xfs: scrub directory metadata
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 18 Oct 2017 04:37:44 +0000 (21:37 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 26 Oct 2017 22:38:25 +0000 (15:38 -0700)
Scrub the hash tree and all the entries in a directory.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
fs/xfs/Makefile
fs/xfs/libxfs/xfs_dir2.c
fs/xfs/libxfs/xfs_dir2.h
fs/xfs/libxfs/xfs_fs.h
fs/xfs/scrub/common.c
fs/xfs/scrub/common.h
fs/xfs/scrub/dir.c [new file with mode: 0644]
fs/xfs/scrub/scrub.c
fs/xfs/scrub/scrub.h
fs/xfs/xfs_dir2_readdir.c
fs/xfs/xfs_file.c

index f4eca58..d6522e8 100644 (file)
@@ -149,6 +149,7 @@ xfs-y                               += $(addprefix scrub/, \
                                   btree.o \
                                   common.o \
                                   dabtree.o \
+                                  dir.o \
                                   ialloc.o \
                                   inode.o \
                                   refcount.o \
index ee5e916..41ea6d4 100644 (file)
@@ -39,7 +39,9 @@ struct xfs_name xfs_name_dotdot = { (unsigned char *)"..", 2, XFS_DIR3_FT_DIR };
 /*
  * Convert inode mode to directory entry filetype
  */
-unsigned char xfs_mode_to_ftype(int mode)
+unsigned char
+xfs_mode_to_ftype(
+       int             mode)
 {
        switch (mode & S_IFMT) {
        case S_IFREG:
index 21c8f8b..1a8f2cf 100644 (file)
@@ -324,4 +324,21 @@ xfs_dir2_leaf_tail_p(struct xfs_da_geometry *geo, struct xfs_dir2_leaf *lp)
                  sizeof(struct xfs_dir2_leaf_tail));
 }
 
+/*
+ * The Linux API doesn't pass down the total size of the buffer
+ * we read into down to the filesystem.  With the filldir concept
+ * it's not needed for correct information, but the XFS dir2 leaf
+ * code wants an estimate of the buffer size to calculate it's
+ * readahead window and size the buffers used for mapping to
+ * physical blocks.
+ *
+ * Try to give it an estimate that's good enough, maybe at some
+ * point we can change the ->readdir prototype to include the
+ * buffer size.  For now we use the current glibc buffer size.
+ * musl libc hardcodes 2k and dietlibc uses PAGE_SIZE.
+ */
+#define XFS_READDIR_BUFSIZE    (32768)
+
+unsigned char xfs_dir3_get_dtype(struct xfs_mount *mp, uint8_t filetype);
+
 #endif /* __XFS_DIR2_H__ */
index 02ae58b..b16d004 100644 (file)
@@ -498,9 +498,10 @@ struct xfs_scrub_metadata {
 #define XFS_SCRUB_TYPE_BMBTD   12      /* data fork block mapping */
 #define XFS_SCRUB_TYPE_BMBTA   13      /* attr fork block mapping */
 #define XFS_SCRUB_TYPE_BMBTC   14      /* CoW fork block mapping */
+#define XFS_SCRUB_TYPE_DIR     15      /* directory */
 
 /* Number of scrub subcommands. */
-#define XFS_SCRUB_TYPE_NR      15
+#define XFS_SCRUB_TYPE_NR      16
 
 /* i: Repair this metadata. */
 #define XFS_SCRUB_IFLAG_REPAIR         (1 << 0)
index 415c6a9..318dd97 100644 (file)
@@ -542,3 +542,31 @@ xfs_scrub_get_inode(
        sc->ip = ip;
        return 0;
 }
+
+/* Set us up to scrub a file's contents. */
+int
+xfs_scrub_setup_inode_contents(
+       struct xfs_scrub_context        *sc,
+       struct xfs_inode                *ip,
+       unsigned int                    resblks)
+{
+       struct xfs_mount                *mp = sc->mp;
+       int                             error;
+
+       error = xfs_scrub_get_inode(sc, ip);
+       if (error)
+               return error;
+
+       /* Got the inode, lock it and we're ready to go. */
+       sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
+       xfs_ilock(sc->ip, sc->ilock_flags);
+       error = xfs_scrub_trans_alloc(sc->sm, mp, &sc->tp);
+       if (error)
+               goto out;
+       sc->ilock_flags |= XFS_ILOCK_EXCL;
+       xfs_ilock(sc->ip, XFS_ILOCK_EXCL);
+
+out:
+       /* scrub teardown will unlock and release the inode for us */
+       return error;
+}
index b3cf4a2..7cd4a78 100644 (file)
@@ -93,6 +93,8 @@ int xfs_scrub_setup_inode_bmap(struct xfs_scrub_context *sc,
                               struct xfs_inode *ip);
 int xfs_scrub_setup_inode_bmap_data(struct xfs_scrub_context *sc,
                                    struct xfs_inode *ip);
+int xfs_scrub_setup_directory(struct xfs_scrub_context *sc,
+                             struct xfs_inode *ip);
 
