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