kernel/kexec_file.c

   1 /*
   2  * kexec: kexec_file_load system call
   3  *
   4  * Copyright (C) 2014 Red Hat Inc.
   5  * Authors:
   6  *      Vivek Goyal <vgoyal@redhat.com>
   7  *
   8  * This source code is licensed under the GNU General Public License,
   9  * Version 2.  See the file COPYING for more details.
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/capability.h>
  15 #include <linux/mm.h>
  16 #include <linux/file.h>
  17 #include <linux/slab.h>
  18 #include <linux/kexec.h>
  19 #include <linux/mutex.h>
  20 #include <linux/list.h>
  21 #include <linux/fs.h>
  22 #include <linux/ima.h>
  23 #include <crypto/hash.h>
  24 #include <crypto/sha.h>
  25 #include <linux/elf.h>
  26 #include <linux/elfcore.h>
  27 #include <linux/kernel.h>
  28 #include <linux/kexec.h>
  29 #include <linux/slab.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/vmalloc.h>
  32 #include "kexec_internal.h"
  33
  34 static int kexec_calculate_store_digests(struct kimage *image);
  35
  36 /*
  37  * Currently this is the only default function that is exported as some
  38  * architectures need it to do additional handlings.
  39  * In the future, other default functions may be exported too if required.
  40  */
  41 int kexec_image_probe_default(struct kimage *image, void *buf,
  42                               unsigned long buf_len)
  43 {
  44         const struct kexec_file_ops * const *fops;
  45         int ret = -ENOEXEC;
  46
  47         for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
  48                 ret = (*fops)->probe(buf, buf_len);
  49                 if (!ret) {
  50                         image->fops = *fops;
  51                         return ret;
  52                 }
  53         }
  54
  55         return ret;
  56 }
  57
  58 /* Architectures can provide this probe function */
  59 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  60                                          unsigned long buf_len)
  61 {
  62         return kexec_image_probe_default(image, buf, buf_len);
  63 }
  64
  65 static void *kexec_image_load_default(struct kimage *image)
  66 {
  67         if (!image->fops || !image->fops->load)
  68                 return ERR_PTR(-ENOEXEC);
  69
  70         return image->fops->load(image, image->kernel_buf,
  71                                  image->kernel_buf_len, image->initrd_buf,
  72                                  image->initrd_buf_len, image->cmdline_buf,
  73                                  image->cmdline_buf_len);
  74 }
  75
  76 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
  77 {
  78         return kexec_image_load_default(image);
  79 }
  80
  81 static int kexec_image_post_load_cleanup_default(struct kimage *image)
  82 {
  83         if (!image->fops || !image->fops->cleanup)
  84                 return 0;
  85
  86         return image->fops->cleanup(image->image_loader_data);
  87 }
  88
  89 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
  90 {
  91         return kexec_image_post_load_cleanup_default(image);
  92 }
  93
  94 #ifdef CONFIG_KEXEC_VERIFY_SIG
  95 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
  96                                           unsigned long buf_len)
  97 {
  98         if (!image->fops || !image->fops->verify_sig) {
  99                 pr_debug("kernel loader does not support signature verification.\n");
 100                 return -EKEYREJECTED;
 101         }
 102
 103         return image->fops->verify_sig(buf, buf_len);
 104 }
 105
 106 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
 107                                         unsigned long buf_len)
 108 {
 109         return kexec_image_verify_sig_default(image, buf, buf_len);
 110 }
 111 #endif
 112
 113 /*
 114  * Free up memory used by kernel, initrd, and command line. This is temporary
 115  * memory allocation which is not needed any more after these buffers have
 116  * been loaded into separate segments and have been copied elsewhere.
 117  */
 118 void kimage_file_post_load_cleanup(struct kimage *image)
 119 {
 120         struct purgatory_info *pi = &image->purgatory_info;
 121
 122         vfree(image->kernel_buf);
 123         image->kernel_buf = NULL;
 124
 125         vfree(image->initrd_buf);
 126         image->initrd_buf = NULL;
 127
 128         kfree(image->cmdline_buf);
 129         image->cmdline_buf = NULL;
 130
 131         vfree(pi->purgatory_buf);
 132         pi->purgatory_buf = NULL;
 133
 134         vfree(pi->sechdrs);
 135         pi->sechdrs = NULL;
 136
 137 #ifdef CONFIG_IMA_KEXEC
 138         vfree(image->ima_buffer);
 139         image->ima_buffer = NULL;
 140 #endif /* CONFIG_IMA_KEXEC */
 141
 142         /* See if architecture has anything to cleanup post load */
 143         arch_kimage_file_post_load_cleanup(image);
 144
 145         /*
 146          * Above call should have called into bootloader to free up
 147          * any data stored in kimage->image_loader_data. It should
 148          * be ok now to free it up.
