- BZ#677449
DM Multipath removed a device if it failed to check the device status due to insufficient memory. This happened because the command checking if the device map existed failed as the system returned an error. With this update, Multipath no longer returns an error under these circumstances and no devices are removed if the system runs out of memory while checking device status.
- BZ#678673
If a device-mapper-multipath device was open but all attached device paths had been lost, the device was unable to create a new table with no device paths. As a concequence the multipath -ll command returned output indicating that no paths to the device were available with confusing "failed faulty running" rows presenting the missing paths. Multipath devices now reload tables with no device paths correctly.
- BZ#689504
Device paths could fail even if unavailable only temporarily. This happened because the RDAC (Redundant Disk Array Controller) checker function did not recheck the status of hosts if it had received a temporary error code. The function now rechecks the path after it has received such error codes and the path failures are transient as expected.
- BZ#697386
A previous bug fix introduced a race condition between the main thread and the thread running the checkerloop routine as the checkerloop thread was created with deferred cancellation type. The checkerloop thread continued running and attempted to access a property, which had been previously unallocated by the main thread. This caused the multipathd daemon to shutdown with a segmentation fault. Now the checkerloop thread checks if a shutdown is in progress and the deamon shuts down gracefully.
- BZ#700169
The Multipath daemon failed to include some ghost paths when counting the number of active paths; however, when the ghost paths failed, they were subtracted from the number of active paths. This caused multipathd to fail IO requests even though some paths were still available. The Multipath daemon now counts ghost paths correctly and no longer fails IO requests while there are still active paths available.
- BZ#705854
If the user set dev_loss_tmo to a value greater than 600 in multipath.conf without setting the fast_io_fail_tmo value, the multipathd daemon did not notify the user that fast_io_fail_tmo was not set. Multipath now issues a warning that fast_io_fail_tmo is not set under such circumstances.
- BZ#706555
On shared-storage multipath setups that set failback to manual, multipath could keep alternating from the failover pathgroup to the primary pathgroup infinitely. This happened because multipath was incorrectly failing back to the primary pathgroup whenever a path priority changed. With this update, multipath no longer fails back to the primary pathgroup when a path's priority changes under such circumstances.
- BZ#707560
If the multipath device was deleted while a path was being checked, multipathd did not abort the path check and terminated unexpectedly when trying to access the multipath device information. The Multipath daemon now aborts any path checks when the multipath device is removed and the problem no longer occurs.
- BZ#714821
The Multipath daemon was removing a multipath device twice. This could cause multipathd to access memory already used for another purpose, and caused the multipathd daemon to terminate unexpectedly. The multipathd daemon now removes the device once and the problem no longer occurs.
- BZ#719571
The kpartx utility built partition devices for invalid GUID partition tables (GPT) because it did not validate the size of GUID partitions. The kpartx utility now checks the partition size, and does not build devices for invalid GPTs.
- BZ#723168
Multipath previously returned an unclear error message when it failed to find rport_id. The returned message and its severity have been adjusted.
- BZ#725541
Several upstream commits have been included in the device-mapper-multipath package providing a number of bug fixes and enhancements over the previous version.
- BZ#738298
Anaconda failed to recognize an existing filesystem on a zSeries Linux fibre-channel adapter (zFCP) LUN and marked it as 'Unknown' when reinstalling the system. This happened due to an incorrect setting of the DM_UDEV_DISABLE_DISK_RULES_FLAG property. Filesystem on a multipath zFCP LUN is now correctly recognized during the installation.
- BZ#747604
The asynchronous TUR path checker caused multipathd to terminate unexpectedly due to memory corruption. This happened if multipathd attempted to delete a path while the asynchronous TUR checker was running on the path. The asynchronous TUR checker code has been removed, and multipathd no longer crashes on path removal.
