cmds-inspect.c

   1 /*
   2  * This program is free software; you can redistribute it and/or
   3  * modify it under the terms of the GNU General Public
   4  * License v2 as published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   9  * General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public
  12  * License along with this program; if not, write to the
  13  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  14  * Boston, MA 021110-1307, USA.
  15  */
  16
  17 #include <stdio.h>
  18 #include <stdlib.h>
  19 #include <unistd.h>
  20 #include <stdint.h>
  21 #include <sys/ioctl.h>
  22 #include <errno.h>
  23 #include <getopt.h>
  24 #include <limits.h>
  25
  26 #include "kerncompat.h"
  27 #include "ioctl.h"
  28 #include "utils.h"
  29 #include "ctree.h"
  30 #include "send-utils.h"
  31 #include "disk-io.h"
  32 #include "commands.h"
  33 #include "btrfs-list.h"
  34 #include "cmds-inspect-dump-tree.h"
  35
  36 static const char * const inspect_cmd_group_usage[] = {
  37         "btrfs inspect-internal <command> <args>",
  38         NULL
  39 };
  40
  41 static int __ino_to_path_fd(u64 inum, int fd, int verbose, const char *prepend)
  42 {
  43         int ret;
  44         int i;
  45         struct btrfs_ioctl_ino_path_args ipa;
  46         struct btrfs_data_container fspath[PATH_MAX];
  47
  48         memset(fspath, 0, sizeof(*fspath));
  49         ipa.inum = inum;
  50         ipa.size = PATH_MAX;
  51         ipa.fspath = ptr_to_u64(fspath);
  52
  53         ret = ioctl(fd, BTRFS_IOC_INO_PATHS, &ipa);
  54         if (ret < 0) {
  55                 printf("ioctl ret=%d, error: %s\n", ret, strerror(errno));
  56                 goto out;
  57         }
  58
  59         if (verbose)
  60                 printf("ioctl ret=%d, bytes_left=%lu, bytes_missing=%lu, "
  61                         "cnt=%d, missed=%d\n", ret,
  62                         (unsigned long)fspath->bytes_left,
  63                         (unsigned long)fspath->bytes_missing,
  64                         fspath->elem_cnt, fspath->elem_missed);
  65
  66         for (i = 0; i < fspath->elem_cnt; ++i) {
  67                 u64 ptr;
  68                 char *str;
  69                 ptr = (u64)(unsigned long)fspath->val;
  70                 ptr += fspath->val[i];
  71                 str = (char *)(unsigned long)ptr;
  72                 if (prepend)
  73                         printf("%s/%s\n", prepend, str);
  74                 else
  75                         printf("%s\n", str);
  76         }
  77
  78 out:
  79         return !!ret;
  80 }
  81
  82 static const char * const cmd_inspect_inode_resolve_usage[] = {
  83         "btrfs inspect-internal inode-resolve [-v] <inode> <path>",
  84         "Get file system paths for the given inode",
  85         "",
  86         "-v   verbose mode",
  87         NULL
  88 };
  89
  90 static int cmd_inspect_inode_resolve(int argc, char **argv)
  91 {
  92         int fd;
  93         int verbose = 0;
  94         int ret;
  95         DIR *dirstream = NULL;
  96
  97         optind = 1;
  98         while (1) {
  99                 int c = getopt(argc, argv, "v");
 100                 if (c < 0)
 101                         break;
 102
 103                 switch (c) {
 104                 case 'v':
 105                         verbose = 1;
 106                         break;
 107                 default:
 108                         usage(cmd_inspect_inode_resolve_usage);
 109                 }
 110         }
 111
 112         if (check_argc_exact(argc - optind, 2))
 113                 usage(cmd_inspect_inode_resolve_usage);
 114
 115         fd = btrfs_open_dir(argv[optind + 1], &dirstream, 1);
 116         if (fd < 0)
 117                 return 1;
 118
 119         ret = __ino_to_path_fd(arg_strtou64(argv[optind]), fd, verbose,
 120                                argv[optind+1]);
 121         close_file_or_dir(fd, dirstream);
 122         return !!ret;
 123
 124 }
 125
 126 static const char * const cmd_inspect_logical_resolve_usage[] = {
 127         "btrfs inspect-internal logical-resolve [-Pv] [-s bufsize] <logical> <path>",
 128         "Get file system paths for the given logical address",
 129         "-P          skip the path resolving and print the inodes instead",
 130         "-v          verbose mode",
 131         "-s bufsize  set inode container's size. This is used to increase inode",
 132         "            container's size in case it is not enough to read all the ",
 133         "            resolved results. The max value one can set is 64k",
 134         NULL
 135 };
 136
 137 static int cmd_inspect_logical_resolve(int argc, char **argv)
 138 {
 139         int ret;
 140         int fd;
 141         int i;
 142         int verbose = 0;
 143         int getpath = 1;
 144         int bytes_left;
 145         struct btrfs_ioctl_logical_ino_args loi;
 146         struct btrfs_data_container *inodes;
 147         u64 size = 4096;
 148         char full_path[4096];
 149         char *path_ptr;
 150         DIR *dirstream = NULL;
 151
 152         optind = 1;
 153         while (1) {
 154                 int c = getopt(argc, argv, "Pvs:");
 155                 if (c < 0)
 156                         break;
 157
 158                 switch (c) {
 159                 case 'P':
 160                         getpath = 0;
 161                         break;
 162                 case 'v':
 163                         verbose = 1;
 164                         break;
 165                 case 's':
 166                         size = arg_strtou64(optarg);
 167                         break;
 168                 default:
 169                         usage(cmd_inspect_logical_resolve_usage);
 170                 }
 171         }
 172
 173         if (check_argc_exact(argc - optind, 2))
 174                 usage(cmd_inspect_logical_resolve_usage);
 175
 176         size = min(size, (u64)64 * 1024);
 177         inodes = malloc(size);
 178         if (!inodes)
 179                 return 1;
 180
 181         memset(inodes, 0, sizeof(*inodes));
 182         loi.logical = arg_strtou64(argv[optind]);
 183         loi.size = size;
 184         loi.inodes = ptr_to_u64(inodes);
 185
 186         fd = btrfs_open_dir(argv[optind + 1], &dirstream, 1);
 187         if (fd < 0) {
 188                 ret = 12;
 189                 goto out;
 190         }
 191
 192         ret = ioctl(fd, BTRFS_IOC_LOGICAL_INO, &loi);
 193         if (ret < 0) {
 194                 printf("ioctl ret=%d, error: %s\n", ret, strerror(errno));
 195                 goto out;
 196         }
 197
 198         if (verbose)
 199                 printf("ioctl ret=%d, total_size=%llu, bytes_left=%lu, "
 200                         "bytes_missing=%lu, cnt=%d, missed=%d\n",
 201                         ret, size,
 202                         (unsigned long)inodes->bytes_left,
 203                         (unsigned long)inodes->bytes_missing,
 204                         inodes->elem_cnt, inodes->elem_missed);
 205
 206         bytes_left = sizeof(full_path);
 207         ret = snprintf(full_path, bytes_left, "%s/", argv[optind+1]);
 208         path_ptr = full_path + ret;
 209         bytes_left -= ret + 1;
 210         BUG_ON(bytes_left < 0);
 211
 212         for (i = 0; i < inodes->elem_cnt; i += 3) {
 213                 u64 inum = inodes->val[i];
 214                 u64 offset = inodes->val[i+1];
 215                 u64 root = inodes->val[i+2];
 216                 int path_fd;
 217                 char *name;
 218                 DIR *dirs = NULL;
 219
 220                 if (getpath) {
 221                         name = btrfs_list_path_for_root(fd, root);
 222                         if (IS_ERR(name)) {
 223                                 ret = PTR_ERR(name);
 224                                 goto out;
 225                         }
 226                         if (!name) {
 227                                 path_ptr[-1] = '\0';
 228                                 path_fd = fd;
 229                         } else {
 230                                 path_ptr[-1] = '/';
 231                                 ret = snprintf(path_ptr, bytes_left, "%s",
 232                                                 name);
 233                                 BUG_ON(ret >= bytes_left);
 234                                 free(name);
 235                                 path_fd = btrfs_open_dir(full_path, &dirs, 1);
 236                                 if (path_fd < 0) {
 237                                         ret = -ENOENT;
 238                                         goto out;
 239                                 }
 240                         }
 241                         __ino_to_path_fd(inum, path_fd, verbose, full_path);
 242                         if (path_fd != fd)
 243                                 close_file_or_dir(path_fd, dirs);
 244                 } else {
 245                         printf("inode %llu offset %llu root %llu\n", inum,
 246                                 offset, root);
 247                 }
 248         }
 249
 250 out:
 251         close_file_or_dir(fd, dirstream);
 252         free(inodes);
 253         return !!ret;
 254 }
 255
 256 static const char * const cmd_inspect_subvolid_resolve_usage[] = {
 257         "btrfs inspect-internal subvolid-resolve <subvolid> <path>",
 258         "Get file system paths for the given subvolume ID.",
 259         NULL
