Whamcloud - gitweb
pschwan [Wed, 27 Feb 2002 04:58:37 +0000 (04:58 +0000)]
Fix the waitqueue race in the mdc, I hope.
pschwan [Wed, 27 Feb 2002 04:26:22 +0000 (04:26 +0000)]
- llmount.sh creates for you a /tmp/ogdb in the hostfs, and waits for you to
press enter before it continues. This lets you set breakpoints and debug
the mount/startup process.
- Run 'mynid localhost' instead of 'mynid'
pschwan [Wed, 27 Feb 2002 04:16:27 +0000 (04:16 +0000)]
Add many debug messages to the RPC path
pschwan [Tue, 26 Feb 2002 22:32:29 +0000 (22:32 +0000)]
Added paths to obdctl's 'modules' command, making it immediately useful:
/r/usr/src/obd/utils/obdctl modules > /r/root/ogdb
braam [Mon, 25 Feb 2002 23:34:54 +0000 (23:34 +0000)]
- wrong offset in commit page
braam [Mon, 25 Feb 2002 20:45:32 +0000 (20:45 +0000)]
Wrong object id sent to obd - fixed
braam [Mon, 25 Feb 2002 18:57:03 +0000 (18:57 +0000)]
- uninitialized error code
- uninitialized mode of data objects
both fixed
braam [Mon, 25 Feb 2002 18:13:41 +0000 (18:13 +0000)]
minor changes to debug prepw and prepare page
remove some debug crud from the OST handler
braam [Mon, 25 Feb 2002 15:49:21 +0000 (15:49 +0000)]
- add the truncate obd call
pschwan [Mon, 25 Feb 2002 08:00:48 +0000 (08:00 +0000)]
Getting rid of tabs makes me happy, just humour me.
braam [Mon, 25 Feb 2002 07:43:46 +0000 (07:43 +0000)]
- minor debugging fixes
- write event BEFORE calling callback
pschwan [Sun, 24 Feb 2002 07:09:01 +0000 (07:09 +0000)]
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.
pschwan [Sun, 24 Feb 2002 06:36:26 +0000 (06:36 +0000)]
- Added DEBUG_SUBSYSTEMs
- started using CERROR, which goes both to the console and the ringbuffer,
instead of printk
pschwan [Sun, 24 Feb 2002 03:57:55 +0000 (03:57 +0000)]
Added a ``modules [path]'' command to obdctl, which is complete modulo the
fact that our directory names do not currently match our module names.
The best part? I did it with zero lines of kernel code and a function that
fits on one page.
braam [Sun, 24 Feb 2002 00:01:11 +0000 (00:01 +0000)]
slightly more debugging in rpc.c - something is wrong there.
pschwan [Sat, 23 Feb 2002 19:06:25 +0000 (19:06 +0000)]
- Converted lots of debugging printks to CDEBUGs
- Removed duplicated debug macros and defines; they live in portals now
- All of OBD's debug information now lives in the Portals debug buffer.
Use 'ptlctl get_debug' to view the buffer. The buffer's size is currently
hardcoded at 1 megabyte in module.c, but that's easy enough to change to a
module argument.
braam [Fri, 22 Feb 2002 23:35:51 +0000 (23:35 +0000)]
MDS wasn't handling refcount on vfsmount's correctly,
resulting in oops.
behlendo [Fri, 22 Feb 2002 02:25:59 +0000 (02:25 +0000)]
- I had set the rpc buffer size down to 4k to ease testing, the smaller
buffer size accidentally got commited with the other changes. This
change just resets it to the original 64k size.
behlendo [Fri, 22 Feb 2002 02:05:30 +0000 (02:05 +0000)]
- Added some fields for rpc double buffering
- Added a NEXT_INDEX() macro
- Adjusted mds/handler.c and ost/ost_handler.c to notify
the rpc layer when finished processing an event, this
ensures the buffers are recycled.
