Whamcloud - gitweb
deen [Thu, 7 May 2009 08:25:01 +0000 (08:25 +0000)]
Fix for conf-sanity test_50e, implement correct waiting.
Original patch by Alexey Lyashkov.
b=19280
i=robert.read
jxiong [Thu, 7 May 2009 01:09:21 +0000 (01:09 +0000)]
b=19264
r=ericm,fanyong
Fixed an issue in __mds_lov_synchronize(), where mds_propagate_capa_keys() tried to propagate the capa keys to all OSTs. This is not good because if one of the OST is not ready, it will cause the mds.osc to be set to inactive, and don't have a chane to get over.
adilger [Wed, 6 May 2009 23:24:00 +0000 (23:24 +0000)]
Branch HEAD
Change LINK_EA_MAGIC to avoid confusion with on-disk ext3 xattr magic 0xEA01.
i=robert.read
i=nathan
nathan [Wed, 6 May 2009 17:46:07 +0000 (17:46 +0000)]
b=19400
i=rread
i=manoj
pack the structs shared across machines
girish [Wed, 6 May 2009 17:09:58 +0000 (17:09 +0000)]
b=14109
update to new e2fsprogs release
isaac [Wed, 6 May 2009 16:07:37 +0000 (16:07 +0000)]
- fixed a comment typo.
pravins [Wed, 6 May 2009 13:24:08 +0000 (13:24 +0000)]
b=16721
i=rahul.deshmukh
i=rread
this patch add check in mdd rename operation for target object. return error if object exist but
dirent does not. it solve problem of double object deletion.
pravins [Wed, 6 May 2009 12:44:36 +0000 (12:44 +0000)]
b=16721
i=zam
i=rread
create separate context to execute osd specific transaction to avoid mdd
callback related to stop/commit transaction.
pravins [Wed, 6 May 2009 12:41:07 +0000 (12:41 +0000)]
b=16721
i=fanyong
i=rahul
return -ESTALE for dead inodes.
pravins [Wed, 6 May 2009 12:39:10 +0000 (12:39 +0000)]
b=16721
i=yong.fan
i=rahul.deshmukh
check object before starting mdd operation.
grev [Wed, 6 May 2009 12:07:31 +0000 (12:07 +0000)]
b=18921
i=Robert.Read
acc-sm type2 tests: PARALLEL_TESTS new test suite
grev [Wed, 6 May 2009 10:12:31 +0000 (10:12 +0000)]
b=19107
i=Maxim.Patlasov
acc-sm liblustre skip list
girish [Wed, 6 May 2009 03:20:48 +0000 (03:20 +0000)]
b=19058
i=adilger
i=johann
return proper error when MMP fails during remount
fanyong [Wed, 6 May 2009 01:47:56 +0000 (01:47 +0000)]
Branch HEAD
b=19308
i=jinshan.xiong
i=tianzy
1) convert "struct lustre_disk_dqblk_v2 *" to "char *" when memcmp for quota information
2) more sanity check when set brw_page flags.
3) alignment fix.
anserper [Tue, 5 May 2009 20:49:07 +0000 (20:49 +0000)]
�� b=19319
�� i=Alexander Zarochentsev
isaac [Tue, 5 May 2009 20:46:42 +0000 (20:46 +0000)]
b=15332,i=liang:
- add a new LND optiion to control peer buffer credits on routers.
stinson1947 [Tue, 5 May 2009 18:54:22 +0000 (18:54 +0000)]
Updated version string for 1.8.2.
johann [Tue, 5 May 2009 17:21:21 +0000 (17:21 +0000)]
Branch HEAD
b=18674
i=shadow
i=bobijam
print the correct number of active rpcs when denying reconnection.
johann [Tue, 5 May 2009 17:18:59 +0000 (17:18 +0000)]
Branch HEAD
b=18674
i=adilger
i=bobijam
print op code in ptlrpc_expire_one_request().
