scrub/phase6.c

   1 /*
   2  * Copyright (C) 2018 Oracle.  All Rights Reserved.
   3  *
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version 2
   9  * of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it would be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write the Free Software Foundation,
  18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
  19  */
  20 #include "xfs.h"
  21 #include <stdint.h>
  22 #include <dirent.h>
  23 #include <sys/statvfs.h>
  24 #include "handle.h"
  25 #include "path.h"
  26 #include "ptvar.h"
  27 #include "workqueue.h"
  28 #include "xfs_scrub.h"
  29 #include "common.h"
  30 #include "bitmap.h"
  31 #include "disk.h"
  32 #include "filemap.h"
  33 #include "fscounters.h"
  34 #include "inodes.h"
  35 #include "read_verify.h"
  36 #include "spacemap.h"
  37 #include "vfs.h"
  38
  39 /*
  40  * Phase 6: Verify data file integrity.
  41  *
  42  * Identify potential data block extents with GETFSMAP, then feed those
  43  * extents to the read-verify pool to get the verify commands batched,
  44  * issued, and (if there are problems) reported back to us.  If there
  45  * are errors, we'll record the bad regions and (if available) use rmap
  46  * to tell us if metadata are now corrupt.  Otherwise, we'll scan the
  47  * whole directory tree looking for files that overlap the bad regions
  48  * and report the paths of the now corrupt files.
  49  */
  50
  51 /* Find the fd for a given device identifier. */
  52 static struct disk *
  53 xfs_dev_to_disk(
  54         struct scrub_ctx        *ctx,
  55         dev_t                   dev)
  56 {
  57         if (dev == ctx->fsinfo.fs_datadev)
  58                 return ctx->datadev;
  59         else if (dev == ctx->fsinfo.fs_logdev)
  60                 return ctx->logdev;
  61         else if (dev == ctx->fsinfo.fs_rtdev)
  62                 return ctx->rtdev;
  63         abort();
  64 }
  65
  66 /* Find the device major/minor for a given file descriptor. */
  67 static dev_t
  68 xfs_disk_to_dev(
  69         struct scrub_ctx        *ctx,
  70         struct disk             *disk)
  71 {
  72         if (disk == ctx->datadev)
  73                 return ctx->fsinfo.fs_datadev;
  74         else if (disk == ctx->logdev)
  75                 return ctx->fsinfo.fs_logdev;
  76         else if (disk == ctx->rtdev)
  77                 return ctx->fsinfo.fs_rtdev;
  78         abort();
  79 }
  80
  81 struct owner_decode {
  82         uint64_t                owner;
  83         const char              *descr;
  84 };
  85
  86 static const struct owner_decode special_owners[] = {
  87         {XFS_FMR_OWN_FREE,      "free space"},
  88         {XFS_FMR_OWN_UNKNOWN,   "unknown owner"},
  89         {XFS_FMR_OWN_FS,        "static FS metadata"},
  90         {XFS_FMR_OWN_LOG,       "journalling log"},
  91         {XFS_FMR_OWN_AG,        "per-AG metadata"},
  92         {XFS_FMR_OWN_INOBT,     "inode btree blocks"},
  93         {XFS_FMR_OWN_INODES,    "inodes"},
  94         {XFS_FMR_OWN_REFC,      "refcount btree"},
  95         {XFS_FMR_OWN_COW,       "CoW staging"},
  96         {XFS_FMR_OWN_DEFECTIVE, "bad blocks"},
  97         {0, NULL},
  98 };
  99
 100 /* Decode a special owner. */
 101 static const char *
 102 xfs_decode_special_owner(
 103         uint64_t                        owner)
 104 {
 105         const struct owner_decode       *od = special_owners;
 106
 107         while (od->descr) {
 108                 if (od->owner == owner)
 109                         return od->descr;
 110                 od++;
 111         }
 112
 113         return NULL;
 114 }
 115
 116 /* Routines to translate bad physical extents into file paths and offsets. */
 117
 118 struct xfs_verify_error_info {
 119         struct bitmap                   *d_bad;         /* bytes */
 120         struct bitmap                   *r_bad;         /* bytes */
 121 };
 122
 123 /* Report if this extent overlaps a bad region. */
 124 static bool
 125 xfs_report_verify_inode_bmap(
 126         struct scrub_ctx                *ctx,
