHT_NOBITS) {
1102                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1103                        get_value ? "get" : "set");
1104                 return -EINVAL;
1105         }
1106
1107         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1108
1109         if (get_value)
1110                 memcpy((void *)buf, sym_buf, size);
1111         else
1112                 memcpy((void *)sym_buf, buf, size);
1113
1114         return 0;
1115 }
1116 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1117
1118 int crash_exclude_mem_range(struct crash_mem *mem,
1119                             unsigned long long mstart, unsigned long long mend)
1120 {
1121         int i, j;
1122         unsigned long long start, end;
1123         struct crash_mem_range temp_range = {0, 0};
1124
1125         for (i = 0; i < mem->nr_ranges; i++) {
1126                 start = mem->ranges[i].start;
1127                 end = mem->ranges[i].end;
1128
1129                 if (mstart > end || mend < start)
1130                         continue;
1131
1132                 /* Truncate any area outside of range */
1133                 if (mstart < start)
1134                         mstart = start;
1135                 if (mend > end)
1136                         mend = end;
1137
1138                 /* Found completely overlapping range */
1139                 if (mstart == start && mend == end) {
1140                         mem->ranges[i].start = 0;
1141                         mem->ranges[i].end = 0;
1142                         if (i < mem->nr_ranges - 1) {
1143                                 /* Shift rest of the ranges to left */
1144                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1145                                         mem->ranges[j].start =
1146                                                 mem->ranges[j+1].start;
1147                                         mem->ranges[j].end =
1148                                                         mem->ranges[j+1].end;
1149                                 }
1150                         }
1151                         mem->nr_ranges--;
1152                         return 0;
1153                 }
1154
1155                 if (mstart > start && mend < end) {
1156                         /* Split original range */
1157                         mem->ranges[i].end = mstart - 1;
1158                         temp_range.start = mend + 1;
1159                         temp_range.end = end;
1160                 } else if (mstart != start)
1161                         mem->ranges[i].end = mstart - 1;
1162                 else
1163                         mem->ranges[i].start = mend + 1;
1164                 break;
1165         }
1166
1167         /* If a split happened, add the split to array */
1168         if (!temp_range.end)
1169                 return 0;
1170
1171         /* Split happened */
1172         if (i == mem->max_nr_ranges - 1)
1173                 return -ENOMEM;
1174
1175         /* Location where new range should go */
1176         j = i + 1;
1177         if (j < mem->nr_ranges) {
1178                 /* Move over all ranges one slot towards the end */
1179                 for (i = mem->nr_ranges - 1; i >= j; i--)
1180                         mem->ranges[i + 1] = mem->ranges[i];
1181         }
1182
1183         mem->ranges[j].start = temp_range.start;
1184         mem->ranges[j].end = temp_range.end;
1185         mem->nr_ranges++;
1186         return 0;
1187 }
1188
1189 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1190                           void **addr, unsigned long *sz)
1191 {
1192         Elf64_Ehdr *ehdr;
1193         Elf64_Phdr *phdr;
1194         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1195         unsigned char *buf;
1196         unsigned int cpu, i;
1197         unsigned long long notes_addr;
1198         unsigned long mstart, mend;
1199
1200         /* extra phdr for vmcoreinfo elf note */
1201         nr_phdr = nr_cpus + 1;
1202         nr_phdr += mem->nr_ranges;
1203
1204         /*
1205          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1206          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1207          * I think this is required by tools like gdb. So same physical
1208          * memory will be mapped in two elf headers. One will contain kernel
1209          * text virtual addresses and other will have __va(physical) addresses.
1210          */
1211
1212         nr_phdr++;
1213         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1214         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1215
1216         buf = vzalloc(elf_sz);
1217         if (!buf)
1218                 return -ENOMEM;
1219
1220         ehdr = (Elf64_Ehdr *)buf;
1221         phdr = (Elf64_Phdr *)(ehdr + 1);
1222         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1223         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1224         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1225         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1226         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1227         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1228         ehdr->e_type = ET_CORE;
1229         ehdr->e_machine = ELF_ARCH;
1230         ehdr->e_version = EV_CURRENT;
1231         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1232         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1233         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1234
1235         /* Prepare one phdr of type PT_NOTE for each present cpu */
1236         for_each_present_cpu(cpu) {
1237                 phdr->p_type = PT_NOTE;
1238                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1239                 phdr->p_offset = phdr->p_paddr = notes_addr;
1240                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1241                 (ehdr->e_phnum)++;
1242                 phdr++;
1243         }
1244
1245         /* Prepare one PT_NOTE header for vmcoreinfo */
1246         phdr->p_type = PT_NOTE;
1247         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1248         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1249         (ehdr->e_phnum)++;
1250         phdr++;
1251
1252         /* Prepare PT_LOAD type program header for kernel text region */
1253         if (kernel_map) {
1254                 phdr->p_type = PT_LOAD;
1255                 phdr->p_flags = PF_R|PF_W|PF_X;
1256                 phdr->p_vaddr = (Elf64_Addr)_text;
1257                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1258                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1259                 ehdr->e_phnum++;
1260                 phdr++;
1261         }
1262
1263         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1264         for (i = 0; i < mem->nr_ranges; i++) {
1265                 mstart = mem->ranges[i].start;
1266                 mend = mem->ranges[i].end;
1267
1268                 phdr->p_type = PT_LOAD;
1269                 phdr->p_flags = PF_R|PF_W|PF_X;
1270                 phdr->p_offset  = mstart;
1271
1272                 phdr->p_paddr = mstart;
1273                 phdr->p_vaddr = (unsigned long long) __va(mstart);
1274                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1275                 phdr->p_align = 0;
1276                 ehdr->e_phnum++;
1277                 phdr++;
1278                 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",
1279                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1280                         ehdr->e_phnum, phdr->p_offset);
1281         }
1282
1283         *addr = buf;
1284         *sz = elf_sz;
1285         return 0;
1286 }