Whamcloud - gitweb
rread [Tue, 2 Jul 2002 20:27:25 +0000 (20:27 +0000)]
- initial version
braam [Tue, 2 Jul 2002 08:41:34 +0000 (08:41 +0000)]
- add missing unlocks.
rread [Tue, 2 Jul 2002 02:41:17 +0000 (02:41 +0000)]
- create device
- make note of need to modify /etc/modules.conf
rread [Tue, 2 Jul 2002 01:36:41 +0000 (01:36 +0000)]
- experiment: attempt to add the cvs version to formatted documentation.
Added a rule to create 'lustre.lyx' which is a copy of master.lyx,
modified to include the cvs version.
Normally, "make lustre.pdf" should produce a pdf with the proper version on
the cover page, but on my system the whole Date tag is not showing up. Not sure why.
rread [Tue, 2 Jul 2002 01:32:42 +0000 (01:32 +0000)]
- remove unneeded parameter to obd_create/destroy
behlendo [Mon, 1 Jul 2002 23:45:32 +0000 (23:45 +0000)]
-Minor fixes to ensure the DTD is valid. net-local.xml passes validation
with the provided validation tool.
>xmllint --valid --noout net-local.xml
braam [Mon, 1 Jul 2002 23:24:26 +0000 (23:24 +0000)]
- other half of the buffer fix.
braam [Mon, 1 Jul 2002 22:55:42 +0000 (22:55 +0000)]
- fix for variable size buffer handling.
pschwan [Mon, 1 Jul 2002 21:23:34 +0000 (21:23 +0000)]
- more LDLM refcount locking infrastructure
- fixed the error handling paths of unlink and rmdir, so that unlinking a
non-existant file returns an error
- fixed the DLM bug that caused the perl script to hang (there's a new bug now
that it triggers under heavier load that causes it to crash)
- fixed a second crashing bug that occurred if a blocked lock gets completed
and then cancelled all before the server-side enqueue finishes (amazing but
true)
- osc_open() now returns the actual rc instead of 0
- fixed a race condition in ptlrpc/service.c related to thread shutdown (nice
catch, Mike)
shaver [Mon, 1 Jul 2002 19:34:20 +0000 (19:34 +0000)]
Fixed some path-slash badness in common.sh, and added embryonic runfailure-ost.
pschwan [Mon, 1 Jul 2002 15:26:52 +0000 (15:26 +0000)]
A Perl script to hammer concurrently at multiple mount points
braam [Mon, 1 Jul 2002 07:30:40 +0000 (07:30 +0000)]
- much of the striping configuration managment and setup builds up
this patch
- the mds has a new lovconfig command to tell it the UUID's and
default striping pattern of the targets it needs to use (these are
the UUID's of the OSC's typically).
- To make this scalable I changed some of the memory management in the
class ioctl handling
- The LOV device has a trivial attach method and setup only tells it
what MDC to use to get its information, by giving it the MDC-UUID.
- As discussed before, the MDS really provides the persistent storage
for the LOV, what little it needs. So during the obd_connect call
for the object storage target (which is made from read_super) the
storage target learns how it is striped and then connects to all the
targets.
- We are in need of better configuration scripts for this stuff and
tomorrow we will push the XML configurations a little further.
- Updated the documentation
- Began to cleanup the /proc/lustre/ stuff -- have some neat ideas
about that and SNMP now.
pschwan [Mon, 1 Jul 2002 06:35:51 +0000 (06:35 +0000)]
- updated LDLM_DEBUG to give more refcount info
- made fixme a macro, so that it shows us where it's called from
- fixed a DLM deadlock (unbalanced l_lock)
- fixed the refcount bug in ldlm_lock_decref
- fixed the refcount bug in ldlm_cli_enqueue in the failed/aborted case
- the lock slab cleans up now!
- fixed the ``connection foo has refcount -61'' bug
- found, but have not yet fixed, a subtle ctrl-c-during-aborted-ldlm-enqueue
bug that can be triggered if you abort the hanging Perl test at _just_ the right
time.
