fs/xfs/scrub/findparent.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (c) 2020-2024 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <djwong@kernel.org>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_trans_resv.h"
  11 #include "xfs_mount.h"
  12 #include "xfs_defer.h"
  13 #include "xfs_bit.h"
  14 #include "xfs_log_format.h"
  15 #include "xfs_trans.h"
  16 #include "xfs_sb.h"
  17 #include "xfs_inode.h"
  18 #include "xfs_icache.h"
  19 #include "xfs_da_format.h"
  20 #include "xfs_da_btree.h"
  21 #include "xfs_dir2.h"
  22 #include "xfs_bmap_btree.h"
  23 #include "xfs_dir2_priv.h"
  24 #include "xfs_trans_space.h"
  25 #include "xfs_health.h"
  26 #include "xfs_exchmaps.h"
  27 #include "xfs_parent.h"
  28 #include "scrub/xfs_scrub.h"
  29 #include "scrub/scrub.h"
  30 #include "scrub/common.h"
  31 #include "scrub/trace.h"
  32 #include "scrub/repair.h"
  33 #include "scrub/iscan.h"
  34 #include "scrub/findparent.h"
  35 #include "scrub/readdir.h"
  36 #include "scrub/tempfile.h"
  37 #include "scrub/listxattr.h"
  38
  39 /*
  40  * Finding the Parent of a Directory
  41  * =================================
  42  *
  43  * Directories have parent pointers, in the sense that each directory contains
  44  * a dotdot entry that points to the single allowed parent.  The brute force
  45  * way to find the parent of a given directory is to scan every directory in
  46  * the filesystem looking for a child dirent that references this directory.
  47  *
  48  * This module wraps the process of scanning the directory tree.  It requires
  49  * that @sc->ip is the directory whose parent we want to find, and that the
  50  * caller hold only the IOLOCK on that directory.  The scan itself needs to
  51  * take the ILOCK of each directory visited.
  52  *
  53  * Because we cannot hold @sc->ip's ILOCK during a scan of the whole fs, it is
  54  * necessary to use dirent hook to update the parent scan results.  Callers
  55  * must not read the scan results without re-taking @sc->ip's ILOCK.
  56  *
  57  * There are a few shortcuts that we can take to avoid scanning the entire
  58  * filesystem, such as noticing directory tree roots and querying the dentry
  59  * cache for parent information.
  60  */
  61
  62 struct xrep_findparent_info {
  63         /* The directory currently being scanned. */
  64         struct xfs_inode        *dp;
  65
  66         /*
  67          * Scrub context.  We're looking for a @dp containing a directory
  68          * entry pointing to sc->ip->i_ino.
  69          */
  70         struct xfs_scrub        *sc;
  71
  72         /* Optional scan information for a xrep_findparent_scan call. */
  73         struct xrep_parent_scan_info *parent_scan;
  74
  75         /*
  76          * Parent that we've found for sc->ip.  If we're scanning the entire
  77          * directory tree, we need this to ensure that we only find /one/
  78          * parent directory.
  79          */
  80         xfs_ino_t               found_parent;
  81
  82         /*
  83          * This is set to true if @found_parent was not observed directly from
  84          * the directory scan but by noticing a change in dotdot entries after
  85          * cycling the sc->ip IOLOCK.
  86          */
  87         bool                    parent_tentative;
  88 };
  89
  90 /*
  91  * If this directory entry points to the scrub target inode, then the directory
  92  * we're scanning is the parent of the scrub target inode.
  93  */
  94 STATIC int
  95 xrep_findparent_dirent(
  96         struct xfs_scrub                *sc,
  97         struct xfs_inode                *dp,
  98         xfs_dir2_dataptr_t              dapos,
  99         const struct xfs_name           *name,
 100         xfs_ino_t                       ino,
 101         void                            *priv)
 102 {
 103         struct xrep_findparent_info     *fpi = priv;
 104         int                             error = 0;
 105
 106         if (xchk_should_terminate(fpi->sc, &error))
 107                 return error;
 108
 109         if (ino != fpi->sc->ip->i_ino)
 110                 return 0;
 111
 112         /* Ignore garbage directory entry names. */
 113         if (name->len == 0 || !xfs_dir2_namecheck(name->name, name->len))
 114                 return -EFSCORRUPTED;
 115
 116         /*
 117          * Ignore dotdot and dot entries -- we're looking for parent -> child
 118          * links only.