 149          */
 150         kfree(image->image_loader_data);
 151         image->image_loader_data = NULL;
 152 }
 153
 154 /*
 155  * In file mode list of segments is prepared by kernel. Copy relevant
 156  * data from user space, do error checking, prepare segment list
 157  */
 158 static int
 159 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 160                              const char __user *cmdline_ptr,
 161                              unsigned long cmdline_len, unsigned flags)
 162 {
 163         int ret = 0;
 164         void *ldata;
 165         loff_t size;
 166
 167         ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
 168                                        &size, INT_MAX, READING_KEXEC_IMAGE);
 169         if (ret)
 170                 return ret;
 171         image->kernel_buf_len = size;
 172
 173         /* IMA needs to pass the measurement list to the next kernel. */
 174         ima_add_kexec_buffer(image);
 175
 176         /* Call arch image probe handlers */
 177         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
 178                                             image->kernel_buf_len);
 179         if (ret)
 180                 goto out;
 181
 182 #ifdef CONFIG_KEXEC_VERIFY_SIG
 183         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
 184                                            image->kernel_buf_len);
 185         if (ret) {
 186                 pr_debug("kernel signature verification failed.\n");
 187                 goto out;
 188         }
 189         pr_debug("kernel signature verification successful.\n");
 190 #endif
 191         /* It is possible that there no initramfs is being loaded */
 192         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
 193                 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
 194                                                &size, INT_MAX,
 195                                                READING_KEXEC_INITRAMFS);
 196                 if (ret)
 197                         goto out;
 198                 image->initrd_buf_len = size;
 199         }
 200
 201         if (cmdline_len) {
 202                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
 203                 if (IS_ERR(image->cmdline_buf)) {
 204                         ret = PTR_ERR(image->cmdline_buf);
 205                         image->cmdline_buf = NULL;
 206                         goto out;
 207                 }
 208
 209                 image->cmdline_buf_len = cmdline_len;
 210
 211                 /* command line should be a string with last byte null */
 212                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
 213                         ret = -EINVAL;
 214                         goto out;
 215                 }
 216         }
 217
 218         /* Call arch image load handlers */
 219         ldata = arch_kexec_kernel_image_load(image);
 220
 221         if (IS_ERR(ldata)) {
 222                 ret = PTR_ERR(ldata);
 223                 goto out;
 224         }
 225
 226         image->image_loader_data = ldata;
 227 out:
 228         /* In case of error, free up all allocated memory in this function */
 229         if (ret)
 230                 kimage_file_post_load_cleanup(image);
 231         return ret;
 232 }
 233
 234 static int
 235 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
 236                        int initrd_fd, const char __user *cmdline_ptr,
 237                        unsigned long cmdline_len, unsigned long flags)
 238 {
 239         int ret;
 240         struct kimage *image;
 241         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
 242
 243         image = do_kimage_alloc_init();
 244         if (!image)
 245                 return -ENOMEM;
 246
 247         image->file_mode = 1;
 248
 249         if (kexec_on_panic) {
 250                 /* Enable special crash kernel control page alloc policy. */
 251                 image->control_page = crashk_res.start;
 252                 image->type = KEXEC_TYPE_CRASH;
 253         }
 254
 255         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
 256                                            cmdline_ptr, cmdline_len, flags);
 257         if (ret)
 258                 goto out_free_image;
 259
 260         ret = sanity_check_segment_list(image);
 261         if (ret)
 262                 goto out_free_post_load_bufs;
 263
 264         ret = -ENOMEM;
 265         image->control_code_page = kimage_alloc_control_pages(image,
 266                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
 267         if (!image->control_code_page) {
 268                 pr_err("Could not allocate control_code_buffer\n");
 269                 goto out_free_post_load_bufs;
 270         }
 271
 272         if (!kexec_on_panic) {
 273                 image->swap_page = kimage_alloc_control_pages(image, 0);
 274                 if (!image->swap_page) {
 275                         pr_err("Could not allocate swap buffer\n");
 276                         goto out_free_control_pages;
 277                 }
 278         }
 279
 280         *rimage = image;
 281         return 0;
 282 out_free_control_pages:
 283         kimage_free_page_list(&image->control_pages);
 284 out_free_post_load_bufs:
 285         kimage_file_post_load_cleanup(image);
 286 out_free_image:
 287         kfree(image);
 288         return ret;
 289 }
 290
 291 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
 292                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
 293                 unsigned long, flags)
 294 {
 295         int ret = 0, i;
 296         struct kimage **dest_image, *image;
 297
 298         /* We only trust the superuser with rebooting the system. */
 299         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 300                 return -EPERM;
 301
 302         /* Make sure we have a legal set of flags */
 303         if (flags != (flags & KEXEC_FILE_FLAGS))
 304                 return -EINVAL;
 305
 306         image = NULL;
 307
 308         if (!mutex_trylock(&kexec_mutex))
 309                 return -EBUSY;
 310
 311         dest_image = &kexec_image;
 312         if (flags & KEXEC_FILE_ON_CRASH) {
 313                 dest_image = &kexec_crash_image;
 314                 if (kexec_crash_image)
 315                         arch_kexec_unprotect_crashkres();
 316         }
 317
 318         if (flags & KEXEC_FILE_UNLOAD)
 319                 goto exchange;
 320
 321         /*
 322          * In case of crash, new kernel gets loaded in reserved region. It is
 323          * same memory where old crash kernel might be loaded. Free any
 324          * current crash dump kernel before we corrupt it.
