Whamcloud - gitweb
Jian Yu [Mon, 8 Nov 2010 10:09:36 +0000 (18:09 +0800)]
b=23289 lprocfs_counter cleanup
Eliminate unnecessary atomic operations in lprocfs_counter.
o=liang
i=andreas.dilger
i=maxim.patlasov
Elena Gryaznova [Mon, 8 Nov 2010 16:09:06 +0000 (19:09 +0300)]
b=23049 various t-f.sh patches
o=Brian.Murrell
i=grev
rundbench is a bash script;
obdfilter-survey is a bash script;
don't su if MPI_USER == "";
Elena Gryaznova [Mon, 8 Nov 2010 15:57:40 +0000 (18:57 +0300)]
b=23049 set path to truncate
i=Brian.Murrell
Andreas Dilger [Mon, 8 Nov 2010 09:21:34 +0000 (12:21 +0300)]
b=20101 a fix for lfs getstripe --pool
a=Jacques-Charles Lafoucriere (CEA)
a=Andreas Dilger
i=Andreas Dilger
i=Andrew Perepechko
Brian J. Murrell [Mon, 8 Nov 2010 11:20:51 +0000 (19:20 +0800)]
b=24094 add kver; update kdir, kobjdir
Add a new macro to allow the caller to simply define which kernel
he wants to build for.
Refactor the kdir and kobjdir to function more properly.
Remove krequires. Nothing is using it anymore.
i=wangyb
i=cliff
adilger [Mon, 8 Nov 2010 11:20:50 +0000 (19:20 +0800)]
b=24088 Quiet unused "path" variable warning.
o=adilger
i=brian
i=panda
Brian J. Murrell [Mon, 8 Nov 2010 11:20:49 +0000 (19:20 +0800)]
b=24092 build src.rpm for lustre if requested
Fix a bug where we were not building the lustre SRPM even if
requested.
i=wangyb
i=cliff
Vladimir Saveliev [Fri, 5 Nov 2010 15:22:14 +0000 (18:22 +0300)]
b=19669 update LOV object version only on data change
LOV object version is changed only when LOV object content is changed, either
on write or size change (truncate). No more on metadata change like
chmod/chown/chgrp. Needed for HSM.
a=aurelien.degremont
i=mikhail.pershin
Vladimir Saveliev [Fri, 5 Nov 2010 15:22:01 +0000 (18:22 +0300)]
b=20631 grouplock use from MDT for HSM
https://bugzilla.lustre.org/attachment.cgi?id=31976
a=jc.lafoucriere
i=oleg.drokin
Vladimir Saveliev [Fri, 5 Nov 2010 15:21:44 +0000 (18:21 +0300)]
b=15599 hsm infrastructure
https://bugzilla.lustre.org/attachment.cgi?id=31961
and wiretest update (https://bugzilla.lustre.org/attachment.cgi?id=31991)
a=jc.lafoucriere
i=andreas.dilger
i=nathan.rutman
i=vladimir.saveliev
Terry Rutledge [Fri, 5 Nov 2010 17:11:20 +0000 (10:11 -0700)]
Updated for build 07.
Vitaly Fertman [Thu, 4 Nov 2010 20:53:32 +0000 (23:53 +0300)]
b=23064 Adopt for new backing_dev_info interface.
o=yangsheng
i=dmitry
Elena Gryaznova [Tue, 2 Nov 2010 13:55:52 +0000 (16:55 +0300)]
b=23402 mmp_fini () multiple oss fix
i=Yu.Jian
LiuYing [Tue, 2 Nov 2010 01:18:04 +0000 (09:18 +0800)]
b=23399 Deadlock probably due to statahead
partly revoke patch (id=28310)
o=fanyong
i=tappro
i=wangdi
Jian Yu [Tue, 2 Nov 2010 14:05:34 +0000 (22:05 +0800)]
b=23289 cleanup unnecessary spinlock dance in ldlm
o=liang
i=andreas.dilger
i=oleg.drokin
Brian J. Murrell [Tue, 2 Nov 2010 20:14:44 +0000 (04:14 +0800)]
b=22281 refactor packaging and versioning
Configure "--with-dist" switch to optimize configure to do just
enough to enable "make dist" to work. There is likely more optimizations
that could be done
Provide reasonable defaults for the --with-linux and --with-linux-obj
configure arguments.
