corrupted ext2 filesystem with raidtools 0.42 and nbd 1.4

[Francesc Oller <francesc@startrek.upc.es>]

Hi all,

I'm gathering together 15 sparse 1GB files distributed in 5 PCs with
3 files each to get a 15GB block device. I'm using nbd 1.4 and raidtools
0.42 in linear mode. Kernel is 2.2.15

I intend to use the block device as a debian archive mirror but after
some amount of successful downloading the file system gets badly corrupted
and I've to fse2ck it. I don't know whose fault it is raid?, nbd? Are
they stable enough for this kind of work? Is anybody doing similar things?
Can anybody help?

My setup is the following:

To start:

losetup /dev/loop0 .debian0
losetup /dev/loop1 .debian1
losetup /dev/loop2 .debian2
./nbd-client soft8 60200 /dev/nd0
./nbd-client soft7 60200 /dev/nd1
./nbd-client soft6 60200 /dev/nd2
./nbd-client soft5 60200 /dev/nd3

/sbin/mdadd /dev/md0 /dev/loop0 /dev/loop1 /dev/loop2 /dev/nd0 \
  /dev/nd1 /dev/nd2 /dev/nd3 
/sbin/mdrun -pl /dev/md0

nbd-server is started from inetd with a line:

nbd stream tcp nowait nbd /francesc0/local/bin/nbd-server nbd-server 0\
  /francesc0/local/bin/.debian%%d 3072M -m

To stop:

/sbin/mdstop /dev/md0

# kill nbd-servers in soft5, soft6, soft7, soft8

losetup -d /dev/loop2
losetup -d /dev/loop1
losetup -d /dev/loop0

Cheers

Francesc Oller