- Added support for double buffering
pschwan [Thu, 21 Feb 2002 23:15:18 +0000 (23:15 +0000)]
Pass 1 of debugging and leak cleanup:
- use OBD_ALLOC/OBD_FREE everywhere
- fixed a few leaks
- cleaned up a few macros
- load obdclass.o before ptlrpc.o now, since obdclass exports the memory counter
- fixed broken strdup()
- made llmount.sh a little more robust
braam [Thu, 21 Feb 2002 23:02:57 +0000 (23:02 +0000)]
*** empty log message ***
braam [Thu, 21 Feb 2002 23:02:32 +0000 (23:02 +0000)]
- 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
behlendo [Thu, 21 Feb 2002 00:19:31 +0000 (00:19 +0000)]
Updated 2.4.17 obd kernel patch.
braam [Wed, 20 Feb 2002 00:56:39 +0000 (00:56 +0000)]
*** empty log message ***
braam [Tue, 19 Feb 2002 16:43:53 +0000 (16:43 +0000)]
This adds most of the metadata infrastructure: I think that all commands
(mkdir, link, symlink, rm, rmdir, mknod, mv etc) now work. Probably the
server code is not multithread safe (particularly for renames, the others
might be OK). I expect other fragilities and perhaps some errors with
setting {a,m,c}times, owners etc.
- add the beginning of an obd command that prints the information
for UML gdb (class/class_obd.c)
- try to ignore errors on creating symbolic links in Makefiles
- the patch is 2.4.17 again for ext2obd, ghees, let's fix that.
- to avoid confusion, null out the buffer of result nio's in prepw
- add other reintegration functions.
braam [Thu, 14 Feb 2002 05:33:26 +0000 (05:33 +0000)]
Oops; I added the wrong page.c
braam [Thu, 14 Feb 2002 05:31:24 +0000 (05:31 +0000)]
- changed back to using no optimization for UML modules, and no -g2
- set inode operations in object create
- s/WRITE/OBD_BRW_WRITE/ where applicable
- added much of the infrastructure for page reads/writes
- update inode->i_size on the MDS if extending the file
- removed extra mntget() which hung the loop device
- misc bugfixes
- changed notify_change to i_op->setattr; otherwise we get the mtime and ctime
from the mds
- set reply lengths (or your replies get dropped!)
braam [Wed, 13 Feb 2002 23:17:05 +0000 (23:17 +0000)]
class/class_obd.c: small OBD_ATTACHED sanity cleanup; OBD_SET_UP fix.
include/linux/lustre_net.h: set the LUSTRE_NAL back to socknal
include/linux/class_obd.h: added full license boilerplate, #define cleanup
include/linux/obd_osc: there was some confusion with obd_ost; fixed.
- added bits to structures for ost_request
- added OSC/OST portals communication
- added osc_setattr()
- added rpc_unregister_service in an attempt to get a handle on cleanup
- brought tests/ostreq.sh into the 1990s
- fixed a trivial obdctl setattr bug
braam [Wed, 13 Feb 2002 20:31:30 +0000 (20:31 +0000)]
- set_page_dirty ifdef for kernels on either side of 2.4.10
braam [Sat, 9 Feb 2002 05:12:51 +0000 (05:12 +0000)]
Minor fixes to
- setup scripts to do Elan stuff
- a few changes to page cache stuff for 2.4.9 (we finally know how to #if it)
- the usual battle the ext2obd patch
- removed aclocal.m4 (enhances immunity from automake versions)
pschwan [Thu, 7 Feb 2002 21:24:33 +0000 (21:24 +0000)]
New scripts to aid in testing; on one machine you should be able to run
sh tests/llmount-server.sh
and on another machine:
sh tests/llmount-client.sh
mount -t lustre_light -o device=3 none /mnt/obd
and do things like:
ls -ld /mnt/obd
chmod 777 /mnt/obd
This works now--even over the socknal, on an SMP kernel (although I haven't
tested it on an SMP machine)
You'll need to edit the SERVER variable in llmount-client.sh. If you want
to get really fancy, you'll make it a command-line option.