zhanghc [Tue, 5 May 2009 13:36:06 +0000 (13:36 +0000)]
b=16818
fix a bug in previous patch for the bug
i=nathan.rutman
i=robert.read
pravins [Tue, 5 May 2009 09:49:07 +0000 (09:49 +0000)]
b=19142
i=rread
i=adilger
print free inode count in case of error in inode create.
liuy [Tue, 5 May 2009 08:18:24 +0000 (08:18 +0000)]
Branch HEAD
some improvements:
-use fd->hints structure instead of MPI Info routines
-change hints name as romio requires
-check lustre/lustre_user.h header file in configure instead of giving
the lustre structs/constants
-add new MPE logging
-revert the hint settings in ad_lustre_open.c
-remove the two confusing hints "contig_data" and "samesize"
-fix avail_cb_node calculation in ad_lustre_aggregate.c
fanyong [Tue, 5 May 2009 02:23:03 +0000 (02:23 +0000)]
Branch HEAD
b=19153
i=jinshan.xiong
i=vitaly
update inode before request buffer freed.
bobijam [Tue, 5 May 2009 01:51:55 +0000 (01:51 +0000)]
Branch HEAD
b=17402
i=vitaly.fertman
i=eric.mei
Clear up nid_stat's nid_exp_ref_count usage.
brian [Mon, 4 May 2009 21:30:56 +0000 (21:30 +0000)]
b=17671
i=zhen.liang
i=sheng.yang
Enable Lustre to be built with OFED 1.4.1.
This is just the glue needed for the build with OFED 1.4.1 to actually work.
OFED 1.4.1 won't actually get built with a Lustre branch until a target file
specifies:
OFED_VERSION=1.4.1-rc4
Note that OFED 1.4.1 only supports RHEL4, RHEL5 and SLES10 targets. SLES9
users are out of luck (but have been for some time with OFED).
brian [Mon, 4 May 2009 19:10:39 +0000 (19:10 +0000)]
b=16312
i=wangyb
i=jack
- allow the download functions to take a "force" argument to force a download
even if we have the file locally (maybe it's corrupt or something)
- download_srpm() should respect the $DOWNLOAD variable
- better detect failed downloads
- return failure if the untar of the OFED tarball fails
- gather diagnostic info if configure for lustre fails
- handle OFED RC and daily release filenames which have an extra tag appended
to them
- move the kernel-ib-devel unpacking to outside the building function so that
it's unpacked even if a cached build is used
- properly return failure if build_lustre() fails
- wipe the KABI whitelist when building the RHEL5 kernel as we do indeed alter
the KABI by using an external OFED stack instead of the inkernel OFED stack
- replace a perl s/// usage with sed for efficiency and portability
brian [Mon, 4 May 2009 19:05:05 +0000 (19:05 +0000)]
b=16312
i=wangyb
i=jack
- add "lnxmin" to the sles10 target file to deal with the breakage that bug
19336 will aim to fix properly
- this meets up with the recent commit to the build module
brian [Mon, 4 May 2009 18:50:24 +0000 (18:50 +0000)]
b=16312
i=wangyb
i=jack
- fixes vanilla build: typo
- enhanced incomplete cache detection
- put ext3 source into kernel-devel for rhel5 to get rid of ugly lbuild hack
- clean up the RPM smp type specification as well as .config selection
- we can actually get rid of the rhel5 -smp .configs in the lustre tree now
- find the kernel rpm by looking at rpm attributes instead of just filenames
- fix a bug where some old code was setting global variables it is no longer
able to set
- add "lnxmin" to the sles10 target file to deal with the breakage that bug
19336 will aim to fix properly
grev [Mon, 4 May 2009 12:26:42 +0000 (12:26 +0000)]
b=19143
i=Robert.Read
t-f pri-facet sec-facet different devs names fix
liangzhen [Mon, 4 May 2009 05:01:09 +0000 (05:01 +0000)]
Branch HEAD
Fix memory leak in obdecho
b=18096
i=wangdi
i=johann
liangzhen [Mon, 4 May 2009 04:51:16 +0000 (04:51 +0000)]
Branch HEAD
Enable obdecho connecting
b=18096
o=ericm
i=liang
girish [Sun, 3 May 2009 18:36:35 +0000 (18:36 +0000)]
b=19184
i=adilger
i=girish
o=brian behlendorf
add proper locking for creation of jbd_4k slab cache
fanyong [Sat, 2 May 2009 09:17:04 +0000 (09:17 +0000)]
Branch HEAD
b=15691
i=eric.mei
i=peng.ye
Another diagnostic patch to collect more log on both client and OSS.
