xfs: scrub/repair should update filesystem metadata health
authorDarrick J. Wong <darrick.wong@oracle.com>
Tue, 16 Apr 2019 15:22:00 +0000 (08:22 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Tue, 16 Apr 2019 17:01:57 +0000 (10:01 -0700)
Now that we have the ability to track sick metadata in-core, make scrub
and repair update those health assessments after doing work.

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

index 786379c..b20964e 100644 (file)
@@ -143,6 +143,7 @@ xfs-y                               += $(addprefix scrub/, \
                                   common.o \
                                   dabtree.o \
                                   dir.o \
+                                  health.o \
                                   ialloc.o \
                                   inode.o \
                                   parent.o \
diff --git a/fs/xfs/scrub/health.c b/fs/xfs/scrub/health.c
new file mode 100644 (file)
index 0000000..df99e00
--- /dev/null
@@ -0,0 +1,176 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (C) 2019 Oracle.  All Rights Reserved.
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ */
+#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_health.h"
+#include "scrub/scrub.h"
+#include "scrub/health.h"
+
+/*
+ * Scrub and In-Core Filesystem Health Assessments
+ * ===============================================
+ *
+ * Online scrub and repair have the time and the ability to perform stronger
+ * checks than we can do from the metadata verifiers, because they can
+ * cross-reference records between data structures.  Therefore, scrub is in a
+ * good position to update the online filesystem health assessments to reflect
+ * the good/bad state of the data structure.
+ *
+ * We therefore extend scrub in the following ways to achieve this:
+ *
+ * 1. Create a "sick_mask" field in the scrub context.  When we're setting up a
+ * scrub call, set this to the default XFS_SICK_* flag(s) for the selected
+ * scrub type (call it A).  Scrub and repair functions can override the default
+ * sick_mask value if they choose.
+ *
+ * 2. If the scrubber returns a runtime error code, we exit making no changes
+ * to the incore sick state.
+ *
+ * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore
+ * sick flags before exiting.
+ *
+ * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore
+ * sick flags.  If the user didn't want to repair then we exit, leaving the
+ * metadata structure unfixed and the sick flag set.
+ *
+ * 5. Now we know that A is corrupt and the user wants to repair, so run the
+ * repairer.  If the repairer returns an error code, we exit with that error
+ * code, having made no further changes to the incore sick state.
+ *
+ * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean,
+ * use sick_mask to clear the incore sick flags.  This should have the effect
+ * that A is no longer marked sick.
+ *
+ * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and
+ * use sick_mask to set the incore sick flags.  This should have no externally
+ * visible effect since we already set them in step (4).
+ *
+ * There are some complications to this story, however.  For certain types of
+ * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild
+ * both structures at the same time.  The following principles apply to this
+ * type of repair strategy:
+ *
+ * 8. Any repair function that rebuilds multiple structures should update
+ * sick_mask_visible to reflect whatever other structures are rebuilt, and
+ * verify that all the rebuilt structures can pass a scrub check.  The outcomes
+ * of 5-7 still apply, but with a sick_mask that covers everything being
+ * rebuilt.
+ */
+
+/* Map our scrub type to a sick mask and a set of health update functions. */
+
+enum xchk_health_group {
+       XHG_FS = 1,
+       XHG_RT,
+       XHG_AG,
+       XHG_INO,
+};
+
+struct xchk_health_map {
+       enum xchk_health_group  group;
+       unsigned int            sick_mask;
+};
+
+static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {
+       [XFS_SCRUB_TYPE_SB]             = { XHG_AG,  XFS_SICK_AG_SB },
+       [XFS_SCRUB_TYPE_AGF]            = { XHG_AG,  XFS_SICK_AG_AGF },
+       [XFS_SCRUB_TYPE_AGFL]           = { XHG_AG,  XFS_SICK_AG_AGFL },
+       [XFS_SCRUB_TYPE_AGI]            = { XHG_AG,  XFS_SICK_AG_AGI },
+       [XFS_SCRUB_TYPE_BNOBT]          = { XHG_AG,  XFS_SICK_AG_BNOBT },
+       [XFS_SCRUB_TYPE_CNTBT]          = { XHG_AG,  XFS_SICK_AG_CNTBT },
+       [XFS_SCRUB_TYPE_INOBT]          = { XHG_AG,  XFS_SICK_AG_INOBT },
+       [XFS_SCRUB_TYPE_FINOBT]         = { XHG_AG,  XFS_SICK_AG_FINOBT },
+       [XFS_SCRUB_TYPE_RMAPBT]         = { XHG_AG,  XFS_SICK_AG_RMAPBT },
+       [XFS_SCRUB_TYPE_REFCNTBT]       = { XHG_AG,  XFS_SICK_AG_REFCNTBT },
+       [XFS_SCRUB_TYPE_INODE]          = { XHG_INO, XFS_SICK_INO_CORE },