braam [Wed, 6 Feb 2002 18:25:41 +0000 (18:25 +0000)]
Minor cleanups and debugging statements added.
braam [Fri, 1 Feb 2002 19:32:21 +0000 (19:32 +0000)]
- zero out the request structure after allocation
- do readpage via portals
- added bulk moving functionality to rpc.c
- cleaned up one small MD leak
braam [Fri, 1 Feb 2002 18:20:06 +0000 (18:20 +0000)]
lllocalmount: startup and mount a lustre lite FS with no networking
llmountcleanup: indeed.
braam [Fri, 1 Feb 2002 18:18:06 +0000 (18:18 +0000)]
Minor bugfix to allow multiple networked operations:
- small fix to avoid crash in getattr
- properly unlink MDs and MEs after we send buffers
- fixups to the test scripts
braam [Thu, 31 Jan 2002 21:27:21 +0000 (21:27 +0000)]
- new RPC infrastructure:
- struct service with callback functions, buffers and Portals support
- services are Portal aware, but the old null transport may still work.
- lustre peer moved to Portals, as was all connection management
-
- fixed ptl entry constants for incoming REQUESTS, REPLIES for MDS, OST etc.
- all rpc's now have a peer destination (which can be NULL)
- replies are now pre-allocated
- the XID's are used to number transactions _and_ as match entries
- packet movement differentiates Portals and non-portals case.
- reception of packets uses the new Portals callbacks
- module initialization finds peer's by UUID. For now "mds" is a UUID.
- callbacks are used for (1) incoming requests (2) freeing outgoing packets
(3) incomding reply. Code reuse was possible.
- adapted all the test scripts (mdc and llmount)
braam [Wed, 30 Jan 2002 21:36:01 +0000 (21:36 +0000)]
- and the new files
braam [Wed, 30 Jan 2002 21:33:40 +0000 (21:33 +0000)]
Changes to request processing:
- all requests are now ptlrpc requests.
- better packing/unpacking once more
- new directory with Portals RPC infrastructure.
- wait and restart hooks event handlers
braam [Mon, 28 Jan 2002 19:06:15 +0000 (19:06 +0000)]
Changes for file creation and small fixes.
braam [Thu, 24 Jan 2002 08:04:57 +0000 (08:04 +0000)]
- more systematic unpacking/packing and handling of update records
- create working on MDS (needs some refinements to store data object id)
braam [Wed, 23 Jan 2002 18:20:23 +0000 (18:20 +0000)]
Add missing files, cleanup .cvsignore's
braam [Wed, 23 Jan 2002 04:59:12 +0000 (04:59 +0000)]
It's been a good day: chmod/chown and friends now work for Lustre Light.
- most of the changes are in the packing. It will take a little
bit more time before this completely settles I think, but we have
a pretty efficient system now.
- added initial infrastructure for MDS update records.
- the MDS server has a new routine to change the attributes: the associated
RPC call is MDS_REINT.
- Lustre Light is simpler than OBDFS since it doesn't have to
change inodes when it creates them: that's part of the MDS transactions we
send over -- removed many "change inode" and "setattr" variants.
braam [Mon, 21 Jan 2002 23:58:26 +0000 (23:58 +0000)]
assorted bug fixes and a working directory readpage routine
for Lustre Light.
braam [Mon, 21 Jan 2002 05:52:56 +0000 (05:52 +0000)]
Nasty cflag confusion in configure.ac.
braam [Mon, 21 Jan 2002 02:33:30 +0000 (02:33 +0000)]
defaults back to format
braam [Mon, 21 Jan 2002 02:31:40 +0000 (02:31 +0000)]
small fix.
braam [Mon, 21 Jan 2002 01:39:01 +0000 (01:39 +0000)]
fix the obdfssetup script to use obdext2 (not filter) for now
patch - revision one: pagecache_lock was missing. What else is missing?
braam [Mon, 21 Jan 2002 01:36:04 +0000 (01:36 +0000)]
cleanups coming from the new functions/macros
braam [Mon, 21 Jan 2002 01:17:17 +0000 (01:17 +0000)]
Fix a broken macro.