brian [Fri, 1 May 2009 19:08:25 +0000 (19:08 +0000)]
b=13908
i=wangyb
i=jack
Now that we have tightened up dependencies between lustre-modules and
the kernel, we need to drop the -[big]smp from the required kernel version
for SLES kernels.
girish [Fri, 1 May 2009 15:23:01 +0000 (15:23 +0000)]
b=19058
b=19184
i=adilger,girish
o=Bernd Schubert
o=Brian Behlendorf
ext3_remount() should return correct error code
add locking for creation of jbd_4k slab cache. Only needed for RHEL5 kernel
manoj [Thu, 30 Apr 2009 20:55:08 +0000 (20:55 +0000)]
b=19051
r=yong.fan, jinshan.xiong
Use 64bits (loff_t) for holding file size instead of size_t
johann [Thu, 30 Apr 2009 08:58:01 +0000 (08:58 +0000)]
Branch HEAD
b=18577
i=shadow
i=hongchao
create_count always drops to the min value (=32) because grow_count
is being changed before the precreate RPC completes.
shadow [Thu, 30 Apr 2009 03:21:59 +0000 (03:21 +0000)]
don't leak transno for recovery.
Branch HEAD
b=19085
i=rread
i=adilger
fanyong [Thu, 30 Apr 2009 01:43:19 +0000 (01:43 +0000)]
Branch HEAD
b=19171
i=tianzy
i=robert.read
Hold obd reference count when quota recovery.
yangsheng [Thu, 30 Apr 2009 01:30:39 +0000 (01:30 +0000)]
Branch HEAD
b=18790
i=robert, panda
Check the types for PPC64 platform.
vitaly [Wed, 29 Apr 2009 21:58:51 +0000 (21:58 +0000)]
Branch HEAD
b=18851
i=tappro
i=vs
walk through abort_list in abort_lock_replay_queue()
brian [Wed, 29 Apr 2009 18:06:28 +0000 (18:06 +0000)]
b=16312
i=wangyb
i=jack
Bug fixes to fix:
- always naming the lustre packages as if they were patchless clients
- detect a partial cache commit and back it out
Provide a more clear e-mail subject line in the case of an untrapped error.
Remove an irrelevent comment.
vitaly [Wed, 29 Apr 2009 15:37:44 +0000 (15:37 +0000)]
Branch HEAD
b=18571
i=girish
i=adilger
do not take an extra reference on journal bh which prevents it to be released.
zhanghc [Wed, 29 Apr 2009 10:15:24 +0000 (10:15 +0000)]
b=18733
add protection (spin_lock(&lock->l_lock)) before
using lock->l_resource in "lock_res_and_lock"
i=johann
i=green
ericm [Wed, 29 Apr 2009 03:47:57 +0000 (03:47 +0000)]
branch: HEAD
fix replayed meta sequence handling.
b=18982
r=rread
r=fanyong
bobijam [Wed, 29 Apr 2009 02:07:28 +0000 (02:07 +0000)]
Branch HEAD
b=18690
o=adilger
i=johann
i=zhenyu.xu (bobijam)
Increase some hash table sizes.
bobijam [Wed, 29 Apr 2009 01:45:10 +0000 (01:45 +0000)]
Branch HEAD
b=18073
i=johann
i=adilger
i=sheng.yang
use "\\" as escape character to prevent '.' from being converted to '/' in
lctl get/set_param.
