xfs: scrub inode block mappings
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 18 Oct 2017 04:37:43 +0000 (21:37 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 26 Oct 2017 22:38:25 +0000 (15:38 -0700)
Scrub an individual inode's block mappings to make sure they make sense.

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

index 1186e1b..df86fc5 100644 (file)
@@ -145,6 +145,7 @@ xfs-y                               += $(addprefix scrub/, \
                                   trace.o \
                                   agheader.o \
                                   alloc.o \
+                                  bmap.o \
                                   btree.o \
                                   common.o \
                                   ialloc.o \
index f8463e0..02ae58b 100644 (file)
@@ -495,9 +495,12 @@ struct xfs_scrub_metadata {
 #define XFS_SCRUB_TYPE_RMAPBT  9       /* reverse mapping btree */
 #define XFS_SCRUB_TYPE_REFCNTBT        10      /* reference count btree */
 #define XFS_SCRUB_TYPE_INODE   11      /* inode record */
+#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 */
 
 /* Number of scrub subcommands. */
-#define XFS_SCRUB_TYPE_NR      12
+#define XFS_SCRUB_TYPE_NR      15
 
 /* i: Repair this metadata. */
 #define XFS_SCRUB_IFLAG_REPAIR         (1 << 0)
diff --git a/fs/xfs/scrub/bmap.c b/fs/xfs/scrub/bmap.c
new file mode 100644 (file)
index 0000000..3c17b18
--- /dev/null
@@ -0,0 +1,366 @@
+/*
+ * 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_inode_fork.h"
+#include "xfs_alloc.h"
+#include "xfs_rtalloc.h"
+#include "xfs_bmap.h"
+#include "xfs_bmap_util.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_rmap.h"
+#include "scrub/xfs_scrub.h"
+#include "scrub/scrub.h"
+#include "scrub/common.h"
+#include "scrub/btree.h"
+#include "scrub/trace.h"
+
+/* Set us up with an inode's bmap. */
+int
+xfs_scrub_setup_inode_bmap(
+       struct xfs_scrub_context        *sc,
+       struct xfs_inode                *ip)
+{
+       struct xfs_mount                *mp = sc->mp;
+       int                             error;
+
+       error = xfs_scrub_get_inode(sc, ip);
+       if (error)
+               goto out;
+
+       sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
+       xfs_ilock(sc->ip, sc->ilock_flags);
+
+       /*
+        * We don't want any ephemeral data fork updates sitting around
+        * while we inspect block mappings, so wait for directio to finish
+        * and flush dirty data if we have delalloc reservations.
+        */
+       if (S_ISREG(VFS_I(sc->ip)->i_mode) &&
+           sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) {
+               inode_dio_wait(VFS_I(sc->ip));
+               error = filemap_write_and_wait(VFS_I(sc->ip)->i_mapping);
+               if (error)
+                       goto out;
+       }
+
+       /* Got the inode, lock it and we're ready to go. */
+       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 */
+       return error;
+}
+
+/*
+ * Inode fork block mapping (BMBT) scrubber.
+ * More complex than the others because we have to scrub
+ * all the extents regardless of whether or not the fork
+ * is in btree format.
+ */
+
+struct xfs_scrub_bmap_info {
+       struct xfs_scrub_context        *sc;
+       xfs_fileoff_t                   lastoff;
+       bool                            is_rt;
+       bool                            is_shared;
+       int                             whichfork;
+};
+
+/* Scrub a single extent record. */
+STATIC int
+xfs_scrub_bmap_extent(
+       struct xfs_inode                *ip,
+       struct xfs_btree_cur            *cur,
+       struct xfs_scrub_bmap_info      *info,
+       struct xfs_bmbt_irec            *irec)
+{
+       struct xfs_mount                *mp = info->sc->mp;
+       struct xfs_buf                  *bp = NULL;
+       int                             error = 0;
+
+       if (cur)
+               xfs_btree_get_block(cur, 0, &bp);
+
+       /*
+        * Check for out-of-order extents.  This record could have come
+        * from the incore list, for which there is no ordering check.
+        */
+       if (irec->br_startoff < info->lastoff)
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+
+       /* There should never be a "hole" extent in either extent list. */
+       if (irec->br_startblock == HOLESTARTBLOCK)
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+
+       /*
+        * Check for delalloc extents.  We never iterate the ones in the
+        * in-core extent scan, and we should never see these in the bmbt.
+        */
+       if (isnullstartblock(irec->br_startblock))
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+
+       /* Make sure the extent points to a valid place. */
+       if (irec->br_startblock + irec->br_blockcount <= irec->br_startblock)