+       [XFS_SCRUB_TYPE_BMBTD]          = { XHG_INO, XFS_SICK_INO_BMBTD },
+       [XFS_SCRUB_TYPE_BMBTA]          = { XHG_INO, XFS_SICK_INO_BMBTA },
+       [XFS_SCRUB_TYPE_BMBTC]          = { XHG_INO, XFS_SICK_INO_BMBTC },
+       [XFS_SCRUB_TYPE_DIR]            = { XHG_INO, XFS_SICK_INO_DIR },
+       [XFS_SCRUB_TYPE_XATTR]          = { XHG_INO, XFS_SICK_INO_XATTR },
+       [XFS_SCRUB_TYPE_SYMLINK]        = { XHG_INO, XFS_SICK_INO_SYMLINK },
+       [XFS_SCRUB_TYPE_PARENT]         = { XHG_INO, XFS_SICK_INO_PARENT },
+       [XFS_SCRUB_TYPE_RTBITMAP]       = { XHG_RT,  XFS_SICK_RT_BITMAP },
+       [XFS_SCRUB_TYPE_RTSUM]          = { XHG_RT,  XFS_SICK_RT_SUMMARY },
+       [XFS_SCRUB_TYPE_UQUOTA]         = { XHG_FS,  XFS_SICK_FS_UQUOTA },
+       [XFS_SCRUB_TYPE_GQUOTA]         = { XHG_FS,  XFS_SICK_FS_GQUOTA },
+       [XFS_SCRUB_TYPE_PQUOTA]         = { XHG_FS,  XFS_SICK_FS_PQUOTA },
+};
+
+/* Return the health status mask for this scrub type. */
+unsigned int
+xchk_health_mask_for_scrub_type(
+       __u32                   scrub_type)
+{
+       return type_to_health_flag[scrub_type].sick_mask;
+}
+
+/*
+ * Update filesystem health assessments based on what we found and did.
+ *
+ * If the scrubber finds errors, we mark sick whatever's mentioned in
+ * sick_mask, no matter whether this is a first scan or an
+ * evaluation of repair effectiveness.
+ *
+ * Otherwise, no direct corruption was found, so mark whatever's in
+ * sick_mask as healthy.
+ */
+void
+xchk_update_health(
+       struct xfs_scrub        *sc)
+{
+       struct xfs_perag        *pag;
+       bool                    bad;
+
+       if (!sc->sick_mask)
+               return;
+
+       bad = (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT);
+       switch (type_to_health_flag[sc->sm->sm_type].group) {
+       case XHG_AG:
+               pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
+               if (bad)
+                       xfs_ag_mark_sick(pag, sc->sick_mask);
+               else
+                       xfs_ag_mark_healthy(pag, sc->sick_mask);
+               xfs_perag_put(pag);
+               break;
+       case XHG_INO:
+               if (!sc->ip)
+                       return;
+               if (bad)
+                       xfs_inode_mark_sick(sc->ip, sc->sick_mask);
+               else
+                       xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
+               break;
+       case XHG_FS:
+               if (bad)
+                       xfs_fs_mark_sick(sc->mp, sc->sick_mask);
+               else
+                       xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
+               break;
+       case XHG_RT:
+               if (bad)
+                       xfs_rt_mark_sick(sc->mp, sc->sick_mask);
+               else
+                       xfs_rt_mark_healthy(sc->mp, sc->sick_mask);
+               break;
+       default:
+               ASSERT(0);
+               break;
+       }
+}
diff --git a/fs/xfs/scrub/health.h b/fs/xfs/scrub/health.h
new file mode 100644 (file)
index 0000000..fd0d466
--- /dev/null
@@ -0,0 +1,12 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (C) 2019 Oracle.  All Rights Reserved.
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ */
+#ifndef __XFS_SCRUB_HEALTH_H__
+#define __XFS_SCRUB_HEALTH_H__
+
+unsigned int xchk_health_mask_for_scrub_type(__u32 scrub_type);
+void xchk_update_health(struct xfs_scrub *sc);
+
+#endif /* __XFS_SCRUB_HEALTH_H__ */
index 02d278b..93b0f07 100644 (file)
@@ -40,6 +40,7 @@
 #include "scrub/trace.h"
 #include "scrub/btree.h"
 #include "scrub/repair.h"
+#include "scrub/health.h"
 
 /*
  * Online Scrub and Repair
@@ -498,6 +499,7 @@ xfs_scrub_metadata(
        xchk_experimental_warning(mp);
 
        sc.ops = &meta_scrub_ops[sm->sm_type];
+       sc.sick_mask = xchk_health_mask_for_scrub_type(sm->sm_type);
 retry_op:
        /* Set up for the operation. */
        error = sc.ops->setup(&sc, ip);
@@ -520,6 +522,8 @@ retry_op:
        } else if (error)
                goto out_teardown;
 
+       xchk_update_health(&sc);
+
        if ((sc.sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) &&
            !(sc.flags & XREP_ALREADY_FIXED)) {
                bool needs_fix;
index 9e8d3d7..1b280f8 100644 (file)
@@ -66,6 +66,13 @@ struct xfs_scrub {
        /* See the XCHK/XREP state flags below. */
        unsigned int                    flags;
 
+       /*
+        * The XFS_SICK_* flags that correspond to the metadata being scrubbed
+        * or repaired.  We will use this mask to update the in-core fs health
+        * status with whatever we find.
+        */
+       unsigned int                    sick_mask;
+
        /* State tracking for single-AG operations. */
        struct xchk_ag                  sa;
 };