On Thu, 1 Aug 2013 12:32:50 +0000 "Dorau, Lukasz" <lukasz.dorau@xxxxxxxxx>
wrote:
>
> There is another, more serious, problem.
> When we stop the array during initial resync (mdadm -Ss)
> and the function is_resync_complete() is entered for the last time,
> array->array.raid_disks already equals 0, because it is zero'ed by manager:
> a->info.array.raid_disks = mdstat->raid_disks;
> at managemon.c:454.
> As a result sync_size equals 0 and is_resync_complete() incorrectly returns 1 and resync finishes...
>
> It seems to be a race condition between monitor and manager - manager changes value of array.raid_disks too fast.
Yes - that is a serious problem. Thanks for reporting it.
I think this is the correct fix.
Thanks,
NeilBrown
From e49a8a80265ab2150c96b636450f5825bcd69d4a Mon Sep 17 00:00:00 2001
From: NeilBrown <neilb@xxxxxxx>
Date: Mon, 5 Aug 2013 15:40:16 +1000
Subject: [PATCH] mdmon: don't use 'ghost' values from an inactive array.
It is possible for mdmon to see (in /proc/mdstat) and array
in 'inactive' state, "mdadm -S" has written "inactive" to
"array_state".
In this state values such as "raid_disk" are not meaningful
and so should be ignored by manage_member().
Reported-by: "Dorau, Lukasz" <lukasz.dorau@xxxxxxxxx>
Signed-off-by: NeilBrown <neilb@xxxxxxx>
diff --git a/managemon.c b/managemon.c
index c245655..f40bbdb 100644
--- a/managemon.c
+++ b/managemon.c
@@ -450,9 +450,11 @@ static void manage_member(struct mdstat_ent *mdstat,
/* Raced with something */
return;
- // FIXME
- a->info.array.raid_disks = mdstat->raid_disks;
- // MORE
+ if (mdstat->active) {
+ // FIXME
+ a->info.array.raid_disks = mdstat->raid_disks;
+ // MORE
+ }
if (sysfs_get_ll(&a->info, NULL, "component_size", &component_size) >= 0)
a->info.component_size = component_size << 1;
Attachment:
signature.asc
Description: PGP signature
