RHBA-2014:1853 - Bug Fix Advisory

Topic

Updated glusterfs, gluster-nfs, glusterfs-fuse, glusterfs-geo-replication, and
redhat-storage-server packages that fix multiple bugs are now
available for use with the Red Hat Storage Server 2.1.

Description

Red Hat Storage is software-only, scale-out storage that provides flexible
and affordable unstructured data storage for an enterprise. GlusterFS, a
key building block of Red Hat Storage, is based on a stackable user-space
design and can deliver exceptional performance for diverse workloads.
GlusterFS aggregates various storage servers over network interconnections
into one large, parallel network file system.

This advisory addresses the following bugs:

• Previously, running the setfacl command on files and directories on

the NFS mount point caused a memory leak in the Gluster NFS server.
When the command was run on a large number of files, the out of memory
killer would get invoked, eventually terminating the Gluster NFS server.
With this fix, the possibility of a memory leak is fixed and the command
works as expected. (BZ#1125658)

• Previously, upon the Red Hat Storage server reboot, the Gluster

Management Daemon checked for the "rebalance command" value instead of
the "rebalance operation status" value to determine whether to restart
the rebalance process. With this fix, the Gluster Management Daemon
checks the rebalance operation status and determines an appropriate
action. As a result, in the event of a server reboot, a rebalance
process that is complete, would not restart.(BZ#1136310)

• Previously, if a file was linked by one client and removed by another,

the subsequent lookup operation from the first client did not override
the cache in the absence of the unlinked name on the bricks, leading it
to conclude that the file name exists. With this fix, the stale inode
mapping is deleted when the lookup operation fails with the ENOENT error
in the first client. (BZ#1122649)

• Previously, When a file was deleted from a master volume after being

renamed, the rename system call processed it internally as an UNLINK
operation in the slave volume. As a result, if a file was created with
the same name in the master volume, it did not get synchronized to
the slave volume and the file had the old GFID associated with
it. With this release, the rename system call is handled appropriately
and the rename operation synchronizes from the master volume to the
slave volume as expected. (BZ#1060683)

• Previously, if there was an attempt to delete a non-existent file,

the GFID-based access returned a ESTALE error code instead of ENOENT.
The ESTALE error was not handled properly in Geo-replication causing
the Geo-replication worker thread to fail upon such an attempt. With
this release, the error codes are handled appropriately and the
Geo-replication worker thread does not fail with an
ESTALE error.(BZ#1129392)

• Previously, due to a race condition between the lookup operations

of multiple rebalance processes spawned on each of the servers, each
rebalance process migrated a file only if the file was present on the
server where it is spawned. The rebalance process copied a file from
the source to destination and on the destination directory, it created
a linkto file. During a small window of time, the lookup operation of
the other rebalance process assumed the linkto file to be a dangling
reference and unlinked it; during which the linkto file was converted
to a data file leading to data loss. With this fix, the race conditions
are handled properly and the data loss is not observed. (BZ#1115937)

• Previously, there was a 100% CPU utilization and continuous memory

allocation that made the Gluster process unusable and caused a very
high load on the Red Hat Storage Server and possibly rendering it
unresponsive to other requests. This was due to the parsing of a Remote
Procedure call (RPC) packet containing a continuation RPC-record causing
an infinite loop in the receiving Gluster process. With this release,
such RPC-records are handled appropriately and do not lead to service
disruptions. (BZ#1146466)

Users of Red Hat Storage are advised to upgrade to these updated packages,
which fix these issues.

Solution

Before applying this update, make sure all previously released errata
relevant to your system have been applied.

This update is available via the Red Hat Network. Details on how to
use the Red Hat Network to apply this update are available at
https://access.redhat.com/articles/11258

Affected Products

• Red Hat Gluster Storage Server for On-premise 2.1 x86_64
• Red Hat Storage for Public Cloud (via RHUI) 2.1 x86_64

Fixes

• BZ - 1122649 - Hard links not deleted completely in fuse-based client
• BZ - 1144413 - High memory usage by rebalance process
• BZ - 1146889 - DHT :- data loss - file is missing on renaming same file from multiple client at same time
• BZ - 1146895 - Renaming file while rebalance is in progress causes data loss
• BZ - 1154019 - DHT: Rebalance- Rebalance doesn't start if we have any mounts of the volume
• BZ - 1157705 - BVT: Connectathon i.e. Cthon basic test fails over NFS when quota is enabled
• BZ - 1159279 - DHT: Rebalance + quota:- glusterfs crash after remove- brick on a quota enabled volume
• BZ - 1159280 - DHT: Rebalance- Rebalance process crash after remove-brick

CVEs

(none)

References

• https://access.redhat.com/site/documentation/en-US/Red_Hat_Storage/2.1/html/Technical_Notes/index.html