RELEASE 1.0.15 To upgrade, do: # rpm -Uvh # depmod -a The latter rebuilds the module dependencies. The MINimum kernel version required for the Red Hat kernels are as follows: 1. Red Hat AS 2.1: 2.4.9-e.57 (U6) 2. Red Hat EL 3.0: 2.4.21-27.EL (U4) For SuSE, the module needs to be built for each kernel version. OCFS rpms for the latest SLES8 kernels, 2.4.21-295 and 2.4.21-304, on all supported platforms (i686, IA64, x86_64) have been provided. Running a mix of nodes with 1.0.9 or earlier and 1.0.15 is NOT recommended. So, when upgrading from 1.0.9 or earlier, please umount all the volumes on all the nodes, before mounting them with 1.0.15. Please also run fsck.ocfs during the upgrade. The same restriction does not apply for 1.0.10 or later. That is, if all the nodes are on 1.0.10 or later, the nodes could be upgraded to 1.0.15 in a rolling manner. This release does not affect the ocfs-support or ocfs-tools packages. The latest release of ocfs-support is 1.1.5 and of ocfs-tools is 1.0.10. (There is no difference between ocfs-tools 1.0.10 and 1.1.2.) BUG# DESCRIPTION ------- ----------------------------------------------------------- 4590449 OFFLINE BACKUP OF 10.1 RAC DB TO OCFS LOCATION GIVES ORA-19502 AND ORA-27061 4871665 (DUP 4590449) RMAN FAILS TO WRITE ON ASYNC OCFS PARTITIONS IN RHAS 3 (U6) AMD64BIT ORA-19502 4440899 READING LAST BLOCK ON A RECENTLY EXTENDED FILE RETURNS EOF =============================================================================== RELEASE 1.0.14 With this release, OCFS fully supports the use of async io (aio) with the Oracle RDBMS. Apart from this, there has been no other real change in 1.0.14. So, if you are not planning to use aio, an upgrade from 1.0.13 is not necessary. However, if you are planning to use aio, please ensure that you also upgrade your kernel to the one listed. For Redhat AS 2.1, the minimum kernel supported is 2.4.9-e.57 (U6). For Redhat EL 3.0, it is 2.4.21-27.EL (U4). For Suse, the appropriate kernel version(s) is listed on the download page. Running a mix of nodes with 1.0.9 or earlier and 1.0.14 is NOT recommended. So, when upgrading from 1.0.9 or earlier, please umount all the volumes on all the nodes, before mounting them with 1.0.14. Please also run fsck.ocfs during the upgrade. The same restriction does not apply for 1.0.10 or later. That is, if all the nodes are on 1.0.10 or later, the nodes could be upgraded to 1.0.14 in a rolling manner. =============================================================================== RELEASE 1.0.13 The 1.0.12 upgrade instructions apply to 1.0.13 too. There are two changes in this release, both concerning the size of memory chunks allocated. First concerns the memory block allocated to r/w directories. Whereas the older release allocated 128K block of memory, the newer release allocs PAGE_SIZE blocks. Whereas the overall size of memory allocated has not changed, the number of contiguous pages required for each allocation has. In the newer scheme, the 128K is made up of 32 4K pages (on the ia32 systems). This has been done to take care of the numerous ENOMEMs experienced due to lowmem fragmentation on heavily loaded systems. The second change concerns the amount of memory allocated to hold the lockres hashtable. Now, instead of 32 4K pages, it requires 8 4K pages. Once again, this has been done to take care of the ENOMEM error one encounters when mounting ocfs volumes on heavily loaded systems. =============================================================================== RELEASE 1.0.12 Running a mix of nodes with 1.0.9 or earlier and 1.0.12 is NOT recommended. So, when upgrading from 1.0.9 or earlier, please umount all the volumes on all the nodes, before mounting them with 1.0.12. Please also run fsck.ocfs during the upgrade. The same restriction does not apply for 1.0.10 or later. That is, if all the nodes are on 1.0.10 or later, the nodes could be upgraded to 1.0.12 in a rolling manner. Though it is not required, it would be advisable to reformat the archive log volume. This release fixes two small races which could have resulted in corrupting the directory index on disk. This will result is lost space (128k per directory node). In 1.0.12, we do not add new objects to dirnodes having corrupted dirindex. Instead we add the file to a new dirnode linked to the previous. For support and tools rpms, you can either continue with 1.0.10-1 or upgrade to the 1.1.2-1 packages available in the OCFS Tools project. BUG# DESCRIPTION ------- ----------------------------------------------------------- 3619877 OS CRASHES WITH KERNEL PANIC