I recently set up a brand new 8-STABLE box with UFS2 + gjournal. After a
hang and hard reboot, I noticed that the setup is not working. File
systems came up inconsistent without the system even noticing. I had to
manually run fsck in single user mode to get back to a consistent state.
The setup is nested as follows:
Slice (MBR partition): ada0s2
geli full drive encryption: ada0s2.eli
bsdlabel: ada0s2.eli.journala - ada0s2.eli.journalg
Partitions a, d, e, f and g are UFS2 file systems. Partition b is swap.
As the hard drive is fully encrypted, the system boots off a USB key
containing the boot loader, kernel and fstab. In the fstab, file systems
are addressed by their volume labels and swap by a label set with glabel.
As it appeared gjournal was working fine, I had set the gjournal flag on
all UFS2 file systems and mounted them async. After the crash and
reboot, gjournal reported "consistent" state. Then, a message informed
me that / had not been unmounted cleanly and fsck ran on that one file
system. For the other file systems (located in partitions d, e, f and g
above), I got no messages of any sort. However, once the system was up,
they were clearly inconsistent.
A manual fsck run revealed serious inconsistencies such as unallocated
inodes. The journal had clearly failed to do its job.
Could anyone point me to what I am doing wrong? I know I have nested
several GEOM providers inside each other. I was under the impression
this was the very concept behind GEOM. Is it that I cannot have a single
journal with multiple bsdlabel partitions inside? Again, it would seem
GEOM allows this - and I have found mailing list messages suggesting
others have done this before.
On a related note, it would appear that there is no way to verify
gjournal is actually active on a file system. I can set the journaling
flag on any UFS2 partition - whether gjournal actually works or not. Is
there some way to ensure that all file systems with journaling flag set
are actually journaled?
The "gjournal list" command tells me that ada0s2.eli is a provider and
ada0s2.eli.journal a consumer. But are all bsdlabel partitions inside
the consumer covered as well? There seems to be no tool that can tell me
that. I am hoping someone on the list can shed some light on this.
As a stop-gap measure, I have enabled softupdates on the file systems
and removed the async flags from my fstab. I really would like to use
journaling though - if I can find a way to enable it and make sure it is
- Bartosz Fabianowski