 119          */
 120         if (name->name[0] == '.' && (name->len == 1 ||
 121                                      (name->len == 2 && name->name[1] == '.')))
 122                 return 0;
 123
 124         /* Uhoh, more than one parent for a dir? */
 125         if (fpi->found_parent != NULLFSINO &&
 126             !(fpi->parent_tentative && fpi->found_parent == fpi->dp->i_ino)) {
 127                 trace_xrep_findparent_dirent(fpi->sc->ip, 0);
 128                 return -EFSCORRUPTED;
 129         }
 130
 131         /* We found a potential parent; remember this. */
 132         trace_xrep_findparent_dirent(fpi->sc->ip, fpi->dp->i_ino);
 133         fpi->found_parent = fpi->dp->i_ino;
 134         fpi->parent_tentative = false;
 135
 136         if (fpi->parent_scan)
 137                 xrep_findparent_scan_found(fpi->parent_scan, fpi->dp->i_ino);
 138
 139         return 0;
 140 }
 141
 142 /*
 143  * If this is a directory, walk the dirents looking for any that point to the
 144  * scrub target inode.
 145  */
 146 STATIC int
 147 xrep_findparent_walk_directory(
 148         struct xrep_findparent_info     *fpi)
 149 {
 150         struct xfs_scrub                *sc = fpi->sc;
 151         struct xfs_inode                *dp = fpi->dp;
 152         unsigned int                    lock_mode;
 153         int                             error = 0;
 154
 155         /*
 156          * The inode being scanned cannot be its own parent, nor can any
 157          * temporary directory we created to stage this repair.
 158          */
 159         if (dp == sc->ip || dp == sc->tempip)
 160                 return 0;
 161
 162         /*
 163          * Similarly, temporary files created to stage a repair cannot be the
 164          * parent of this inode.
 165          */
 166         if (xrep_is_tempfile(dp))
 167                 return 0;
 168
 169         /*
 170          * Scan the directory to see if there it contains an entry pointing to
 171          * the directory that we are repairing.
 172          */
 173         lock_mode = xfs_ilock_data_map_shared(dp);
 174
 175         /*
 176          * If this directory is known to be sick, we cannot scan it reliably
 177          * and must abort.
 178          */
 179         if (xfs_inode_has_sickness(dp, XFS_SICK_INO_CORE |
 180                                        XFS_SICK_INO_BMBTD |
 181                                        XFS_SICK_INO_DIR)) {
 182                 error = -EFSCORRUPTED;
 183                 goto out_unlock;
 184         }
 185
 186         /*
 187          * We cannot complete our parent pointer scan if a directory looks as
 188          * though it has been zapped by the inode record repair code.
 189          */
 190         if (xchk_dir_looks_zapped(dp)) {
 191                 error = -EBUSY;
 192                 goto out_unlock;
 193         }
 194
 195         error = xchk_dir_walk(sc, dp, xrep_findparent_dirent, fpi);
 196         if (error)
 197                 goto out_unlock;
 198
 199 out_unlock:
 200         xfs_iunlock(dp, lock_mode);
 201         return error;
 202 }
 203
 204 /*
 205  * Update this directory's dotdot pointer based on ongoing dirent updates.
 206  */
 207 STATIC int
 208 xrep_findparent_live_update(
 209         struct notifier_block           *nb,
 210         unsigned long                   action,
 211         void                            *data)
 212 {
 213         struct xfs_dir_update_params    *p = data;
 214         struct xrep_parent_scan_info    *pscan;
 215         struct xfs_scrub                *sc;
 216
 217         pscan = container_of(nb, struct xrep_parent_scan_info,
 218                         dhook.dirent_hook.nb);
 219         sc = pscan->sc;
 220
 221         /*
 222          * If @p->ip is the subdirectory that we're interested in and we've
 223          * already scanned @p->dp, update the dotdot target inumber to the
 224          * parent inode.