 void xfs_scrub_ag_free(struct xfs_scrub_context *sc, struct xfs_scrub_ag *sa);
 int xfs_scrub_ag_init(struct xfs_scrub_context *sc, xfs_agnumber_t agno,
@@ -111,5 +113,7 @@ int xfs_scrub_walk_agfl(struct xfs_scrub_context *sc,
 int xfs_scrub_setup_ag_btree(struct xfs_scrub_context *sc,
                             struct xfs_inode *ip, bool force_log);
 int xfs_scrub_get_inode(struct xfs_scrub_context *sc, struct xfs_inode *ip_in);
+int xfs_scrub_setup_inode_contents(struct xfs_scrub_context *sc,
+                                  struct xfs_inode *ip, unsigned int resblks);
 
 #endif /* __XFS_SCRUB_COMMON_H__ */
diff --git a/fs/xfs/scrub/dir.c b/fs/xfs/scrub/dir.c
new file mode 100644 (file)
index 0000000..da0f4b1
--- /dev/null
@@ -0,0 +1,331 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_btree.h"
+#include "xfs_bit.h"
+#include "xfs_log_format.h"
+#include "xfs_trans.h"
+#include "xfs_sb.h"
+#include "xfs_inode.h"
+#include "xfs_icache.h"
+#include "xfs_itable.h"
+#include "xfs_da_format.h"
+#include "xfs_da_btree.h"
+#include "xfs_dir2.h"
+#include "xfs_dir2_priv.h"
+#include "xfs_ialloc.h"
+#include "scrub/xfs_scrub.h"
+#include "scrub/scrub.h"
+#include "scrub/common.h"
+#include "scrub/trace.h"
+#include "scrub/dabtree.h"
+
+/* Set us up to scrub directories. */
+int
+xfs_scrub_setup_directory(
+       struct xfs_scrub_context        *sc,
+       struct xfs_inode                *ip)
+{
+       return xfs_scrub_setup_inode_contents(sc, ip, 0);
+}
+
+/* Directories */
+
+/* Scrub a directory entry. */
+
+struct xfs_scrub_dir_ctx {
+       /* VFS fill-directory iterator */
+       struct dir_context              dir_iter;
+
+       struct xfs_scrub_context        *sc;
+};
+
+/* Check that an inode's mode matches a given DT_ type. */
+STATIC int
+xfs_scrub_dir_check_ftype(
+       struct xfs_scrub_dir_ctx        *sdc,
+       xfs_fileoff_t                   offset,
+       xfs_ino_t                       inum,
+       int                             dtype)
+{
+       struct xfs_mount                *mp = sdc->sc->mp;
+       struct xfs_inode                *ip;
+       int                             ino_dtype;
+       int                             error = 0;
+
+       if (!xfs_sb_version_hasftype(&mp->m_sb)) {
+               if (dtype != DT_UNKNOWN && dtype != DT_DIR)
+                       xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
+                                       offset);
+               goto out;
+       }
+
+       /*
+        * Grab the inode pointed to by the dirent.  We release the
+        * inode before we cancel the scrub transaction.  Since we're
+        * don't know a priori that releasing the inode won't trigger
+        * eofblocks cleanup (which allocates what would be a nested
+        * transaction), we can't use DONTCACHE here because DONTCACHE
+        * inodes can trigger immediate inactive cleanup of the inode.
+        */
+       error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
+       if (!xfs_scrub_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
+                       &error))
+               goto out;
+
+       /* Convert mode to the DT_* values that dir_emit uses. */
+       ino_dtype = xfs_dir3_get_dtype(mp,
+                       xfs_mode_to_ftype(VFS_I(ip)->i_mode));
+       if (ino_dtype != dtype)
+               xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
+       iput(VFS_I(ip));
+out:
+       return error;
+}
+
+/*
+ * Scrub a single directory entry.
+ *
+ * We use the VFS directory iterator (i.e. readdir) to call this
+ * function for every directory entry in a directory.  Once we're here,
+ * we check the inode number to make sure it's sane, then we check that
+ * we can look up this filename.  Finally, we check the ftype.
+ */
+STATIC int
+xfs_scrub_dir_actor(
+       struct dir_context              *dir_iter,
+       const char                      *name,
+       int                             namelen,
+       loff_t                          pos,
+       u64                             ino,
+       unsigned                        type)
+{
+       struct xfs_mount                *mp;
+       struct xfs_inode                *ip;
+       struct xfs_scrub_dir_ctx        *sdc;
+       struct xfs_name                 xname;
+       xfs_ino_t                       lookup_ino;
+       xfs_dablk_t                     offset;
+       int                             error = 0;
+
+       sdc = container_of(dir_iter, struct xfs_scrub_dir_ctx, dir_iter);
+       ip = sdc->sc->ip;