The LB_BUILDID macro determines the BUILDID to be used in the resulting
packages.
The LB_DOWNSTREAM_RELEASE macro implements the --with-downstream-release
configure option for downstream packages to provide their own versioning
details
Many changes to support allowing packagers to define a value in the release
string.
- allow callers to tell rpm where to find the kernel source/headers and
objects.
- default to /lib/modules/$(uname -r)/build if not specified
- XXX: this should probably be handled completely in configure
- allow callers to tell rpm what the release string should be set to
- the spec will default to the value that configure uses currently if
not specified
- this allows both the release string used in Lustre and the Release:
string used in the RPM to be in sync
- in the make rpms target, strip the configure arguments for --with-linux
and --with-linux-obj and give them to rpmbuild via it's new switches
- also, pass the release defined by the main configure to the RPM spec for
further consistency
- XXX: it is questionable whether this in fact has any value
- put the "downstream release" right into the release as it is defined by
by configure
- bring the ldiskfs spec more into sync with the lustre spec file
version_tag.pl/make_META.pl needed better support for git.
It also needed better support for transporting the information
available in a source code system into the tarball so that when
unpacked, where no sourcecode system was in place, the origins
of the code could still be determined.
Example output:
git: v1_8_3_RC1-g751b802-CHANGED-2.6.18-128.1.1.el5_lustre.1.8.0.50.20090302160821smp
cvs: b1_8-
20100429144101-CHANGED-2.6.18-128.1.1.el5_lustre.1.8.0.50.20090302160821smp
When run in a tree from a tarball, the output will be composed
of the same tag and build id (hash or datestamp) as the source the
tarball came from, but will refelect whatever kernel the unpacked
source was configured with.
Rename the tree_status file to META.
There are times when the kernel should be rebuilt for all distros, but
other times only for certain distros. Make the "all distros" case
easier than having to increment the version in all of the
build/lbuild-$distro files.
i=wangyb
i=mjmac
When the lustre-fix is .0, when a tag is made, it includes the .0 so
just drop it.
Only append the .$BUILDID to the $EXTRA_VERSION when $BUILDID is
actually not empty. Otherwise we get a trailing ".".
It seems that the regex(7) used in bash's [[ ]] is not terribly
portable across bash versions. SLES10 expects (, ), and | (at least)
escaped with \. RHEL5 will accept them either escaped or not.
Ubuntu fails to match if they are escaped.
We need a more consistent test for being a subdir of Lustre.
Differences in how automake passes dirnames made the existing code
unpredictable.
When the tag format is vW_X_Y_Z and lustre-fix (Z) is _0 when a tag
is made, it includes the _0 so just drop it.
i=yangsheng
i=wangyb
The return code from git status seems to be not as one would expect
and it can return both 1 or 0 in a git repository. So change up and
use git branch instead.
Simplify the logic around this determination as well.
Due to some unknown issue with using rpmbuild with the ldiskfs tarball
causing symbol mismatches, do the right thing instead and build ldiskfs
first and then build lustre with the external ldiskfs.
Right now the method for doing this is a bit hacky since we point lustre
at the ldiskfs rpm BUILD product. The right fix is to make an
ldiskfs-devel RPM, which is a future enhancement of this work.
This also fixes two bugs with ldiskfs' external module support building.
One is that the external ldiskfs tree check was using a file that only
exists in ext3 and not ext4. The second is that ldiskfs' configure was
not properly initializing ldiskfs_is_ext4 and instead depending on the
rpmbuild/configure caller to pass flags to tell it to build with ext4.
i=wangyb
i=minh
yangsheng [Wed, 3 Nov 2010 06:55:01 +0000 (14:55 +0800)]
b=21610 Changes for 2.6.32 kernel udpate.
i=Andreas.Dilger
i=Mikhail.Pershin
Cliff White [Tue, 2 Nov 2010 17:02:23 +0000 (10:02 -0700)]
b=20805 rate limit D_NETERR messages
i=isaac
i=liang
change CDEBUG(D_NETERROR) to CNETERR in various places. Port of 1_8 to master.
