Description

Solaris 11, Solaris 11.1  and ZFS Storage Appliances (ZFSSA) running software version 2011.1.3, 2011.1.4 or 2011.1.4.1 may have the size of regular files incorrectly set following a system panic or power failure. This also includes a failover and takeover event for ZFSSA.  In some cases, this may result in data loss on the affected files indicating an EIO return to the caller or a system panic with the message "zfs: accessing past end of object."

The event rarely occurs, and only under the following conditions:

    - An unclean shutdown of the system (panic/power fail/failover event)

    - More than 1 file being written to (at the time of the panic or power failure event)

    - The writes must be synchronous (O_SYNC/O_DSYNC, fsync(3C), sync=always property)

    - The sequence of transaction representing the writes and their offsets must be such as to cause parallel execution of threads replaying the transactions (i.e. a race condition)

    - The writes are not block-aligned (for example Oracle DB will not exhibit this issue)

Occurrence

This issue can occur in the following releases:

SPARC Platform

• Solaris 11
• Solaris 11.1 without SRU 3.4

x86 Platform

• Solaris 11
• Solaris 11.1 without SRU 3.4

ZFSSA

• 2011.1.3  (Version string 2011.04.24.3.0,1-1.19)
• 2011.1.4  (Version string 2011.04.24.4.0,1-1.21)
• 2011.1.4.1  (Version string 2011.04.24.4.1,1-1.26)

Notes:

    1. ZFSSA version 2011.1.4.2 is not affected by this issue.
    2. Solaris 8, 9 and 10 are not affected by this issue.

To determine the ZFSSA software version running, use the following:

From the CLI:

    > configuration version get version

From the BUI, navigate to:

    Maintenance - > System

Symptoms

If the described issue occurs, the affected files will have their size set incorrectly. The effect of this is application dependent. In other cases, accessing the file may also trigger a system panic with a message similar to the following: 

where the numbers aaaa, bbbb, cccc, dddd and eeee will vary.

Workaround

There is no workaround for this issue.

For Solaris, this issue is addressed in the following releases:

SPARC Platform

• Solaris 11.1 with SRU 3.4 or later

x86 Platform

• Solaris 11.1 with SRU 3.4 or later

ZFSSA

• 2011.1.4.2 (version string 2011.04.24.4.2) or later

Note: The current version for ZFSSA software is 2011.1.5.0 (version string 2011.04.24.5.0) which can be downloaded from the Software Updates page in <Document:2021771.1> and as patch 1608900 from My Oracle Support.

History

31-Oct-2012: Date of Workaround release
08-Nov-2012: Correction to version string for ZFSSA in "Occurrence"
15-Nov-2012: Maintenance update; no change in content, Additional Product attributions
19-Nov-2012: Notes added in Occurrence and Workaround for ZFSSA version 2011.1.4.2
15-Jan-2013: Removed A7000 from Product Attribution
22-Jan-2013: Updated for Solaris 11 information, issue is Resolved
25-Jan-2013: Updated for new/additional ZFSSA upgrade information

This regression introduced by the putback for BugID: 7044065

The cause of this regression: 7044065 (<BUG:15765449>):SUNBT7044065) was also putback to Sol 10 and so it looks like S10 may also
be affected. However, 7044065 was only putback to the feature gate and the fix has been included
in Sol 10 in time, before update 11 was finished, so that no affected patch was released.

Questions regarding this document should be addressed to:
sunalertpublication_us_grp@oracle.com and copy the
responsible engineer listed below.

Internal Contributor/Submitter: bhaskar.sarkar@oracle.com, christian.rasmussen@oracle.com
Internal Eng Responsible Engineer: neil.perrin@oracle.com
Oracle Knowledge Analyst: david.mariotto@oracle.com
Internal Eng Business Unit Group: Systems NPE
Internal Pending Patches: TBD

References

This solution has no attachment