4.21. device-mapper-multipath

Bug Fix

BZ#951435

Previously, the multipathd daemon did not stop and wait for all of its threads on shutdown, which caused multipathd to occasionally terminate unexpectedly with a segmentation fault during shutdown. With this update, multipathd now correctly stops and waits for its threads during shutdown, and crashes no longer occur in the described scenario.

Bug Fix

BZ#904106

Previously, the uev_discard() function did not take into account trailing NULL bytes in stack variables. Consequently, on 32-bit systems with the cciss disk controller driver that uses long device names, the multipathd daemon could terminate unexpectedly with a segmentation fault. A patch has been provided to fix uev_discard() and the crashes no longer occur in the described scenario.

Bug Fixes

BZ#522108

Multipath was not correctly getting the vendor, product, and revision strings for certain devices. Consequently, users were unable to correctly configure or use these devices with multipath. With this update, multipath now uses a SCSI inquiry IOCTL request to correctly get this information, and multipath is now able to configure and use these devices.

BZ#855147

When a partition was resized, kpartx reloaded the partition device with the NOFLUSH flag enabled, which device-mapper does not allow, causing the device to be left in a suspended state. Users had to manually resume the devices to restore their use. However, this did not fix the partition device size, and the partitioned device would become unusable again the next time kpartx was run on the device. With this update, kpartx now checks if the partition has changed size, and does not set the NOFLUSH flag when reloading a partitioned device with a new size. Now, users can repartition devices with kpartx without those devices becoming unusable.

BZ#860185

Previously, the multipathd daemon did not stop and wait for all of its threads on shutdown, which caused multipathd to occasionally terminate unexpectedly with a segmentation fault during shutdown. With this update, multipathd now correctly stops and waits for its threads during shutdown, and crashes no longer occur in the described scenario.

BZ#872439

Previously, the multipathd daemon terminated unexpectedly with a segmentation fault on 32-bit systems with a cciss disk controller. This was caused by a stack corruption in uev_discard() function. This update corrects the code in the uev_discard() function, and crashes no longer occur in the described scenario.

BZ#910585

If the /var/cache directory was a separate file system, the multipathd daemon unmounted it in its private namespace on start up. However, multipath requires the /var/cache directory to properly setup its private namespace. Consequently, the multipathd daemon failed to start on machines where the /var/cache directory was a separate file system. With this update, the multipathd daemon no longer unmounts the /var/cache directory from its private namespace, and multipathd will now correctly start up on machines where the /var/cache directory is a separate file system.

BZ#948309

Multipath switched to use the cciss_id callout for HP P2000 G3 SAS storage devices, which did not correctly provide the UUID. Consequently, multipath was unable to correctly configure HP P2000 G3 SAS devices. With this update, multipath switched back to using the scsi_id callout with the "-n" option to enable it to function with cciss devices, and multipath can now correctly configure HP P2000 G3 SAS devices.

Enhancements

BZ#799907

This update adds a built-in configuration for the IBM XIV Storage System.

BZ#839983

With this update, when multipath is set to log level 3, it will now print messages whenever it forks to execute a callout. This update helps debug a segmentation fault with multipath on Xen setups.

BZ#916630

With this update, the default configuration for NetApp devices now includes the "flush_on_last_del yes" parameter. With the default configuration, when all path devices to a NetApp LUN have been removed from the system, the multipath device for that LUN stops queuing I/O.