xfs: scrub inode btrees
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 18 Oct 2017 04:37:40 +0000 (21:37 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 26 Oct 2017 22:38:25 +0000 (15:38 -0700)
Check the records of the inode btrees to make sure that the values
make sense given the inode records themselves.

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

index 9cf0809..09232f6 100644 (file)
@@ -147,6 +147,7 @@ xfs-y                               += $(addprefix scrub/, \
                                   alloc.o \
                                   btree.o \
                                   common.o \
+                                  ialloc.o \
                                   scrub.o \
                                   )
 endif
index 23229f0..154c3dd 100644 (file)
@@ -518,7 +518,7 @@ static inline int xfs_sb_version_hasftype(struct xfs_sb *sbp)
                 (sbp->sb_features2 & XFS_SB_VERSION2_FTYPE));
 }
 
-static inline int xfs_sb_version_hasfinobt(xfs_sb_t *sbp)
+static inline bool xfs_sb_version_hasfinobt(xfs_sb_t *sbp)
 {
        return (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) &&
                (sbp->sb_features_ro_compat & XFS_SB_FEAT_RO_COMPAT_FINOBT);
index 1e23d13..74df6ec 100644 (file)
@@ -490,9 +490,11 @@ struct xfs_scrub_metadata {
 #define XFS_SCRUB_TYPE_AGI     4       /* AG inode header */
 #define XFS_SCRUB_TYPE_BNOBT   5       /* freesp by block btree */
 #define XFS_SCRUB_TYPE_CNTBT   6       /* freesp by length btree */
+#define XFS_SCRUB_TYPE_INOBT   7       /* inode btree */
+#define XFS_SCRUB_TYPE_FINOBT  8       /* free inode btree */
 
 /* Number of scrub subcommands. */
-#define XFS_SCRUB_TYPE_NR      7
+#define XFS_SCRUB_TYPE_NR      9
 
 /* i: Repair this metadata. */
 #define XFS_SCRUB_IFLAG_REPAIR         (1 << 0)
index 018127a..39165c3 100644 (file)
@@ -40,6 +40,8 @@
 #include "xfs_refcount_btree.h"
 #include "xfs_rmap.h"
 #include "xfs_rmap_btree.h"
+#include "xfs_log.h"
+#include "xfs_trans_priv.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
@@ -451,11 +453,38 @@ xfs_scrub_setup_ag_btree(
        struct xfs_inode                *ip,
        bool                            force_log)
 {
+       struct xfs_mount                *mp = sc->mp;
        int                             error;
 
+       /*
+        * If the caller asks us to checkpont the log, do so.  This
+        * expensive operation should be performed infrequently and only
+        * as a last resort.  Any caller that sets force_log should
+        * document why they need to do so.
+        */
+       if (force_log) {
+               error = xfs_scrub_checkpoint_log(mp);
+               if (error)
+                       return error;
+       }
+
        error = xfs_scrub_setup_ag_header(sc, ip);
        if (error)
                return error;
 
        return xfs_scrub_ag_init(sc, sc->sm->sm_agno, &sc->sa);
 }
+
+/* Push everything out of the log onto disk. */
+int
+xfs_scrub_checkpoint_log(
+       struct xfs_mount        *mp)
+{
+       int                     error;
+
+       error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
+       if (error)
+               return error;
+       xfs_ail_push_all_sync(mp->m_ail);
+       return 0;
+}
index 372a844..17830b8 100644 (file)
@@ -73,6 +73,7 @@ void xfs_scrub_fblock_set_warning(struct xfs_scrub_context *sc, int whichfork,
                xfs_fileoff_t offset);
 
 void xfs_scrub_set_incomplete(struct xfs_scrub_context *sc);
+int xfs_scrub_checkpoint_log(struct xfs_mount *mp);
 
 /* Setup functions */
 int xfs_scrub_setup_fs(struct xfs_scrub_context *sc, struct xfs_inode *ip);
@@ -80,6 +81,8 @@ int xfs_scrub_setup_ag_header(struct xfs_scrub_context *sc,
                              struct xfs_inode *ip);
 int xfs_scrub_setup_ag_allocbt(struct xfs_scrub_context *sc,
                               struct xfs_inode *ip);
+int xfs_scrub_setup_ag_iallocbt(struct xfs_scrub_context *sc,
+                               struct xfs_inode *ip);
 
 
 void xfs_scrub_ag_free(struct xfs_scrub_context *sc, struct xfs_scrub_ag *sa);
diff --git a/fs/xfs/scrub/ialloc.c b/fs/xfs/scrub/ialloc.c
new file mode 100644 (file)
index 0000000..496d6f2
--- /dev/null
@@ -0,0 +1,337 @@
+/*
+ * 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_alloc.h"
+#include "xfs_ialloc.h"
+#include "xfs_ialloc_btree.h"
+#include "xfs_icache.h"
+#include "xfs_rmap.h"
+#include "xfs_log.h"
+#include "xfs_trans_priv.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 to scrub inode btrees.
+ * If we detect a discrepancy between the inobt and the inode,
+ * try again after forcing logged inode cores out to disk.
+ */
+int
+xfs_scrub_setup_ag_iallocbt(
+       struct xfs_scrub_context        *sc,
+       struct xfs_inode                *ip)
+{
+       return xfs_scrub_setup_ag_btree(sc, ip, sc->try_harder);
+}
+
+/* Inode btree scrubber. */
+
+/* Is this chunk worth checking? */
+STATIC bool
+xfs_scrub_iallocbt_chunk(
+       struct xfs_scrub_btree          *bs,
+       struct xfs_inobt_rec_incore     *irec,
+       xfs_agino_t                     agino,
+       xfs_extlen_t                    len)
+{
+       struct xfs_mount                *mp = bs->cur->bc_mp;
+       xfs_agnumber_t                  agno = bs->cur->bc_private.a.agno;
+       xfs_agblock_t                   bno;
+
+       bno = XFS_AGINO_TO_AGBNO(mp, agino);
+       if (bno + len <= bno ||
+           !xfs_verify_agbno(mp, agno, bno) ||
+           !xfs_verify_agbno(mp, agno, bno + len - 1))
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+       return true;
+}
+
+/* Count the number of free inodes. */
+static unsigned int
+xfs_scrub_iallocbt_freecount(
+       xfs_inofree_t                   freemask)
+{
+       BUILD_BUG_ON(sizeof(freemask) != sizeof(__u64));
+       return hweight64(freemask);
+}
+
+/* Check a particular inode with ir_free. */
+STATIC int
+xfs_scrub_iallocbt_check_cluster_freemask(
+       struct xfs_scrub_btree          *bs,
+       xfs_ino_t                       fsino,
+       xfs_agino_t                     chunkino,
+       xfs_agino_t                     clusterino,
+       struct xfs_inobt_rec_incore     *irec,
+       struct xfs_buf                  *bp)
+{
+       struct xfs_dinode               *dip;
+       struct xfs_mount                *mp = bs->cur->bc_mp;
+       bool                            inode_is_free = false;
+       bool                            freemask_ok;
+       bool                            inuse;
+       int                             error = 0;
+
+       if (xfs_scrub_should_terminate(bs->sc, &error))
+               return error;
+
+       dip = xfs_buf_offset(bp, clusterino * mp->m_sb.sb_inodesize);
+       if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
+           (dip->di_version >= 3 &&
+            be64_to_cpu(dip->di_ino) != fsino + clusterino)) {
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+               goto out;
+       }
+
+       if (irec->ir_free & XFS_INOBT_MASK(chunkino + clusterino))
+               inode_is_free = true;
+       error = xfs_icache_inode_is_allocated(mp, bs->cur->bc_tp,
+                       fsino + clusterino, &inuse);
+       if (error == -ENODATA) {
+               /* Not cached, just read the disk buffer */
+               freemask_ok = inode_is_free ^ !!(dip->di_mode);
+               if (!bs->sc->try_harder && !freemask_ok)
+                       return -EDEADLOCK;
+       } else if (error < 0) {
+               /*
+                * Inode is only half assembled, or there was an IO error,
+                * or the verifier failed, so don't bother trying to check.
+                * The inode scrubber can deal with this.
+                */
+               goto out;
+       } else {
+               /* Inode is all there. */
+               freemask_ok = inode_is_free ^ inuse;
+       }
+       if (!freemask_ok)
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+out:
