[Top] [All Lists]

Re: umount doesn't seem to really unmount

To: Eric Sandeen <sandeen@xxxxxxxxxxx>
Subject: Re: umount doesn't seem to really unmount
From: Stefan Ring <stefanrin@xxxxxxxxx>
Date: Tue, 28 Aug 2012 07:45:44 +0200
Cc: Linux fs XFS <xfs@xxxxxxxxxxx>
Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; bh=AoGImNrdgkAwWKZcB5VCTORqMFR91VlwNW1fbpaJqtM=; b=lXjlsisfX9u0FC6vz0A23YFDGKT3n6MR/pRV2No/02+ivdAebKgRyVH/Ln/1qg0TdW Q3UZBKEYk5BeMHpKaBylS35C30RSNGDZP/ZV94Ye4lezi4dRJ1MLFfgFSw4QhT8mKVKr k0u1knsoSDVQkMW/3SsyG5zfkvJsTvioV8Cbod4FMrbyLneuzNCXGtuEysw5z/+zvZHn Yk57jPhFWapFsIuJ3+6BMnTBtE22VMn4E3+XH5xr+75aNL1qBDdsUBd5VJQyh8Vl1cxm qWkBNUyL/+lMl5vjFXORaHieDEd6fHx8T6EwzxdNaeKs081WToWODdeIyG4TbfWbyA9z Z4/Q==
In-reply-to: <503C0582.3080009@xxxxxxxxxxx>
References: <CAAxjCEwdYbcDoUvcAV1n8281HCCvrCvReaG2oM5+h+WV5yDZvA@xxxxxxxxxxxxxx> <1104CDC2-47D2-40CF-8294-A17E33125D54@xxxxxxxxxxx> <CAAxjCExx8LB3PyMUR8T215hMYps17D+Lvg4C0aOwLwD=d-NOMQ@xxxxxxxxxxxxxx> <alpine.LRH.2.02.1208270952390.8166@xxxxxxxxxxxxxxx> <CAAxjCEynRoS3AfYer=is725XR+kZEwe_SPxt_iDTJNRMo2UxOA@xxxxxxxxxxxxxx> <alpine.LRH.2.02.1208271016440.10454@xxxxxxxxxxxxxxx> <503C0582.3080009@xxxxxxxxxxx>
> Gah, is umount replaced by something via systemd?  I mean.... umount is
> still umount, no?
> IOW is this just doing a lazy unmount behind your back?

My interpretation is that at the time a service is started, the entire
filesystem "namespace" is copied to a secret place, including the
mounts. Then, when I issue umount, it is actually unmounted from the
main namespace, but from the viewpoint of the filesystem, it's still
mounted, precisely in the hidden namespace that got created for the
running service. Only when this service quits, its namespace will be
destroyed, and with it the filesystem's refcount will drop to 0. At
this point, the real "unmount", or rather "close" will happen.

Apparently, this is not usually a problem, but for me it is, because
at that time, the iSCSI storage device is gone. Gabriel has filed a
bug here: [1]. Unfortunately, I have no hope that this will ever be
"fixed", because it behaves just the way it was designed :(.

At least  it has become clear that this has nothing to do with XFS,
although subjectively my feeling is that the effects (i.e., data loss)
are felt more with XFS than with ext4.

[1] https://bugzilla.redhat.com/show_bug.cgi?id=851970

<Prev in Thread] Current Thread [Next in Thread>