I am having this strange problem with a FAT partition a in large disk. It is a
long story. At first, you'll see me talk a lot about NetBSD, but the problem
also occurs with OpenBSD.
My 120-Gb disk
--------------------
I had OpenBSD and NetBSD in the beginning of a 120-Gb hard disk. They were
followed by a couple of Linux partitions and a few FAT32 partitions. These
FAT32 partitions were labeled D:, E:, F: and G: in Windows. Here is a tree-like
explanation:
Windows 98
BSD
---- extended ----
Linux
Linux swap
Linux
FAT32: D
FAT32: E
FAT32: F
FAT32: G (20 Gb)
I know I wrote data to D: and E: with OpenBSD (not NetBSD IIRC). But I don't
remember ever writing to the G: partition, the last one in the disk. Except for
one time.
I was quite satisfied with OpenBSD, but wanted to give NetBSD another try, a
few months after giving up on it. I didn't have extra space, so I deleted
OpenBSD to install NetBSD. Actually, I can't really say that I *used* OpenBSD
or NetBSD. I just had them installed so I could learn about them. I spent most
of the time reading man pages and experimenting a little.
Before installing NetBSD, I made a "netbsd" directory in my G: partition (20 Gb
in size) while running Windows 98. I downloaded many binary packages into that
directory so I could install them in NetBSD.
When I installed the binary packages, already running NetBSD, I had an idea: I
created a sub-directory named "gone" and, as I installed the packages, I moved
them to the "gone" folder. So this "gone" folder was created with 'mkdir' in
NetBSD. After some time, that "gone" directory had at least 30 or maybe even
100 files.
A few hours later, I was in Windows 98 again and, for some reason, wanted to
check something in that folder. Surprise: the "gone" folder appeared empty from
Windows! Where had all the files gone? I booted into NetBSD and everything was
still there! Only Windows 98 couldn't see it. Weird
Anyway, I just moved everything to a 160-Gb hard disk. And it got worse.
My 160-Gb disk
--------------------
With a bigger disk, I decided to have both NetBSD and OpenBSD:
Windows 98
OpenBSD
NetBSD
---- extended ----
Linux
Linux swap
Linux
FAT32: D
FAT32: E
FAT32: F
FAT32: G (40 Gb)
The last partition got a lot bigger too. Twice as big.
As soon as I had my 160-Gb disk in place, I started with NetBSD, because it is
easier to install than OpenBSD. Once again, I made a "netbsd" directory in my
G: partition (now 40 Gb in size). Don't forget: this directory was created in
Windows 98. All over again, I downloaded many binary packages into that
directory so I could install them in NetBSD. And when I installed them, I
created a sub-directory named "gone" for the packages that had been installed
already. And once again, this "gone" folder was created with 'mkdir' in NetBSD.
Time to check the mysterious disappearance of the "gone" folder's contents. I
booted into Windows and, hey, everything was there! : -)
But not so fast. I made a stupid mistake with NetBSD and decided I wanted to
format its partition and reinstall from scratch. OK. Then I tried to install
all the binary packages from that G: partition again. NOT OK! NetBSD installed
some packages, but most failed. NetBSD told me they were "invalid" and
"probably not a package"! What the hell? So I installed all the packages again,
but now downloading them through FTP. So many of the packages in that directory
were replaced with new copies while I was in NetBSD.
I booted into Windows again. I saw the "gone" folder. When I opened it, YIKES!
Hundreds of folders with totally messed up names! And the status bar said the
folder contained some 220 Gb! That was really ugly.
So I booted into Linux and inspected the folder. Everything looked normal! I
even opened some of the files. So what is the culprit? Windows 98? My new hard
disk? NetBSD? Like I said, it is a long story
Looking for the culprit
----------------------------
I could not delete those strangely named files and folders. Windows 98 won't
handle them. I ran Scandisk on the G: partition. That was not a good idea
either. Scandisk detected errors, but messed up many other files and folders in
G: so badly that I would have lost everything if I didn't have backups. I
decided to reboot into NetBSD and delete those bad folders from there. I could
not. Now NetBSD refused to boot because it wouldn't find the "/usr" partition's
superblock. Oh boy.
At that point, two bad thoughts crossed my mind:
1) I had made some very bad mistake when I partitioned the disk.