 260 };
 261
 262 static int cmd_inspect_subvolid_resolve(int argc, char **argv)
 263 {
 264         int ret;
 265         int fd = -1;
 266         u64 subvol_id;
 267         char path[PATH_MAX];
 268         DIR *dirstream = NULL;
 269
 270         if (check_argc_exact(argc, 3))
 271                 usage(cmd_inspect_subvolid_resolve_usage);
 272
 273         fd = btrfs_open_dir(argv[2], &dirstream, 1);
 274         if (fd < 0) {
 275                 ret = -ENOENT;
 276                 goto out;
 277         }
 278
 279         subvol_id = arg_strtou64(argv[1]);
 280         ret = btrfs_subvolid_resolve(fd, path, sizeof(path), subvol_id);
 281
 282         if (ret) {
 283                 fprintf(stderr,
 284                         "%s: btrfs_subvolid_resolve(subvol_id %llu) failed with ret=%d\n",
 285                         argv[0], (unsigned long long)subvol_id, ret);
 286                 goto out;
 287         }
 288
 289         path[PATH_MAX - 1] = '\0';
 290         printf("%s\n", path);
 291
 292 out:
 293         close_file_or_dir(fd, dirstream);
 294         return ret ? 1 : 0;
 295 }
 296
 297 static const char* const cmd_inspect_rootid_usage[] = {
 298         "btrfs inspect-internal rootid <path>",
 299         "Get tree ID of the containing subvolume of path.",
 300         NULL
 301 };
 302
 303 static int cmd_inspect_rootid(int argc, char **argv)
 304 {
 305         int ret;
 306         int fd = -1;
 307         u64 rootid;
 308         DIR *dirstream = NULL;
 309
 310         if (check_argc_exact(argc, 2))
 311                 usage(cmd_inspect_rootid_usage);
 312
 313         fd = btrfs_open_dir(argv[1], &dirstream, 1);
 314         if (fd < 0) {
 315                 ret = -ENOENT;
 316                 goto out;
 317         }
 318
 319         ret = lookup_ino_rootid(fd, &rootid);
 320         if (ret) {
 321                 fprintf(stderr, "%s: rootid failed with ret=%d\n",
 322                         argv[0], ret);
 323                 goto out;
 324         }
 325
 326         printf("%llu\n", (unsigned long long)rootid);
 327 out:
 328         close_file_or_dir(fd, dirstream);
 329
 330         return !!ret;
 331 }
 332
 333 static const char* const cmd_inspect_min_dev_size_usage[] = {
 334         "btrfs inspect-internal min-dev-size [options] <path>",
 335         "Get the minimum size the device can be shrunk to. The",
 336         "device id 1 is used by default.",
 337         "--id DEVID   specify the device id to query",
 338         NULL
 339 };
 340
 341 struct dev_extent_elem {
 342         u64 start;
 343         /* inclusive end */
 344         u64 end;
 345         struct list_head list;
 346 };
 347
 348 static int add_dev_extent(struct list_head *list,
 349                           const u64 start, const u64 end,
 350                           const int append)
 351 {
 352         struct dev_extent_elem *e;
 353
 354         e = malloc(sizeof(*e));
 355         if (!e)
 356                 return -ENOMEM;
 357
 358         e->start = start;
 359         e->end = end;
 360
 361         if (append)
 362                 list_add_tail(&e->list, list);
 363         else
 364                 list_add(&e->list, list);
 365
 366         return 0;
 367 }
 368
 369 static void free_dev_extent_list(struct list_head *list)
 370 {
 371         while (!list_empty(list)) {
 372                 struct dev_extent_elem *e;
 373
 374                 e = list_first_entry(list, struct dev_extent_elem, list);
 375                 list_del(&e->list);
 376                 free(e);
 377         }
 378 }
 379
 380 static int hole_includes_sb_mirror(const u64 start, const u64 end)
 381 {
 382         int i;
 383         int ret = 0;
 384
 385         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 386                 u64 bytenr = btrfs_sb_offset(i);
 387
 388                 if (bytenr >= start && bytenr <= end) {
 389                         ret = 1;
 390                         break;
 391                 }
 392         }
 393
 394         return ret;
 395 }
 396
 397 static void adjust_dev_min_size(struct list_head *extents,
 398                                 struct list_head *holes,
 399                                 u64 *min_size)
 400 {
 401         /*
 402          * If relocation of the block group of a device extent must happen (see
 403          * below) scratch space is used for the relocation. So track here the
 404          * size of the largest device extent that has to be relocated. We track
 405          * only the largest and not the sum of the sizes of all relocated block
 406          * groups because after each block group is relocated the running
 407          * transaction is committed so that pinned space is released.