+       return 0;
+}
+
+/* Make sure the free mask is consistent with what the inodes think. */
+STATIC int
+xfs_scrub_iallocbt_check_freemask(
+       struct xfs_scrub_btree          *bs,
+       struct xfs_inobt_rec_incore     *irec)
+{
+       struct xfs_owner_info           oinfo;
+       struct xfs_imap                 imap;
+       struct xfs_mount                *mp = bs->cur->bc_mp;
+       struct xfs_dinode               *dip;
+       struct xfs_buf                  *bp;
+       xfs_ino_t                       fsino;
+       xfs_agino_t                     nr_inodes;
+       xfs_agino_t                     agino;
+       xfs_agino_t                     chunkino;
+       xfs_agino_t                     clusterino;
+       xfs_agblock_t                   agbno;
+       int                             blks_per_cluster;
+       uint16_t                        holemask;
+       uint16_t                        ir_holemask;
+       int                             error = 0;
+
+       /* Make sure the freemask matches the inode records. */
+       blks_per_cluster = xfs_icluster_size_fsb(mp);
+       nr_inodes = XFS_OFFBNO_TO_AGINO(mp, blks_per_cluster, 0);
+       xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
+
+       for (agino = irec->ir_startino;
+            agino < irec->ir_startino + XFS_INODES_PER_CHUNK;
+            agino += blks_per_cluster * mp->m_sb.sb_inopblock) {
+               fsino = XFS_AGINO_TO_INO(mp, bs->cur->bc_private.a.agno, agino);
+               chunkino = agino - irec->ir_startino;
+               agbno = XFS_AGINO_TO_AGBNO(mp, agino);
+
+               /* Compute the holemask mask for this cluster. */
+               for (clusterino = 0, holemask = 0; clusterino < nr_inodes;
+                    clusterino += XFS_INODES_PER_HOLEMASK_BIT)
+                       holemask |= XFS_INOBT_MASK((chunkino + clusterino) /
+                                       XFS_INODES_PER_HOLEMASK_BIT);
+
+               /* The whole cluster must be a hole or not a hole. */
+               ir_holemask = (irec->ir_holemask & holemask);
+               if (ir_holemask != holemask && ir_holemask != 0) {
+                       xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+                       continue;
+               }
+
+               /* If any part of this is a hole, skip it. */
+               if (ir_holemask)
+                       continue;
+
+               /* Grab the inode cluster buffer. */
+               imap.im_blkno = XFS_AGB_TO_DADDR(mp, bs->cur->bc_private.a.agno,
+                               agbno);
+               imap.im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
+               imap.im_boffset = 0;
+
+               error = xfs_imap_to_bp(mp, bs->cur->bc_tp, &imap,
+                               &dip, &bp, 0, 0);
+               if (!xfs_scrub_btree_process_error(bs->sc, bs->cur, 0, &error))
+                       continue;
+
+               /* Which inodes are free? */
+               for (clusterino = 0; clusterino < nr_inodes; clusterino++) {
+                       error = xfs_scrub_iallocbt_check_cluster_freemask(bs,
+                                       fsino, chunkino, clusterino, irec, bp);
+                       if (error) {
+                               xfs_trans_brelse(bs->cur->bc_tp, bp);
+                               return error;
+                       }
+               }
+
+               xfs_trans_brelse(bs->cur->bc_tp, bp);
+       }
+
+       return error;
+}
+
+/* Scrub an inobt/finobt record. */
+STATIC int
+xfs_scrub_iallocbt_rec(
+       struct xfs_scrub_btree          *bs,
+       union xfs_btree_rec             *rec)
+{
+       struct xfs_mount                *mp = bs->cur->bc_mp;
+       struct xfs_inobt_rec_incore     irec;
+       uint64_t                        holes;
+       xfs_agnumber_t                  agno = bs->cur->bc_private.a.agno;
+       xfs_agino_t                     agino;
+       xfs_agblock_t                   agbno;
+       xfs_extlen_t                    len;
+       int                             holecount;
+       int                             i;
+       int                             error = 0;
+       unsigned int                    real_freecount;
+       uint16_t                        holemask;
+
+       xfs_inobt_btrec_to_irec(mp, rec, &irec);
+
