Whamcloud - gitweb
fs/lustre-release.git
15 years agob=17839
grev [Thu, 12 Feb 2009 18:49:24 +0000 (18:49 +0000)]
b=17839
i=Brian
recovery-mds-scale for SLOW=no

15 years agorevert b=17839 commit
grev [Thu, 12 Feb 2009 16:14:01 +0000 (16:14 +0000)]
revert b=17839 commit

15 years agob=17839
grev [Thu, 12 Feb 2009 15:27:38 +0000 (15:27 +0000)]
b=17839
i=Brian
recovery-mds-scale for SLOW=no

15 years agoBranch b18
bobijam [Thu, 12 Feb 2009 06:41:10 +0000 (06:41 +0000)]
Branch b18
b=18015

sles9 doesn't define SLAB_DESTROY_BY_RCU.

15 years agob=18417
girish [Wed, 11 Feb 2009 18:07:07 +0000 (18:07 +0000)]
b=18417
o=h.huang
i=girish

Fix typo from last commit which caused test-framework to exit wrongly

15 years agoupdate client to support vanila kernels up to 2.6.27.
shadow [Wed, 11 Feb 2009 08:48:43 +0000 (08:48 +0000)]
update client to support vanila kernels up to 2.6.27.
Branch b1_8
b=14250
i=green
i=deen

15 years agob=18417
girish [Wed, 11 Feb 2009 06:54:52 +0000 (06:54 +0000)]
b=18417
i=kalpak
i=girish
o=adilger

cleanup correctly after sanity 102 tests

15 years agob=15428
jxiong [Tue, 10 Feb 2009 07:11:36 +0000 (07:11 +0000)]
b=15428
r=adilger

Fix a crash issue for raid5 zerocopy

15 years agoBranch b1_8
fanyong [Tue, 10 Feb 2009 04:57:51 +0000 (04:57 +0000)]
Branch b1_8
b=18294
i=tianzy
i=panda

Diagnostic patch to catch what's the garbage left in the system when quota error.

15 years agoBranch b1_8
tianzy [Tue, 10 Feb 2009 04:28:48 +0000 (04:28 +0000)]
Branch b1_8
add an entry under /proc to support what policy(RR or QOS) will be used.
b=18334
i=nathan
i=adilger

15 years agob=18219
grev [Tue, 10 Feb 2009 00:24:07 +0000 (00:24 +0000)]
b=18219
i=Adilger
test_70b one client fix

15 years agob=18366
grev [Mon, 9 Feb 2009 23:42:19 +0000 (23:42 +0000)]
b=18366
i=Johann
test_5 ENOSPC fix

15 years agoBranch b1_8
johann [Mon, 9 Feb 2009 15:19:09 +0000 (15:19 +0000)]
Branch b1_8
b=18374
i=johann (patch from Tappro)
i=shadow

Return only valid cookies, empty slot for failed ones.
This fixes a memory corruption causing random oops.

15 years agoBranch b1_8
johann [Mon, 9 Feb 2009 15:17:18 +0000 (15:17 +0000)]
Branch b1_8
b=18374
i=shadow
i=tappro

Wait for mds_fl_synced to be set during recovery before allowing clients
to reconnect. This is done to make sure that llog is properly set up.

15 years agoBranch b1_8
johann [Mon, 9 Feb 2009 15:16:00 +0000 (15:16 +0000)]
Branch b1_8
b=18067
i=johann
i=adilger

Change oap magic LASSERT into LASSERTF to aid debugging.

15 years agokernel support clarification
anserper [Sun, 8 Feb 2009 23:44:26 +0000 (23:44 +0000)]
kernel support clarification

15 years agob=16823
green [Sun, 8 Feb 2009 23:14:31 +0000 (23:14 +0000)]
b=16823
r=adilger,shadow

Update offset calculations for 32 bit platforms

15 years agob=15134
pravins [Sun, 8 Feb 2009 17:24:39 +0000 (17:24 +0000)]
b=15134
i=adilger
i=h.huang
interop patch for lu_dirent attribute handling.