 325          */
 326         if (flags & KEXEC_FILE_ON_CRASH)
 327                 kimage_free(xchg(&kexec_crash_image, NULL));
 328
 329         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
 330                                      cmdline_len, flags);
 331         if (ret)
 332                 goto out;
 333
 334         ret = machine_kexec_prepare(image);
 335         if (ret)
 336                 goto out;
 337
 338         /*
 339          * Some architecture(like S390) may touch the crash memory before
 340          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 341          */
 342         ret = kimage_crash_copy_vmcoreinfo(image);
 343         if (ret)
 344                 goto out;
 345
 346         ret = kexec_calculate_store_digests(image);
 347         if (ret)
 348                 goto out;
 349
 350         for (i = 0; i < image->nr_segments; i++) {
 351                 struct kexec_segment *ksegment;
 352
 353                 ksegment = &image->segment[i];
 354                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
 355                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
 356                          ksegment->memsz);
 357
 358                 ret = kimage_load_segment(image, &image->segment[i]);
 359                 if (ret)
 360                         goto out;
 361         }
 362
 363         kimage_terminate(image);
 364
 365         /*
 366          * Free up any temporary buffers allocated which are not needed
 367          * after image has been loaded
 368          */
 369         kimage_file_post_load_cleanup(image);
 370 exchange:
 371         image = xchg(dest_image, image);
 372 out:
 373         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
 374                 arch_kexec_protect_crashkres();
 375
 376         mutex_unlock(&kexec_mutex);
 377         kimage_free(image);
 378         return ret;
 379 }
 380
 381 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
 382                                     struct kexec_buf *kbuf)
 383 {
 384         struct kimage *image = kbuf->image;
 385         unsigned long temp_start, temp_end;
 386
 387         temp_end = min(end, kbuf->buf_max);
 388         temp_start = temp_end - kbuf->memsz;
 389
 390         do {
 391                 /* align down start */
 392                 temp_start = temp_start & (~(kbuf->buf_align - 1));
 393
 394                 if (temp_start < start || temp_start < kbuf->buf_min)
 395                         return 0;
 396
 397                 temp_end = temp_start + kbuf->memsz - 1;
 398
 399                 /*
 400                  * Make sure this does not conflict with any of existing
 401                  * segments
 402                  */
 403                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 404                         temp_start = temp_start - PAGE_SIZE;
 405                         continue;
 406                 }
 407
 408                 /* We found a suitable memory range */
 409                 break;
 410         } while (1);
 411
 412         /* If we are here, we found a suitable memory range */
 413         kbuf->mem = temp_start;
 414
 415         /* Success, stop navigating through remaining System RAM ranges */
 416         return 1;
 417 }
 418
 419 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
 420                                      struct kexec_buf *kbuf)
 421 {
 422         struct kimage *image = kbuf->image;
 423         unsigned long temp_start, temp_end;
 424
 425         temp_start = max(start, kbuf->buf_min);
 426
 427         do {
 428                 temp_start = ALIGN(temp_start, kbuf->buf_align);
 429                 temp_end = temp_start + kbuf->memsz - 1;
 430
 431                 if (temp_end > end || temp_end > kbuf->buf_max)
 432                         return 0;
 433                 /*
 434                  * Make sure this does not conflict with any of existing
 435                  * segments
 436                  */
 437                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
 438                         temp_start = temp_start + PAGE_SIZE;
 439                         continue;
 440                 }
 441
 442                 /* We found a suitable memory range */
 443                 break;
 444         } while (1);
 445
 446         /* If we are here, we found a suitable memory range */
 447         kbuf->mem = temp_start;
 448
 449         /* Success, stop navigating through remaining System RAM ranges */
 450         return 1;
 451 }
 452
 453 static int locate_mem_hole_callback(struct resource *res, void *arg)
 454 {
 455         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
 456         u64 start = res->start, end = res->end;
 457         unsigned long sz = end - start + 1;
 458
 459         /* Returning 0 will take to next memory range */
 460         if (sz < kbuf->memsz)
 461                 return 0;
 462
 463         if (end < kbuf->buf_min || start > kbuf->buf_max)
 464                 return 0;
 465
 466         /*
 467          * Allocate memory top down with-in ram range. Otherwise bottom up
 468          * allocation.
