1 |
On Fri, 15 Feb 2008, Uwe Thiem wrote: |
2 |
|
3 |
> Second, no journalled filesystem in the whole wide world can prevent |
4 |
> occurences of inconsisteny in case of a power cut. None, try as they |
5 |
> might. |
6 |
|
7 |
This is correct. |
8 |
|
9 |
> If the journal change still resides in the |
10 |
> harddrive cache while your power cut occurs, boooom - inconsistency. |
11 |
|
12 |
But this isn't the reason. Harddrives know a "flush" command which - |
13 |
when properly used by the filesystem (and I guess reiserfs and ext3 |
14 |
use it properly) - forces the journal to be written before the actual |
15 |
change in the main file system occurs. Whence, no loss of consistency. |
16 |
[Of course, there are some harddrives which ignore the "flush", but |
17 |
this should be counted as faulty hardware. Of course, on broken |
18 |
hardware, no software can work as it should.] |
19 |
|
20 |
If the power loss occurs *during* flushing the journal (and thus |
21 |
the journal might contain nonsense) the filesystem might still use |
22 |
a checksum over the journal to detect this and thus preserves |
23 |
consistency (although I don't know whether any existing filesystem |
24 |
currently does this). |
25 |
|
26 |
The real problem is that during power cut the harddrive might be |
27 |
writing complete nonsense *somewhere* - this is not related with |
28 |
any caching, and no software can safe you from this problem |
29 |
(and what is even worse is that there is no way to detect it...) |
30 |
-- |
31 |
gentoo-user@l.g.o mailing list |