Re: init scripts problems ?

"Also sprach denis bonnenfant:"
> I managed to crash it again !
> 
> I set up the double link, create a fr1 array on A-2, mount,
> 
> then /etc/init.d/enbd stop client, stopping both
> 
> It seems to work, nda is out of raid. but enbd-client is not killed.

Well, one would like to see the state of the client at this point.


> echo 0 > /proc/nbdinfo does nothing

It always does something - it will produce kernel output. If you really
see nothing, then the kernel is dead!

> mdadm /dev/md0 --add /dev/nda is blocked, and there is no way out.

echo 0 always works.

> ps ax :
> 
>   3776 ?        D      0:04 [enbd-client]
>   3777 ?        D      0:04 [enbd-client]
>   3975 pts/0    D      0:00 enbd-client se3-croise 12355 -i A-2 -n 2 -m -b 512 /dev/nda
>   3982 pts/0    D+     0:00 mdadm /dev/md0 --add /dev/nda
>   4094 tty1     D      0:00 enbd-client se3-croise 12355 -i A-2 -n 2 -m -b 512 /dev/nda
> 

Need ps axl (note the "l")! That will show what they are doing.

> it's really stange, it looks like nda is registered as ndb in the kernel !
> 
> dmesg :
> 
> ENBD #2251[35]: enbd_clr_queue unqueued 0 reqs
> ENBD #3288[32]: enbd_soft_reset INVALIDATE DEVICE nda BEGIN
> ENBD #3302[28]: enbd_soft_reset INVALIDATE DEVICE nda DONE
> ENBD #3315[28]: enbd_soft_reset run enbd_request on nda
> ENBD #3193[18]: enbd_disable disabled device ndb
> ENBD #2251[36]: enbd_clr_queue unqueued 0 reqs
> ENBD #3288[33]: enbd_soft_reset INVALIDATE DEVICE ndb BEGIN
> ENBD #3292[4]: enbd_soft_reset invalidate would crash on device 00000000 with inode 00000000
> ENBD #3293[4]: enbd_soft_reset aborting invalidate

Well, there you were saved by the patch I gave you. bdev is NULL for
nbd, as inferred from your OOPS output.

> ldap:~# cat /proc/nbdinfo
> Device a:       Open
> [a] State:      verify, rw, disabled, last error 0, lives 11, bp 0

Strange - has been stopped and started 11 times!

When you get it stuck, show the ps axl.

In any case, I have an openers patch which may work better.

Peter