braam [Mon, 21 Jan 2002 00:48:30 +0000 (00:48 +0000)]
this minimal patch might actually work.
filter obd won't work: we don't need it now.
braam [Sun, 20 Jan 2002 05:03:56 +0000 (05:03 +0000)]
Small fixes to get lustre_light mount working over the request handling
infrastructure.
braam [Sun, 20 Jan 2002 00:49:49 +0000 (00:49 +0000)]
A few missing files...
braam [Sun, 20 Jan 2002 00:45:19 +0000 (00:45 +0000)]
Small fixes to the request processing.
Split off the metadata client into a separate module, probably easier
to test things that way.
braam [Fri, 18 Jan 2002 06:01:30 +0000 (06:01 +0000)]
Fix small typos in ost/osc
braam [Thu, 3 Jan 2002 18:39:07 +0000 (18:39 +0000)]
omitted test script
braam [Thu, 3 Jan 2002 18:37:25 +0000 (18:37 +0000)]
forgotten two files
braam [Thu, 3 Jan 2002 17:10:42 +0000 (17:10 +0000)]
Small bug fixes. OSC/OST almost working.
braam [Sun, 30 Dec 2001 18:18:33 +0000 (18:18 +0000)]
it _compiles_ finally on 7.1 and 7.2, but the solution is not great.
braam [Sun, 30 Dec 2001 06:43:06 +0000 (06:43 +0000)]
Lustre light client request module
braam [Sun, 30 Dec 2001 06:17:15 +0000 (06:17 +0000)]
Most of the code for the OST target.
lib/obd_pack.c - routines for packing/unpacking of OST requests
lib/mds_pack.c - fixes (should be 64bit safe now)
autogen.sh - no more Makefile.in in CVS
tests/testreq.c tests/umreq.sh tests of the RPC infrastructure
braam [Sat, 29 Dec 2001 00:20:44 +0000 (00:20 +0000)]
- fixes to the MDS: it now answers requests from the request test
program.
- beginning of OST client/server code - needs work.
braam [Fri, 28 Dec 2001 04:41:15 +0000 (04:41 +0000)]
A solid part of the MDS request processing infrastructure.
- packing/unpacking and definition of requests
- a server daemon
- getattr call coded
Lustre idl file created:
- move all interface definitions here
- that's ioctl's, mds requests/replies & obd requests/replies
Makefiles:
- configure.in can handle user mode linux now
utils:
- fixed the obdctl library to support old version of readline.
braam [Mon, 17 Dec 2001 15:00:39 +0000 (15:00 +0000)]
small changes and fixes to get the filter OBD working fully.
It seems to work well under ext2, XFS and Reiser at the moment.
braam [Thu, 13 Dec 2001 22:21:05 +0000 (22:21 +0000)]
- 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
braam [Tue, 11 Dec 2001 21:08:42 +0000 (21:08 +0000)]
Numerous changes:
- new control utility - no Perl
- filter obd halfway coded ; requires new patches to kernel
- big cleanup in the class driver with better packed ioctls
- macros to pack other things, like network packets
braam [Thu, 8 Nov 2001 19:54:43 +0000 (19:54 +0000)]
Two small changes to support large files.
- internally open files with O_LARGEFILE
- s_maxbytesize set to 64G (should do info for that).
braam [Thu, 8 Nov 2001 19:11:49 +0000 (19:11 +0000)]
large file printk cleanups
braam [Thu, 8 Nov 2001 18:56:34 +0000 (18:56 +0000)]
fix minor kmap/kunmap mismatch.
braam [Tue, 6 Nov 2001 19:05:51 +0000 (19:05 +0000)]
Minor compilation fixes for RedHat 2.4.9
Fix reference decreasing bug in namei.c
braam [Tue, 6 Nov 2001 18:42:41 +0000 (18:42 +0000)]
minor fix to the patch
braam [Tue, 6 Nov 2001 04:20:34 +0000 (04:20 +0000)]
created
braam [Mon, 5 Nov 2001 22:20:52 +0000 (22:20 +0000)]
If you free a string ... it can cost a lot of time.
