]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commitdiff
xfs: bmap scrub should only scrub records once
authorDarrick J. Wong <darrick.wong@oracle.com>
Tue, 27 Aug 2019 00:06:00 +0000 (17:06 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Tue, 27 Aug 2019 00:43:15 +0000 (17:43 -0700)
The inode block mapping scrub function does more work for btree format
extent maps than is absolutely necessary -- first it will walk the bmbt
and check all the entries, and then it will load the incore tree and
check every entry in that tree, possibly for a second time.

Simplify the code and decrease check runtime by separating the two
responsibilities.  The bmbt walk will make sure the incore extent
mappings are loaded, check the shape of the bmap btree (via xchk_btree)
and check that every bmbt record has a corresponding incore extent map;
and the incore extent map walk takes all the responsibility for checking
the mapping records and cross referencing them with other AG metadata.

This enables us to clean up some messy parameter handling and reduce
redundant code.  Rename a few functions to make the split of
responsibilities clearer.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
fs/xfs/scrub/bmap.c

index 1bd29fdc2ab586945251084eebca671c9ec066b6..7b19c63e12ce0ad36c3048fe06c5b0af4615243a 100644 (file)
@@ -75,6 +75,7 @@ struct xchk_bmap_info {
        xfs_fileoff_t           lastoff;
        bool                    is_rt;
        bool                    is_shared;
+       bool                    was_loaded;
        int                     whichfork;
 };
 
@@ -213,25 +214,20 @@ xchk_bmap_xref_rmap(
 
 /* Cross-reference a single rtdev extent record. */
 STATIC void
-xchk_bmap_rt_extent_xref(
-       struct xchk_bmap_info   *info,
+xchk_bmap_rt_iextent_xref(
        struct xfs_inode        *ip,
-       struct xfs_btree_cur    *cur,
+       struct xchk_bmap_info   *info,
        struct xfs_bmbt_irec    *irec)
 {
-       if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
-               return;
-
        xchk_xref_is_used_rt_space(info->sc, irec->br_startblock,
                        irec->br_blockcount);
 }
 
 /* Cross-reference a single datadev extent record. */
 STATIC void
-xchk_bmap_extent_xref(
-       struct xchk_bmap_info   *info,
+xchk_bmap_iextent_xref(
        struct xfs_inode        *ip,
-       struct xfs_btree_cur    *cur,
+       struct xchk_bmap_info   *info,
        struct xfs_bmbt_irec    *irec)
 {
        struct xfs_mount        *mp = info->sc->mp;
@@ -240,9 +236,6 @@ xchk_bmap_extent_xref(
        xfs_extlen_t            len;
        int                     error;
 
-       if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
-               return;
-
        agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock);
        agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
        len = irec->br_blockcount;
@@ -300,20 +293,15 @@ xchk_bmap_dirattr_extent(
 
 /* Scrub a single extent record. */
 STATIC int
-xchk_bmap_extent(
+xchk_bmap_iextent(
        struct xfs_inode        *ip,
-       struct xfs_btree_cur    *cur,
        struct xchk_bmap_info   *info,
        struct xfs_bmbt_irec    *irec)
 {
        struct xfs_mount        *mp = info->sc->mp;
-       struct xfs_buf          *bp = NULL;
        xfs_filblks_t           end;
        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.
@@ -364,10 +352,13 @@ xchk_bmap_extent(
                xchk_fblock_set_corrupt(info->sc, info->whichfork,
                                irec->br_startoff);
 
+       if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
+               return 0;
+
        if (info->is_rt)
-               xchk_bmap_rt_extent_xref(info, ip, cur, irec);
+               xchk_bmap_rt_iextent_xref(ip, info, irec);
        else
-               xchk_bmap_extent_xref(info, ip, cur, irec);
+               xchk_bmap_iextent_xref(ip, info, irec);
 
        info->lastoff = irec->br_startoff + irec->br_blockcount;
        return error;
@@ -380,10 +371,13 @@ xchk_bmapbt_rec(
        union xfs_btree_rec     *rec)
 {
        struct xfs_bmbt_irec    irec;
+       struct xfs_bmbt_irec    iext_irec;
+       struct xfs_iext_cursor  icur;
        struct xchk_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;
+       struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, info->whichfork);
        uint64_t                owner;
        int                     i;
 
@@ -402,9 +396,26 @@ xchk_bmapbt_rec(
                }
        }
 
-       /* Set up the in-core record and scrub it. */
+       /*
+        * Check that the incore extent tree contains an extent that matches
+        * this one exactly.  We validate those cached bmaps later, so we don't
+        * need to check them here.  If the incore extent tree was just loaded
+        * from disk by the scrubber, we assume that its contents match what's
+        * on disk (we still hold the ILOCK) and skip the equivalence check.
+        */
+       if (!info->was_loaded)
+               return 0;
+
        xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
-       return xchk_bmap_extent(ip, bs->cur, info, &irec);
+       if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur,
+                               &iext_irec) ||
+           irec.br_startoff != iext_irec.br_startoff ||
+           irec.br_startblock != iext_irec.br_startblock ||
+           irec.br_blockcount != iext_irec.br_blockcount ||
+           irec.br_state != iext_irec.br_state)
+               xchk_fblock_set_corrupt(bs->sc, info->whichfork,
+                               irec.br_startoff);
+       return 0;
 }
 
 /* Scan the btree records. */
@@ -415,15 +426,26 @@ xchk_bmap_btree(
        struct xchk_bmap_info   *info)
 {
        struct xfs_owner_info   oinfo;
+       struct xfs_ifork        *ifp = XFS_IFORK_PTR(sc->ip, whichfork);
        struct xfs_mount        *mp = sc->mp;
        struct xfs_inode        *ip = sc->ip;
        struct xfs_btree_cur    *cur;
        int                     error;
 
+       /* Load the incore bmap cache if it's not loaded. */
+       info->was_loaded = ifp->if_flags & XFS_IFEXTENTS;
+       if (!info->was_loaded) {
+               error = xfs_iread_extents(sc->tp, ip, whichfork);
+               if (!xchk_fblock_process_error(sc, whichfork, 0, &error))
+                       goto out;
+       }
+
+       /* Check the btree structure. */
        cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork);
        xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
        error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info);
        xfs_btree_del_cursor(cur, error);
+out:
        return error;
 }
 
@@ -671,13 +693,6 @@ xchk_bmap(
        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 (!xchk_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 (!xchk_fblock_process_error(sc, whichfork, 0, &error))
@@ -689,7 +704,7 @@ xchk_bmap(
        for_each_xfs_iext(ifp, &icur, &irec) {
                if (xchk_should_terminate(sc, &error) ||
                    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
-                       break;
+                       goto out;
                if (isnullstartblock(irec.br_startblock))
                        continue;
                if (irec.br_startoff >= endoff) {
@@ -697,7 +712,7 @@ xchk_bmap(
                                        irec.br_startoff);
                        goto out;
                }
-               error = xchk_bmap_extent(ip, NULL, &info, &irec);
+               error = xchk_bmap_iextent(ip, &info, &irec);
                if (error)
                        goto out;
        }