]> sourceware.org Git - lvm2.git/commitdiff
metadata: _vg_read: check if PV_EXT_USED flag is set correctly for orphan PVs and...
authorPeter Rajnoha <prajnoha@redhat.com>
Wed, 11 Mar 2015 15:18:42 +0000 (16:18 +0100)
committerPeter Rajnoha <prajnoha@redhat.com>
Mon, 15 Feb 2016 11:44:46 +0000 (12:44 +0100)
If we know that the PV is orphan, meaning there's at least one MDA on
that PV which does not reference any VG and at the same time there's
PV_EXT_USED flag set, we're certainly in an inconsistent state and we
need to fix this.

For example, such situation can happen during vgremove/vgreduce if we
removed/reduced the VG, but we haven't written PV headers yet because
vgremove stopped abruptly for whatever reason just before writing new
PV headers with updated state, including PV extension flags (and so the
PV_EXT_USED flag).

However, in case the PV has no MDAs at all, we can't double-check
whether the PV_EXT_USED is correct or not - if that PV is marked
as used, it's either:
  - really used (but other disks with MDAs are missing)
  - or the error state as described above is hit

User needs to overwrite the PV header directly if it's really clear
the PV having no MDAs does not belong to any VG and at the same time
it's still marked as being in use (pvcreate -ff <dev_name> will fix this).

For example - /dev/sda here has 1 MDA, orphan and is incorrectly marked
with PV_EXT_USED flag:

$ pvs --binary -o+pv_in_use
  WARNING: Found inconsistent standalone Physical Volumes.
  WARNING: Repairing flag incorrectly marking Physical Volume /dev/sda as used.
  PV         VG     Fmt  Attr PSize   PFree   InUse
  /dev/sda          lvm2 ---  128.00m 128.00m     0

lib/metadata/metadata.c
tools/toollib.c

index 7af545a11dd0d7a67207c484ee282a017e7ea5ed..61084bcbb1ab81b4dd41b24bef7864fe0776cf9a 100644 (file)
@@ -3328,11 +3328,64 @@ void vg_revert(struct volume_group *vg)
                stack; // FIXME: What should we do?
 }
 
+static int _check_mda_in_use(struct metadata_area *mda, void *_in_use)
+{
+       int *in_use = _in_use;
+       if (!mda_is_ignored(mda))
+               *in_use = 1;
+       return 1;
+}
+
 struct _vg_read_orphan_baton {
+       struct cmd_context *cmd;
        struct volume_group *vg;
        uint32_t warn_flags;
+       int consistent;
+       int repair;
 };
 
+static int _check_or_repair_orphan_pv_ext(struct physical_volume *pv,
+                                         struct lvmcache_info *info,
+                                         struct _vg_read_orphan_baton *b)
+{
+       uint32_t ext_flags = lvmcache_ext_flags(info);
+       int at_least_one_mda_used;
+
+       if (ext_flags & PV_EXT_USED) {
+               if (lvmcache_mda_count(info)) {
+                       at_least_one_mda_used = 0;
+                       lvmcache_foreach_mda(info, _check_mda_in_use, &at_least_one_mda_used);
+
+                       /*
+                        * We've found a PV that is marked as used with PV_EXT_USED flag
+                        * and it's orphan at the same time while it contains MDAs.
+                        * This is incorrect state and it needs to be fixed.
+                        * The PV_EXT_USED flag needs to be dropped!
+                        */
+                       if (b->repair) {
+                               if (at_least_one_mda_used) {
+                                       log_warn("WARNING: Repairing flag incorrectly marking "
+                                                "Physical Volume %s as used.", pv_dev_name(pv));
+
+                                       /* pv_write will set correct ext_flags */
+                                       if (!pv_write(b->cmd, pv, 0)) {
+                                               b->consistent = 0;
+                                               log_error("Failed to repair physical volume \"%s\".",
+                                                         pv_dev_name(pv));
+                                               return 0;
+                                       }
+                               }
+                               b->consistent = 1;
+                       } else if (at_least_one_mda_used) {
+                               /* mark as inconsistent only if there's at least 1 MDA used */
+                               b->consistent = 0;
+                       }
+               }
+       }
+
+       return 1;
+}
+
 static int _vg_read_orphan_pv(struct lvmcache_info *info, void *baton)
 {
        struct _vg_read_orphan_baton *b = baton;
@@ -3352,13 +3405,20 @@ static int _vg_read_orphan_pv(struct lvmcache_info *info, void *baton)
        }
        pvl->pv = pv;
        add_pvl_to_vgs(b->vg, pvl);
+
+       if (!_check_or_repair_orphan_pv_ext(pv, info, baton)) {
+               stack;
+               return 0;
+       }
+
        return 1;
 }
 