tianzy [Wed, 29 Apr 2009 01:29:49 +0000 (01:29 +0000)]
Branch HEAD
fix the printing of rq_xid(change into LPU)
b=18280
i=johann
i=tianzy
vitaly [Tue, 28 Apr 2009 17:31:19 +0000 (17:31 +0000)]
Branch HEAD
b=19154
i=zam
i=rread
properly protect filter_destroy() with a DLM lock
jxiong [Tue, 28 Apr 2009 05:02:36 +0000 (05:02 +0000)]
b=16839
r=shadow,rread
The rpc in __mds_lov_synchronize() should be sync.
shadow [Tue, 28 Apr 2009 04:27:08 +0000 (04:27 +0000)]
correctly skip time estimate if in recovery
Branch HEAD
b=19223
i=johann
i=nathan
stinson1947 [Tue, 28 Apr 2009 01:35:16 +0000 (01:35 +0000)]
Updated for the next 2.0 bi-weekly build.
bobijam [Tue, 28 Apr 2009 01:22:24 +0000 (01:22 +0000)]
Branch HEAD
b=18683
i=joahnn
i=sheng.yang
If no options are given, tunefs.lustre/mkfs.lustre just print info.
robert.read [Mon, 27 Apr 2009 21:49:25 +0000 (21:49 +0000)]
Branch HEAD
b=19104
r=wangdi,fanyong
we have to replace an entirely new environment for ptlrpc callbacks. Otherwise, in liblustre, the env running for IO would be used for callback.
Relanding Jay's patch.
robert.read [Mon, 27 Apr 2009 21:45:04 +0000 (21:45 +0000)]
Branch HEAD
b=18818
i=manoj
i=rread
fix RMF_SETINFO_VAL type and verify all setinfo lengths
Relanding Nathan's patch.
robert.read [Mon, 27 Apr 2009 21:43:36 +0000 (21:43 +0000)]
Branch HEAD
port lazystatfs to HEAD.
Branch HEAD
b=17974
i=rread
Re-landing shadow's patch.
isaac [Mon, 27 Apr 2009 18:56:09 +0000 (18:56 +0000)]
- back out peer_buffer_credits changes, as requested by rread.
grev [Sat, 25 Apr 2009 09:40:28 +0000 (09:40 +0000)]
b=19262
i=Robert.Read
include functions.sh properly
robert.read [Sat, 25 Apr 2009 01:42:31 +0000 (01:42 +0000)]
Branch HEAD
b=17974
Backing out change due to sanity failures.
robert.read [Fri, 24 Apr 2009 23:42:25 +0000 (23:42 +0000)]
Branch HEAD
b=18818
Backing changes due to sanity failure.
robert.read [Fri, 24 Apr 2009 23:26:11 +0000 (23:26 +0000)]
Branch HEAD
b=16312
Revert cause of build failure.
robert.read [Fri, 24 Apr 2009 23:09:28 +0000 (23:09 +0000)]
Branch HEAD
b=19104
Revertting patches due to sanity failure.
brian [Fri, 24 Apr 2009 20:12:13 +0000 (20:12 +0000)]
b=16312
Remedial cleanup of an lbuild feature that changed while this patch was
waiting to land.
brian [Fri, 24 Apr 2009 18:33:17 +0000 (18:33 +0000)]
b=16312
i=robert.read
i=yibin.wang
Make lbuild build kernels based on the vendor's own SRPM rather than all of
the gyrations we used to go through to build the kernel tarballs and so on.
This should give people a kernel (and associated packages) that looks much
more like the kernel (and packages) they are used to getting from their
vendor.
This has resulted in a much easier to understand lbuild and once we finally
fully drop support for rhel4 and sles9 we can get rid of the
lustre-kernel-2.4.spec.in, build/lbuild.old_school and build/lmake files.