- fixed request leaks in osc_connect and mdc_connect
- we create an import in osc_connect but never use or free it -- I didn't
remove this code, assuming it was going to be used soon?
braam [Sun, 30 Jun 2002 20:34:25 +0000 (20:34 +0000)]
- same for mdc
braam [Sun, 30 Jun 2002 20:33:59 +0000 (20:33 +0000)]
fix a blunder in osc_connect that frees the connection before
returning success.
braam [Sun, 30 Jun 2002 08:03:40 +0000 (08:03 +0000)]
Fix a few compiler warnings.
braam [Sun, 30 Jun 2002 07:39:40 +0000 (07:39 +0000)]
- except for fixing the segfault in the documentation build, this is
merely a cleanup checkin. Functions in the obdclass system were
given sensible names, old unused headers were removed, and stuff
that isn't used widely was sometimes move from generic headers to
specific subsystems.
pschwan [Sun, 30 Jun 2002 03:22:06 +0000 (03:22 +0000)]
- Fixes memory and refcount links in the DLM
- Fixes a long-standing request leak
- The lock slab is not yet cleaned up correctly in some cases, but I'm on it
braam [Sun, 30 Jun 2002 03:19:02 +0000 (03:19 +0000)]
- small issues
- magic and versions into Lustre messages
- obdo cleanups, remove unused fields.
- small updates to object protocol documentation.
braam [Sat, 29 Jun 2002 13:10:22 +0000 (13:10 +0000)]
File I/O fix: move the lock name space to the filter itself.
braam [Sat, 29 Jun 2002 12:53:02 +0000 (12:53 +0000)]
Minor bug fixes to get export handles used by the DLM. I believe the
head is pretty close to where it was before in functionality, but the
cleanups can now continue further.
rread [Sat, 29 Jun 2002 00:41:48 +0000 (00:41 +0000)]
- fix make dist
- bring spec file uptodate
- add rpms rule to main Makefile.am
rread [Sat, 29 Jun 2002 00:40:47 +0000 (00:40 +0000)]
- need to add include for PAGE_SIZE
braam [Fri, 28 Jun 2002 20:05:03 +0000 (20:05 +0000)]
- shared MDS/OST connection routines
braam [Fri, 28 Jun 2002 20:04:21 +0000 (20:04 +0000)]
- The same changes made to the OST are now working in the MDS
- Significant simplification is starting to emerge now, as expected.
- Next is the locking subsystem.
- Added new terms to the glossary.
brian [Fri, 28 Jun 2002 19:49:00 +0000 (19:49 +0000)]
- Refined the DTD futher: 'id' changed to 'name'; tree structure rearranged
and brought more inline with the documentation; other general ongoing
cleanup.
- DTD changes reflected in net-local.xml.
- obdctl properly uses of UUIDs during config, on going work here.
behlendo [Fri, 28 Jun 2002 19:48:59 +0000 (19:48 +0000)]
- Refined the DTD futher: 'id' changed to 'name'; tree structure rearranged
and brought more inline with the documentation; other general ongoing
cleanup.
- DTD changes reflected in net-local.xml.
- obdctl properly uses of UUIDs during config, on going work here.
braam [Fri, 28 Jun 2002 06:54:02 +0000 (06:54 +0000)]
Add a new patch for the new LLNL kernel. This contains many serious changes
from the previous one.
braam [Fri, 28 Jun 2002 01:53:38 +0000 (01:53 +0000)]
- more fixes to export handling. Probably tomorrow they can be
completed.
braam [Thu, 27 Jun 2002 23:15:06 +0000 (23:15 +0000)]
- More changes in the connection handle stuff. We are back to where
obdecho works over the net.
braam [Thu, 27 Jun 2002 21:09:07 +0000 (21:09 +0000)]
Don't update your trees yet..... Brian and I are trying to get our
work in sync.
- This is a fairly pervasive change to get away from "device
numbers".
- Introduce connection handles everywhere (treat these as
lustre_handles).
- On targets map UUID's to handles upon connect and use those handles
in the future.
brian [Wed, 26 Jun 2002 22:09:29 +0000 (22:09 +0000)]
Another slightly less rough cut at XML based configuration. There
is still a large reliance on the existing scripts for loopback
support, portals config, and module loading. I'll try and resolve
some of these things after some consultation about what the best
way to do that is.