 127         const char                      *descr,
 128         int                             fd,
 129         int                             whichfork,
 130         struct fsxattr                  *fsx,
 131         struct xfs_bmap                 *bmap,
 132         void                            *arg)
 133 {
 134         struct xfs_verify_error_info    *vei = arg;
 135         struct bitmap                   *bmp;
 136
 137         /* Only report errors for real extents. */
 138         if (bmap->bm_flags & (BMV_OF_PREALLOC | BMV_OF_DELALLOC))
 139                 return true;
 140
 141         if (fsx->fsx_xflags & FS_XFLAG_REALTIME)
 142                 bmp = vei->r_bad;
 143         else
 144                 bmp = vei->d_bad;
 145
 146         if (!bitmap_test(bmp, bmap->bm_physical, bmap->bm_length))
 147                 return true;
 148
 149         str_error(ctx, descr,
 150 _("offset %llu failed read verification."), bmap->bm_offset);
 151         return true;
 152 }
 153
 154 /* Iterate the extent mappings of a file to report errors. */
 155 static bool
 156 xfs_report_verify_fd(
 157         struct scrub_ctx                *ctx,
 158         const char                      *descr,
 159         int                             fd,
 160         void                            *arg)
 161 {
 162         struct xfs_bmap                 key = {0};
 163         bool                            moveon;
 164
 165         /* data fork */
 166         moveon = xfs_iterate_filemaps(ctx, descr, fd, XFS_DATA_FORK, &key,
 167                         xfs_report_verify_inode_bmap, arg);
 168         if (!moveon)
 169                 return false;
 170
 171         /* attr fork */
 172         moveon = xfs_iterate_filemaps(ctx, descr, fd, XFS_ATTR_FORK, &key,
 173                         xfs_report_verify_inode_bmap, arg);
 174         if (!moveon)
 175                 return false;
 176         return true;
 177 }
 178
 179 /* Report read verify errors in unlinked (but still open) files. */
 180 static int
 181 xfs_report_verify_inode(
 182         struct scrub_ctx                *ctx,
 183         struct xfs_handle               *handle,
 184         struct xfs_bstat                *bstat,
 185         void                            *arg)
 186 {
 187         char                            descr[DESCR_BUFSZ];
 188         bool                            moveon;
 189         int                             fd;
 190         int                             error;
 191
 192         snprintf(descr, DESCR_BUFSZ, _("inode %"PRIu64" (unlinked)"),
 193                         (uint64_t)bstat->bs_ino);
 194
 195         /* Ignore linked files and things we can't open. */
 196         if (bstat->bs_nlink != 0)
 197                 return 0;
 198         if (!S_ISREG(bstat->bs_mode) && !S_ISDIR(bstat->bs_mode))
 199                 return 0;
 200
 201         /* Try to open the inode. */
 202         fd = xfs_open_handle(handle);
 203         if (fd < 0) {
 204                 error = errno;
 205                 if (error == ESTALE)
 206                         return error;
 207
 208                 str_info(ctx, descr,
 209 _("Disappeared during read error reporting."));
 210                 return error;
 211         }
 212
 213         /* Go find the badness. */
 214         moveon = xfs_report_verify_fd(ctx, descr, fd, arg);
 215         error = close(fd);
 216         if (error)
 217                 str_errno(ctx, descr);
 218
 219         return moveon ? 0 : XFS_ITERATE_INODES_ABORT;
 220 }
 221
 222 /* Scan a directory for matches in the read verify error list. */
 223 static bool
 224 xfs_report_verify_dir(
 225         struct scrub_ctx        *ctx,
 226         const char              *path,
 227         int                     dir_fd,
 228         void                    *arg)
 229 {
 230         return xfs_report_verify_fd(ctx, path, dir_fd, arg);
 231 }
 232
 233 /*
 234  * Scan the inode associated with a directory entry for matches with
 235  * the read verify error list.