15 years agowarning clarificationlustre-core.m4
anserper [Sun, 8 Feb 2009 11:26:09 +0000 (11:26 +0000)]
warning clarificationlustre-core.m4

15 years agoresolve lots races and use after free.
shadow [Fri, 6 Feb 2009 12:30:33 +0000 (12:30 +0000)]
resolve lots races and use after free.
Branch b1_8
b=18114
i=jc.lafoucriere
i=panda

15 years agoBranch b1_8
bobijam [Fri, 6 Feb 2009 02:20:38 +0000 (02:20 +0000)]
Branch b1_8
b=17011
i=tom.wang
i=joahnn

remove redundant brw_stats on obdfilter.

15 years agoBranch b1_8
huanghua [Fri, 6 Feb 2009 02:11:31 +0000 (02:11 +0000)]
Branch b1_8
b=3633
i=robert.read
i=oleg.drokin
i=tappro

clear the "hold for replay" flag for open request once a file is closed

15 years agoBranch b1_8
huanghua [Thu, 5 Feb 2009 12:30:29 +0000 (12:30 +0000)]
Branch b1_8
b=11824
i=rahul.deshmukh
i=yong.fan

interop - enable recovery between 1.8 client and 2.0 server.

15 years agohandle mds reconnect more fastly.
shadow [Thu, 5 Feb 2009 11:19:01 +0000 (11:19 +0000)]
handle mds reconnect more fastly.
Branch b1_8
b=18304
i=johann
i=green

15 years agoBranch b1_8
bobijam [Thu, 5 Feb 2009 02:09:43 +0000 (02:09 +0000)]
Branch b1_8
b=18225
o=Nic Henke(nic@cray.com)
i=johann
i=adilger

use obd_brw_async in echo_client_kbrw.

15 years agoBranch b1_8
bobijam [Wed, 4 Feb 2009 01:42:30 +0000 (01:42 +0000)]
Branch b1_8
b=18015
i=joahnn
i=sheng.yang

add SLAB_DESTROY_BY_RCU to ldlm_lock_slab's flags for ldlm_locks are freed
with RCU mechanism.

15 years agob=18417
girish [Tue, 3 Feb 2009 19:24:43 +0000 (19:24 +0000)]
b=18417
i=adilger
i=kalpak

missed hunk from last commit

15 years agob=18417
girish [Tue, 3 Feb 2009 19:09:40 +0000 (19:09 +0000)]
b=18417
i=adilger
i=kalpak

remove star tests from sanity.sh

15 years agob=17899
grev [Tue, 3 Feb 2009 13:34:16 +0000 (13:34 +0000)]
b=17899
i=Johann
test_53* fix: rm dirs used at start

15 years agoBranch b1_8
tianzy [Tue, 3 Feb 2009 06:45:03 +0000 (06:45 +0000)]
Branch b1_8
use QUOTA_REQ_RETURNED to replace -EAGAIN when a quota req is finished.
b=18307
i=johann
i=panda

15 years agoBranch b1_8
fanyong [Tue, 3 Feb 2009 02:20:02 +0000 (02:20 +0000)]
Branch b1_8
b=18135
i=tianzy
i=andrew.perepechko

Wake up all wait threads on the "lq_waitq" when "lqc_import" is invalid.

15 years agodon't call obd_disconnect under lov_lock.
shadow [Mon, 2 Feb 2009 17:33:32 +0000 (17:33 +0000)]
don't call obd_disconnect under lov_lock.
Branch b1_8
b=17310
i=johann
i=rread

15 years agoremove debug line
grev [Mon, 2 Feb 2009 13:50:08 +0000 (13:50 +0000)]
remove debug line

15 years agob=16818
grev [Mon, 2 Feb 2009 12:53:26 +0000 (12:53 +0000)]
b=16818
i=Nathan
get rid of set_obd_timeout(), improve test_26*

15 years agob=17895
kalpak [Mon, 2 Feb 2009 09:05:05 +0000 (09:05 +0000)]
b=17895
i=adilger
i=alex

add files for last commit

15 years agob=17895
kalpak [Mon, 2 Feb 2009 08:59:35 +0000 (08:59 +0000)]
b=17895
i=adilger
i=alex

While using HA for Lustre servers with Linux RAID, it is possible that MMP will not detect multiple mounts. To make this work we need to unplug the device queue in RAID when the MMP block is being written. Also while reading the MMP block, we should read it from disk and not the cached one.

