| 1 |
On 6/24/06, Mark Knecht <markknecht@×××××.com> wrote: |
| 2 |
> That said, I seem to have file system problems on my external 1394 |
| 3 |
> hard drives. I do not know if this is due to a recent move to |
| 4 |
> 2.6.17-rt1, or bad maintenance on my part, or just bad luck. I've |
| 5 |
|
| 6 |
Well, I've had various problems with external drives. I have 3 |
| 7 |
1394/USB2 combo drives that I use for backups. If I connect them via |
| 8 |
1394, they will start to have problems at some point during my |
| 9 |
backups....sometimes several Gb will transfer fine, and then it will |
| 10 |
hiccup with various read/write failures. I *never* have a problem |
| 11 |
with these same drives connected via USB. |
| 12 |
|
| 13 |
In another event, I moved a 2.5" drive from a slightly damaged USB2 |
| 14 |
case to a new one. The new case started corrupting data written to |
| 15 |
it, without any warning or errors. Filesystem checks would always |
| 16 |
report no errors, but for example I could transfer kernel sources to |
| 17 |
the drive, but they would fail to extract from there. I eventually |
| 18 |
replaced the case, (same drive) and that cleared up all problems, but |
| 19 |
I had to restore the files from backups as they were actually |
| 20 |
corrupted on write (vs just not being able to read them correctly). |
| 21 |
|
| 22 |
So, what I really recommend you do for awhile is to do an md5sum of |
| 23 |
any files that you transfer to the drive before transferring them. |
| 24 |
And then periodically do a md5sum -c to verify the file contents. |
| 25 |
That is really the best way to make sure your drives are working |
| 26 |
right, and that your data hasn't been corrupted. |
| 27 |
|
| 28 |
And if they are combo drives, try the USB2 connection instead. |
| 29 |
|
| 30 |
> I wondered if there is really any value to a journaled file system? |
| 31 |
> Did it protect me? I cannot tell. |
| 32 |
|
| 33 |
With one exception, journaled filesystems are designed to protect |
| 34 |
themselves, not your data. That is, if a crash occurs in the middle |
| 35 |
of an operation, the filesystem meta-data will not be corrupted. It |
| 36 |
will know how large all files should be, which blocks are allocated to |
| 37 |
which files, whether any files should be deleted, and so on. It does |
| 38 |
not guarantee the actual file data however. But this 'guarantee' is |
| 39 |
useless if a hardware problem is corrupting the filesystem, which |
| 40 |
appears to be the problem here. |
| 41 |
|
| 42 |
The one exception is ext3 when mounted with the 'data=journal' option. |
| 43 |
But this makes the filesystem fairly slow, and still doesn't help |
| 44 |
with hardware problems.... |
| 45 |
|
| 46 |
-Richard |
| 47 |
-- |
| 48 |
gentoo-user@g.o mailing list |
Archives