 236  */
 237 static bool
 238 xfs_report_verify_dirent(
 239         struct scrub_ctx        *ctx,
 240         const char              *path,
 241         int                     dir_fd,
 242         struct dirent           *dirent,
 243         struct stat             *sb,
 244         void                    *arg)
 245 {
 246         bool                    moveon;
 247         int                     fd;
 248         int                     error;
 249
 250         /* Ignore things we can't open. */
 251         if (!S_ISREG(sb->st_mode) && !S_ISDIR(sb->st_mode))
 252                 return true;
 253
 254         /* Ignore . and .. */
 255         if (!strcmp(".", dirent->d_name) || !strcmp("..", dirent->d_name))
 256                 return true;
 257
 258         /*
 259          * If we were given a dirent, open the associated file under
 260          * dir_fd for badblocks scanning.  If dirent is NULL, then it's
 261          * the directory itself we want to scan.
 262          */
 263         fd = openat(dir_fd, dirent->d_name,
 264                         O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NOCTTY);
 265         if (fd < 0)
 266                 return true;
 267
 268         /* Go find the badness. */
 269         moveon = xfs_report_verify_fd(ctx, path, fd, arg);
 270         if (moveon)
 271                 goto out;
 272
 273 out:
 274         error = close(fd);
 275         if (error)
 276                 str_errno(ctx, path);
 277         return moveon;
 278 }
 279
 280 /* Given bad extent lists for the data & rtdev, find bad files. */
 281 static bool
 282 xfs_report_verify_errors(
 283         struct scrub_ctx                *ctx,
 284         struct bitmap                   *d_bad,
 285         struct bitmap                   *r_bad)
 286 {
 287         struct xfs_verify_error_info    vei;
 288         bool                            moveon;
 289
 290         vei.d_bad = d_bad;
 291         vei.r_bad = r_bad;
 292
 293         /* Scan the directory tree to get file paths. */
 294         moveon = scan_fs_tree(ctx, xfs_report_verify_dir,
 295                         xfs_report_verify_dirent, &vei);
 296         if (!moveon)
 297                 return false;
 298
 299         /* Scan for unlinked files. */
 300         return xfs_scan_all_inodes(ctx, xfs_report_verify_inode, &vei);
 301 }
 302
 303 /* Verify disk blocks with GETFSMAP */
 304
 305 struct xfs_verify_extent {
 306         struct read_verify_pool *readverify;
 307         struct ptvar            *rvstate;
 308         struct bitmap           *d_bad;         /* bytes */
 309         struct bitmap           *r_bad;         /* bytes */
 310 };
 311
 312 /* Report an IO error resulting from read-verify based off getfsmap. */
 313 static bool
 314 xfs_check_rmap_error_report(
 315         struct scrub_ctx        *ctx,
 316         const char              *descr,
 317         struct fsmap            *map,
 318         void                    *arg)
 319 {
 320         const char              *type;
 321         char                    buf[32];
 322         uint64_t                err_physical = *(uint64_t *)arg;
 323         uint64_t                err_off;
 324
 325         if (err_physical > map->fmr_physical)
 326                 err_off = err_physical - map->fmr_physical;
 327         else
 328                 err_off = 0;
 329
 330         snprintf(buf, 32, _("disk offset %"PRIu64),
 331                         (uint64_t)BTOBB(map->fmr_physical + err_off));
 332
 333         if (map->fmr_flags & FMR_OF_SPECIAL_OWNER) {
 334                 type = xfs_decode_special_owner(map->fmr_owner);
 335                 str_error(ctx, buf,
 336 _("%s failed read verification."),
 337                                 type);
 338         }
 339
 340         /*
 341          * XXX: If we had a getparent() call we could report IO errors
 342          * efficiently.  Until then, we'll have to scan the dir tree
 343          * to find the bad file's pathname.
 344          */
 345
 346         return true;
 347 }
 348
 349 /*
 350  * Remember a read error for later, and see if rmap will tell us about the
 351  * owner ahead of time.