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+       if (info->is_rt &&
+           (!xfs_verify_rtbno(mp, irec->br_startblock) ||
+            !xfs_verify_rtbno(mp, irec->br_startblock +
+                               irec->br_blockcount - 1)))
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+       if (!info->is_rt &&
+           (!xfs_verify_fsbno(mp, irec->br_startblock) ||
+            !xfs_verify_fsbno(mp, irec->br_startblock +
+                               irec->br_blockcount - 1)))
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+
+       /* We don't allow unwritten extents on attr forks. */
+       if (irec->br_state == XFS_EXT_UNWRITTEN &&
+           info->whichfork == XFS_ATTR_FORK)
+               xfs_scrub_fblock_set_corrupt(info->sc, info->whichfork,
+                               irec->br_startoff);
+
+       info->lastoff = irec->br_startoff + irec->br_blockcount;
+       return error;
+}
+
+/* Scrub a bmbt record. */
+STATIC int
+xfs_scrub_bmapbt_rec(
+       struct xfs_scrub_btree          *bs,
+       union xfs_btree_rec             *rec)
+{
+       struct xfs_bmbt_rec_host        ihost;
+       struct xfs_bmbt_irec            irec;
+       struct xfs_scrub_bmap_info      *info = bs->private;
+       struct xfs_inode                *ip = bs->cur->bc_private.b.ip;
+       struct xfs_buf                  *bp = NULL;
+       struct xfs_btree_block          *block;
+       uint64_t                        owner;
+       int                             i;
+
+       /*
+        * Check the owners of the btree blocks up to the level below
+        * the root since the verifiers don't do that.
+        */
+       if (xfs_sb_version_hascrc(&bs->cur->bc_mp->m_sb) &&
+           bs->cur->bc_ptrs[0] == 1) {
+               for (i = 0; i < bs->cur->bc_nlevels - 1; i++) {
+                       block = xfs_btree_get_block(bs->cur, i, &bp);
+                       owner = be64_to_cpu(block->bb_u.l.bb_owner);
+                       if (owner != ip->i_ino)
+                               xfs_scrub_fblock_set_corrupt(bs->sc,
+                                               info->whichfork, 0);
+               }
+       }
+
+       /* Set up the in-core record and scrub it. */
+       ihost.l0 = be64_to_cpu(rec->bmbt.l0);
+       ihost.l1 = be64_to_cpu(rec->bmbt.l1);
+       xfs_bmbt_get_all(&ihost, &irec);
+       return xfs_scrub_bmap_extent(ip, bs->cur, info, &irec);
+}
+
+/* Scan the btree records. */
+STATIC int
+xfs_scrub_bmap_btree(
+       struct xfs_scrub_context        *sc,
+       int                             whichfork,
+       struct xfs_scrub_bmap_info      *info)
+{
+       struct xfs_owner_info           oinfo;
+       struct xfs_mount                *mp = sc->mp;
+       struct xfs_inode                *ip = sc->ip;
+       struct xfs_btree_cur            *cur;
+       int                             error;
+
+       cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
+       xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
+       error = xfs_scrub_btree(sc, cur, xfs_scrub_bmapbt_rec, &oinfo, info);
+       xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR :
+                                         XFS_BTREE_NOERROR);
+       return error;
+}
+
+/*
+ * Scrub an inode fork's block mappings.
+ *
+ * First we scan every record in every btree block, if applicable.
+ * Then we unconditionally scan the incore extent cache.
+ */
+STATIC int
+xfs_scrub_bmap(
+       struct xfs_scrub_context        *sc,
+       int                             whichfork)
+{
+       struct xfs_bmbt_irec            irec;
+       struct xfs_scrub_bmap_info      info = {0};
+       struct xfs_mount                *mp = sc->mp;
+       struct xfs_inode                *ip = sc->ip;
+       struct xfs_ifork                *ifp;
+       xfs_fileoff_t                   endoff;
+       xfs_extnum_t                    idx;
+       bool                            found;
+       int                             error = 0;
+
+       ifp = XFS_IFORK_PTR(ip, whichfork);
+
+       info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip);
+       info.whichfork = whichfork;
+       info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip);
+       info.sc = sc;
+
+       switch (whichfork) {
+       case XFS_COW_FORK:
+               /* Non-existent CoW forks are ignorable. */
+               if (!ifp)
+                       goto out;
+               /* No CoW forks on non-reflink inodes/filesystems. */
+               if (!xfs_is_reflink_inode(ip)) {
+                       xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino, NULL);
+                       goto out;
+               }
+               break;
+       case XFS_ATTR_FORK:
+               if (!ifp)
