7-STABLE, gjournal and fsck.

[Eugene Butusov]

Hi,

   Recently I've decided to play with gjournal. Main reason was a 
promise of avoiding full fsck check after unclean shutdown. I've 
successfuly configured gjournal on existing filesystems (all UFS). And 
then it happened - my system had a power failure. After boot, it 
forced me to run fsck manualy. Nothing special, I did it before... But 
this time it failed on gjournaled disks.

So, when I was dropped to the single-user shell, I tried:

fsck /dev/ad4s1g.journal

It said:

CANNOT READ BLK: xxxx
CONTINUE? [yn]

I typed 'y' and nothing happened. Here is the log:

-8<-	
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 4059706613: 
ad4s1g contains data.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 4059706613: 
ad4s1g contains journal.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal ad4s1g clean.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 188084099: ad6s1d 
contains data.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 188084099: ad6s1d 
contains journal.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal ad6s1d clean.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 2559963968: 
ad6s1e contains data.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal 2559963968: 
ad6s1e contains journal.
Aug  2 19:13:43 matrix kernel: GEOM_JOURNAL: Journal ad6s1e clean.
...
Aug  2 19:13:43 matrix kernel: ** /dev/ad4s1g.journal
Aug  2 19:13:43 matrix kernel:
Aug  2 19:13:43 matrix kernel: CANNOT READ BLK: 727112224
Aug  2 19:13:43 matrix kernel: CONTINUE? [yn]
Aug  2 19:13:43 matrix kernel:
Aug  2 19:13:43 matrix kernel: THE FOLLOWING DISK SECTORS COULD NOT BE 
READ: 727112224, 727112225, 727112226, 727112227,
Aug  2 19:13:43 matrix kernel: /dev/ad4s1g.journal: CANNOT FIGURE OUT 
FILE SYSTEM PARTITION
->8-

After ctrl+d the system tried to continue boot, and again threw me 
into shell because of the same reason:

-8<-
Aug  2 19:13:43 matrix kernel:
WARNING: R/W mount of /home denied.  Filesystem is not clean - run fsck
Aug  2 19:13:43 matrix kernel: mount:
->8-

Like I mentioned, only gjournaled filesystems failed to pass fsck. 
Other labels passed. I was in a hurry, because the machine acts as a 
local file server, and I was standing against the wall, because one of 
gjournaled disks was the share itself...

What I did was mounting gjournaled partitions in ro mode (it warned me 
that they were not cleanly unmounted) and doing some backup. Then I 
removed gjournal (gjournal clear, tunefs -J disable) from journaled 
disks, ran fsck (few errors of type: PARTIALLY ALLOCATED INODE), and 
then I was able to turn on softupdates back and mount the fs in rw 
mode. I've double checked the disk's SMART results in case of hardware 
failure, but they were ok.

My question is: what could cause such problem? Why only gjournaled fs 
are affected? Is there a solution?

Best regards,
-- 
_/_/   .. Eugene Butusov
  _/_/  ... www.devilka.info
   _/_/ .... ebutusov(at)gmail(dot)com