 469          */
 470         if (kbuf->top_down)
 471                 return locate_mem_hole_top_down(start, end, kbuf);
 472         return locate_mem_hole_bottom_up(start, end, kbuf);
 473 }
 474
 475 /**
 476  * arch_kexec_walk_mem - call func(data) on free memory regions
 477  * @kbuf:       Context info for the search. Also passed to @func.
 478  * @func:       Function to call for each memory region.
 479  *
 480  * Return: The memory walk will stop when func returns a non-zero value
 481  * and that value will be returned. If all free regions are visited without
 482  * func returning non-zero, then zero will be returned.
 483  */
 484 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
 485                                int (*func)(struct resource *, void *))
 486 {
 487         if (kbuf->image->type == KEXEC_TYPE_CRASH)
 488                 return walk_iomem_res_desc(crashk_res.desc,
 489                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
 490                                            crashk_res.start, crashk_res.end,
 491                                            kbuf, func);
 492         else
 493                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
 494 }
 495
 496 /**
 497  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
 498  * @kbuf:       Parameters for the memory search.
 499  *
 500  * On success, kbuf->mem will have the start address of the memory region found.
 501  *
 502  * Return: 0 on success, negative errno on error.
 503  */
 504 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
 505 {
 506         int ret;
 507
 508         ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
 509
 510         return ret == 1 ? 0 : -EADDRNOTAVAIL;
 511 }
 512
 513 /**
 514  * kexec_add_buffer - place a buffer in a kexec segment
 515  * @kbuf:       Buffer contents and memory parameters.
 516  *
 517  * This function assumes that kexec_mutex is held.
 518  * On successful return, @kbuf->mem will have the physical address of
 519  * the buffer in memory.
 520  *
 521  * Return: 0 on success, negative errno on error.
 522  */
 523 int kexec_add_buffer(struct kexec_buf *kbuf)
 524 {
 525
 526         struct kexec_segment *ksegment;
 527         int ret;
 528
 529         /* Currently adding segment this way is allowed only in file mode */
 530         if (!kbuf->image->file_mode)
 531                 return -EINVAL;
 532
 533         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
 534                 return -EINVAL;
 535
 536         /*
 537          * Make sure we are not trying to add buffer after allocating
 538          * control pages. All segments need to be placed first before
 539          * any control pages are allocated. As control page allocation
 540          * logic goes through list of segments to make sure there are
 541          * no destination overlaps.
 542          */
 543         if (!list_empty(&kbuf->image->control_pages)) {
 544                 WARN_ON(1);
 545                 return -EINVAL;
 546         }
 547
 548         /* Ensure minimum alignment needed for segments. */
 549         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
 550         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
 551
 552         /* Walk the RAM ranges and allocate a suitable range for the buffer */
 553         ret = kexec_locate_mem_hole(kbuf);
 554         if (ret)
 555                 return ret;
 556
 557         /* Found a suitable memory range */
 558         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
 559         ksegment->kbuf = kbuf->buffer;
 560         ksegment->bufsz = kbuf->bufsz;
 561         ksegment->mem = kbuf->mem;
 562         ksegment->memsz = kbuf->memsz;
 563         kbuf->image->nr_segments++;
 564         return 0;
 565 }
 566
 567 /* Calculate and store the digest of segments */
 568 static int kexec_calculate_store_digests(struct kimage *image)
 569 {
 570         struct crypto_shash *tfm;
 571         struct shash_desc *desc;
 572         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
 573         size_t desc_size, nullsz;
 574         char *digest;
 575         void *zero_buf;
 576         struct kexec_sha_region *sha_regions;
 577         struct purgatory_info *pi = &image->purgatory_info;
 578
 579         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
 580                 return 0;
 581
 582         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
 583         zero_buf_sz = PAGE_SIZE;
 584
 585         tfm = crypto_alloc_shash("sha256", 0, 0);
 586         if (IS_ERR(tfm)) {
 587                 ret = PTR_ERR(tfm);
 588                 goto out;
 589         }
 590
 591         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
 592         desc = kzalloc(desc_size, GFP_KERNEL);
 593         if (!desc) {
 594                 ret = -ENOMEM;
 595                 goto out_free_tfm;
 596         }
 597
 598         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
 599         sha_regions = vzalloc(sha_region_sz);
 600         if (!sha_regions) {
 601                 ret = -ENOMEM;
 602                 goto out_free_desc;
 603         }
 604
 605         desc->tfm   = tfm;
 606         desc->flags = 0;
 607
 608         ret = crypto_shash_init(desc);
 609         if (ret < 0)
 610                 goto out_free_sha_regions;
 611
 612         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
 613         if (!digest) {
 614                 ret = -ENOMEM;
 615                 goto out_free_sha_regions;
 616         }
 617
 618         for (j = i = 0; i < image->nr_segments; i++) {
 619                 struct kexec_segment *ksegment;
 620
 621                 ksegment = &image->segment[i];
 622                 /*
 623                  * Skip purgatory as it will be modified once we put digest
 624                  * info in purgatory.