+       mp = ip->i_mount;
+       offset = xfs_dir2_db_to_da(mp->m_dir_geo,
+                       xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
+
+       /* Does this inode number make sense? */
+       if (!xfs_verify_dir_ino(mp, ino)) {
+               xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
+               goto out;
+       }
+
+       if (!strncmp(".", name, namelen)) {
+               /* If this is "." then check that the inum matches the dir. */
+               if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
+                       xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
+                                       offset);
+               if (ino != ip->i_ino)
+                       xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
+                                       offset);
+       } else if (!strncmp("..", name, namelen)) {
+               /*
+                * If this is ".." in the root inode, check that the inum
+                * matches this dir.
+                */
+               if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
+                       xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
+                                       offset);
+               if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
+                       xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
+                                       offset);
+       }
+
+       /* Verify that we can look up this name by hash. */
+       xname.name = name;
+       xname.len = namelen;
+       xname.type = XFS_DIR3_FT_UNKNOWN;
+
+       error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
+       if (!xfs_scrub_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
+                       &error))
+               goto fail_xref;
+       if (lookup_ino != ino) {
+               xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
+               goto out;
+       }
+
+       /* Verify the file type.  This function absorbs error codes. */
+       error = xfs_scrub_dir_check_ftype(sdc, offset, lookup_ino, type);
+       if (error)
+               goto out;
+out:
+       return error;
+fail_xref:
+       return error;
+}
+
+/* Scrub a directory btree record. */
+STATIC int
+xfs_scrub_dir_rec(
+       struct xfs_scrub_da_btree       *ds,
+       int                             level,
+       void                            *rec)
+{
+       struct xfs_mount                *mp = ds->state->mp;
+       struct xfs_dir2_leaf_entry      *ent = rec;
+       struct xfs_inode                *dp = ds->dargs.dp;
+       struct xfs_dir2_data_entry      *dent;
+       struct xfs_buf                  *bp;
+       xfs_ino_t                       ino;
+       xfs_dablk_t                     rec_bno;
+       xfs_dir2_db_t                   db;
+       xfs_dir2_data_aoff_t            off;
+       xfs_dir2_dataptr_t              ptr;
+       xfs_dahash_t                    calc_hash;
+       xfs_dahash_t                    hash;
+       unsigned int                    tag;
+       int                             error;
+
+       /* Check the hash of the entry. */
+       error = xfs_scrub_da_btree_hash(ds, level, &ent->hashval);
+       if (error)
+               goto out;
+
+       /* Valid hash pointer? */
+       ptr = be32_to_cpu(ent->address);
+       if (ptr == 0)
+               return 0;
+
+       /* Find the directory entry's location. */
+       db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
+       off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
+       rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
+
+       if (rec_bno >= mp->m_dir_geo->leafblk) {
+               xfs_scrub_da_set_corrupt(ds, level);
+               goto out;
+       }
+       error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
+       if (!xfs_scrub_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
+                       &error))
+               goto out;
+       if (!bp) {
+               xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
+               goto out;
+       }
+
+       /* Retrieve the entry, sanity check it, and compare hashes. */
+       dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
+       ino = be64_to_cpu(dent->inumber);
+       hash = be32_to_cpu(ent->hashval);
+       tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
+       if (!xfs_verify_dir_ino(mp, ino) || tag != off)
+               xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
+       if (dent->namelen == 0) {