 408          */
 409         u64 scratch_space = 0;
 410
 411         /*
 412          * List of device extents is sorted by descending order of the extent's
 413          * end offset. If some extent goes beyond the computed minimum size,
 414          * which initially matches the sum of the lenghts of all extents,
 415          * we need to check if the extent can be relocated to an hole in the
 416          * device between [0, *min_size[ (which is what the resize ioctl does).
 417          */
 418         while (!list_empty(extents)) {
 419                 struct dev_extent_elem *e;
 420                 struct dev_extent_elem *h;
 421                 int found = 0;
 422                 u64 extent_len;
 423                 u64 hole_len = 0;
 424
 425                 e = list_first_entry(extents, struct dev_extent_elem, list);
 426                 if (e->end <= *min_size)
 427                         break;
 428
 429                 /*
 430                  * Our extent goes beyond the computed *min_size. See if we can
 431                  * find a hole large enough to relocate it to. If not we must stop
 432                  * and set *min_size to the end of the extent.
 433                  */
 434                 extent_len = e->end - e->start + 1;
 435                 list_for_each_entry(h, holes, list) {
 436                         hole_len = h->end - h->start + 1;
 437                         if (hole_len >= extent_len) {
 438                                 found = 1;
 439                                 break;
 440                         }
 441                 }
 442
 443                 if (!found) {
 444                         *min_size = e->end + 1;
 445                         break;
 446                 }
 447
 448                 /*
 449                  * If the hole found contains the location for a superblock
 450                  * mirror, we are pessimistic and require allocating one
 451                  * more extent of the same size. This is because the block
 452                  * group could be in the worst case used by a single extent
 453                  * with a size >= (block_group.length - superblock.size).
 454                  */
 455                 if (hole_includes_sb_mirror(h->start,
 456                                             h->start + extent_len - 1))
 457                         *min_size += extent_len;
 458
 459                 if (hole_len > extent_len) {
 460                         h->start += extent_len;
 461                 } else {
 462                         list_del(&h->list);
 463                         free(h);
 464                 }
 465
 466                 list_del(&e->list);
 467                 free(e);
 468
 469                 if (extent_len > scratch_space)
 470                         scratch_space = extent_len;
 471         }
 472
 473         if (scratch_space) {
 474                 *min_size += scratch_space;
 475                 /*
 476                  * Chunk allocation requires inserting/updating items in the
 477                  * chunk tree, so often this can lead to the need of allocating
 478                  * a new system chunk too, which has a maximum size of 32Mb.
 479                  */
 480                 *min_size += 32 * 1024 * 1024;
 481         }
 482 }
 483
 484 static int print_min_dev_size(int fd, u64 devid)
 485 {
 486         int ret = 1;
 487         /*
 488          * Device allocations starts at 1Mb or at the value passed through the
 489          * mount option alloc_start if it's bigger than 1Mb. The alloc_start
 490          * option is used for debugging and testing only, and recently the
 491          * possibility of deprecating/removing it has been discussed, so we
 492          * ignore it here.
 493          */
 494         u64 min_size = 1 * 1024 * 1024ull;
 495         struct btrfs_ioctl_search_args args;
 496         struct btrfs_ioctl_search_key *sk = &args.key;
 497         u64 last_pos = (u64)-1;
 498         LIST_HEAD(extents);
 499         LIST_HEAD(holes);
 500
 501         memset(&args, 0, sizeof(args));
 502         sk->tree_id = BTRFS_DEV_TREE_OBJECTID;
 503         sk->min_objectid = devid;
 504         sk->max_objectid = devid;
 505         sk->max_type = BTRFS_DEV_EXTENT_KEY;
 506         sk->min_type = BTRFS_DEV_EXTENT_KEY;
 507         sk->min_offset = 0;
 508         sk->max_offset = (u64)-1;
 509         sk->min_transid = 0;
 510         sk->max_transid = (u64)-1;
 511         sk->nr_items = 4096;
 512
 513         while (1) {
 514                 int i;
 515                 struct btrfs_ioctl_search_header *sh;
 516                 unsigned long off = 0;
 517
 518                 ret = ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args);
 519                 if (ret < 0) {
 520                         fprintf(stderr,
 521                                 "Error invoking tree search ioctl: %s\n",
 522                                 strerror(errno));
 523                         ret = 1;
 524                         goto out;
 525                 }
 526