- Several minor changes were made to the DTD, we're still not
validating against it but it should be consistent with the example.
- Simple code added for ldlm, mds, obd, ost, osc, and mdc configuration.
- Other misc support and fixes.
A local filesystem should be mountable with the net-local.xml config:
~/tests/llxmlsetup.sh net-local.xml
behlendo [Wed, 26 Jun 2002 22:09:28 +0000 (22:09 +0000)]
Another slightly less rough cut at XML based configuration. There
is still a large reliance on the existing scripts for loopback
support, portals config, and module loading. I'll try and resolve
some of these things after some consultation about what the best
way to do that is.
- Several minor changes were made to the DTD, we're still not
validating against it but it should be consistent with the example.
- Simple code added for ldlm, mds, obd, ost, osc, and mdc configuration.
- Other misc support and fixes.
A local filesystem should be mountable with the net-local.xml config:
~/tests/llxmlsetup.sh net-local.xml
braam [Wed, 26 Jun 2002 00:44:59 +0000 (00:44 +0000)]
Start maintaining the release changelog file.
pschwan [Tue, 25 Jun 2002 22:09:20 +0000 (22:09 +0000)]
- ldlm fixes to get it to pass intent-test.sh, although it doesn't quite clean
up correctly afterwards yet.
adilger [Tue, 25 Jun 2002 07:06:17 +0000 (07:06 +0000)]
Merge trivial changes from branch to head.
The only real change is the filter multi-page BRW failure cleanup code.
braam [Tue, 25 Jun 2002 04:49:57 +0000 (04:49 +0000)]
- Cleanup of the lock callback infrastructure - not quite functional
yet.
- API improvements.
- Documentation update.
adilger [Mon, 24 Jun 2002 20:11:56 +0000 (20:11 +0000)]
Remove obdext2.cfg as we no longer have obdext2
braam [Mon, 24 Jun 2002 18:08:00 +0000 (18:08 +0000)]
- fixes to let a client do correct lock cancellation during namespace
free.
braam [Mon, 24 Jun 2002 17:22:36 +0000 (17:22 +0000)]
status: on ./mcreate I believe the client lock on the file and the mds
lock on the directory are both correctly enqueued.
- updated documentation regarding lock / resource refcounting
- removed unnecessary resource_getref before policy (thanks phil)
- local locks to addref, server locks don't
- variety of small fixes
- print lock/type names
- add one reference on lock for creation, put it away once for destroy
[otherwise 0 ref locks can "live" and that causes the resources grief]
- small fix to change_resource
adilger [Mon, 24 Jun 2002 16:57:47 +0000 (16:57 +0000)]
Fix o_valid and o_size for truncate through osc/ost/filterobd (on HEAD).
This should also remove the need to explicitly set mtime/ctime in
ll_truncate(), but that has not yet been tested and causes no real harm.
braam [Mon, 24 Jun 2002 07:43:13 +0000 (07:43 +0000)]
- first fixes to get things working again: current state:
ldlm_cli_enqueue returns a lock, but it misses one refcount.
braam [Mon, 24 Jun 2002 06:32:05 +0000 (06:32 +0000)]
last compile fixes: WARNING nothing works probably.
braam [Mon, 24 Jun 2002 06:25:23 +0000 (06:25 +0000)]
- more fixes
braam [Mon, 24 Jun 2002 06:24:19 +0000 (06:24 +0000)]
- mds compile fixes
pschwan [Mon, 24 Jun 2002 06:22:30 +0000 (06:22 +0000)]
*** empty log message ***
pschwan [Mon, 24 Jun 2002 06:20:03 +0000 (06:20 +0000)]
*** empty log message ***
pschwan [Mon, 24 Jun 2002 06:13:37 +0000 (06:13 +0000)]
*** empty log message ***
braam [Mon, 24 Jun 2002 06:11:21 +0000 (06:11 +0000)]
*** empty log message ***
braam [Mon, 24 Jun 2002 06:05:19 +0000 (06:05 +0000)]
- compile fix.