 625                  */
 626                 if (ksegment->kbuf == pi->purgatory_buf)
 627                         continue;
 628
 629                 ret = crypto_shash_update(desc, ksegment->kbuf,
 630                                           ksegment->bufsz);
 631                 if (ret)
 632                         break;
 633
 634                 /*
 635                  * Assume rest of the buffer is filled with zero and
 636                  * update digest accordingly.
 637                  */
 638                 nullsz = ksegment->memsz - ksegment->bufsz;
 639                 while (nullsz) {
 640                         unsigned long bytes = nullsz;
 641
 642                         if (bytes > zero_buf_sz)
 643                                 bytes = zero_buf_sz;
 644                         ret = crypto_shash_update(desc, zero_buf, bytes);
 645                         if (ret)
 646                                 break;
 647                         nullsz -= bytes;
 648                 }
 649
 650                 if (ret)
 651                         break;
 652
 653                 sha_regions[j].start = ksegment->mem;
 654                 sha_regions[j].len = ksegment->memsz;
 655                 j++;
 656         }
 657
 658         if (!ret) {
 659                 ret = crypto_shash_final(desc, digest);
 660                 if (ret)
 661                         goto out_free_digest;
 662                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
 663                                                      sha_regions, sha_region_sz, 0);
 664                 if (ret)
 665                         goto out_free_digest;
 666
 667                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
 668                                                      digest, SHA256_DIGEST_SIZE, 0);
 669                 if (ret)
 670                         goto out_free_digest;
 671         }
 672
 673 out_free_digest:
 674         kfree(digest);
 675 out_free_sha_regions:
 676         vfree(sha_regions);
 677 out_free_desc:
 678         kfree(desc);
 679 out_free_tfm:
 680         kfree(tfm);
 681 out:
 682         return ret;
 683 }
 684
 685 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
 686 /*
 687  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
 688  * @pi:         Purgatory to be loaded.
 689  * @kbuf:       Buffer to setup.
 690  *
 691  * Allocates the memory needed for the buffer. Caller is responsible to free
 692  * the memory after use.
 693  *
 694  * Return: 0 on success, negative errno on error.
 695  */
 696 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
 697                                       struct kexec_buf *kbuf)
 698 {
 699         const Elf_Shdr *sechdrs;
 700         unsigned long bss_align;
 701         unsigned long bss_sz;
 702         unsigned long align;
 703         int i, ret;
 704
 705         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 706         kbuf->buf_align = bss_align = 1;
 707         kbuf->bufsz = bss_sz = 0;
 708
 709         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 710                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 711                         continue;
 712
 713                 align = sechdrs[i].sh_addralign;
 714                 if (sechdrs[i].sh_type != SHT_NOBITS) {
 715                         if (kbuf->buf_align < align)
 716                                 kbuf->buf_align = align;
 717                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
 718                         kbuf->bufsz += sechdrs[i].sh_size;
 719                 } else {
 720                         if (bss_align < align)
 721                                 bss_align = align;
 722                         bss_sz = ALIGN(bss_sz, align);
 723                         bss_sz += sechdrs[i].sh_size;
 724                 }
 725         }
 726         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
 727         kbuf->memsz = kbuf->bufsz + bss_sz;
 728         if (kbuf->buf_align < bss_align)
 729                 kbuf->buf_align = bss_align;
 730
 731         kbuf->buffer = vzalloc(kbuf->bufsz);
 732         if (!kbuf->buffer)
 733                 return -ENOMEM;
 734         pi->purgatory_buf = kbuf->buffer;
 735
 736         ret = kexec_add_buffer(kbuf);
 737         if (ret)
 738                 goto out;
 739
 740         return 0;
 741 out:
 742         vfree(pi->purgatory_buf);
 743         pi->purgatory_buf = NULL;
 744         return ret;
 745 }
 746
 747 /*
 748  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
 749  * @pi:         Purgatory to be loaded.
 750  * @kbuf:       Buffer prepared to store purgatory.
 751  *
 752  * Allocates the memory needed for the buffer. Caller is responsible to free
 753  * the memory after use.
 754  *
 755  * Return: 0 on success, negative errno on error.
 756  */
 757 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
 758                                          struct kexec_buf *kbuf)
 759 {
 760         unsigned long bss_addr;
 761         unsigned long offset;
 762         Elf_Shdr *sechdrs;
 763         int i;
 764
 765         /*
 766          * The section headers in kexec_purgatory are read-only. In order to
 767          * have them modifiable make a temporary copy.
