Remove ancient and entirely non-working obdfs code from the HEAD.
Use the OBD_MD_FLTYPE and OBD_MD_FLMODE flags where appropriate.
This commit contains probably 92% of the striping infrastructure we need initially. The most pervasive change is the introduction of "lov_stripe_md" throughout the code. In addtion several small little bugs were nailed in the locking -- more are outstanding. The setup scripts are not yet capable of running this code. Kernel patches were updated to include LOOKUP (to let runtests.sh work).
Pass 64-bit object numbers wherever possible, instead of truncating to 32-bit.
64-bit warning fixes. Someone should take a closer look at ext2_obd.c
- name switch: "light" --> "lite" "/mnt/obd" --> "/mnt/lustre" - small changes to RPC code to handle request packet freeing later, separate connection initialization and connect. - rename event handlers and queues, add separate request_out and reply_out queue
We touched everything, but it's not as scary as it looks. - dramatically reorganized and simplified the headers. Most of the nasty recursive definitions are gone, and obd_class.h no longer drags in the entirety of Lustre. - .cvsignores updated and cleaned - added another chunk of DLM code and a test, which passes (note that it doesn't free the locks yet, so don't expect to cleanup afterwards) - fixed OBD_ALLOC/OBD_FREE in the same ways as the Portals macros - removed a lot of dead obd_class code
The most tedious patch in history. Changes almost all of the 'printk's to 'CERROR's. Also contains a lot of cleanup, removing things like __FUNCTION__ from CDEBUG and CERROR statements which already display that information.
- Added DEBUG_SUBSYSTEMs - started using CERROR, which goes both to the console and the ringbuffer, instead of printk
- elimininate the system calls from filter obd - use module_init macros not old style init_module functions - remove unused crud from lustre_light.h use the good stuff more - move directory readpage operations to dir.c - cleanup rw.c significantly; it's actually understandable now. - attribute handling: indicate to setattr to call truncate on OBD or not - add osc destroy - fix mds_reint_unlink (checked mode wrongly) - add ll_delete_inode (to also go to the OST to remove data object) - add truncate.c test program to tests - avoid using ambigous READ/WRITE constants
cleanups coming from the new functions/macros
Fix a broken macro.
small changes and fixes to get the filter OBD working fully. It seems to work well under ext2, XFS and Reiser at the moment.
- obd filter works over ext2, but bonnie++ is buggie - obd filter doesn't work over reiser - obdctl now works and is taking over from obdcontrol
Minor compilation fixes for RedHat 2.4.9 Fix reference decreasing bug in namei.c
many changes: seems ok on 2.4.9
Back out size/blocks changes.
Segregate block and size counting between obdfs and obdext2.
Another very major cleanup: - new lookup/new_inode code based on iget4 (saves RPC's) - reorganized the layout of ext2_obd.c - the bonnie++ bug can be fixed by patching the 2.4.3 kernel.
This fixes a race between killing inodes in obdfs and in ext2obd. When files are kept open in obdfs it is important not to propagate the 0 link count to ext2obd since that would delete the inode prematurely. This one seems to run dbench 50 ok.