+               xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
+               goto out_relse;
+       }
+       calc_hash = xfs_da_hashname(dent->name, dent->namelen);
+       if (calc_hash != hash)
+               xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
+
+out_relse:
+       xfs_trans_brelse(ds->dargs.trans, bp);
+out:
+       return error;
+}
+
+/* Scrub a whole directory. */
+int
+xfs_scrub_directory(
+       struct xfs_scrub_context        *sc)
+{
+       struct xfs_scrub_dir_ctx        sdc = {
+               .dir_iter.actor = xfs_scrub_dir_actor,
+               .dir_iter.pos = 0,
+               .sc = sc,
+       };
+       size_t                          bufsize;
+       loff_t                          oldpos;
+       int                             error;
+
+       if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
+               return -ENOENT;
+
+       /* Plausible size? */
+       if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
+               xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino, NULL);
+               goto out;
+       }
+
+       /* Check directory tree structure */
+       error = xfs_scrub_da_btree(sc, XFS_DATA_FORK, xfs_scrub_dir_rec);
+       if (error)
+               return error;
+
+       if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
+               return error;
+
+       /*
+        * Check that every dirent we see can also be looked up by hash.
+        * Userspace usually asks for a 32k buffer, so we will too.
+        */
+       bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
+                       sc->ip->i_d.di_size);
+
+       /*
+        * Look up every name in this directory by hash.
+        *
+        * Use the xfs_readdir function to call xfs_scrub_dir_actor on
+        * every directory entry in this directory.  In _actor, we check
+        * the name, inode number, and ftype (if applicable) of the
+        * entry.  xfs_readdir uses the VFS filldir functions to provide
+        * iteration context.
+        *
+        * The VFS grabs a read or write lock via i_rwsem before it reads
+        * or writes to a directory.  If we've gotten this far we've
+        * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
+        * getting a write lock on i_rwsem.  Therefore, it is safe for us
+        * to drop the ILOCK here in order to reuse the _readdir and
+        * _dir_lookup routines, which do their own ILOCK locking.
+        */
+       oldpos = 0;
+       sc->ilock_flags &= ~XFS_ILOCK_EXCL;
+       xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
+       while (true) {
+               error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
+               if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0,
+                               &error))
+                       goto out;
+               if (oldpos == sdc.dir_iter.pos)
+                       break;
+               oldpos = sdc.dir_iter.pos;
+       }
+
+out:
+       return error;
+}
index cb66919..68daedf 100644 (file)
@@ -227,6 +227,10 @@ static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
                .setup  = xfs_scrub_setup_inode_bmap,
                .scrub  = xfs_scrub_bmap_cow,
        },
+       { /* directory */
+               .setup  = xfs_scrub_setup_directory,
+               .scrub  = xfs_scrub_directory,
+       },
 };
 
 /* This isn't a stable feature, warn once per day. */
index 8920ccf..844506e 100644 (file)
@@ -82,5 +82,6 @@ int xfs_scrub_inode(struct xfs_scrub_context *sc);
 int xfs_scrub_bmap_data(struct xfs_scrub_context *sc);
 int xfs_scrub_bmap_attr(struct xfs_scrub_context *sc);
 int xfs_scrub_bmap_cow(struct xfs_scrub_context *sc);
+int xfs_scrub_directory(struct xfs_scrub_context *sc);
 
 #endif /* __XFS_SCRUB_SCRUB_H__ */
index ba2638d..238e365 100644 (file)
@@ -41,7 +41,7 @@ static unsigned char xfs_dir3_filetype_table[] = {
        DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
 };
 
-static unsigned char
+unsigned char
 xfs_dir3_get_dtype(
        struct xfs_mount        *mp,
        uint8_t                 filetype)
index 6526ef0..1814687 100644 (file)
@@ -984,7 +984,7 @@ xfs_file_readdir(
         * point we can change the ->readdir prototype to include the
         * buffer size.  For now we use the current glibc buffer size.
         */
-       bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
+       bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
 
        return xfs_readdir(NULL, ip, ctx, bufsize);
 }