 225          */
 226         if (p->ip->i_ino == sc->ip->i_ino &&
 227             xchk_iscan_want_live_update(&pscan->iscan, p->dp->i_ino)) {
 228                 if (p->delta > 0) {
 229                         xrep_findparent_scan_found(pscan, p->dp->i_ino);
 230                 } else {
 231                         xrep_findparent_scan_found(pscan, NULLFSINO);
 232                 }
 233         }
 234
 235         return NOTIFY_DONE;
 236 }
 237
 238 /*
 239  * Set up a scan to find the parent of a directory.  The provided dirent hook
 240  * will be called when there is a dotdot update for the inode being repaired.
 241  */
 242 int
 243 __xrep_findparent_scan_start(
 244         struct xfs_scrub                *sc,
 245         struct xrep_parent_scan_info    *pscan,
 246         notifier_fn_t                   custom_fn)
 247 {
 248         int                             error;
 249
 250         if (!(sc->flags & XCHK_FSGATES_DIRENTS)) {
 251                 ASSERT(sc->flags & XCHK_FSGATES_DIRENTS);
 252                 return -EINVAL;
 253         }
 254
 255         pscan->sc = sc;
 256         pscan->parent_ino = NULLFSINO;
 257
 258         mutex_init(&pscan->lock);
 259
 260         xchk_iscan_start(sc, 30000, 100, &pscan->iscan);
 261
 262         /*
 263          * Hook into the dirent update code.  The hook only operates on inodes
 264          * that were already scanned, and the scanner thread takes each inode's
 265          * ILOCK, which means that any in-progress inode updates will finish
 266          * before we can scan the inode.
 267          */
 268         if (custom_fn)
 269                 xfs_dir_hook_setup(&pscan->dhook, custom_fn);
 270         else
 271                 xfs_dir_hook_setup(&pscan->dhook, xrep_findparent_live_update);
 272         error = xfs_dir_hook_add(sc->mp, &pscan->dhook);
 273         if (error)
 274                 goto out_iscan;
 275
 276         return 0;
 277 out_iscan:
 278         xchk_iscan_teardown(&pscan->iscan);
 279         mutex_destroy(&pscan->lock);
 280         return error;
 281 }
 282
 283 /*
 284  * Scan the entire filesystem looking for a parent inode for the inode being
 285  * scrubbed.  @sc->ip must not be the root of a directory tree.  Callers must
 286  * not hold a dirty transaction or any lock that would interfere with taking
 287  * an ILOCK.
 288  *
 289  * Returns 0 with @pscan->parent_ino set to the parent that we found.
 290  * Returns 0 with @pscan->parent_ino set to NULLFSINO if we found no parents.
 291  * Returns the usual negative errno if something else happened.
 292  */
 293 int
 294 xrep_findparent_scan(
 295         struct xrep_parent_scan_info    *pscan)
 296 {
 297         struct xrep_findparent_info     fpi = {
 298                 .sc                     = pscan->sc,
 299                 .found_parent           = NULLFSINO,
 300                 .parent_scan            = pscan,
 301         };
 302         struct xfs_scrub                *sc = pscan->sc;
 303         int                             ret;
 304
 305         ASSERT(S_ISDIR(VFS_IC(sc->ip)->i_mode));
 306
 307         while ((ret = xchk_iscan_iter(&pscan->iscan, &fpi.dp)) == 1) {
 308                 if (S_ISDIR(VFS_I(fpi.dp)->i_mode))
 309                         ret = xrep_findparent_walk_directory(&fpi);
 310                 else
 311                         ret = 0;
 312                 xchk_iscan_mark_visited(&pscan->iscan, fpi.dp);
 313                 xchk_irele(sc, fpi.dp);
 314                 if (ret)
 315                         break;
 316
 317                 if (xchk_should_terminate(sc, &ret))
 318                         break;
 319         }
 320         xchk_iscan_iter_finish(&pscan->iscan);
 321
 322         return ret;
 323 }
 324
 325 /* Tear down a parent scan. */
 326 void
 327 xrep_findparent_scan_teardown(
 328         struct xrep_parent_scan_info    *pscan)
 329 {
 330         xfs_dir_hook_del(pscan->sc->mp, &pscan->dhook);
 331         xchk_iscan_teardown(&pscan->iscan);
 332         mutex_destroy(&pscan->lock);
 333 }
 334
 335 /* Finish a parent scan early. */
 336 void
 337 xrep_findparent_scan_finish_early(
 338         struct xrep_parent_scan_info    *pscan,
 339         xfs_ino_t                       ino)
 340 {
 341         xrep_findparent_scan_found(pscan, ino);
 342         xchk_iscan_finish_early(&pscan->iscan);
 343 }
 344
 345 /*
 346  * Confirm that the directory @parent_ino actually contains a directory entry
 347  * pointing to the child @sc->ip->ino.  This function returns one of several
 348  * ways:
 349  *
 350  * Returns 0 with @parent_ino unchanged if the parent was confirmed.