15 years agoBranch b1_8
huanghua [Thu, 29 Jan 2009 08:34:57 +0000 (08:34 +0000)]
Branch b1_8
b=18318
i=rahul
i=pravin

reserve larger reply buffer for close and unlink:
2.0 mds wire protocol returns capabilities.

15 years agofix typo
shadow [Wed, 28 Jan 2009 09:32:44 +0000 (09:32 +0000)]
fix typo

15 years agodon't lose wakeups for imp_recovery_waitq
shadow [Wed, 28 Jan 2009 09:12:00 +0000 (09:12 +0000)]
don't lose wakeups for imp_recovery_waitq

Branch b1_8
b=18154
i=nathan
i=rread

15 years agob=16895
girish [Wed, 28 Jan 2009 03:52:30 +0000 (03:52 +0000)]
b=16895
i=adilger (o=johann)
i=girish

dont assert on lvfs_check_rdonly()

15 years ago- added comment about LDLM_FL_* flags at Adilger's request;
yury [Tue, 27 Jan 2009 14:25:03 +0000 (14:25 +0000)]
- added comment about LDLM_FL_* flags at Adilger's request;
- ordered LDLM_FL_* flags correctly.

15 years ago b=17282
yury [Tue, 27 Jan 2009 13:17:48 +0000 (13:17 +0000)]
 b=17282
 r=johann,shadow

 - simplifies a lot lru resize memory presssure handling. No estimation pass is needed right now as it was too CPU consuming.

15 years agoBranch b1_8
johann [Mon, 26 Jan 2009 13:59:40 +0000 (13:59 +0000)]
Branch b1_8
b=18304
i=shadow
i=nathan

MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
As a conseqence, the patch from bug 18224 is not operational
and the MDS export cannot be reused on the OSTs until it gets
evicted.

15 years agoBranch b1_8
johann [Mon, 26 Jan 2009 13:42:42 +0000 (13:42 +0000)]
Branch b1_8
b=18307
i=adilger
i=shadow

send OST_CONNECT to OST_REQUEST_PORTAL instead of OST_IO_PORTAL (was used
for compatibility with releases < 1.4.6)

15 years agoBranch b1_8
yangsheng [Sat, 24 Jan 2009 18:11:17 +0000 (18:11 +0000)]
Branch b1_8
b=15593
i=nikita, shadow

Force to remove the suid in interop environment.

15 years agoBranch b1_8
yangsheng [Sat, 24 Jan 2009 17:56:44 +0000 (17:56 +0000)]
Branch b1_8
b=18091
i=adilger, tappro

Clean up IOC_{INIT, PACK, UNPACK} macros.

15 years agob=18045
grev [Sat, 24 Jan 2009 09:18:00 +0000 (09:18 +0000)]
b=18045
i=Adilger
-- get rid of Hendrix rates values
-- create NUM_FILES -le IFree
-- use mdsrate for cleanup instead of rm

15 years agoBranch b1_8
huanghua [Sat, 24 Jan 2009 02:07:36 +0000 (02:07 +0000)]
Branch b1_8
b=17777
i=robert.read

as Robert suggests, get version with more precise match.

15 years agoBranch b1_8
yangsheng [Fri, 23 Jan 2009 17:38:19 +0000 (17:38 +0000)]
Branch b1_8
b=17464

Fixed previous commit.

15 years agoBranch b1_8
huanghua [Fri, 23 Jan 2009 16:45:56 +0000 (16:45 +0000)]
Branch b1_8
b=17777
i=grev
i=adilger
i=robert.read

test scripts improvement to enable interop testing between 1.8 and 2.0

15 years agob=15524
girish [Fri, 23 Jan 2009 16:39:55 +0000 (16:39 +0000)]
b=15524
i=huanghua
i=kalpak

fix test 102f

15 years agoadd workaround for leak space in ESTALE case.
shadow [Fri, 23 Jan 2009 14:31:58 +0000 (14:31 +0000)]
add workaround for leak space in ESTALE case.
Branch b1_8
b=16417

15 years agob=18279
grev [Fri, 23 Jan 2009 13:27:32 +0000 (13:27 +0000)]
b=18279
i=Nathan
AT tests cleanup: run setup/cleanup only if AT tests run

15 years agoBranch b1_8
yangsheng [Fri, 23 Jan 2009 10:24:38 +0000 (10:24 +0000)]
Branch b1_8
b=17464
i=johann, bobijam

Swab the lov_mds_md when ioctl() be called directly in liblustre.

15 years agoBranch b1_8
tianzy [Fri, 23 Jan 2009 09:50:54 +0000 (09:50 +0000)]
Branch b1_8
when there is any inactive ost, refuse to do quotacheck
b=18151
i=johann
i=andrew.perepechko

15 years agoBranch b1_8
tianzy [Fri, 23 Jan 2009 06:21:47 +0000 (06:21 +0000)]
Branch b1_8
fix the test issue of test_12 in sanity-quota.sh
b=18151
i=johann
i=panda

15 years agoBranch b1_8
vitaly [Thu, 22 Jan 2009 20:52:17 +0000 (20:52 +0000)]
Branch b1_8
b=17748
i=grev
i=adilger

sanityN test fix

15 years agob=16895
girish [Wed, 21 Jan 2009 18:38:33 +0000 (18:38 +0000)]
b=16895
i=adilger (o=johann)
i=girish

don't LASSERT on read-only device

15 years agoBranch b1_8
huanghua [Wed, 21 Jan 2009 17:32:45 +0000 (17:32 +0000)]
Branch b1_8
b=18267

i=yong.fan
i=rahul.deshmukh

pass default group value, that is 0,  to ost:
this is for such interop: 1.8 client + 1.8 mds + 2.0 ost.

15 years agob=17310
yury [Wed, 21 Jan 2009 16:46:20 +0000 (16:46 +0000)]
b=17310
r=johann

- tmp fix for long connect rpc causing this assert in umount time

15 years agoBranch b1_8
fanyong [Wed, 21 Jan 2009 08:56:10 +0000 (08:56 +0000)]
Branch b1_8
b=17800
i=adilger
i=huanghua

Add missing ll_rpc_opcode_table entries for "MDS_WRITEPAGE" and "MDS_IS_SUBDIR".

15 years agoBranch b1_8
fanyong [Wed, 21 Jan 2009 03:26:52 +0000 (03:26 +0000)]
Branch b1_8
b=15691
i=huanghua
i=yep

Another diagnostic patch to collect more log when error.

15 years agoBranch b1_8
tianzy [Wed, 21 Jan 2009 02:56:51 +0000 (02:56 +0000)]
Branch b1_8
take refcount of export when quotacheck is going on
b=18126
i=shadow
i=panda

15 years agoBranch b1_8
adilger [Tue, 20 Jan 2009 19:22:32 +0000 (19:22 +0000)]
Branch b1_8
Minor updates to lustre_idl.h to make it largely match the version in HEAD.
A number of new #defines from HEAD are added to avoid potential future
conflicts, along with whitespace cleanups.
b=17800

15 years agoalways handle mds connect as reconnect.
shadow [Tue, 20 Jan 2009 13:02:28 +0000 (13:02 +0000)]
always handle mds connect as reconnect.

Branch b1_8
b=18224
i=johhan
i=green
i=adilger
i=wangdi

15 years agodon't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
shadow [Tue, 20 Jan 2009 08:57:54 +0000 (08:57 +0000)]
don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)

Branch b1_8
b=18238
i=green
i=umka

15 years agoBranch b1_8
adilger [Tue, 20 Jan 2009 05:16:10 +0000 (05:16 +0000)]
Branch b1_8
Minor updates to lustre_idl.h to make it largely match the version in HEAD.
A number of new #defines from HEAD are added to avoid potential future
conflicts, along with whitespace cleanups.
b=17800

15 years agob=17942
kalpak [Tue, 20 Jan 2009 04:43:42 +0000 (04:43 +0000)]
b=17942
i=adilger
i=girish

update fiemap structures and API to match upstream

15 years agoBranch b1_8
fanyong [Tue, 20 Jan 2009 01:38:17 +0000 (01:38 +0000)]
Branch b1_8
b=18182
i=tianzy
i=panda

Set stripe count as 1 and stripte index as 0 to make sure the file is created on first ost.