 352  */
 353 static void
 354 xfs_check_rmap_ioerr(
 355         struct scrub_ctx                *ctx,
 356         struct disk                     *disk,
 357         uint64_t                        start,
 358         uint64_t                        length,
 359         int                             error,
 360         void                            *arg)
 361 {
 362         struct fsmap                    keys[2];
 363         char                            descr[DESCR_BUFSZ];
 364         struct xfs_verify_extent        *ve = arg;
 365         struct bitmap                   *tree;
 366         dev_t                           dev;
 367         bool                            moveon;
 368
 369         dev = xfs_disk_to_dev(ctx, disk);
 370
 371         /*
 372          * If we don't have parent pointers, save the bad extent for
 373          * later rescanning.
 374          */
 375         if (dev == ctx->fsinfo.fs_datadev)
 376                 tree = ve->d_bad;
 377         else if (dev == ctx->fsinfo.fs_rtdev)
 378                 tree = ve->r_bad;
 379         else
 380                 tree = NULL;
 381         if (tree) {
 382                 moveon = bitmap_set(tree, start, length);
 383                 if (!moveon)
 384                         str_errno(ctx, ctx->mntpoint);
 385         }
 386
 387         snprintf(descr, DESCR_BUFSZ, _("dev %d:%d ioerr @ %"PRIu64":%"PRIu64" "),
 388                         major(dev), minor(dev), start, length);
 389
 390         /* Go figure out which blocks are bad from the fsmap. */
 391         memset(keys, 0, sizeof(struct fsmap) * 2);
 392         keys->fmr_device = dev;
 393         keys->fmr_physical = start;
 394         (keys + 1)->fmr_device = dev;
 395         (keys + 1)->fmr_physical = start + length - 1;
 396         (keys + 1)->fmr_owner = ULLONG_MAX;
 397         (keys + 1)->fmr_offset = ULLONG_MAX;
 398         (keys + 1)->fmr_flags = UINT_MAX;
 399         xfs_iterate_fsmap(ctx, descr, keys, xfs_check_rmap_error_report,
 400                         &start);
 401 }
 402
 403 /* Schedule a read-verify of a (data block) extent. */
 404 static bool
 405 xfs_check_rmap(
 406         struct scrub_ctx                *ctx,
 407         const char                      *descr,
 408         struct fsmap                    *map,
 409         void                            *arg)
 410 {
 411         struct xfs_verify_extent        *ve = arg;
 412         struct disk                     *disk;
 413
 414         dbg_printf("rmap dev %d:%d phys %"PRIu64" owner %"PRId64
 415                         " offset %"PRIu64" len %"PRIu64" flags 0x%x\n",
 416                         major(map->fmr_device), minor(map->fmr_device),
 417                         (uint64_t)map->fmr_physical, (int64_t)map->fmr_owner,
 418                         (uint64_t)map->fmr_offset, (uint64_t)map->fmr_length,
 419                         map->fmr_flags);
 420
 421         /* "Unknown" extents should be verified; they could be data. */
 422         if ((map->fmr_flags & FMR_OF_SPECIAL_OWNER) &&
 423                         map->fmr_owner == XFS_FMR_OWN_UNKNOWN)
 424                 map->fmr_flags &= ~FMR_OF_SPECIAL_OWNER;
 425
 426         /*
 427          * We only care about read-verifying data extents that have been
 428          * written to disk.  This means we can skip "special" owners
 429          * (metadata), xattr blocks, unwritten extents, and extent maps.
 430          * These should all get checked elsewhere in the scrubber.
 431          */
 432         if (map->fmr_flags & (FMR_OF_PREALLOC | FMR_OF_ATTR_FORK |
 433                               FMR_OF_EXTENT_MAP | FMR_OF_SPECIAL_OWNER))
 434                 goto out;
 435
 436         /* XXX: Filter out directory data blocks. */
 437
 438         /* Schedule the read verify command for (eventual) running. */
 439         disk = xfs_dev_to_disk(ctx, map->fmr_device);
 440
 441         read_verify_schedule_io(ve->readverify, ptvar_get(ve->rvstate), disk,
 442                         map->fmr_physical, map->fmr_length, ve);
 443
 444 out:
 445         /* Is this the last extent?  Fire off the read. */
 446         if (map->fmr_flags & FMR_OF_LAST)
 447                 read_verify_force_io(ve->readverify, ptvar_get(ve->rvstate));
 448
 449         return true;
 450 }
 451
 452 /*
 453  * Read verify all the file data blocks in a filesystem.  Since XFS doesn't
 454  * do data checksums, we trust that the underlying storage will pass back
 455  * an IO error if it can't retrieve whatever we previously stored there.
