Journal abortion in ext3 and its relation with remounting

[Theodore Ts'o]

On Mon, Jan 21, 2013 at 08:48:23PM +0530, shubham wrote:
> I have RHEL-5.8 installed server.
> 
> Due to some inconsistent read write operation, ext3 journal got
> corrupted and aborted but filesystem was not remounted read only.
> In my understanding when there is ext3 journal has some corruption
> then it should mount the filesystem read only .
> So, I want to know in what cases this will not happen.

It will be remounted read-only if the file system has been configured
to do so.  See the man page for tune2fs, and it's documentation abouts
its -e option, which allows you to set the behavior when a file system
corruption is detected.  You can configure the superblock so that on
detection of a fs corruption, it will either continue (this is the
"don't worry, be happy" option), remount the file system read-only, or
panic the kernel.  (The last might be appropriate if you have a high
availability setup where a backup system can take cover immediately;
in that case, a panic and reboot so the file system can be repaired
using fsck might be the best thing to do.)

> For unlinked inodes, I want to know the logic of handling unlinked
> inodes in filesystem.
> 
> For an ex:
> - Situation where unlinked inode will be found.
> - How ext3 identifies unlinked inode ?
> - What ext3 does when it sees an unlinked inode.

I assume what you mean by this is when a directory discovers a file
name which points at a deleted inode?  In that case, it will log a
file system error, which will cause an EXT3-fs error message to appear
in the system log, and then either reboot, remount read-only, or
continue.

Regards,

						- Ted