filesystem information

[Bob McConnell]

On Behalf Of Jim

>> Just a thought, but in normal circumstances files *are* written to,
>> even when they are just being read: the access time is updated
(unless
>> you mount the fs with the noatime flag).
>>
> 
> quite true, but isn't that file metadata and not the actual file? I
> thought most filesystems had a file-entry section, with all the
> metadata, permissions, etc, and a file data section, which contains
> the information of the actual file.
> 
> I guess I wouldn't be surprised if the metadata being edited were
> corrupted, but to corrupt the file data/location seems odd to me.
> Especially with soft updates; unless I'm mistaken, that's what it is
> supposed to minimise.
> -Jim

But if the power failure interrupts an update, all files with data in
the sector(s) it was writing are at risk. A corrupted sector can contain
multiple file entries, and any or all of their entries may be lost. It
is up to fsck and friends to determine which of them can still be safely
accessed and restore those entries. It is not only possible, but likely
that some will be lost each time this occurs.

On the other hand, I thought this was one of the problems that journal
led file systems were invented to solve.

Bob McConnell