braam [Mon, 5 Nov 2001 07:35:40 +0000 (07:35 +0000)]
many changes: seems ok on 2.4.9
gord-fig [Thu, 25 Oct 2001 16:50:17 +0000 (16:50 +0000)]
Back out size/blocks changes.
gord-fig [Wed, 24 Oct 2001 20:51:25 +0000 (20:51 +0000)]
Segregate block and size counting between obdfs and obdext2.
gord-fig [Wed, 24 Oct 2001 19:38:29 +0000 (19:38 +0000)]
Remove debugging clutter.
braam [Sun, 21 Oct 2001 02:52:05 +0000 (02:52 +0000)]
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.
gord-fig [Fri, 19 Oct 2001 22:54:38 +0000 (22:54 +0000)]
Again, only use loop device for testing.
braam [Fri, 19 Oct 2001 18:54:07 +0000 (18:54 +0000)]
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.
braam [Fri, 19 Oct 2001 16:06:28 +0000 (16:06 +0000)]
fairly massive change to remove write_inode methods and switch
to setattr everywhere -- dbench 50 completes, so does bonnie++.
braam [Fri, 19 Oct 2001 05:17:34 +0000 (05:17 +0000)]
fix statfs bug and a truncation issue.
braam [Fri, 19 Oct 2001 04:13:53 +0000 (04:13 +0000)]
fix setattr bug; check in symlink changes
gord-fig [Thu, 18 Oct 2001 23:38:29 +0000 (23:38 +0000)]
Don't use printk unconditionally.
gord-fig [Thu, 18 Oct 2001 22:11:07 +0000 (22:11 +0000)]
Add code to help debug the rmmod hang.
gord-fig [Wed, 17 Oct 2001 22:36:28 +0000 (22:36 +0000)]
Consolidate variables into a structure.
gord-fig [Wed, 17 Oct 2001 17:11:15 +0000 (17:11 +0000)]
Convert signals in pupdated into wait queues.
gord-fig [Wed, 17 Oct 2001 00:28:29 +0000 (00:28 +0000)]
Remove the obdo->o_rdev member, and store rdev numbers in the inline data.
gord-fig [Wed, 17 Oct 2001 00:02:40 +0000 (00:02 +0000)]
Mark potentially obsolete functions.
braam [Tue, 16 Oct 2001 05:27:35 +0000 (05:27 +0000)]
fix for the symlink problem.
braam [Tue, 16 Oct 2001 03:51:52 +0000 (03:51 +0000)]
debugging fixes and small memory leak fix.
braam [Mon, 15 Oct 2001 07:31:34 +0000 (07:31 +0000)]
one more high memory kmap missing.
braam [Mon, 15 Oct 2001 05:51:19 +0000 (05:51 +0000)]
more page cache changes
braam [Mon, 15 Oct 2001 05:33:22 +0000 (05:33 +0000)]
ext2 page cache updates
braam [Sun, 14 Oct 2001 23:50:11 +0000 (23:50 +0000)]
highmemory changes
braam [Sun, 14 Oct 2001 23:49:06 +0000 (23:49 +0000)]
high memory fixes
braam [Fri, 12 Oct 2001 20:27:16 +0000 (20:27 +0000)]
attempted bug fixes for 2.4.9.
gord-fig [Fri, 12 Oct 2001 19:44:57 +0000 (19:44 +0000)]
Update RPM building.
braam [Wed, 10 Oct 2001 18:02:22 +0000 (18:02 +0000)]
- ext2_obd.c --- fix the bugs in read/write for Linux 2.4.3
gord-fig [Mon, 8 Oct 2001 16:53:53 +0000 (16:53 +0000)]
Convert to Automake.
gord-fig [Mon, 1 Oct 2001 17:06:58 +0000 (17:06 +0000)]
Create lustre-tools and lustre-modules packages.