braam [Mon, 24 Jun 2002 06:04:06 +0000 (06:04 +0000)]
compile fixes.
pschwan [Mon, 24 Jun 2002 06:02:26 +0000 (06:02 +0000)]
extern/static fix
pschwan [Mon, 24 Jun 2002 06:01:24 +0000 (06:01 +0000)]
lock descriptor fixups
pschwan [Mon, 24 Jun 2002 05:52:20 +0000 (05:52 +0000)]
compilation cleanups
braam [Mon, 24 Jun 2002 05:46:44 +0000 (05:46 +0000)]
- compilation fixes.
pschwan [Mon, 24 Jun 2002 05:37:14 +0000 (05:37 +0000)]
handful of conflict-causing DLM changes
pschwan [Mon, 24 Jun 2002 05:13:52 +0000 (05:13 +0000)]
compilation fixups
braam [Mon, 24 Jun 2002 05:10:40 +0000 (05:10 +0000)]
Lots of reorg in the lock manager, dealing with the updated design.
See the Lustre documentation for more details.
pschwan [Mon, 24 Jun 2002 04:56:33 +0000 (04:56 +0000)]
structure changes for LDLM revamp
pschwan [Mon, 24 Jun 2002 03:58:36 +0000 (03:58 +0000)]
more DLM lock cleanups
braam [Mon, 24 Jun 2002 03:34:18 +0000 (03:34 +0000)]
- minor renaming of handling functions
pschwan [Mon, 24 Jun 2002 03:32:43 +0000 (03:32 +0000)]
more locking hackery
braam [Mon, 24 Jun 2002 03:32:18 +0000 (03:32 +0000)]
- rework lock handling functions with get/put and refcounts
braam [Mon, 24 Jun 2002 03:17:18 +0000 (03:17 +0000)]
- changes for lock handles and refcounts
pschwan [Mon, 24 Jun 2002 03:17:07 +0000 (03:17 +0000)]
First swing at DLM lock cleanup. Do not update your tree if you value
compilation.
pschwan [Sun, 23 Jun 2002 21:21:01 +0000 (21:21 +0000)]
Fix a handful of unsafe spin_unlock()s
pschwan [Sun, 23 Jun 2002 20:15:26 +0000 (20:15 +0000)]
Fixed lock_mode thinko
braam [Sun, 23 Jun 2002 19:18:10 +0000 (19:18 +0000)]
- improve the explanatory comment
braam [Sun, 23 Jun 2002 19:16:22 +0000 (19:16 +0000)]
- minor fix
pschwan [Sun, 23 Jun 2002 19:13:44 +0000 (19:13 +0000)]
- If the DLM server returns ELDLM_LOCK_ABORTED, pass that status up to the
caller
- If ldlm_cli_enqueue returns ELDLM_LOCK_ABORTED, set it->it_lock_mode to 0 to
avoid doing a decref on a non-existent lock
- Add a couple debugging messages
braam [Sun, 23 Jun 2002 19:12:57 +0000 (19:12 +0000)]
- add reentrant locks
pschwan [Sun, 23 Jun 2002 18:30:18 +0000 (18:30 +0000)]
- The first unfinished bits of the DLM regression test. Move along, nothing to
see here.
- Treat the completion ASTs like the blocking ASTs: hold the requests for
later dispatching, after we've finished walking all of the lists.
braam [Sun, 23 Jun 2002 08:16:02 +0000 (08:16 +0000)]
turn off directory indexing pending the resolution of the htree bug.
braam [Sun, 23 Jun 2002 04:37:10 +0000 (04:37 +0000)]
- remainder of rmdir changes
braam [Sun, 23 Jun 2002 04:32:15 +0000 (04:32 +0000)]
automatic detection of OST/MDS
\18\18#
adilger [Sun, 23 Jun 2002 03:16:32 +0000 (03:16 +0000)]
Allow specifying > 4GB sizes for truncate via strtoull().
adilger [Sun, 23 Jun 2002 02:48:30 +0000 (02:48 +0000)]
llite and MDC statfs operations.