 /* Make orphan PVs look like a VG. */
 static struct volume_group *_vg_read_orphans(struct cmd_context *cmd,
                                             uint32_t warn_flags,
-                                            const char *orphan_vgname)
+                                            const char *orphan_vgname,
+                                            int *consistent)
 {
        const struct format_type *fmt;
        struct lvmcache_vginfo *vginfo;
@@ -3390,8 +3450,11 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd,
        vg->extent_count = 0;
        vg->free_count = 0;
 
+       baton.cmd = cmd;
        baton.warn_flags = warn_flags;
        baton.vg = vg;
+       baton.consistent = 1;
+       baton.repair = *consistent;
 
        while ((pvl = (struct pv_list *) dm_list_first(&head.list))) {
                dm_list_del(&pvl->list);
@@ -3403,6 +3466,7 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd,
        if (!lvmcache_foreach_pv(vginfo, _vg_read_orphan_pv, &baton))
                return_NULL;
 
+       *consistent = baton.consistent;
        return vg;
 }
 
@@ -3526,14 +3590,6 @@ static int _repair_inconsistent_vg(struct volume_group *vg)
        return 1;
 }
 
-static int _check_mda_in_use(struct metadata_area *mda, void *_in_use)
-{
-       int *in_use = _in_use;
-       if (!mda_is_ignored(mda))
-               *in_use = 1;
-       return 1;
-}
-
 static int _wipe_outdated_pvs(struct cmd_context *cmd, struct volume_group *vg, struct dm_list *to_check)
 {
        struct pv_list *pvl, *pvl2;
@@ -3631,8 +3687,7 @@ static struct volume_group *_vg_read(struct cmd_context *cmd,
                                  "with pre-commit.");
                        return NULL;
                }
-               *consistent = 1;
-               return _vg_read_orphans(cmd, warn_flags, vgname);
+               return _vg_read_orphans(cmd, warn_flags, vgname, consistent);
        }
 
        if (lvmetad_active() && !use_precommitted) {
@@ -4131,8 +4186,12 @@ struct volume_group *vg_read_internal(struct cmd_context *cmd, const char *vgnam
        }
 
 out:
-       if (!*consistent && (warn_flags & WARN_INCONSISTENT))
-               log_warn("WARNING: Volume Group %s is not consistent.", vgname);
+       if (!*consistent && (warn_flags & WARN_INCONSISTENT)) {
+               if (is_orphan_vg(vgname))
+                       log_warn("WARNING: Found inconsistent standalone Physical Volumes.");
+               else
+                       log_warn("WARNING: Volume Group %s is not consistent.", vgname);
+       }
 
        return vg;
 }
@@ -5126,8 +5185,11 @@ static struct volume_group *_vg_lock_and_read(struct cmd_context *cmd, const cha
        if (!consistent && !failure) {
                release_vg(vg);
                if (!(vg = _recover_vg(cmd, vg_name, vgid))) {
-                       log_error("Recovery of volume group \"%s\" failed.",
-                                 vg_name);
+                       if (is_orphan_vg(vg_name))
+                               log_error("Recovery of standalone physical volumes failed.");
+                       else
+                               log_error("Recovery of volume group \"%s\" failed.",
+                                         vg_name);
                        failure |= FAILED_RECOVERY;
                        goto bad_no_unlock;
                }
index f9ac259394ea0f0905749710648fc4b6db7e2ce1..7f8c2eb68ce9800ec1d1292c90750783a8121e99 100644 (file)
@@ -258,7 +258,10 @@ static int _ignore_vg(struct volume_group *vg, const char *vg_name,
 
        if (read_error != SUCCESS) {
                *skip = 0;
-               log_error("Cannot process volume group %s", vg_name);
+               if (is_orphan_vg(vg_name))
+                       log_error("Cannot process standalone physical volumes");
+               else
+                       log_error("Cannot process volume group %s", vg_name);
                return 1;
        }
 
This page took 0.069996 seconds and 5 git commands to generate.