+       if (irec.ir_count > XFS_INODES_PER_CHUNK ||
+           irec.ir_freecount > XFS_INODES_PER_CHUNK)
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+       real_freecount = irec.ir_freecount +
+                       (XFS_INODES_PER_CHUNK - irec.ir_count);
+       if (real_freecount != xfs_scrub_iallocbt_freecount(irec.ir_free))
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+       agino = irec.ir_startino;
+       /* Record has to be properly aligned within the AG. */
+       if (!xfs_verify_agino(mp, agno, agino) ||
+           !xfs_verify_agino(mp, agno, agino + XFS_INODES_PER_CHUNK - 1)) {
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+               goto out;
+       }
+
+       /* Make sure this record is aligned to cluster and inoalignmnt size. */
+       agbno = XFS_AGINO_TO_AGBNO(mp, irec.ir_startino);
+       if ((agbno & (xfs_ialloc_cluster_alignment(mp) - 1)) ||
+           (agbno & (xfs_icluster_size_fsb(mp) - 1)))
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+       /* Handle non-sparse inodes */
+       if (!xfs_inobt_issparse(irec.ir_holemask)) {
+               len = XFS_B_TO_FSB(mp,
+                               XFS_INODES_PER_CHUNK * mp->m_sb.sb_inodesize);
+               if (irec.ir_count != XFS_INODES_PER_CHUNK)
+                       xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+               if (!xfs_scrub_iallocbt_chunk(bs, &irec, agino, len))
+                       goto out;
+               goto check_freemask;
+       }
+
+       /* Check each chunk of a sparse inode cluster. */
+       holemask = irec.ir_holemask;
+       holecount = 0;
+       len = XFS_B_TO_FSB(mp,
+                       XFS_INODES_PER_HOLEMASK_BIT * mp->m_sb.sb_inodesize);
+       holes = ~xfs_inobt_irec_to_allocmask(&irec);
+       if ((holes & irec.ir_free) != holes ||
+           irec.ir_freecount > irec.ir_count)
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+       for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; i++) {
+               if (holemask & 1)
+                       holecount += XFS_INODES_PER_HOLEMASK_BIT;
+               else if (!xfs_scrub_iallocbt_chunk(bs, &irec, agino, len))
+                       break;
+               holemask >>= 1;
+               agino += XFS_INODES_PER_HOLEMASK_BIT;
+       }
+
+       if (holecount > XFS_INODES_PER_CHUNK ||
+           holecount + irec.ir_count != XFS_INODES_PER_CHUNK)
+               xfs_scrub_btree_set_corrupt(bs->sc, bs->cur, 0);
+
+check_freemask:
+       error = xfs_scrub_iallocbt_check_freemask(bs, &irec);
+       if (error)
+               goto out;
+
+out:
+       return error;
+}
+
+/* Scrub the inode btrees for some AG. */
+STATIC int
+xfs_scrub_iallocbt(
+       struct xfs_scrub_context        *sc,
+       xfs_btnum_t                     which)
+{
+       struct xfs_btree_cur            *cur;
+       struct xfs_owner_info           oinfo;
+
+       xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INOBT);
+       cur = which == XFS_BTNUM_INO ? sc->sa.ino_cur : sc->sa.fino_cur;
+       return xfs_scrub_btree(sc, cur, xfs_scrub_iallocbt_rec, &oinfo, NULL);
+}
+
+int
+xfs_scrub_inobt(
+       struct xfs_scrub_context        *sc)
+{
+       return xfs_scrub_iallocbt(sc, XFS_BTNUM_INO);
+}
+
+int
+xfs_scrub_finobt(
+       struct xfs_scrub_context        *sc)
+{
+       return xfs_scrub_iallocbt(sc, XFS_BTNUM_FINO);
+}
index 4398b34..1303194 100644 (file)
@@ -182,6 +182,15 @@ static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
                .setup  = xfs_scrub_setup_ag_allocbt,
                .scrub  = xfs_scrub_cntbt,
        },
+       { /* inobt */
+               .setup  = xfs_scrub_setup_ag_iallocbt,
+               .scrub  = xfs_scrub_inobt,
+       },
+       { /* finobt */
+               .setup  = xfs_scrub_setup_ag_iallocbt,
+               .scrub  = xfs_scrub_finobt,
+               .has    = xfs_sb_version_hasfinobt,
+       },
 };
 
 /* This isn't a stable feature, warn once per day. */
index a4af99c..5d97453 100644 (file)
@@ -73,5 +73,7 @@ int xfs_scrub_agfl(struct xfs_scrub_context *sc);
 int xfs_scrub_agi(struct xfs_scrub_context *sc);
 int xfs_scrub_bnobt(struct xfs_scrub_context *sc);
 int xfs_scrub_cntbt(struct xfs_scrub_context *sc);
+int xfs_scrub_inobt(struct xfs_scrub_context *sc);
+int xfs_scrub_finobt(struct xfs_scrub_context *sc);
 
 #endif /* __XFS_SCRUB_SCRUB_H__ */