5.48. device-mapper-multipath

Bug Fix

BZ#837594

When a multipath vector (a dynamically allocated array) was resized to a smaller size, device-mapper-multipath did not reassign the pointer to the array. If the array location was changed by reducing its size, device-mapper-multipath corrupted its memory. With this update, device-mapper-multipath correctly reassigns the pointer in this scenario, and memory corruption no longer occurs.

Bug Fixes

BZ#812832

The multipathd daemon was not correctly stopping waiter threads during shutdown. The waiter threads could access freed memory and cause the daemon to terminate unexpectedly during shutdown. With this update, the mutlipathd daemon now correctly stops the waiter threads before they can access any freed memory and no longer crashes during shutdown.

BZ#662433

When Device Mapper Multipath was stopped, multipathd did not disable the queue_if_no_path option on multipath devices by default. When multipathd was stopped during shutdown, I/O of the device was added to the queue if all paths to a device were lost, and the shutdown process became unresponsive. With this update, multipathd now sets the queue_without_daemon option to no by default. As a result, all multipath devices stop queueing when multipathd is stopped and multipath now shuts down as expected.

BZ#752989

Device Mapper Multipath uses regular expressions in built-in device configurations to determine a multipath device so as to apply the correct configuration to the device. Previously, some regular expressions for resolving the device vendor name and product ID were not specific enough. As a consequence, some devices could be matched with incorrect device configurations. With this update, the product and vendor regular expressions have been modified so that all multipath devices are now configured properly.

BZ#754586

After renaming a device, there was a race condition between multipathd and udev to rename the new multipath device nodes. If udev renamed the device node first, multipathd removed the device created by udev and failed to create the new device node. With this update, multipathd immediately creates the new device nodes, and the race condition no longer occurs. As a result, the renamed device is now available as expected.

BZ#769527

Previously, the flush_on_last_dev handling code did not implement handling of the queue feature properly. Consequently, even though the flush_on_last_del feature was activated, multipathd re-enabled queueing on multipath devices that could not be removed immediately after the last path device was deleted. With this update, the code has been fixed and when the user sets flush_on_last_del, their multipath devices correctly disable queueing, even if the devices cannot be closed immediately.

BZ#796384

Previously, Device Mapper Multipath used a fixed-size buffer to read the Virtual Device Identification page [0x83]. The buffer size was sometimes insufficient to accommodate the data sent by devices and the ALUA (Asymmetric Logical Unit Access) prioritizer failed. Device Mapper Multipath now dynamically allocates a buffer large enough for the Virtual Device Identification page and the ALUA prioritizer no longer fails in the scenario described.

BZ#744210

Previously, multipathd did not set the max_fds option by default, which sets the maximum number of file descriptors that multipathd can open. Also, the user_friendly_names setting could only be configured in the defaults section of /etc/multipath.conf. The user had to set max_fds manually and override the default user_friendly_names value in their device-specific configurations. With this update, multipath now sets max_fds to the system maximum by default, and user_friendly_names can be configured in the devices section of multipath.conf. Users no longer need to set max_fds for large setups, and they are able to select user_friendly_names per device type.

BZ#744756

Previously, to modify a built-in configuration, the vendor and product strings of the user's configuration had to be identical to the vendor and product strings of the built-in configuration. The vendor and product strings are regular expressions, and the user did not always know the correct vendor and product strings needed to modify a built-in configuration. With this update, the hwtable_regex_match option was added to the defaults section of multipath.conf. If it is set to yes, Multipath uses regular-expression matching to determine if the user's vendor and product strings match the built-in device configuration strings: the user can use the actual vendor and product information from their hardware in their device configuration, and it will modify the default configuration for that device. The option is set to no by default.

BZ#750132

Previously, multipathd was using a deprecated Out-of-Memory (OOM) adjustment interface. Consequently, the daemon was not protected from the OOM killer properly; the OOM killer could kill the daemon when memory was low and the user was unable to restore failed paths. With this update, multipathd now uses the new Out-of-Memory adjustment interface and can no longer be killed by the Out-of-Memory killer.

BZ#702222

The multipath.conf file now contains a comment which informs the user that the configuration must be reloaded for any changes to take effect.

BZ#751938

The multipathd daemon incorrectly exited with code 1 when multipath -h (print usage) was run. With this update, the underlying code has been modified and multipathd now returns code 0 as expected in the scenario described.

BZ#751039

Some multipathd threads did not check if multipathd was shutting down before they started their execution. Consequently, the multipathd daemon could terminate unexpectedly with a segmentation fault on shutdown. With this update, the multipathd threads now check if multipathd is shutting down before triggering their execution, and multipathd no longer terminates with a segmentation fault on shutdown.

BZ#467709

The multipathd daemon did not have a failover method to handle switching of path groups when multiple nodes were using the same storage. Consequently, if one node lost access to the preferred paths to a logical unit, while the preferred path of the other node was preserved, multipathd could end up switching back and forth between path groups. This update adds the followover failback method to device-mapper-multipath. If the followover failback method is set, multipathd does not fail back to the preferred path group, unless it just came back online. When multiple nodes are using the same storage, a path failing on one machine now no longer causes the path groups to continually switch back and forth.

Enhancements

BZ#737051

The NetApp brand name has been added to the documentation about the RDAC (Redundant Disk Array Controller) checker and prioritizer.

BZ#788963

The built-in device configuration for Fujitsu ETERNUS has been added.

BZ#760852

If the multipath checker configuration was set to tur, the checks were not performed asynchronously. If a device failed and the checker was waiting for the SCSI layer to fail back, the checks on other paths were kept waiting. The checker has been rewritten so as to check the paths asynchronously, and the path checking on other paths continues as expected.

BZ#799908

A built-in configuration for IBM XIV Storage System has been added.

BZ#799842

The NetApp LUN built-in configuration now uses the tur path checker by default. Also flush_on_last_del has been enabled, dev_loss_tmo has been set to infinity, fast_io_fail_tmo has been set to 5, and pg_init_retries has been set to 50.