Bug 160227

There is work in progress to integrate this into the initrd, to support a dm
multipath root device.

Two proposed patches to initscripts for you to try out if you wish.

Created attachment 119548 [details]
call kpartx from rc.sysinit for dm-multipath devices

Created attachment 119549 [details]
call kpartx from hotplug script for dm-multipath

dev.d is deprecated in favor of direct udev rules, FWIW.

(That is, deprecated in the post-FC4 world.)

Added to initscripts CVS, will be in 8.17-1. Bouncing over to udev.

FYI, upstream uses the following rules file :

1 #
2 # multipath and multipath partitions nodes are created in /dev/mapper/
3 # this file should be installed in /etc/udev/rules.d
4 #
5 # !! udev must not discard DM events !!
6 # !! check the other installed rules !!
7 #
8
9 # lookup the devmap name
10 #ACTION=="add", SUBSYSTEM=="block", KERNEL=="dm-*", \
11 # PROGRAM="/sbin/devmap_name %M %m"
12 ACTION=="add", SUBSYSTEM=="block", KERNEL=="dm-*", \
13 PROGRAM="/sbin/dmsetup -j %M -m %m --noopencount --noheadings -c -o name info"
14
15 # take care of devmap partitioning
16 ACTION=="add", SUBSYSTEM=="block", KERNEL=="dm-*", \
17 RUN+="/sbin/kpartx -a /dev/mapper/%c"
18
19 # insert new paths in multipath topology
20 ACTION=="add", SUBSYSTEM=="block", KERNEL!="dm-*", \
21 RUN+="/sbin/multipath -v0 %r/%k"
22

And that's not ideal for a distribution because it assumes you're using
multipath on your system.  Most people will have the package installed, but not
actually be using it, and the impact on their systems needs to be as close to
nothing as possible.

1) I only want to run kpartx against dm multipath devices not other types of
device-mapper devices: see the dmsetup options used in the script.

2) I don't want to run /sbin/multipath whenever a new device appears on the
system if I'm not using multipath.  And if I am, I just want to notify the
daemon using mpath_ctl, and let the daemon take care of things.

For 2, this is the sort of thing that needs translating to udev:

if /sbin/pidof /sbin/multipathd > /dev/null 2>&1 ; then
        if [ "${DEVPATH:7:3}" = "dm-" ] ; then
                /sbin/mpath_ctl "${ACTION} map ${DEVPATH:7}" > /dev/null 2>&1
        else
                /sbin/mpath_ctl "${ACTION} path ${DEVPATH:7}" > /dev/null 2>&1
        fi
fi

And 'multipath' ought to go away I think: the daemon should manage everything
internally - so state is only held in one place - taking messages and returning
status info via mpath_ctl.  

[More precisely, the 'multipath' command can stay but its functionality moves
into the daemon and merges with mpath_ctl: all it does is send/receive data
to/from the daemon.  The hard-coded '1000' limit on data transfer (with buffer
overflow if exceeded) in the daemon also needs fixing so it either works out the
buffer size it needs before filling it, or else it 'reallocs' the buffer if it
needs more space.]

This report targets the FC3 or FC4 products, which have now been EOL'd.

Could you please check that it still applies to a current Fedora release, and
either update the target product or close it ?

Thanks.

reassigning to device-mapper-multipath, which can provide udev rules for that.

This all works for current releases