Whamcloud - gitweb
maxim [Sat, 5 Dec 2009 11:09:28 +0000 (11:09 +0000)]
b=21079
i=rread
i=liang
Sanity test_215 is rewritten to eliminate too optimistic assumptions
about values LNET exports to /proc/sys/lnet/*. Also, rewritten test
reflects latest modifications of /proc/sys/lnet/routers and
/proc/sys/lnet/nis (made in the scope of bug #20171).
dzogin [Sat, 5 Dec 2009 06:19:00 +0000 (06:19 +0000)]
Branch HEAD
b=21513
i=alexey.lyashkov
i=johann
----------------------------------------------------------------------
Modified Files:
lustre/ChangeLog lustre/ptlrpc/import.c
----------------------------------------------------------------------
Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
jxiong [Sat, 5 Dec 2009 03:36:54 +0000 (03:36 +0000)]
b=21460
r=wangdi,eric.mei
- in ll_prepare_write, it initialises an env, but not hold a refcount against it, so it would have been freed in up-coming ll_commit_write.
- remove cui_oneshot
brian [Fri, 4 Dec 2009 21:48:50 +0000 (21:48 +0000)]
b=21514
i=wangyb
i=yangsheng
If lbuild is interrupted (i.e. LBATS "Cancel") then it should not call
the ccache -s exit trap action because experience has shown that the
ccache binary is gone by the time lbuild gets to calling it.
Also, extend the API for the delete_exit_trap() to handle multiple
action handles in a single call. This makes it easy to create a stack
of actions and then delete them all in a single command.
What would be even nicer would be (shell and pdsh style) glob handling
so that one could do delete_exit_trap foo[1-10], assuming one pushed 10
actions on a stack to be unwound during an operation that is supposed to
be atomic. Or more simply delete_exit_trap foo*.
grev [Fri, 4 Dec 2009 21:01:18 +0000 (21:01 +0000)]
b=19616
o=Walter.Poxon
i=grev
read the list of tests to except from the file if it exist
grev [Fri, 4 Dec 2009 20:58:44 +0000 (20:58 +0000)]
cat /tmp/msg
brian [Fri, 4 Dec 2009 20:22:53 +0000 (20:22 +0000)]
b=21041
i=dmitry.zogin
i=johann
On RHEL5 at least, the sg_set_page autoconf macro needs the addition of
<asm/types.h> in order to properly work.
nathan [Fri, 4 Dec 2009 19:35:03 +0000 (19:35 +0000)]
b=21516
i=rread
fix acc-sm for new lustre_rsync name
fix stripe count determination in test_7
panda [Fri, 4 Dec 2009 11:30:30 +0000 (11:30 +0000)]
b=21348
i=Mike Pershin
i=Alexander Zarochentsev
allow adding several instances to evictor list
yangsheng [Fri, 4 Dec 2009 10:30:49 +0000 (10:30 +0000)]
Branch b1_8
b=21375
i=johann, adilger
Cleanup for FS_HAS_FIEMAP.
brian [Thu, 3 Dec 2009 19:44:59 +0000 (19:44 +0000)]
b=21426
Re-land now that the missing directory is created in CVS (and a .cvsignore
file placed in it so that a checkout will create the directory).
stinson1947 [Thu, 3 Dec 2009 19:36:46 +0000 (19:36 +0000)]
Updated version for 1.8.2 Build 06.
brian [Thu, 3 Dec 2009 18:59:19 +0000 (18:59 +0000)]
Create a .cvsignore for this directory.
brian [Thu, 3 Dec 2009 18:44:00 +0000 (18:44 +0000)]
Back out 20315 landing due to breakage with SLES10.
I suspect that we should be using RPMs %configure macro in our lustre.spec
and we are not.
brian [Thu, 3 Dec 2009 18:41:43 +0000 (18:41 +0000)]
Back out 20315 landing due to breakage with SLES10.