 768          */
 769         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
 770         if (!sechdrs)
 771                 return -ENOMEM;
 772         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
 773                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
 774         pi->sechdrs = sechdrs;
 775
 776         offset = 0;
 777         bss_addr = kbuf->mem + kbuf->bufsz;
 778         kbuf->image->start = pi->ehdr->e_entry;
 779
 780         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 781                 unsigned long align;
 782                 void *src, *dst;
 783
 784                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
 785                         continue;
 786
 787                 align = sechdrs[i].sh_addralign;
 788                 if (sechdrs[i].sh_type == SHT_NOBITS) {
 789                         bss_addr = ALIGN(bss_addr, align);
 790                         sechdrs[i].sh_addr = bss_addr;
 791                         bss_addr += sechdrs[i].sh_size;
 792                         continue;
 793                 }
 794
 795                 offset = ALIGN(offset, align);
 796
 797                 /*
 798                  * Check if the segment contains the entry point, if so,
 799                  * calculate the value of image->start based on it.
 800                  * If the compiler has produced more than one .text section
 801                  * (Eg: .text.hot), they are generally after the main .text
 802                  * section, and they shall not be used to calculate
 803                  * image->start. So do not re-calculate image->start if it
 804                  * is not set to the initial value, and warn the user so they
 805                  * have a chance to fix their purgatory's linker script.
 806                  */
 807                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
 808                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
 809                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
 810                                          + sechdrs[i].sh_size) &&
 811                     !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
 812                         kbuf->image->start -= sechdrs[i].sh_addr;
 813                         kbuf->image->start += kbuf->mem + offset;
 814                 }
 815
 816                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
 817                 dst = pi->purgatory_buf + offset;
 818                 memcpy(dst, src, sechdrs[i].sh_size);
 819
 820                 sechdrs[i].sh_addr = kbuf->mem + offset;
 821                 sechdrs[i].sh_offset = offset;
 822                 offset += sechdrs[i].sh_size;
 823         }
 824
 825         return 0;
 826 }
 827
 828 static int kexec_apply_relocations(struct kimage *image)
 829 {
 830         int i, ret;
 831         struct purgatory_info *pi = &image->purgatory_info;
 832         const Elf_Shdr *sechdrs;
 833
 834         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
 835
 836         for (i = 0; i < pi->ehdr->e_shnum; i++) {
 837                 const Elf_Shdr *relsec;
 838                 const Elf_Shdr *symtab;
 839                 Elf_Shdr *section;
 840
 841                 relsec = sechdrs + i;
 842
 843                 if (relsec->sh_type != SHT_RELA &&
 844                     relsec->sh_type != SHT_REL)
 845                         continue;
 846
 847                 /*
 848                  * For section of type SHT_RELA/SHT_REL,
 849                  * ->sh_link contains section header index of associated
 850                  * symbol table. And ->sh_info contains section header
 851                  * index of section to which relocations apply.
 852                  */
 853                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
 854                     relsec->sh_link >= pi->ehdr->e_shnum)
 855                         return -ENOEXEC;
 856
 857                 section = pi->sechdrs + relsec->sh_info;
 858                 symtab = sechdrs + relsec->sh_link;
 859
 860                 if (!(section->sh_flags & SHF_ALLOC))
 861                         continue;
 862
 863                 /*
 864                  * symtab->sh_link contain section header index of associated
 865                  * string table.
 866                  */
 867                 if (symtab->sh_link >= pi->ehdr->e_shnum)
 868                         /* Invalid section number? */
 869                         continue;
 870
 871                 /*
 872                  * Respective architecture needs to provide support for applying
 873                  * relocations of type SHT_RELA/SHT_REL.
 874                  */
 875                 if (relsec->sh_type == SHT_RELA)
 876                         ret = arch_kexec_apply_relocations_add(pi, section,
 877                                                                relsec, symtab);
 878                 else if (relsec->sh_type == SHT_REL)
 879                         ret = arch_kexec_apply_relocations(pi, section,
 880                                                            relsec, symtab);
 881                 if (ret)
 882                         return ret;
 883         }
 884
 885         return 0;
 886 }
 887
 888 /*
 889  * kexec_load_purgatory - Load and relocate the purgatory object.
 890  * @image:      Image to add the purgatory to.
 891  * @kbuf:       Memory parameters to use.
 892  *
 893  * Allocates the memory needed for image->purgatory_info.sechdrs and
 894  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
 895  * to free the memory after use.
 896  *
 897  * Return: 0 on success, negative errno on error.
