Red Hat Bugzilla – Bug 601842
Device mapper-multipath resizing problems.
Last modified: 2013-10-15 15:24:53 EDT
Description of problem:
multipathd failed to resize a multipath device with:
Jun 7 23:01:49 cornucopia multipathd: mpath4: resize map (operator)
Jun 7 23:01:49 cornucopia multipathd: mpath4: failed to resize map : Invalid argument
Why it failed to resize is unknown. Also, failing to resize the device caused other problems. After a failed resize, multipathd did not resync its state with the kernel, so it though that the device was the new size. This meant that future resize operations were ignored because it though that the user was simply trying to resize the device to its current size.
Trying to correct this by reconfiguring multipathd, caused multipathd to remove the device completely. Rerunning multipath picked the device back up, with
the correct size.
I've seen this on a machine now, and run a systemtap script to capture the information when it failed. Here is what I see.
1276632312: multipathd entered dev_suspend for 14f504e46494c450045636534686e2d695575522d49524568 with flags 1540
1276632312: multipathd entered dm_swap_table with flags 3, suspended_bdev 0 and capacity 61472768
1276632312: multipathd exitted dm_table_get_size with 409600000
1276632312: multipathd exitted dm_swap_table with -22
1276632312: multipathd exitted dev_suspend with -22
I'm not sure if this initially failed for the same reason as the initial comment. In this case, the initial failure happened because the device was already suspended when I tried the initial
multipathd -k"resize map <mapname>" command
However it was suspended with noflush set, so multipath couldn't change the size.
It turns out that this is easy to cause. If you simply run
# multipathd -k"resize map <mapname>"
multipath will try to reload the device with noflush set, causing this problem if you don't first resume the device before trying to resize it. That is getting fixed in 584742.
I'm leaving this open for making multipathd resync it's state with the kernel after a failed operation.
No additional minor releases are planned for Production Phase 2 in Red Hat Enterprise Linux 5, and therefore Red Hat is closing this bugzilla as it does not meet the inclusion criteria as stated in: