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