Whamcloud - gitweb
gord-fig [Fri, 26 Jul 2002 05:14:54 +0000 (05:14 +0000)]
Another shot at CVS tag substitution.
rread [Fri, 26 Jul 2002 05:06:28 +0000 (05:06 +0000)]
- use llmodules.sh to load modules now
- all networking config (except for the modules) is now in lconf
gord-fig [Fri, 26 Jul 2002 03:14:45 +0000 (03:14 +0000)]
Properly munge CVS/Tag.
pschwan [Fri, 26 Jul 2002 00:44:18 +0000 (00:44 +0000)]
- Fix some dangling lock references in error paths
- If ldlm_cli_cancel fails in the cleanup path, force the lock cleanup
pschwan [Fri, 26 Jul 2002 00:06:00 +0000 (00:06 +0000)]
Don't LBUG() in resource_cleanup() when cancel fails
pschwan [Thu, 25 Jul 2002 23:34:29 +0000 (23:34 +0000)]
set b_page in osc_brw_write, so that it will get kunmapped when complete
shaver [Thu, 25 Jul 2002 23:34:18 +0000 (23:34 +0000)]
Only interrupt after timeout in ptlrpc_queue_wait. Fixed bug 583214.
rread [Thu, 25 Jul 2002 21:48:16 +0000 (21:48 +0000)]
add ldlm
braam [Thu, 25 Jul 2002 21:35:24 +0000 (21:35 +0000)]
Fix some simple errors and warnings
- LOV's mount again.
- set some unitialized variables to 0
adilger [Thu, 25 Jul 2002 21:35:21 +0000 (21:35 +0000)]
Updates to the extN patch that should resolve the corrpution problem with
indexed directories. We do not enable indexed directories in new_fs yet.
pschwan [Thu, 25 Jul 2002 21:02:33 +0000 (21:02 +0000)]
- the LBUG() after ll_lock "fails" is no longer necessary.
- avoid crashing in ll_create_node if there's no intent
gord-fig [Thu, 25 Jul 2002 20:20:23 +0000 (20:20 +0000)]
Transform *.lin into *.lyx, updating @VERSION@, @TAG@, and @DATE@.
Make generated .lyx files intermediates, so that they don't confuse people.
Regenerate documentation if $(VERSION), or CVS/Tag changes.
braam [Thu, 25 Jul 2002 19:20:36 +0000 (19:20 +0000)]
- update change logs
braam [Thu, 25 Jul 2002 19:07:37 +0000 (19:07 +0000)]
- align the UUID in the ost requests
- document the LDLM wire protocol, first draft.
braam [Thu, 25 Jul 2002 18:03:54 +0000 (18:03 +0000)]
- Remaining documentation updates for Client - OST protocol in 0.5.1
- Move type of lustre_msg to correct spot (alignment)
- Change ost statfs field to uuid.
This version will be tagged as v0_5_1
braam [Thu, 25 Jul 2002 16:05:45 +0000 (16:05 +0000)]
- remove two almost unused fields from the OST request packets
- this makes separate opcodes out of OST_BRW: OST_WRITE and OST_READ
- documentation updates, still in progress for 0.5.1
pschwan [Wed, 24 Jul 2002 20:34:08 +0000 (20:34 +0000)]
fix ldlm_cli_cancel return value
b=583873
adilger [Wed, 24 Jul 2002 20:05:29 +0000 (20:05 +0000)]
Fix RPC request leak and potential refcounting problems on bulk descriptor.
adilger [Wed, 24 Jul 2002 19:04:47 +0000 (19:04 +0000)]
Missed one...
adilger [Wed, 24 Jul 2002 19:04:19 +0000 (19:04 +0000)]
Initialize bulk descriptor refcount to 1, and call it "desc" instead of
"bulk" for consistency.
adilger [Wed, 24 Jul 2002 16:43:48 +0000 (16:43 +0000)]
Make a distinction between bulk callbacks and brw callbacks - the bulk
callbacks take a ptlrpc_bulk_desc as a parameter, and brw in general has
no idea about that (not that we use the brw callback yet). Also add a
data parameter for the brw callback, otherwise it is mostly useless.
Some minor prep work for fixing the osc_brw_{read,write} request leak.
adilger [Wed, 24 Jul 2002 16:10:41 +0000 (16:10 +0000)]
Actually use the COUNT value.
shaver [Wed, 24 Jul 2002 16:02:11 +0000 (16:02 +0000)]
Don't deref desc after we drop our reference to it.
shaver [Wed, 24 Jul 2002 15:48:16 +0000 (15:48 +0000)]
Unmap and decref bulk descriptors from a bottom half, to fix I/O on
highmem machines.
adilger [Tue, 23 Jul 2002 23:53:19 +0000 (23:53 +0000)]
Allow runtests to be configured for a maximum number of files to copy.