15 years agoBranch b1_8
adilger [Tue, 20 Jan 2009 01:00:13 +0000 (01:00 +0000)]
Branch b1_8
Minor updates to lustre_idl.h to make it largely match the version in HEAD.
A number of new #defines from HEAD are added to avoid potential future
conflicts, along with whitespace cleanups.
b=17800

15 years agorelease lock refernce only after add page to pages list.
shadow [Mon, 19 Jan 2009 21:19:04 +0000 (21:19 +0000)]
release lock refernce only after add page to pages list.

Branch b1_8
b=17972
i=shadow
i=johann

15 years agoBranch b1_8
johann [Mon, 19 Jan 2009 12:28:40 +0000 (12:28 +0000)]
Branch b1_8
b=18044
i=bobijam
i=vitaly

copy DISCARD_DATA flag if cp ast includes bl ast.

15 years agoBranch b1_8
yangsheng [Thu, 15 Jan 2009 02:31:35 +0000 (02:31 +0000)]
Branch b1_8
b=18056
i=adilger, tappro

Fix warnings during automake-1.9.

15 years agob=17887
zhanghc [Tue, 13 Jan 2009 15:41:20 +0000 (15:41 +0000)]
b=17887

1, print the actual time to wait for
   lock cancellation in "waiting_locks_callback"

2, compare resource in "ost_rw_hpreq_lock_match"

3, replace "LDLM_ERROR" with "LDLM_WARN" if a lock
   is completed after waiting more than obd_timeout

i=vitaly.fertman
i=johann

15 years ago- warning instead of assert when gap is found in objids without recovery. That
tappro [Tue, 13 Jan 2009 07:06:37 +0000 (07:06 +0000)]
- warning instead of assert when gap is found in objids without recovery. That
may happen due to inode reuse problem
b:18168
i:shadow, adilger

15 years agowait until mds<>ost connected.
shadow [Sun, 11 Jan 2009 08:26:18 +0000 (08:26 +0000)]
wait until mds<>ost connected.
Branch b1_8
b=18150
i=grev

15 years agoresolve compile issue after 16919 landing.
shadow [Sun, 11 Jan 2009 07:16:06 +0000 (07:16 +0000)]
resolve compile issue after 16919 landing.
b=16919

15 years agob=18016
green [Sat, 10 Jan 2009 06:47:14 +0000 (06:47 +0000)]
b=18016
r=deen,wangdi

page index type in ll_read_ahead_page should be pgoff_t, not int

15 years agoChange sanity.sh to test "tar --xattr"
girish [Fri, 9 Jan 2009 18:06:07 +0000 (18:06 +0000)]
Change sanity.sh to test "tar --xattr"
b=15524
i=adilger
i=kalpak.shah

15 years agob=16919
green [Fri, 9 Jan 2009 04:03:08 +0000 (04:03 +0000)]
b=16919
r=adilger,shadow

Implement replay of bulk write RPCs, allow servers to reply to write RPCs before
all the metadata was finally committed to disk

15 years agoAdd configure option to allow repackaging of e2fsprogs as ldiskfsprogs. (by Jim Garlick)
girish [Thu, 8 Jan 2009 18:09:15 +0000 (18:09 +0000)]
Add configure option to allow repackaging of e2fsprogs as ldiskfsprogs. (by Jim Garlick)
b=17953
i=adilger
i=girish.shilamkar

15 years agoBranch b1_8
yangsheng [Thu, 8 Jan 2009 06:26:53 +0000 (06:26 +0000)]
Branch b1_8
b=17968
i=yangsheng, deen

Fixed various compiler warnings on ppc/ppc64 sles9.

