Another loop bug? (Was: 2.4.17rc1aa1)

[Gergely Nagy]

> Only in 2.4.17rc1aa1: 00_loop-deadlock-1
> 
> 	Fixed loop deadlock (balance_dirty() must not be called within the
> 	request_fn or it can deadlock or recurse way too much). It is
> 	enough to recall balance_dirty() at the highest layer so this won't
> 	introduce any stability problem.
> 

Maybe my problem is unrelated to this, but when I try to mount an ext2
image (from an ext3 filesystem, if that counts), loop0 starts to eat
CPU, and brings up the load to around 2.90. After a short time (about
a minute after mounting the image), everything that tries to access
the mounted area hangs. I can't umount it, and can't kill the
processes which are trying to access it.

I have this problem since 2.4.16, 2.4.14 (with ext3 0.9.15) works
fine.