Anyone know knows anything about the build system will understand what kind
of simplicity in understanding of how the build works that will yield.
brian [Fri, 24 Apr 2009 17:59:56 +0000 (17:59 +0000)]
*** empty log message ***
brian [Fri, 24 Apr 2009 17:10:03 +0000 (17:10 +0000)]
With 1.6 heading towards maintenance mode, use a dedicated branch of the
build system for it to isolate future build changes away from it.
shadow [Fri, 24 Apr 2009 09:04:38 +0000 (09:04 +0000)]
fix typo founded by maxim.
Branch HEAD
fanyong [Fri, 24 Apr 2009 02:03:43 +0000 (02:03 +0000)]
Branch HEAD
b=19149
i=tianzy
i=panda
Make sure quota has been released to master in test_6.
robert.read [Fri, 24 Apr 2009 01:18:01 +0000 (01:18 +0000)]
Branch HEAD
b=13908
Also build the source package, for the same reason.
jxiong [Fri, 24 Apr 2009 01:05:55 +0000 (01:05 +0000)]
b=19104
r=wangdi,fanyong
we have to replace an entirely new environment for ptlrpc callbacks. Otherwise, in liblustre, the env running for IO would be used for callback.
robert.read [Fri, 24 Apr 2009 00:57:07 +0000 (00:57 +0000)]
Branch HEAD
Whitespace cleanup and silence persistent error message.
robert.read [Fri, 24 Apr 2009 00:41:59 +0000 (00:41 +0000)]
Branch HEAD
b=13908
Re-enable building lustre-client-tests, so oslo can find tests for all clients,
especially clients we don't build a server for like ia64.
Tested with "make rpms"
robert.read [Fri, 24 Apr 2009 00:11:26 +0000 (00:11 +0000)]
Branch HEAD
b=19190
i=nathan
Wait for recovery to complete before finishing test.
robert.read [Thu, 23 Apr 2009 21:01:22 +0000 (21:01 +0000)]
Branch HEAD
b=19280,17974
Temporarily disable test 50e and 50f, which are broken in multi-node configurations.
nathan [Thu, 23 Apr 2009 20:54:00 +0000 (20:54 +0000)]
b=19047
i=adilger
i=nathan
add llog_reader man page
nathan [Thu, 23 Apr 2009 19:35:49 +0000 (19:35 +0000)]
b=18818
i=manoj
i=rread
fix RMF_SETINFO_VAL type and verify all setinfo lengths
shadow [Thu, 23 Apr 2009 18:14:02 +0000 (18:14 +0000)]
port lazystatfs to HEAD.
Branch HEAD
b=17974
i=rread
adilger [Thu, 23 Apr 2009 08:01:43 +0000 (08:01 +0000)]
Branch HEAD
Fix error message to match actual filenames.
Fix OBD_FAIL_TGT_REPLAY_DROP comment to have correct fail_loc value.
b=19114
isaac [Thu, 23 Apr 2009 04:29:19 +0000 (04:29 +0000)]
b=15332,i=liang:
- add a new LND optiion to control peer buffer credits on routers.
shadow [Wed, 22 Apr 2009 18:03:59 +0000 (18:03 +0000)]
allocate inodes for llog files in last inodes group, for avoid broking
recovery.
Branch HEAD
b=18192
i=adilger
i=shadow
i=girish
brian [Wed, 22 Apr 2009 17:55:29 +0000 (17:55 +0000)]
b=13908
i=yibin.wang
i=sheng.yang
Allow the name, version, kernel release and release of the lustre packages
to be defined on the command line.
With such a feature, actually properly name the patchless client packages in
our own build.