This aids in tracking leaks and such, so the logs don't overflow.
adilger [Tue, 23 Jul 2002 23:44:51 +0000 (23:44 +0000)]
Fix patch context.
adilger [Mon, 22 Jul 2002 18:45:22 +0000 (18:45 +0000)]
Fix minor leak of import struct, which we do not use yet.
adilger [Mon, 22 Jul 2002 18:37:52 +0000 (18:37 +0000)]
Remove some cruft that got into chaos10 patch
adilger [Mon, 22 Jul 2002 18:36:29 +0000 (18:36 +0000)]
Include lustre_mds.h to have definition for mdc_getlovinfo.
adilger [Mon, 22 Jul 2002 18:35:56 +0000 (18:35 +0000)]
Cosmetic cleanup to use local variable for ll_inode_info.
adilger [Mon, 22 Jul 2002 18:35:19 +0000 (18:35 +0000)]
Add beginning of symlink support. Not totally functional yet, but not totally
broken either.
adilger [Mon, 22 Jul 2002 18:34:12 +0000 (18:34 +0000)]
Add helper function ptlrpc_bulk_addref() to match ptlrpc_bulk_decref().
adilger [Mon, 22 Jul 2002 18:33:22 +0000 (18:33 +0000)]
We need an mdc_ref for a mountpoint, and not an mds_ref.
adilger [Mon, 22 Jul 2002 18:32:52 +0000 (18:32 +0000)]
Call bulk descriptor "desc" instead of "bulk" for consistency.
adilger [Mon, 22 Jul 2002 18:27:07 +0000 (18:27 +0000)]
This should fix the extN bug. I have not enabled indexed directories by
default yet, so it should not affect anything either way.
adilger [Mon, 22 Jul 2002 17:22:12 +0000 (17:22 +0000)]
Add module refcounts for filesystem interface modules.
Minor cleanups in MDS LOV handling (i.e. endianness, code duplication)
gord-fig [Sat, 20 Jul 2002 23:48:22 +0000 (23:48 +0000)]
HOWTO updates.
gord-fig [Sat, 20 Jul 2002 23:47:23 +0000 (23:47 +0000)]
Make the usermode patch apply cleanly.
adilger [Sat, 20 Jul 2002 16:21:15 +0000 (16:21 +0000)]
Add intent support for MKNOD.
Still to be done - SYMLINK and LINK.
braam [Sat, 20 Jul 2002 00:36:30 +0000 (00:36 +0000)]
- new patch for new LLNL kernel. No changes.
braam [Fri, 19 Jul 2002 21:53:06 +0000 (21:53 +0000)]
Patch was missing IT_LOOKUP.
adilger [Fri, 19 Jul 2002 09:02:39 +0000 (09:02 +0000)]
Fixups - should work for single ost.
adilger [Fri, 19 Jul 2002 09:00:08 +0000 (09:00 +0000)]
Change mountpoint to be /mnt/lustre for LOV XML config.
adilger [Fri, 19 Jul 2002 07:09:58 +0000 (07:09 +0000)]
Fix linux version checking so that modules install correctly under UML also.
adilger [Fri, 19 Jul 2002 06:33:57 +0000 (06:33 +0000)]
So, it turns out that the CVS head was wrong - kptrouter is unloaded last.
adilger [Fri, 19 Jul 2002 05:40:58 +0000 (05:40 +0000)]
Get the correct return code from the setup, not from echo.
adilger [Fri, 19 Jul 2002 05:21:24 +0000 (05:21 +0000)]
Remove intersync stuff from /etc/init.d/lustre.
adilger [Fri, 19 Jul 2002 05:15:03 +0000 (05:15 +0000)]
Change the default config so that it doesn't do anything out of the box.
adilger [Fri, 19 Jul 2002 04:49:14 +0000 (04:49 +0000)]
Same fixes on head as branch.
pschwan [Fri, 19 Jul 2002 04:19:07 +0000 (04:19 +0000)]
Peter's module fixes on the tip
braam [Thu, 18 Jul 2002 23:52:04 +0000 (23:52 +0000)]
- fix / problem
pschwan [Thu, 18 Jul 2002 23:37:14 +0000 (23:37 +0000)]
- added recovery.lyx
- changed the 'date' commands to use %x
- added DOCVERSION which will hopefully be included in make dist
pschwan [Thu, 18 Jul 2002 23:14:13 +0000 (23:14 +0000)]
add config files to EXTRA_DIST
braam [Thu, 18 Jul 2002 23:12:10 +0000 (23:12 +0000)]
revert configure.in
braam [Thu, 18 Jul 2002 22:49:09 +0000 (22:49 +0000)]
Makefile now puts the version and date on front of lustre.pdf
pschwan [Thu, 18 Jul 2002 22:46:19 +0000 (22:46 +0000)]
Split the documentation and sample config files into a separate lustre-doc RPM
gord-fig [Thu, 18 Jul 2002 22:42:29 +0000 (22:42 +0000)]
Add Phil's patch-2.4.18-um change.