 351  * Returns 0 with @parent_ino set to NULLFSINO if the parent was not valid.
 352  * Returns the usual negative errno if something else happened.
 353  */
 354 int
 355 xrep_findparent_confirm(
 356         struct xfs_scrub        *sc,
 357         xfs_ino_t               *parent_ino)
 358 {
 359         struct xrep_findparent_info fpi = {
 360                 .sc             = sc,
 361                 .found_parent   = NULLFSINO,
 362         };
 363         int                     error;
 364
 365         /*
 366          * The root directory always points to itself.  Unlinked dirs can point
 367          * anywhere, so we point them at the root dir too.
 368          */
 369         if (sc->ip == sc->mp->m_rootip || VFS_I(sc->ip)->i_nlink == 0) {
 370                 *parent_ino = sc->mp->m_sb.sb_rootino;
 371                 return 0;
 372         }
 373
 374         /* Reject garbage parent inode numbers and self-referential parents. */
 375         if (*parent_ino == NULLFSINO)
 376                return 0;
 377         if (!xfs_verify_dir_ino(sc->mp, *parent_ino) ||
 378             *parent_ino == sc->ip->i_ino) {
 379                 *parent_ino = NULLFSINO;
 380                 return 0;
 381         }
 382
 383         error = xchk_iget(sc, *parent_ino, &fpi.dp);
 384         if (error)
 385                 return error;
 386
 387         if (!S_ISDIR(VFS_I(fpi.dp)->i_mode)) {
 388                 *parent_ino = NULLFSINO;
 389                 goto out_rele;
 390         }
 391
 392         error = xrep_findparent_walk_directory(&fpi);
 393         if (error)
 394                 goto out_rele;
 395
 396         *parent_ino = fpi.found_parent;
 397 out_rele:
 398         xchk_irele(sc, fpi.dp);
 399         return error;
 400 }
 401
 402 /*
 403  * If we're the root of a directory tree, we are our own parent.  If we're an
 404  * unlinked directory, the parent /won't/ have a link to us.  Set the parent
 405  * directory to the root for both cases.  Returns NULLFSINO if we don't know
 406  * what to do.
 407  */
 408 xfs_ino_t
 409 xrep_findparent_self_reference(
 410         struct xfs_scrub        *sc)
 411 {
 412         if (sc->ip->i_ino == sc->mp->m_sb.sb_rootino)
 413                 return sc->mp->m_sb.sb_rootino;
 414
 415         if (VFS_I(sc->ip)->i_nlink == 0)
 416                 return sc->mp->m_sb.sb_rootino;
 417
 418         return NULLFSINO;
 419 }
 420
 421 /* Check the dentry cache to see if knows of a parent for the scrub target. */
 422 xfs_ino_t
 423 xrep_findparent_from_dcache(
 424         struct xfs_scrub        *sc)
 425 {
 426         struct inode            *pip = NULL;
 427         struct dentry           *dentry, *parent;
 428         xfs_ino_t               ret = NULLFSINO;
 429
 430         dentry = d_find_alias(VFS_I(sc->ip));
 431         if (!dentry)
 432                 goto out;
 433
 434         parent = dget_parent(dentry);
 435         if (!parent)
 436                 goto out_dput;
 437
 438         ASSERT(parent->d_sb == sc->ip->i_mount->m_super);
 439
 440         pip = igrab(d_inode(parent));
 441         dput(parent);
 442
 443         if (S_ISDIR(pip->i_mode)) {
 444                 trace_xrep_findparent_from_dcache(sc->ip, XFS_I(pip)->i_ino);
 445                 ret = XFS_I(pip)->i_ino;
 446         }
 447
 448         xchk_irele(sc, XFS_I(pip));
 449
 450 out_dput:
 451         dput(dentry);
 452 out:
 453         return ret;
 454 }