 456  * If we hit an IO error, we'll record the bad blocks in a bitmap and then
 457  * scan the extent maps of the entire fs tree to figure (and the unlinked
 458  * inodes) out which files are now broken.
 459  */
 460 bool
 461 xfs_scan_blocks(
 462         struct scrub_ctx                *ctx)
 463 {
 464         struct xfs_verify_extent        ve;
 465         bool                            moveon;
 466
 467         ve.rvstate = ptvar_init(scrub_nproc(ctx), sizeof(struct read_verify));
 468         if (!ve.rvstate) {
 469                 str_errno(ctx, ctx->mntpoint);
 470                 return false;
 471         }
 472
 473         moveon = bitmap_init(&ve.d_bad);
 474         if (!moveon) {
 475                 str_errno(ctx, ctx->mntpoint);
 476                 goto out_ve;
 477         }
 478
 479         moveon = bitmap_init(&ve.r_bad);
 480         if (!moveon) {
 481                 str_errno(ctx, ctx->mntpoint);
 482                 goto out_dbad;
 483         }
 484
 485         ve.readverify = read_verify_pool_init(ctx, ctx->geo.blocksize,
 486                         xfs_check_rmap_ioerr, disk_heads(ctx->datadev));
 487         if (!ve.readverify) {
 488                 moveon = false;
 489                 str_info(ctx, ctx->mntpoint,
 490 _("Could not create media verifier."));
 491                 goto out_rbad;
 492         }
 493         moveon = xfs_scan_all_spacemaps(ctx, xfs_check_rmap, &ve);
 494         if (!moveon)
 495                 goto out_pool;
 496         read_verify_pool_flush(ve.readverify);
 497         ctx->bytes_checked += read_verify_bytes(ve.readverify);
 498         read_verify_pool_destroy(ve.readverify);
 499
 500         /* Scan the whole dir tree to see what matches the bad extents. */
 501         if (!bitmap_empty(ve.d_bad) || !bitmap_empty(ve.r_bad))
 502                 moveon = xfs_report_verify_errors(ctx, ve.d_bad, ve.r_bad);
 503
 504         bitmap_free(&ve.r_bad);
 505         bitmap_free(&ve.d_bad);
 506         ptvar_free(ve.rvstate);
 507         return moveon;
 508
 509 out_pool:
 510         read_verify_pool_destroy(ve.readverify);
 511 out_rbad:
 512         bitmap_free(&ve.r_bad);
 513 out_dbad:
 514         bitmap_free(&ve.d_bad);
 515 out_ve:
 516         ptvar_free(ve.rvstate);
 517         return moveon;
 518 }
 519
 520 /* Estimate how much work we're going to do. */
 521 bool
 522 xfs_estimate_verify_work(
 523         struct scrub_ctx        *ctx,
 524         uint64_t                *items,
 525         unsigned int            *nr_threads,
 526         int                     *rshift)
 527 {
 528         unsigned long long      d_blocks;
 529         unsigned long long      d_bfree;
 530         unsigned long long      r_blocks;
 531         unsigned long long      r_bfree;
 532         unsigned long long      f_files;
 533         unsigned long long      f_free;
 534         bool                    moveon;
 535
 536         moveon = xfs_scan_estimate_blocks(ctx, &d_blocks, &d_bfree,
 537                                 &r_blocks, &r_bfree, &f_files, &f_free);
 538         if (!moveon)
 539                 return moveon;
 540
 541         *items = ((d_blocks - d_bfree) + (r_blocks - r_bfree)) << ctx->blocklog;
 542         *nr_threads = disk_heads(ctx->datadev);
 543         *rshift = 20;
 544         return moveon;
 545 }