 898  */
 899 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
 900 {
 901         struct purgatory_info *pi = &image->purgatory_info;
 902         int ret;
 903
 904         if (kexec_purgatory_size <= 0)
 905                 return -EINVAL;
 906
 907         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
 908
 909         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
 910         if (ret)
 911                 return ret;
 912
 913         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
 914         if (ret)
 915                 goto out_free_kbuf;
 916
 917         ret = kexec_apply_relocations(image);
 918         if (ret)
 919                 goto out;
 920
 921         return 0;
 922 out:
 923         vfree(pi->sechdrs);
 924         pi->sechdrs = NULL;
 925 out_free_kbuf:
 926         vfree(pi->purgatory_buf);
 927         pi->purgatory_buf = NULL;
 928         return ret;
 929 }
 930
 931 /*
 932  * kexec_purgatory_find_symbol - find a symbol in the purgatory
 933  * @pi:         Purgatory to search in.
 934  * @name:       Name of the symbol.
 935  *
 936  * Return: pointer to symbol in read-only symtab on success, NULL on error.
 937  */
 938 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
 939                                                   const char *name)
 940 {
 941         const Elf_Shdr *sechdrs;
 942         const Elf_Ehdr *ehdr;
 943         const Elf_Sym *syms;
 944         const char *strtab;
 945         int i, k;
 946
 947         if (!pi->ehdr)
 948                 return NULL;
 949
 950         ehdr = pi->ehdr;
 951         sechdrs = (void *)ehdr + ehdr->e_shoff;
 952
 953         for (i = 0; i < ehdr->e_shnum; i++) {
 954                 if (sechdrs[i].sh_type != SHT_SYMTAB)
 955                         continue;
 956
 957                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
 958                         /* Invalid strtab section number */
 959                         continue;
 960                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
 961                 syms = (void *)ehdr + sechdrs[i].sh_offset;
 962
 963                 /* Go through symbols for a match */
 964                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
 965                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
 966                                 continue;
 967
 968                         if (strcmp(strtab + syms[k].st_name, name) != 0)
 969                                 continue;
 970
 971                         if (syms[k].st_shndx == SHN_UNDEF ||
 972                             syms[k].st_shndx >= ehdr->e_shnum) {
 973                                 pr_debug("Symbol: %s has bad section index %d.\n",
 974                                                 name, syms[k].st_shndx);
 975                                 return NULL;
 976                         }
 977
 978                         /* Found the symbol we are looking for */
 979                         return &syms[k];
 980                 }
 981         }
 982
 983         return NULL;
 984 }
 985
 986 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
 987 {
 988         struct purgatory_info *pi = &image->purgatory_info;
 989         const Elf_Sym *sym;
 990         Elf_Shdr *sechdr;
 991
 992         sym = kexec_purgatory_find_symbol(pi, name);
 993         if (!sym)
 994                 return ERR_PTR(-EINVAL);
 995
 996         sechdr = &pi->sechdrs[sym->st_shndx];
 997
 998         /*
 999          * Returns the address where symbol will finally be loaded after
1000          * kexec_load_segment()
1001          */
1002         return (void *)(sechdr->sh_addr + sym->st_value);
1003 }
1004
1005 /*
1006  * Get or set value of a symbol. If "get_value" is true, symbol value is
1007  * returned in buf otherwise symbol value is set based on value in buf.
1008  */
1009 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1010                                    void *buf, unsigned int size, bool get_value)
1011 {
1012         struct purgatory_info *pi = &image->purgatory_info;
1013         const Elf_Sym *sym;
1014         Elf_Shdr *sec;
1015         char *sym_buf;
1016
1017         sym = kexec_purgatory_find_symbol(pi, name);
1018         if (!sym)
1019                 return -EINVAL;
1020
1021         if (sym->st_size != size) {
1022                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1023                        name, (unsigned long)sym->st_size, size);
1024                 return -EINVAL;
1025         }
1026
1027         sec = pi->sechdrs + sym->st_shndx;
1028
1029         if (sec->sh_type == SHT_NOBITS) {
1030                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1031                        get_value ? "get" : "set");
1032                 return -EINVAL;
1033         }
1034
1035         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1036
1037         if (get_value)
1038                 memcpy((void *)buf, sym_buf, size);
1039         else
1040                 memcpy((void *)sym_buf, buf, size);
1041
1042         return 0;
1043 }
1044 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1045
1046 int crash_exclude_mem_range(struct crash_mem *mem,
1047                             unsigned long long mstart, unsigned long long mend)
1048 {
1049         int i, j;
1050         unsigned long long start, end;
1051         struct crash_mem_range temp_range = {0, 0};
1052
1053         for (i = 0; i < mem->nr_ranges; i++) {
1054                 start = mem->ranges[i].start;
1055                 end = mem->ranges[i].end;
1056
1057                 if (mstart > end || mend < start)
1058                         continue;
1059
1060                 /* Truncate any area outside of range */
1061                 if (mstart < start)
1062                         mstart = start;
1063                 if (mend > end)
1064                         mend = end;
1065
1066                 /* Found completely overlapping range */
1067                 if (mstart == start && mend == end) {
1068                         mem->ranges[i].start = 0;
1069                         mem->ranges[i].end = 0;
1070                         if (i < mem->nr_ranges - 1) {
