[dm-devel] Deadlock on RHEL4 with kpartx/queue_if_no_path - any way out?

[Tore Anderson]

Hi.  Due to a PEBCAK and wrong paths being removed I've now ended up
in this situation:

$ multipath -ll multimedia
multimedia (3600a0b80002984ae00001da546c27de1)
[size=200 GB][features="1 queue_if_no_path"][hwhandler="0"]
\_ round-robin 0 [enabled]
 \_ #:#:#:#  -    65:16  [failed][faulty]

I readded the paths with 'echo - - - > /sys/class/scsi_host/host*/scan',
but the map does not get healthy - all attempts to reload the map with
the multipath tools ends up in blocking I/O, using dmsetup directly I
only get failures due to the map being busy.

This is how the process list looks:

$ ps axf
  PID TTY      STAT   TIME COMMAND
[...]
11598 ?        SLl    0:01 /sbin/multipathd
12080 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
12087 ?        S<     0:00  \_ /sbin/dmsetup ls -j 253 -m 3 --target multipath --exec /sbin/kpartx -a
12088 ?        D<     0:00      \_ /sbin/kpartx -a /dev/mapper/multimedia
12714 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
12720 ?        S<     0:00  \_ /sbin/multipath -v0 /dev/sdb
12794 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
12800 ?        D<     0:00  \_ /sbin/multipath -v0 /dev/sdc
12831 ?        D      0:00 multipath -v 2
14132 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
14139 ?        S<     0:00  \_ /sbin/dmsetup ls -j 253 -m 3 --target multipath --exec /sbin/kpartx -a
14140 ?        D<     0:00      \_ /sbin/kpartx -a /dev/mapper/multimedia
14219 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
14225 ?        D<     0:00  \_ /sbin/multipath -v0 /dev/sdd
14255 ?        D      0:00 multipath -v 2
15312 ?        D      0:00 multipath /dev/sdd
17032 ?        D      0:00 multipath
 1008 ?        S<     0:00 /bin/sh /etc/dev.d/block/multipath.dev block
 1014 ?        D<     0:00  \_ /sbin/multipath -v0 /dev/sde

...I assume the kpartx process is locking the map so no changes can be
done, but since there's no alive paths and queue_if_no_path is active,
it will never return, and I'm stuck in a deadlock.

Is there any way out of this save a reboot?

Regards,
-- 
Tore Anderson