Add --with-linux and --with-portals as aliases for --enable-linuxdir and --enable-portalsdir.
gord-fig [Thu, 18 Jul 2002 22:41:15 +0000 (22:41 +0000)]
Add version information to lustre-HOWTO.
pschwan [Thu, 18 Jul 2002 21:42:32 +0000 (21:42 +0000)]
changelog update, head
pschwan [Thu, 18 Jul 2002 17:15:52 +0000 (17:15 +0000)]
a small coherence test with 2 mountpoints
pschwan [Thu, 18 Jul 2002 14:18:56 +0000 (14:18 +0000)]
One more.
pschwan [Thu, 18 Jul 2002 14:13:46 +0000 (14:13 +0000)]
Small cleanups to make it look and behave more like the runfailure-net
adilger [Thu, 18 Jul 2002 10:07:04 +0000 (10:07 +0000)]
A mostly-fix for "mknod /mnt/lustre/foofo p". It doesn't fail outright
anymore (S_IFREG only for regular files), it doesn't oops (pass in
dentry->d_it to ll_create), it only complains "file already exists" after
it has already created the file. Definitely an intent thing, but it is
too late to fix that part.
Also includes the LOV changes from today, as they don't seem to harm
the current setup (runtests finishes).
pschwan [Thu, 18 Jul 2002 07:12:16 +0000 (07:12 +0000)]
More connection-level hackery for the DLM
pschwan [Thu, 18 Jul 2002 07:03:45 +0000 (07:03 +0000)]
Be more explicit to aid debugging
pschwan [Thu, 18 Jul 2002 06:26:43 +0000 (06:26 +0000)]
small cleanup
adilger [Wed, 17 Jul 2002 20:26:53 +0000 (20:26 +0000)]
Big diff, mostly cosmetic.
The two fixes are:
- osc_con2dlmcl() now fills in rconn properly.
- lov_connect() gets the mdc_conn data passed in via the conn param
The cosmetic:
- add a new field to the export struct exp_rconnh. Don't use the
import field, it is for the LDLM to use later.
- make the export struct elements exp_foo instead of export_foo.
pschwan [Wed, 17 Jul 2002 18:47:59 +0000 (18:47 +0000)]
- Attempted fix for the crash-during-unload-after-ldlm-test bug
pschwan [Wed, 17 Jul 2002 15:55:10 +0000 (15:55 +0000)]
- Temporary fix for the problem that the MDS needs to make lock requests to
itself, but doesn't go through the whole connection setup nonsense, so its
connection never advances above LUSTRE_CONN_NEW level
- Right after we failed to get a lock in the I/O patch, we tried to debug-print
it. Fixed.
gord-fig [Wed, 17 Jul 2002 13:26:34 +0000 (13:26 +0000)]
Tweak autogen.sh and Makefiles to be compatible with Automake 1.6.2.
rread [Wed, 17 Jul 2002 08:18:46 +0000 (08:18 +0000)]
- fix default rule
braam [Wed, 17 Jul 2002 02:07:21 +0000 (02:07 +0000)]
- move LLNL kernel patch forward to chaos9
rread [Wed, 17 Jul 2002 01:31:13 +0000 (01:31 +0000)]
- mountpoint uses mdc, not mds
- still not getting connectoin to lov, though
rread [Wed, 17 Jul 2002 00:57:48 +0000 (00:57 +0000)]
- mountpoints are coded, but ost uuid is not found, at least for lov
rread [Wed, 17 Jul 2002 00:10:09 +0000 (00:10 +0000)]
- lov config works now
rread [Tue, 16 Jul 2002 22:55:06 +0000 (22:55 +0000)]
- clean up usage help
- use normal traceback
rread [Tue, 16 Jul 2002 22:42:15 +0000 (22:42 +0000)]
- look for profile for the current hostname, or localhost for default
rread [Tue, 16 Jul 2002 21:50:12 +0000 (21:50 +0000)]
- look for lctl in ../utils or in path
rread [Tue, 16 Jul 2002 21:06:44 +0000 (21:06 +0000)]
- simple config, mostly to load modules
rread [Tue, 16 Jul 2002 20:57:57 +0000 (20:57 +0000)]
- config and cleanup is working from obd to lov
- modules, rpc, and mountpoints not working yet
rread [Tue, 16 Jul 2002 20:56:51 +0000 (20:56 +0000)]
- initial checkin
- lov test config for lconf
rread [Tue, 16 Jul 2002 20:52:12 +0000 (20:52 +0000)]
- use stdout for regular messages
adilger [Tue, 16 Jul 2002 17:20:41 +0000 (17:20 +0000)]
Fix cleanup in llite/super.c
rread [Tue, 16 Jul 2002 16:32:01 +0000 (16:32 +0000)]
- improve error handling for command line use
adilger [Tue, 16 Jul 2002 11:52:07 +0000 (11:52 +0000)]
Fix problem with multiple connections to the same local device.