Includes fix for permission problems.
adilger [Sun, 23 Jun 2002 02:47:20 +0000 (02:47 +0000)]
Move cleanup of service struct into ptlrpc_unregister_service().
Includes MDS statfs method.
Includes fixes for sgid directories on MDS.
adilger [Sun, 23 Jun 2002 02:39:05 +0000 (02:39 +0000)]
Merge from posix_stable.
braam [Sun, 23 Jun 2002 00:49:33 +0000 (00:49 +0000)]
patches for the previous fixes
braam [Sun, 23 Jun 2002 00:47:37 +0000 (00:47 +0000)]
- fixes for truncating new files with intent locks
- fixes for truncating the data objects ** Andreas: check branch too
please!
- fixes for mode setting in open_namei path
- remove useless setattr from mds
- change test truncates to bytes, not kilo bytes
- improve the intent-test.sh
pschwan [Sat, 22 Jun 2002 22:28:10 +0000 (22:28 +0000)]
LBUG() if we get a NULL intent in ll_lookup2
pschwan [Sat, 22 Jun 2002 20:28:07 +0000 (20:28 +0000)]
Bring small ha_assist2.sh change from branch to trunk
braam [Sat, 22 Jun 2002 18:36:42 +0000 (18:36 +0000)]
do panic dump logs with the notifier list
braam [Sat, 22 Jun 2002 17:13:54 +0000 (17:13 +0000)]
Page must be mapped before prepare write, since on ENOSPACE we fall
into lustre_put_page and do kunmap (this crashes on highmem machines
if the page wasn't mapped).
braam [Sat, 22 Jun 2002 00:27:01 +0000 (00:27 +0000)]
new images and editorial updates to documentation
braam [Fri, 21 Jun 2002 23:52:41 +0000 (23:52 +0000)]
kernel patches that dump the Lustre debug log when the kernel panics
braam [Fri, 21 Jun 2002 23:46:08 +0000 (23:46 +0000)]
- commit mkdir fix to head as well. Head has slightly better error
handling in make empty dir now.
pschwan [Fri, 21 Jun 2002 20:37:25 +0000 (20:37 +0000)]
give the intent test some easily-changed variables
pschwan [Fri, 21 Jun 2002 20:04:44 +0000 (20:04 +0000)]
Fix for cleanup deadlock
pschwan [Fri, 21 Jun 2002 19:38:34 +0000 (19:38 +0000)]
Small typo fix
pschwan [Fri, 21 Jun 2002 18:43:43 +0000 (18:43 +0000)]
- If a completion callback arrives before the reply for the enqueue, AND the
completion callback specifies a different resource than the original enqueue,
fixup the resource.
- When the enqueue reply does come back, recognize that we've already fixed up
the resource and avoid doing it again.
- I am ---> <--- this close to replacing all of the DLM locking, which is
almost certainly a cesspool of deadlock opportunities, with lock_kernel() until
lustre lite performance
braam [Fri, 21 Jun 2002 07:47:05 +0000 (07:47 +0000)]
tiny changes to accompany the Portals update which follows.
adilger [Fri, 21 Jun 2002 06:42:42 +0000 (06:42 +0000)]
Merge from posix_stable.
adilger [Fri, 21 Jun 2002 06:35:54 +0000 (06:35 +0000)]
Merge from posix_stable
adilger [Fri, 21 Jun 2002 06:29:52 +0000 (06:29 +0000)]
Merge from posix_stable.
adilger [Fri, 21 Jun 2002 06:24:52 +0000 (06:24 +0000)]
Use OBD_BRW_{READ,WRITE} for BRW calls.
braam [Fri, 21 Jun 2002 05:15:03 +0000 (05:15 +0000)]
fix the path problem in common.sh
pschwan [Fri, 21 Jun 2002 03:48:11 +0000 (03:48 +0000)]
Fixed an MDS crash bug discovered as a result of earlier warning cleanup
pschwan [Thu, 20 Jun 2002 23:52:35 +0000 (23:52 +0000)]
Fix many (mostly un-serious) build warnings