 527                 if (sk->nr_items == 0)
 528                         break;
 529
 530                 for (i = 0; i < sk->nr_items; i++) {
 531                         struct btrfs_dev_extent *extent;
 532                         u64 len;
 533
 534                         sh = (struct btrfs_ioctl_search_header *)(args.buf +
 535                                                                   off);
 536                         off += sizeof(*sh);
 537                         extent = (struct btrfs_dev_extent *)(args.buf + off);
 538                         off += sh->len;
 539
 540                         sk->min_objectid = sh->objectid;
 541                         sk->min_type = sh->type;
 542                         sk->min_offset = sh->offset + 1;
 543
 544                         if (sh->objectid != devid ||
 545                             sh->type != BTRFS_DEV_EXTENT_KEY)
 546                                 continue;
 547
 548                         len = btrfs_stack_dev_extent_length(extent);
 549                         min_size += len;
 550                         ret = add_dev_extent(&extents, sh->offset,
 551                                              sh->offset + len - 1, 0);
 552
 553                         if (!ret && last_pos != (u64)-1 &&
 554                             last_pos != sh->offset)
 555                                 ret = add_dev_extent(&holes, last_pos,
 556                                                      sh->offset - 1, 1);
 557                         if (ret) {
 558                                 fprintf(stderr, "Error: %s\n", strerror(-ret));
 559                                 ret = 1;
 560                                 goto out;
 561                         }
 562
 563                         last_pos = sh->offset + len;
 564                 }
 565
 566                 if (sk->min_type != BTRFS_DEV_EXTENT_KEY ||
 567                     sk->min_objectid != devid)
 568                         break;
 569         }
 570
 571         adjust_dev_min_size(&extents, &holes, &min_size);
 572         printf("%llu bytes (%s)\n", min_size, pretty_size(min_size));
 573         ret = 0;
 574 out:
 575         free_dev_extent_list(&extents);
 576         free_dev_extent_list(&holes);
 577
 578         return ret;
 579 }
 580
 581 static int cmd_inspect_min_dev_size(int argc, char **argv)
 582 {
 583         int ret;
 584         int fd = -1;
 585         DIR *dirstream = NULL;
 586         u64 devid = 1;
 587
 588         while (1) {
 589                 int c;
 590                 enum { GETOPT_VAL_DEVID = 256 };
 591                 static const struct option long_options[] = {
 592                         { "id", required_argument, NULL, GETOPT_VAL_DEVID },
 593                         {NULL, 0, NULL, 0}
 594                 };
 595
 596                 c = getopt_long(argc, argv, "", long_options, NULL);
 597                 if (c < 0)
 598                         break;
 599
 600                 switch (c) {
 601                 case GETOPT_VAL_DEVID:
 602                         devid = arg_strtou64(optarg);
 603                         break;
 604                 default:
 605                         usage(cmd_inspect_min_dev_size_usage);
 606                 }
 607         }
 608         if (check_argc_exact(argc - optind, 1))
 609                 usage(cmd_inspect_min_dev_size_usage);
 610
 611         fd = btrfs_open_dir(argv[optind], &dirstream, 1);
 612         if (fd < 0) {
 613                 ret = -ENOENT;
 614                 goto out;
 615         }
 616
 617         ret = print_min_dev_size(fd, devid);
 618         close_file_or_dir(fd, dirstream);
 619 out:
 620         return !!ret;
 621 }
 622
 623 static int cmd_inspect_dump_tree_hook(int ac, char **av)
 624 {
 625         return cmd_inspect_dump_tree(ac, av);
 626 }
 627
 628 static const char inspect_cmd_group_info[] =
 629 "query various internal information";
 630
 631 const struct cmd_group inspect_cmd_group = {
 632         inspect_cmd_group_usage, inspect_cmd_group_info, {
 633                 { "inode-resolve", cmd_inspect_inode_resolve,
 634                         cmd_inspect_inode_resolve_usage, NULL, 0 },
 635                 { "logical-resolve", cmd_inspect_logical_resolve,
 636                         cmd_inspect_logical_resolve_usage, NULL, 0 },
 637                 { "subvolid-resolve", cmd_inspect_subvolid_resolve,
 638                         cmd_inspect_subvolid_resolve_usage, NULL, 0 },
 639                 { "rootid", cmd_inspect_rootid, cmd_inspect_rootid_usage, NULL,
 640                         0 },
 641                 { "min-dev-size", cmd_inspect_min_dev_size,
 642                         cmd_inspect_min_dev_size_usage, NULL, 0 },
 643                 { "dump-tree", cmd_inspect_dump_tree_hook,
 644                                 cmd_inspect_dump_tree_usage, NULL, 0 },
 645                 NULL_CMD_STRUCT
 646         }
 647 };
 648
 649 int cmd_inspect(int argc, char **argv)
 650 {
 651         return handle_command_group(&inspect_cmd_group, argc, argv);
 652 }