I suspect that we should be using RPMs %configure macro in our lustre.spec
and we are not.
johann [Thu, 3 Dec 2009 17:14:53 +0000 (17:14 +0000)]
Branch b1_8
b=20668
i=johann (patch from eeb)
Remove trailing semicolons in do{}while(0) macro definitions.
johann [Thu, 3 Dec 2009 17:14:47 +0000 (17:14 +0000)]
Branch b1_8
b=20668
i=johann (patch from eeb)
Remove trailing semicolons in do{}while(0) macro definitions.
panda [Thu, 3 Dec 2009 16:55:49 +0000 (16:55 +0000)]
b=21477
i=Johann
disable pinger when appropriate
panda [Thu, 3 Dec 2009 16:52:52 +0000 (16:52 +0000)]
b=21477
i=Johann Lombardi
i=Dmitry Zogin
check connection for NULL (for the obdecho client)
dzogin [Thu, 3 Dec 2009 16:01:37 +0000 (16:01 +0000)]
Branch b1_8
b=21513
i=alexey.lyashkov
i=johann
----------------------------------------------------------------------
Modified Files:
Tag: b1_8
lustre/ChangeLog lustre/ptlrpc/import.c
----------------------------------------------------------------------
Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
tappro [Thu, 3 Dec 2009 15:03:45 +0000 (15:03 +0000)]
Branch b1_8
b=21430
i=rread
i=johann
seeds PRNG with current time on server, check export is connected before request
handling
tappro [Thu, 3 Dec 2009 10:28:50 +0000 (10:28 +0000)]
Branch HEAD
b=21240
i=rread
don't check result of lctl because it is allowed to fail during that test
tappro [Thu, 3 Dec 2009 10:27:01 +0000 (10:27 +0000)]
Branch HEAD
b=21430
i=adilger
i=johann
seeds PRNG with current time on server, checks export is connected during
request handling
grev [Thu, 3 Dec 2009 09:35:57 +0000 (09:35 +0000)]
b=21265
i=Robert.Read
skip last sleep if ELAPSED + sleep > DURATION
grev [Thu, 3 Dec 2009 08:33:40 +0000 (08:33 +0000)]
b=21142
i=Andrew.Perepechko
skip cascading_rw for NFSCLIENT mode
yangsheng [Thu, 3 Dec 2009 06:20:14 +0000 (06:20 +0000)]
Branch HEAD
b=21375
i=robert, johann
Cleanup code.
deshmukh [Thu, 3 Dec 2009 05:37:23 +0000 (05:37 +0000)]
Reverting the following patch to avoid the lbat head build failure.
b=21426
[lustre-cvs] HEAD: lustre-build autoMakefile.am.toplevel lbuild
jxiong [Thu, 3 Dec 2009 03:28:30 +0000 (03:28 +0000)]
b=16721
r=wangdi,tappro
Use global rename lock at mdt to avoid deadlock
jxiong [Thu, 3 Dec 2009 03:22:59 +0000 (03:22 +0000)]
b=21434
r=wangdi,eric.mei
We can't use env and io in osc page operations because they are not correctly setup. To support lockless IO, I choose to set srvlock flag at page initialization time.
isaac [Wed, 2 Dec 2009 23:35:59 +0000 (23:35 +0000)]
b=21156,i=liang:
- simple code cleanup.
isaac [Wed, 2 Dec 2009 23:31:16 +0000 (23:31 +0000)]
b=21156,i=liang:
- simple code cleanup.
brian [Wed, 2 Dec 2009 22:00:56 +0000 (22:00 +0000)]
b=21426
i=wangyb
i=yangsheng
Allow the easy dropping in of patches to OFED. This is mainly meant as a
mechanism to test patches from OFA prior to them committing them to their
upstream.
We should likely never actually release a version of Lustre and OFED built
with any OFED patches.
panda [Wed, 2 Dec 2009 21:42:20 +0000 (21:42 +0000)]
b=20101
i=Andreas Dilger
i=Robert Read
a=James Simmons
Fixed lfs to maintain old behavior and support new functionality
brian [Wed, 2 Dec 2009 21:35:56 +0000 (21:35 +0000)]
b=20315
i=adilger
o=Christopher Morrone
Use the more standard libexecdir for scripts.
brian [Wed, 2 Dec 2009 21:33:10 +0000 (21:33 +0000)]
b=20315
i=adilger
o=Christopher Morrone
Use the more standard libexecdir for scripts.
brian [Wed, 2 Dec 2009 20:44:37 +0000 (20:44 +0000)]
b=21457
i=<inspected_by>
i=<inspected_by>
Refactor the tarball downloading code so that we remove the duplicated
(in fact, at least triplicated) code and avoid races between multiple
lbuilds downloading the same thing. Now parallel lbuilds wait for the
first lbuild to complete the download and use it's product.
