Re: [ros-dev] FullFAT replacement for Fastfat.sys

[Sarocet]

James Walmsley wrote:
>   Also addressing the questions regarding a journaling
> system, we can
>  simply have the driver
>  create a ROS.journal file in the root dir. This file won't mean
>  anything to other systems,
>     
I'd prefer to see it named journal.fat
.journal extension would need a LFN. Since this would be a critical file
for the fs, we should
stick to 8.3 (it's more compatible, allow for more robust code and keeps
the gate open for also
journaling implementations without long filenames).
Journaled fat doesn't need to be reactos specific so a vendor neutral
name like journal.fat instead
of journal.ros look better.

>   and will simply be ignored. Causing no compatibility
> issues. If
>  another system messes up the
>  FAT table or directory structures thats not the fault of ReactOS, and
>  would require a full chkdsk.
>     
For example, the following correct behavior:
-ReactOS create a file, allocate and use several blocks and note that at
the journal before writing the directory entry.
-System hangs.
-ReactOS reboots, check the log and free those entries.


Can produce corruption if there's a implementation not aware of
journaling in the mix:
-ReactOS create a file, allocate and use several blocks and note that at
the journal before writing the directory entry.
-System hangs.
-User reboots in FooOS
-FooOS detect the partition incorrectly umounted, chkdsk it and frees
those blocks.
-FooOS create a new file, using those sectors. This file gets correctly
commited.
-System hangs again.
-User reboot on ReactOS, check the log and free those entries.

Here, by relying on the journal contents, ROS has just corrupted what
was a correct FS.