Author: Jim Garlick(LLNL)

15 years agob=17694
vs [Wed, 7 Jan 2009 23:37:57 +0000 (23:37 +0000)]
b=17694
i=adilger

run the test in separate directory

15 years agoBranch b1_8
adilger [Wed, 7 Jan 2009 20:55:55 +0000 (20:55 +0000)]
Branch b1_8
Use __GFP_HIGHMEM so that the pages can use all of the available memory
on 32-bit machines.

b=17576
i=adilger (alex.zhuravlev original patch)
i=nikita

15 years agob=17710
zhanghc [Wed, 7 Jan 2009 08:04:18 +0000 (08:04 +0000)]
b=17710

uncommit the patch for further investigation

i=Andreas

15 years ago- fix for 18148
tappro [Tue, 6 Jan 2009 07:41:21 +0000 (07:41 +0000)]
- fix for 18148
b:18148
i:shadow,bzzz

15 years agoBranch b1_8
adilger [Tue, 6 Jan 2009 03:34:26 +0000 (03:34 +0000)]
Branch b1_8
Remove obsolete test.

15 years ago- synchronize test with HEAD version and rename it to test_35 (just like in HEAD);
yury [Mon, 5 Jan 2009 18:09:35 +0000 (18:09 +0000)]
- synchronize test with HEAD version and rename it to test_35 (just like in HEAD);

- OBD_RACE in all execution branches in ldlm_lock_decref_internal() to balance OBD_RACE in bl_ast
and prevent hanging.

15 years agob=18127
yury [Wed, 31 Dec 2008 16:10:56 +0000 (16:10 +0000)]
b=18127

 - reformat in test 47 after test 46 which uses 6 OST

15 years agob=17797
grev [Tue, 30 Dec 2008 12:13:55 +0000 (12:13 +0000)]
b=17797
i=H.Huang
test_57b remote mds fix

15 years agob=17710
zhanghc [Tue, 30 Dec 2008 06:23:19 +0000 (06:23 +0000)]
b=17710

add "sys.ldlm_timeout" as an option to the
MKFS.LUSTRE for setting "ldlm_timeout"

i=Johann
i=Nathan.Rutman

15 years agosend ACTIVATE event only if connect finished and import have state FULL.
shadow [Mon, 29 Dec 2008 09:46:46 +0000 (09:46 +0000)]
send ACTIVATE event only if connect finished and import have state FULL.

Branch b1_8
i=addilger
i=tappro
i=green

15 years agoBranch b1_8
tianzy [Mon, 29 Dec 2008 07:19:04 +0000 (07:19 +0000)]
Branch b1_8
fix test problems in test_18 of sanity-quota.sh
b=17985
i=shadow

15 years agorevert patch from bug 16417
shadow [Mon, 29 Dec 2008 06:15:08 +0000 (06:15 +0000)]
revert patch from bug 16417

15 years agoBranch b1_8
yangsheng [Sun, 28 Dec 2008 16:18:21 +0000 (16:18 +0000)]
Branch b1_8
b=17153
i=adilger, tappro

Remove 2.4 compatibility.

15 years agob:18103
tappro [Thu, 25 Dec 2008 07:32:11 +0000 (07:32 +0000)]
b:18103
i:adilger, bzzz

15 years agochecking maximal acl size before set.
shadow [Wed, 24 Dec 2008 21:20:45 +0000 (21:20 +0000)]
checking maximal acl size before set.

Branch b1_8
b=17636
i=johann
i=deen

15 years agoBranch b1_8
yangsheng [Wed, 24 Dec 2008 17:45:20 +0000 (17:45 +0000)]
Branch b1_8
b=17153
i=adilger, tappro

Remove 2.4 compatibility from build process.

15 years agob=18083
anserper [Wed, 24 Dec 2008 17:30:48 +0000 (17:30 +0000)]
b=18083
i=Yong Fan
i=Andrew Perepechko

Enforce op. quota file format for MDS too

15 years agor=shadow
yury [Wed, 24 Dec 2008 13:00:31 +0000 (13:00 +0000)]
r=shadow

- check for OBD_FAIL_OBD_LLOG_SETUP fail_loc regardless ->lop_setup definition. This is required because
  after changes in mds_ost llog setup, which is now done on behalf of separate thread, we need to make
  llog_setup() fail in context of mds start thread to make the test actualy work. And those llogs which
  get initialized in mds start thread do not have ->lop_setup defined.