panda [Wed, 2 Dec 2009 17:40:14 +0000 (17:40 +0000)]
b=20101
i=Andreas Dilger
i=Robert Read
a=James Simmons
Fixed lfs to maintain old behavior and support new functionality
grev [Wed, 2 Dec 2009 07:20:50 +0000 (07:20 +0000)]
b=20863
i=Brian
i=Manoj
cleanup pools created
isaac [Wed, 2 Dec 2009 00:44:12 +0000 (00:44 +0000)]
- removed unused header mxlnd_wire.h.
isaac [Wed, 2 Dec 2009 00:29:58 +0000 (00:29 +0000)]
- removed unused header mxlnd_wire.h.
brian [Tue, 1 Dec 2009 21:53:00 +0000 (21:53 +0000)]
b=20617
i=rread
Remove a really naive attempt in the mptlinux.spec to enforce that the kernel
modules package is only installed on the architecture it was built on. This
of course falls down when you are installing (say) i686 modules into an i686
chroot on an x86_64 machine.
robert.read [Tue, 1 Dec 2009 20:32:24 +0000 (20:32 +0000)]
Branch HEAD
b=19669
i=nathan
i=adilger
Fix memcpy bug.
grev [Tue, 1 Dec 2009 17:33:19 +0000 (17:33 +0000)]
b=20863
i=Brian
i=Manoj
cleanup pools created
deshmukh [Tue, 1 Dec 2009 17:13:49 +0000 (17:13 +0000)]
b=20655
Reverting HEAD to ext3 as default.
ericm [Tue, 1 Dec 2009 16:14:02 +0000 (16:14 +0000)]
branch: HEAD
getxattr should return -ENODATA in case of no xattr present.
b=19908
r=wangdi
r=fanyong
isaac [Tue, 1 Dec 2009 15:10:57 +0000 (15:10 +0000)]
i=liang,b=15332,b=21103:
- LNet router shuffler.
isaac [Tue, 1 Dec 2009 15:05:28 +0000 (15:05 +0000)]
i=liang,b=15332,b=21103:
- LNet fine grain routing support.
isaac [Tue, 1 Dec 2009 15:00:55 +0000 (15:00 +0000)]
i=maxim,b=18460,b=20171:
- avoid asymmetrical router failures; monotonic timing source for router checker.
isaac [Tue, 1 Dec 2009 13:53:55 +0000 (13:53 +0000)]
i=liang,b=21103,b=19735:
- multiple-instance support for kptllnd.
johann [Tue, 1 Dec 2009 13:18:05 +0000 (13:18 +0000)]
Branch b1_8
b=20978
i=adilger
i=liang
use LICFS_ALLOC instead of vmalloc in the lustre hash code.
johann [Tue, 1 Dec 2009 13:18:04 +0000 (13:18 +0000)]
Branch b1_8
b=20978
i=adilger
i=liang
use LICFS_ALLOC instead of vmalloc in the lustre hash code.
deshmukh [Tue, 1 Dec 2009 12:57:45 +0000 (12:57 +0000)]
b=20655
i=robert.read
i=sheng.yang
With this patch, ldiskfs will be ext4 by default on HEAD for RHEL5. To disable ext4 i.e.
to build ext3 ldiskfs --disable-ext4 configure option needs to be given.
There is no change for b1_8.
grev [Tue, 1 Dec 2009 11:07:28 +0000 (11:07 +0000)]
b=21265
i=Robert.Read
skip last sleep if ELAPSED + sleep > DURATION
zam [Tue, 1 Dec 2009 09:22:12 +0000 (09:22 +0000)]
Branch HEAD
b=20816
i=johann
i=robert.read
fixing replay-single test 67b: exhaust precreating before
checking effects of long file creation on OST.
yangsheng [Tue, 1 Dec 2009 09:19:15 +0000 (09:19 +0000)]
Branch HEAD
b=20892
i=robert, yangsheng
Author:johann
Cleanup the VFS_RENAME config check.
liuy [Tue, 1 Dec 2009 02:24:23 +0000 (02:24 +0000)]
Branch HEAD
b=19518
o=liangzhen
i=adilger
i=vitaly
Second step change to move lustre hashes to libcfs:
-move lustre hashes from obdclass to libcfs
nathan [Tue, 1 Dec 2009 00:18:04 +0000 (00:18 +0000)]
b=21275
i=manoj
rename lreplicate to lustre_rsync since "replicate" means many things
johann [Mon, 30 Nov 2009 16:57:06 +0000 (16:57 +0000)]
Branch b1_8
Remove LC_VFS_RENAME_MUTEX since this configure check is no
longer used.