Tighten up some dependencies.
nathan [Wed, 22 Apr 2009 17:12:27 +0000 (17:12 +0000)]
b=18676
i=johann
i=manoj
replace obd ioctl with dir ioctl for changelogs
vitaly [Wed, 22 Apr 2009 13:37:05 +0000 (13:37 +0000)]
Branch HEAD
b=15977
i=deen
i=vs
lctl dk should zero the file before writing the debug log into,
otherwise, if a file was long enough and is not completely overwritten
we get a mix of several logs.
deen [Wed, 22 Apr 2009 12:08:34 +0000 (12:08 +0000)]
This patch implements separate function for handling OPEN requests from Lustre clients acting as
NFS servers, named mdt_open_anon_by_fid(). It searches for existing object and, if succeed, takes
an open lock on the object and its attributes. If a client didn't ask for the open lock, it releases
it at the end. We must always take an open lock at the beginning in order to protect object's
attributes.
b=16715
i=oleg.drokin
i=alexey.lyashkov
zhanghc [Wed, 22 Apr 2009 02:07:14 +0000 (02:07 +0000)]
b=18689
change thread name "ll_log_commit_*" in
llog_recov_thread_init into "lcm_*" so it
can be fitted into task struct
i=adilger
anserper [Wed, 22 Apr 2009 01:08:07 +0000 (01:08 +0000)]
b=16890
the missing part from a previous commit
jxiong [Wed, 22 Apr 2009 00:45:23 +0000 (00:45 +0000)]
b=18711
r=jay, adilger
in lu_context_key_deregister
- to not print message inside a spinlock
- assert when the key has extra references
anserper [Wed, 22 Apr 2009 00:26:47 +0000 (00:26 +0000)]
b=16890
i=Johann Lombardi
i=ZhiYong Tian
avoid several unnecessary memsets
grev [Tue, 21 Apr 2009 19:03:40 +0000 (19:03 +0000)]
b=19116
i=Robert.Read
convert "-" to X to create client variable name
yangsheng [Tue, 21 Apr 2009 14:42:04 +0000 (14:42 +0000)]
Branch HEAD
b=18810
i=bzzz, adilger
Use sb_bgl_lock to protect bitmaps.
Author: adilger
tianzy [Tue, 21 Apr 2009 09:22:08 +0000 (09:22 +0000)]
Branch HEAD
1. delete the lc_watchdog_touch_ms()
2. changing to seconds instead of ms
3. delete lcw_time in lc_watchdog
b=18582
i=nathan
i=johann
grev [Mon, 20 Apr 2009 21:11:47 +0000 (21:11 +0000)]
b=19175
i=Robert.Read
check_catastrophe fix
anserper [Mon, 20 Apr 2009 21:02:04 +0000 (21:02 +0000)]
b=18951
grev [Mon, 20 Apr 2009 18:49:35 +0000 (18:49 +0000)]
b=19140
i=Alexander.Zarochentsev
PTLDEBUG/SUBSYSTEM t-f set_param fix:
accept the string value
anserper [Mon, 20 Apr 2009 18:29:09 +0000 (18:29 +0000)]
b=18951
i=Nathan Rutman
i=Vladimir Saveliev
fix the race between fflush and pause
zam [Thu, 16 Apr 2009 07:54:44 +0000 (07:54 +0000)]
Branch HEAD
b=18988
i=Alexey.Lyashkov
new sanity test for for O_APPEND.
deshmukh [Thu, 16 Apr 2009 05:42:38 +0000 (05:42 +0000)]
b=17931
i=fonyoung
i=grev
conf_sanity_33a test-case is fixed now.
maxim [Wed, 15 Apr 2009 19:38:54 +0000 (19:38 +0000)]
b=18844
i=isaac
Landing a patch fixing deadlock in usocklnd on HEAD. The patch also includes:
- minor code cleanup suggested by Isaac in comment #6 of bug #18844
- trivial cleanup patch for acceptor.c (att #21983 of bug #14132)
- minor one-line fix for handling EINTR error of poll(2) - inspected by Shadow.
maxim [Wed, 15 Apr 2009 19:22:44 +0000 (19:22 +0000)]
b=18414
A few words to lnet/ChangeLog about 'running out of ports' fix.
yangsheng [Wed, 15 Apr 2009 10:35:49 +0000 (10:35 +0000)]
Branch HEAD
b=18033
Fix the typo.