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