The problem was that we stored the remote export addr+cookie in the
device array. When we disconnected any connection, it would drop the
remote export from the most recent connection.
What we do now is store the remote export addr+cookie in the local
export/import struct and always use that (the old mdc_connh and osc_connh
are gone). We could do some further cleanup of the API at this point,
but I will leave it until there is some discussion on the matter. I was
able to successfully do multiple simultaneous connect and disconnect steps,
as well as run multi-thread test_getattrs with the change. I'm just doing
a full runregression-net test.
I'm still not 100% confident that this is what Peter wants, so I tagged
the pre-commit tree with "before_rconn" to make it easy to generate
a diff (I will also tag the post-commit with "after_rconn").
rread [Tue, 16 Jul 2002 01:53:37 +0000 (01:53 +0000)]
- remove obdfs/Makefile
- get kernel version from version.h, fix uml kversion bug
rread [Tue, 16 Jul 2002 01:52:16 +0000 (01:52 +0000)]
- configuring an obd seems to work
rread [Tue, 16 Jul 2002 01:49:31 +0000 (01:49 +0000)]
- enable command line flags for lctl
- device <name> now works
pschwan [Mon, 15 Jul 2002 20:55:57 +0000 (20:55 +0000)]
Fixed brw_finish to kunmap the pointers in the bulk descriptor instead of the
callback data. The cb_data had a pointer to stack data which might not be
valid anymore, and instead of allocating a new array to hold it, we can just
walk the bulk pages.
pschwan [Mon, 15 Jul 2002 20:32:19 +0000 (20:32 +0000)]
The ioctl handler for test_brw was corrupting kernel memory and giving an
empty array to osc_brw_read, which cheerfully tried to kmap(0). Fixed.
rread [Mon, 15 Jul 2002 19:23:33 +0000 (19:23 +0000)]
- change id to ref
- change ref attributes to just uuidref
- changing service_ref to more specific references where appropriate.
pschwan [Mon, 15 Jul 2002 18:04:24 +0000 (18:04 +0000)]
- Fix for the 16kb page directory handling (thanks Andreas, it appears to
work perfectly)
- Fix for Eric's page_array compilation error
- Increase event queue sizes to 1024
- Re-enable bits of the runfailure-ost test
braam [Mon, 15 Jul 2002 15:40:45 +0000 (15:40 +0000)]
- add Andreas's suggestion for the LLNL Oops's to the patch
braam [Sun, 14 Jul 2002 17:27:12 +0000 (17:27 +0000)]
- see previous message
braam [Sun, 14 Jul 2002 17:26:49 +0000 (17:26 +0000)]
- lookup2: drop a lock in an unlikely error case; clarify interrupt
handling in lookup2
- osc_request.c: bring connection level to full after connect
- mds_reint.c: in unlink, don't cancel locks that you don't have for
negative dentries (bug fix).
braam [Sun, 14 Jul 2002 00:13:54 +0000 (00:13 +0000)]
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).
pschwan [Sat, 13 Jul 2002 23:42:21 +0000 (23:42 +0000)]
Don't load /etc/lustre/lustre.cfg if config files were passed on the command
line.
shaver [Sat, 13 Jul 2002 23:38:45 +0000 (23:38 +0000)]
- Use refcounting to control lifetime of bulk descriptors.
- Add some LDLM timeout #warnings for future reference.
- Make all bulk-desc cleanup happen asynchronously, for cleaner interrupt (and
later timeout) handling.
- Have OSC wait for brw_write reply before beginning bulk operations, for
symmetry with brw_read and general niceness.
- Fix reply-for-freed-request check to use proper XID width.
- Use l_killable_pending in ptlrpc_check_bulk_sent, for consistency. (Also in the
new ptlrpc_check_bulk_received sibling.)
pschwan [Sat, 13 Jul 2002 19:10:22 +0000 (19:10 +0000)]
osc_connect needs to go with level LUSTRE_CONN_NEW. Fixed.
pschwan [Sat, 13 Jul 2002 19:03:23 +0000 (19:03 +0000)]
- Added match_or_enqueue helper function
- fixed userspace build problem in lustre_lib.h
- added intent-based lookup code
- fixed intent-based setattr
- added mds_fid2locked_dentry and mds_name2locked_dentry helpers
- don't crash in ptlrpc_reply, just warn of API violation
- update create.pl to open instead of mcreate