tianzy [Mon, 30 Nov 2009 03:47:53 +0000 (03:47 +0000)]
Branch b1_8
when dqacq_interpret() can't unpack qunit_data, it will use original one
so that no qunit will hang on there forever.
b=21372
i=johann
i=yong.fan
brian [Fri, 27 Nov 2009 17:20:55 +0000 (17:20 +0000)]
b=20617
Fix a typo that leaked into the commit.
brian [Fri, 27 Nov 2009 14:45:18 +0000 (14:45 +0000)]
b=20617
i=yibin.wang
i=sheng.yang
Build desired RDAC and MPTLINUX drivers with lbuild.
brian [Fri, 27 Nov 2009 14:31:07 +0000 (14:31 +0000)]
b=20617
i=yibin.wang
i=sheng.yang
Build desired RDAC and MPTLINUX drivers with lbuild.
brian [Fri, 27 Nov 2009 14:31:06 +0000 (14:31 +0000)]
b=20617
i=yibin.wang
i=sheng.yang
Build desired RDAC and MPTLINUX drivers with lbuild.
brian [Fri, 27 Nov 2009 14:20:50 +0000 (14:20 +0000)]
b=19721
i=adilger
Try to be somewhat intelligent about the need to autogen.sh or not by
seeing if any of the patches touch an autoconf file.
tappro [Fri, 27 Nov 2009 10:24:58 +0000 (10:24 +0000)]
Branch HEAD
b=19023
i=rread
missed client_df in ost-pools test
tappro [Fri, 27 Nov 2009 10:22:41 +0000 (10:22 +0000)]
Branch b1_8
b=19023
i=rread
i=johann
don't use df to determine recovery result
yangsheng [Thu, 26 Nov 2009 15:42:33 +0000 (15:42 +0000)]
Branch b1_8
b=19325
Revert the patch since it expand extent lock improper.
zhanghc [Thu, 26 Nov 2009 11:22:49 +0000 (11:22 +0000)]
b=19069
save the string returned from libcfs_nid2str into a local alloced buffer,
then to avoid problem caused by the buffer storing the string is modified
during proc_mkdir.
i=johann@sun.com
i=zhen.liang@sun.com
zhanghc [Thu, 26 Nov 2009 11:22:48 +0000 (11:22 +0000)]
b=19069
save the string returned from libcfs_nid2str into a local alloced buffer,
then to avoid problem caused by the buffer storing the string is modified
during proc_mkdir.
i=johann@sun.com
i=zhen.liang@sun.com
tianzy [Thu, 26 Nov 2009 02:58:13 +0000 (02:58 +0000)]
Branch b1_8
fix some problem in test_23 of ost_pools.sh
b=20696
i=manoj.joseph
i=johann
isaac [Wed, 25 Nov 2009 20:17:31 +0000 (20:17 +0000)]
- corrected a trivial comment typo.
isaac [Wed, 25 Nov 2009 20:16:14 +0000 (20:16 +0000)]
- corrected a trivial comment typo.
brian [Wed, 25 Nov 2009 19:59:38 +0000 (19:59 +0000)]
b=21041
i=dmitry.zogin
i=johann
On RHEL5 at least, the sg_set_page autoconf macro needs the addition of <asm/types.h> in order to properly work.
robert.read [Wed, 25 Nov 2009 17:35:00 +0000 (17:35 +0000)]
Branch HEAD
Update version number.
zhanghc [Wed, 25 Nov 2009 10:08:32 +0000 (10:08 +0000)]
b=16818
1, the test will be skipped if the MGS and OST1 are on the same node
2, the test will be interrputed if "zconf_mount $MOUNT2" failed
i=nathan.rutman@sun.com
i=robert.read@sun.com
i=grev@sun.com
zam [Wed, 25 Nov 2009 09:11:33 +0000 (09:11 +0000)]
Branch HEAD
b=20816
i=robert.read
i=tappro
i=johann
improve simulation of late reply: ignore obd_fail_timeout for ping replies.
robert.read [Wed, 25 Nov 2009 03:10:51 +0000 (03:10 +0000)]
Branch HEAD
b=19742
The FIEMAP tests need to be updated, so there's seems to be no point in running them right now.
tianzy [Wed, 25 Nov 2009 03:03:00 +0000 (03:03 +0000)]
Branch HEAD
fix some problem in test_23 of ost_pools.sh
b=20696
i=manoj.joseph
i=johann
jxiong [Wed, 25 Nov 2009 02:20:28 +0000 (02:20 +0000)]
b=19906
cancel wait lock - small fixes.