DURING MOUNTING OCFS 1.0.10-1 3553340 ARCHIVE LOGS ARE BEING LOST ON OCFS 3530519 DATAFILE RESIZE DOES NOT RELEASE SPACE IN OCFS PARTITION 3464802 EXCLUSIVE LOCK ON DIRNODE NOT RELEASED AFTER FILECREATE 3437651 ERROR STATUS = -2 WHEN GENERATING SMALL ARCHIVELOGS ON OCFS =============================================================================== RELEASE 1.0.11 Running a mix of nodes with 1.0.9 or earlier and 1.0.11 is NOT recommended. So, when upgrading from 1.0.9 or earlier, please umount all the volumes on all the nodes, before mounting them with 1.0.11. Please also run fsck.ocfs during the upgrade. The same restriction does not apply for 1.0.10. That is, if all the nodes are on 1.0.10, the nodes could be upgraded to 1.0.11 in a rolling manner. BUG# DESCRIPTION ------- ------------------------------------------------ 3467544 CSS FAILS TO FLUSH WRITES TO VOTING FILE ON OCFS =============================================================================== RELEASE 1.0.10 The 1.0.10 release of ocfs tools automatically bind the device (/dev/sdXX) to raw (/dev/raw/rawX). This has been done to ensure consistent direct io behavior on all flavors of linux. So, if you encounter an error, like, "Error opening /dev/rawctl" or "unable to find a free raw device /dev/raw/rawXX", please check the raw device setup (raw -qa). The tools will not do the binding if the device provided is a raw device (/dev/raw/rawX). BUG# DESCRIPTION ------- ----------------------------------------------------------------------- 3041091 DD AND CP HANDLE ENOSPC DIFFERENT WITH OCFS 3041094 MV ENDS UP COPYING FILES 3068104 ERROR: STATUS = -999, COMMON/OCFSGENALLOC.C DURING RAC STRESS 3134746 TRYING TO MOUNT OCFS DRIVE CAUSES KERNEL PANIC 3390683 CANNOT DELETE DIRECTORY ON OCFS MOUNTED VOLUME 3393027 UNMOUNT OF OCFS VOLUME GIVES STATUS= -999 3401940 OOPS IN FREE_KIOVEC_SZ [KERNEL] 0X26 3181864 MOUNT OCFS VOLUME CAUSES KERNEL PANIC: UNABLE TO HANDLE KERNEL NULL PTR 3215119 DO_IRQ WARNING MESSAGES COMBINED WITH CAT /PROC/OCFS/HASHSTAT 3319129 OCFS HEADER CORRUPTION ON SEVERAL FILESYSTEMS 3329565 OCFS ERROR SHOULD BE REPORTED AS A WARNING =============================================================================== RELEASE 1.0.9-12 Fixes a potential corruption with large, aligned, direct I/Os, for example Oracle redo logs or direct path SQL loader. =============================================================================== RELEASE 1.0.9-11 Fixes a potential race condition in lockresources on a local node. performance improvement in aio code (improved blockgrouping) =============================================================================== RELEASE 1.0.9-9 BUG# DESCRIPTION ------- ---------------------------------------------------------------------- 3198353 DO_IRQ Stackoverflow warnings, unexpected hangs This release limits kernel stack usage by a large amount, which will reduce the possibilities of ocfs causing kernel stack overflow warnings or even potential actual stack overflows (critical recommended) =============================================================================== RELEASE 1.0.9-8 Fix for space bitmap corruption when having highly concurrent filecreate/extended on multiple nodes. (critical) If you archive from multiple nodes to the same filesystem it is very much adviced to apply this bugfix update BUG# DESCRIPTION ------- ---------------------------------------------------------------------- 3188271 OCFS INCOMPLETELY ARCHIVED LOG =============================================================================== RELEASE 1.0.9-7 BUG# DESCRIPTION ------- ---------------------------------------------------------------------- 3038923 LOADING OF MODULE OCFS 109 FAILS WITH UNRESOLVED SYMBOL BRW_KVEC_ASYNC_RSMP_DA53 3151969 /ETC/INIT.D/OCFS LINK FAILS ON UNIPROCESSOR 2.4.9-E.24 KERNEL 3143177 STAT() CALL ON OCFS FILES RETURNS NON-ZERO ST_RDEV VALUE FOR NON-RAW DEVICES 3132635 LOAD_OCFS DOESN'T WORK IF LANG<>'EN_US' ERRORS: IP ADDRESS X IS NOT ASSOCIATED. 3152010 LOAD_OCFS FAILS W/ UNRESOLVED SYMBOLS ON 2.4.9-E.24 UNIPROCESSOR KERNEL 3146671 OCFS ARCHIVE DESTINATION NOT RESPONDING TO UNIX COMMANDS 3160222 NET DLM DOES NOT FAILOVER TO DISK ON SOME CONDITIONS =============================================================================== RELEASE 1.0.9-6 Fix for re-use datafiles in oracle with large fragmented datafiles Fix in ocfs.init script during upgrades Fix for space allocation on files that are fragmented Fix for race condition in parallel allocates across nodes ===============================================================================