NANDFS eats itself up

[Boris Astardzhiev]

Hi,

I've been playing with the NANDFS implementation in 10.0-CURRENT on a
SHEEVAPLUG device. So far i've noticed a strange problem having a simple
test on it. Firstly I have a partition that is 32MBs and it is mounted in
/mnt (ROOTFS is over NFS).

I've made an repetitive scp transfer of a 2MB file to the /mnt on the
sheevaplug. After a delay of 100s the file is rm'ed from /mnt and this is
repeating for quite awhile now. It seems the FS is eating itself up.

# uname -a:
FreeBSD smartcpe 10.0-CURRENT FreeBSD 10.0-CURRENT #32 r243606:243615M: Tue
Nov 27 15:34:50 EET 2012 root at freebsd9:/usr/obj/arm.arm/usr/src2/sys/SHEEVAPLUG
arm

# sysctl vfs.nandfs
vfs.nandfs.cleaner_segments: 5
vfs.nandfs.cleaner_interval: 5
vfs.nandfs.cleaner_enable: 1
vfs.nandfs.cps_between_sblocks: 5
vfs.nandfs.max_dirty_segs: 5
vfs.nandfs.sync_interval: 5
vfs.nandfs.verbose: 0

root at smartcpe:/mnt % df -h
Filesystem Size Used Avail Capacity Mounted on
10.100.100.34:/usr/sheeva-fs 7.5G 5.9G 1.0G 85% /
devfs 1.0k 1.0k 0B 100% /dev
/dev/gnand0s.root 31M *896k* 30M 3% /mnt

In the beginning the "Used" value was 128k. 4 hours later we have a "Used"
value of "2.4MB":

root at smartcpe:/mnt % df -h
Filesystem Size Used Avail Capacity Mounted on
10.100.100.34:/usr/sheeva-fs 7.5G 5.9G 1.0G 85% /
devfs 1.0k 1.0k 0B 100% /dev
/dev/gnand0s.root 31M *2.4M* 28M 8% /mnt

After 14h30min it has drained even more space - 7.8MB:
root at smartcpe:/mnt % df -h
Filesystem Size Used Avail Capacity Mounted on
10.100.100.34:/usr/sheeva-fs 7.5G 5.9G 1.0G 85% /
devfs 1.0k 1.0k 0B 100% /dev
/dev/gnand0s.root 31M *7.8M* 23M 25% /mnt

How do I fix this?

Greetings,
Boris