jxiong [Wed, 25 Nov 2009 02:10:39 +0000 (02:10 +0000)]
b=19906
r=wangdi,ericm
Revise osc_lock_cancel_wait. In clio, we tend to cancel the conflicting locks
at the same client before enqueuing a new lock. In the old implementation, it
waits at osc layer with parent lock held, this is not good - so I moved the
wait to lov layer.
jxiong [Wed, 25 Nov 2009 01:27:52 +0000 (01:27 +0000)]
b=21178
r=wangdi,ericm
Fix a race condition in cl_sync_io_note.
jxiong [Wed, 25 Nov 2009 01:05:36 +0000 (01:05 +0000)]
b=19928
r=wangdi,ericm
A workaround fix for 19928 where we found that sometimes MDS takes much long time to finish setattr, which then causes clients can't take inode mutex on time after holding extent lock.
I just added ->lli_trunc_sem to make the above process atomic.
adilger [Tue, 24 Nov 2009 17:48:38 +0000 (17:48 +0000)]
Branch HEAD
Fix whitespace (tabs, trailing spaces).
robert.read [Tue, 24 Nov 2009 00:22:12 +0000 (00:22 +0000)]
Branch HEAD
b=21240,19242
Disabling replay-single 44a, which is using obd_race, which appears to be racy itself, and
also appear to be incorrectly checking the return code of lclt recover, which appears be supposed
to fail in this test. Also disabling 44b which looks like it has the same issue with lctl recover.
yangsheng [Sat, 21 Nov 2009 15:33:17 +0000 (15:33 +0000)]
Branch b1_8
b=21375
i=brian, johann
Config check for FS_HAS_FIEMAP separately. It only exist in RHEL5.4.
stinson1947 [Fri, 20 Nov 2009 17:21:58 +0000 (17:21 +0000)]
Updated for 1.8.2 build 05.
shadow [Thu, 19 Nov 2009 14:18:49 +0000 (14:18 +0000)]
don't fail import after eviction
Branch HEAD
b=20765
i=tappro
i=zam
shadow [Thu, 19 Nov 2009 12:05:53 +0000 (12:05 +0000)]
avoid unnecessary allocation.
Branch HEAD
b=19557
i=vitaly
panda [Thu, 19 Nov 2009 11:28:16 +0000 (11:28 +0000)]
b=21150
i=Johann Lombardi
i=ZhiYong Tian
avoid mds-lov cleanup during dqacq
shadow [Thu, 19 Nov 2009 11:22:40 +0000 (11:22 +0000)]
avoid unnecessary allocation
Branch b1_8
b=19557
i=vitaly
fanyong [Thu, 19 Nov 2009 08:39:57 +0000 (08:39 +0000)]
b=21064
i=robert.read
i=adilger
1) Break reqeust process when service to be stopped.
2) Recovery the process for "ptlrpc_stop_thread()" as suggestion.
tappro [Thu, 19 Nov 2009 07:08:38 +0000 (07:08 +0000)]
*** empty log message ***
jxiong [Thu, 19 Nov 2009 05:58:04 +0000 (05:58 +0000)]
b=19906
r=wangdi,ericm
Revert a bit of cleanup patch.
zhanghc [Thu, 19 Nov 2009 04:08:03 +0000 (04:08 +0000)]
b=20354
the client_uuid(lcd_uuid) stored in last_rcvd on server(MDT or OST)
could be corrupted, so it is needed to check during loading from disk
i=johann@sun.com
i=tappro@sun.com
jxiong [Thu, 19 Nov 2009 03:39:07 +0000 (03:39 +0000)]
b=19906
r=wangdi,ericm
Make racer work at the client side.
jxiong [Thu, 19 Nov 2009 03:06:29 +0000 (03:06 +0000)]
b=19906
r=wangdi,ericm
IO segfault fix
jxiong [Thu, 19 Nov 2009 03:01:03 +0000 (03:01 +0000)]
b=19906
r=wangdi,ericm
cleanup patch, add more debug info.
brian [Wed, 18 Nov 2009 21:12:21 +0000 (21:12 +0000)]
Back out landing that was made in error of flags being mis-set.
johann [Wed, 18 Nov 2009 19:09:31 +0000 (19:09 +0000)]
b=16774,21252
Disable lock cancelling before replay due to bug 21252.
brian [Wed, 18 Nov 2009 14:44:01 +0000 (14:44 +0000)]
b=20617
One hunk missed in landing. Result is benign though as both constructs
result in the same value.