2) Disks have some sort of extra "buffer zone" that is used to make for bad
clusters. The disk automatically remaps bad clusters to some part of that extra
area and provides the illusion that the entire disk is sane. Maybe there was
some sort of "cross-reference" between two areas of the disk that was making
NetBSD corrupt G: and, conversely, changes written to G: corrupt the NetBSD
area. I don't know. I was guessing a lot.
So I booted into Linux and inspected the folder again, after the Scandisk
fiasco. I could see some damage, but not as bad as it looked from within
Windows 98. So I deleted almost everything with Linux. Then the next thing I
did was copy some files and folders from Linux to the G: FAT32 partition. I
also mkdir'ed a couple of folders and touch'ed a few files.
Boot into Windows. Saw and opened files: everything normal! Linux did not cause
any damage.
I formatted G: again, from DOS, with slow formatting. Repeated all tests. Same
results. Ugh!
That encouraged me to rule out disk defects for the time being. Obviously,
neither Windows nor Linux ever caused any problem. Only NetBSD did. Wait. Only
NetBSD? What will happen if I try to write data to that disk with OpenBSD?
Well, the short answer: ALL THE SAME trouble happens when I write data to that
partition with OpenBSD. I get hundreds of strangely named folders when I
inspect the folder with Windows 98, but a very normal and sane listing when I
inspect the folder with Linux.
Then I ran a few more tests and found out something interesting. Nothing wrong
happens when I write data to the D:, E: and F: partitions with NetBSD or
OpenBSD! Only the G: partition is susceptible. And since it is a long reading,
let me remind you: it happened in the 120-GB disk and now it happens in the
160-Gb.
Even so, I decided it was time to check the disk again. First I ran Scandisk
with a full check, surface checking etc. Then I ran Partition Magic. It is a
very old version, but the infinite times it takes to format a partition suggest
that it does quite a lot of checking. Not a problem found. Damn. I went
further. I decided to repartition everything. Instead of cfdisk and QtParted,
this time I used my copy of Partition Magic. It is very old, but heck, it never
failed me before. I created one partition at a time. Slow formatting and
checking. I even changed some partition sizes, i.e. changed the position of
partition boundaries. I was resorting to superstition already.
I reinstalled NetBSD. Then OpenBSD. Arf! Another problem. When I try to install
OpenBSD with NetBSD already present, OpenBSD detects NetBSD's disklabel and
won't let me create its own. The only option I have is overwriting that
disklabel. Sigh Whatever. I tested a little with NetBSD, removed NetBSD and
installed OpenBSD.
The test: I copied a couple of HTML folders to D:, E:, F: and G:. Same thing
all over again: the D:, E: and F: partitions are okay, but G: throws a horrible
fit whenever I write to it with NetBSD or OpenBSD. And this time I am testing
with an empty G: partition. So instead of writing to "the middle" of the
partition, now I am writing data to the very beginning of it. I really don't
think there is anything wrong with this disk anyway.
OK, so here is a summary of the problem with that last FAT32 partition:
Write with Read with Result
-------- -------- --------
Windows 98 Windows 98 OK
Windows 98 Linux OK
Windows 98 Net/Open BSD OK
Linux Linux OK
Linux Windows 98 OK
Linux Net/Open BSD OK
Net/Open BSD Linux looks OK
Net/Open BSD Windows 98 garbled files and directories
Net/Open BSD Net/Open BSD looks OK at first, but I had
problems reading binaries back
from that partition and installing them.
Epilogue
------------
And this is the end of my story, at least for a while. Now I hope that someone
can:
- give me suggestions to find out what the problem is and maybe solve it
completely, short of avoiding that partition altogether when running OpenBSD or
NetBSD;
- run a similar test and confirm that it does not happen to me only;
- not crack silly jokes or throw in derisive comments because I still use
Windows 98 - I've heard a lot of them already and none of them has changed my
needs or preferences;
- instruct me on the best course of action in case this actually proves to be a
bug in the BSDs' kernels - which I find unlikely but not impossible.
At any rate, I wish to thank everyone that was kind and patient enough to read
the entire lengthy account of my problems.
A little more information follows:
Machine: i386 - Sempron 2.2
Disk: Seagate ST3160021A (160 GB)
Operating systems: Windows 98 SE, NetBSD 2.0.2, OpenBSD 3.7 and Slackware Linux
10.2.
Disklabel and dmesg obtained from OpenBSD:
http://tinyurl.com/93thb
http://tinyurl.com/7wwdg