1071                                 /* Shift rest of the ranges to left */
1072                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1073                                         mem->ranges[j].start =
1074                                                 mem->ranges[j+1].start;
1075                                         mem->ranges[j].end =
1076                                                         mem->ranges[j+1].end;
1077                                 }
1078                         }
1079                         mem->nr_ranges--;
1080                         return 0;
1081                 }
1082
1083                 if (mstart > start && mend < end) {
1084                         /* Split original range */
1085                         mem->ranges[i].end = mstart - 1;
1086                         temp_range.start = mend + 1;
1087                         temp_range.end = end;
1088                 } else if (mstart != start)
1089                         mem->ranges[i].end = mstart - 1;
1090                 else
1091                         mem->ranges[i].start = mend + 1;
1092                 break;
1093         }
1094
1095         /* If a split happened, add the split to array */
1096         if (!temp_range.end)
1097                 return 0;
1098
1099         /* Split happened */
1100         if (i == mem->max_nr_ranges - 1)
1101                 return -ENOMEM;
1102
1103         /* Location where new range should go */
1104         j = i + 1;
1105         if (j < mem->nr_ranges) {
1106                 /* Move over all ranges one slot towards the end */
1107                 for (i = mem->nr_ranges - 1; i >= j; i--)
1108                         mem->ranges[i + 1] = mem->ranges[i];
1109         }
1110
1111         mem->ranges[j].start = temp_range.start;
1112         mem->ranges[j].end = temp_range.end;
1113         mem->nr_ranges++;
1114         return 0;
1115 }
1116
1117 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1118                           void **addr, unsigned long *sz)
1119 {
1120         Elf64_Ehdr *ehdr;
1121         Elf64_Phdr *phdr;
1122         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1123         unsigned char *buf;
1124         unsigned int cpu, i;
1125         unsigned long long notes_addr;
1126         unsigned long mstart, mend;
1127
1128         /* extra phdr for vmcoreinfo elf note */
1129         nr_phdr = nr_cpus + 1;
1130         nr_phdr += mem->nr_ranges;
1131
1132         /*
1133          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1134          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1135          * I think this is required by tools like gdb. So same physical
1136          * memory will be mapped in two elf headers. One will contain kernel
1137          * text virtual addresses and other will have __va(physical) addresses.
1138          */
1139
1140         nr_phdr++;
1141         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1142         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1143
1144         buf = vzalloc(elf_sz);
1145         if (!buf)
1146                 return -ENOMEM;
1147
1148         ehdr = (Elf64_Ehdr *)buf;
1149         phdr = (Elf64_Phdr *)(ehdr + 1);
1150         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1151         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1152         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1153         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1154         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1155         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1156         ehdr->e_type = ET_CORE;
1157         ehdr->e_machine = ELF_ARCH;
1158         ehdr->e_version = EV_CURRENT;
1159         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1160         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1161         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1162
1163         /* Prepare one phdr of type PT_NOTE for each present cpu */
1164         for_each_present_cpu(cpu) {
1165                 phdr->p_type = PT_NOTE;
1166                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1167                 phdr->p_offset = phdr->p_paddr = notes_addr;
1168                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1169                 (ehdr->e_phnum)++;
1170                 phdr++;
1171         }
1172
1173         /* Prepare one PT_NOTE header for vmcoreinfo */
1174         phdr->p_type = PT_NOTE;
1175         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1176         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1177         (ehdr->e_phnum)++;
1178         phdr++;
1179
1180         /* Prepare PT_LOAD type program header for kernel text region */
1181         if (kernel_map) {
1182                 phdr->p_type = PT_LOAD;
1183                 phdr->p_flags = PF_R|PF_W|PF_X;
1184                 phdr->p_vaddr = (Elf64_Addr)_text;
1185                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1186                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1187                 ehdr->e_phnum++;
1188                 phdr++;
1189         }
1190
1191         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1192         for (i = 0; i < mem->nr_ranges; i++) {
1193                 mstart = mem->ranges[i].start;
1194                 mend = mem->ranges[i].end;
1195
1196                 phdr->p_type = PT_LOAD;
1197                 phdr->p_flags = PF_R|PF_W|PF_X;
1198                 phdr->p_offset  = mstart;
1199
1200                 phdr->p_paddr = mstart;
1201                 phdr->p_vaddr = (unsigned long long) __va(mstart);
1202                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1203                 phdr->p_align = 0;
1204                 ehdr->e_phnum++;
1205                 phdr++;
1206                 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",
1207                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1208                         ehdr->e_phnum, phdr->p_offset);
1209         }
1210
1211         *addr = buf;
1212         *sz = elf_sz;
1213         return 0;
1214 }