+                       goto out;
+               if (!xfs_sb_version_hasattr(&mp->m_sb) &&
+                   !xfs_sb_version_hasattr2(&mp->m_sb))
+                       xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino, NULL);
+               break;
+       default:
+               ASSERT(whichfork == XFS_DATA_FORK);
+               break;
+       }
+
+       /* Check the fork values */
+       switch (XFS_IFORK_FORMAT(ip, whichfork)) {
+       case XFS_DINODE_FMT_UUID:
+       case XFS_DINODE_FMT_DEV:
+       case XFS_DINODE_FMT_LOCAL:
+               /* No mappings to check. */
+               goto out;
+       case XFS_DINODE_FMT_EXTENTS:
+               if (!(ifp->if_flags & XFS_IFEXTENTS)) {
+                       xfs_scrub_fblock_set_corrupt(sc, whichfork, 0);
+                       goto out;
+               }
+               break;
+       case XFS_DINODE_FMT_BTREE:
+               if (whichfork == XFS_COW_FORK) {
+                       xfs_scrub_fblock_set_corrupt(sc, whichfork, 0);
+                       goto out;
+               }
+
+               error = xfs_scrub_bmap_btree(sc, whichfork, &info);
+               if (error)
+                       goto out;
+               break;
+       default:
+               xfs_scrub_fblock_set_corrupt(sc, whichfork, 0);
+               goto out;
+       }
+
+       if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
+               goto out;
+
+       /* Now try to scrub the in-memory extent list. */
+        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
+               error = xfs_iread_extents(sc->tp, ip, whichfork);
+               if (!xfs_scrub_fblock_process_error(sc, whichfork, 0, &error))
+                       goto out;
+       }
+
+       /* Find the offset of the last extent in the mapping. */
+       error = xfs_bmap_last_offset(ip, &endoff, whichfork);
+       if (!xfs_scrub_fblock_process_error(sc, whichfork, 0, &error))
+               goto out;
+
+       /* Scrub extent records. */
+       info.lastoff = 0;
+       ifp = XFS_IFORK_PTR(ip, whichfork);
+       for (found = xfs_iext_lookup_extent(ip, ifp, 0, &idx, &irec);
+            found != 0;
+            found = xfs_iext_get_extent(ifp, ++idx, &irec)) {
+               if (xfs_scrub_should_terminate(sc, &error))
+                       break;
+               if (isnullstartblock(irec.br_startblock))
+                       continue;
+               if (irec.br_startoff >= endoff) {
+                       xfs_scrub_fblock_set_corrupt(sc, whichfork,
+                                       irec.br_startoff);
+                       goto out;
+               }
+               error = xfs_scrub_bmap_extent(ip, NULL, &info, &irec);
+               if (error)
+                       goto out;
+       }
+
+out:
+       return error;
+}
+
+/* Scrub an inode's data fork. */
+int
+xfs_scrub_bmap_data(
+       struct xfs_scrub_context        *sc)
+{
+       return xfs_scrub_bmap(sc, XFS_DATA_FORK);
+}
+
+/* Scrub an inode's attr fork. */
+int
+xfs_scrub_bmap_attr(
+       struct xfs_scrub_context        *sc)
+{
+       return xfs_scrub_bmap(sc, XFS_ATTR_FORK);
+}
+
+/* Scrub an inode's CoW fork. */
+int
+xfs_scrub_bmap_cow(
+       struct xfs_scrub_context        *sc)
+{
+       if (!xfs_is_reflink_inode(sc->ip))
+               return -ENOENT;
+
+       return xfs_scrub_bmap(sc, XFS_COW_FORK);
+}
index fcec11e..b3cf4a2 100644 (file)
@@ -89,7 +89,10 @@ int xfs_scrub_setup_ag_refcountbt(struct xfs_scrub_context *sc,
                                  struct xfs_inode *ip);
 int xfs_scrub_setup_inode(struct xfs_scrub_context *sc,
                          struct xfs_inode *ip);
-
+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);
 
 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,
index 45cc046..cb66919 100644 (file)
@@ -215,6 +215,18 @@ static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
                .setup  = xfs_scrub_setup_inode,
                .scrub  = xfs_scrub_inode,
        },
+       { /* inode data fork */
+               .setup  = xfs_scrub_setup_inode_bmap,
+               .scrub  = xfs_scrub_bmap_data,
+       },
+       { /* inode attr fork */
+               .setup  = xfs_scrub_setup_inode_bmap,
+               .scrub  = xfs_scrub_bmap_attr,
+       },
+       { /* inode CoW fork */
+               .setup  = xfs_scrub_setup_inode_bmap,
+               .scrub  = xfs_scrub_bmap_cow,
+       },
 };
 
 /* This isn't a stable feature, warn once per day. */
index ec635d4..8920ccf 100644 (file)
@@ -79,5 +79,8 @@ int xfs_scrub_finobt(struct xfs_scrub_context *sc);
 int xfs_scrub_rmapbt(struct xfs_scrub_context *sc);
 int xfs_scrub_refcountbt(struct xfs_scrub_context *sc);
 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);
 
 #endif /* __XFS_SCRUB_SCRUB_H__ */