Cleans up klnds, libcfs
Jian Yu [Wed, 3 Nov 2010 01:18:07 +0000 (09:18 +0800)]
b=18551 remove ENTRY & EXIT from hot hash functions
o=liang
i=rahul.deshmukh
i=jian.yu
Elena Gryaznova [Mon, 1 Nov 2010 16:00:12 +0000 (19:00 +0300)]
b=23869 HARD failure mode fixes
i=Brian.Murrel
i=Wei.G.Li
* facet_failover() has to restart only those affected facets which were UP
before the node failure
* replay-single tests which use shutdown_facet() && reboot_facet() instead of
facet_failover() have to take care about the affected facets
Rahul Deshmukh [Mon, 1 Nov 2010 13:23:16 +0000 (18:53 +0530)]
b=24010 lfs fid2path doesn't work for symlinks
o=thomas
i=andreas
i=rahul
Eric Mei [Mon, 1 Nov 2010 16:15:02 +0000 (10:15 -0600)]
b=24007 obj mode should be covered by MA_INODE.
r=di.wang
r=vitaly.fertman
LiuYing [Fri, 22 Oct 2010 09:29:37 +0000 (17:29 +0800)]
b=23924 LBUG: ASSERTION(hs == exp->exp_lock_hash) failed
fix the wrong assertion in lprocfs_exp_print_hash()
i=andreas
i=eric.mei
Landen [Tue, 2 Nov 2010 07:02:52 +0000 (15:02 +0800)]
b=23913 fix "ASSERTION(!cfs_list_empty(&dquot->dq_hash)) failed"
1. In an error handling, free_dquot() should be used.
2. Check if quota is on in dqacq_handler()
i=johann
i=panda
Landen [Tue, 2 Nov 2010 07:24:48 +0000 (15:24 +0800)]
b=23868 fix "sanity-quota test_18c: @@@@@@ FAIL: quotaon failed!"
i=johann
i=mikhail.pershin
Jian Yu [Tue, 2 Nov 2010 06:57:41 +0000 (14:57 +0800)]
b=23289 skip big loop in mds_lov_dump_objids if D_INFO is not set
o=liang
i=andreas.dilger
i=vitaly.fertman
Jian Yu [Tue, 2 Nov 2010 07:05:29 +0000 (15:05 +0800)]
b=23289 remove mc_waitq
Remove legacy.
o=liang
i=mikhail.pershin
i=di.wang
Jian Yu [Tue, 2 Nov 2010 07:12:41 +0000 (15:12 +0800)]
b=23289 don't need call mdd_links_get for the new created file
We know it's the first link for new created file, so we don't need to
call mdd_links_get, which is an xattr_get and overhead, while creating file.
o=liang
i=mikhail.pershin
i=di.wang
Jian Yu [Tue, 2 Nov 2010 06:46:15 +0000 (14:46 +0800)]
b=23289 disable pdo locks in MDD
MDT has already taken care of locks.
o=liang
i=mikhail.pershin
i=alex.zhuravlev
Landen [Tue, 2 Nov 2010 07:29:03 +0000 (15:29 +0800)]
b=22546 fix errors in test_18c
i=johann
i=panda
Mikhail Pershin [Fri, 22 Oct 2010 14:32:23 +0000 (18:32 +0400)]
b=20997 passthrough obd_force from lmv to mdc
i=zam,vs
Terry Rutledge [Fri, 22 Oct 2010 21:13:48 +0000 (14:13 -0700)]
Updated for 2.1.0 build 06.
Brian J. Murrell [Fri, 22 Oct 2010 11:31:43 +0000 (07:31 -0400)]
b=20744 return ofed version to 1.5.1
Since the patch from bug 20744 that updates the client to the
2.6.16.60-0.69.1 kernel has not landed yet, continue to build the
OFA OFED 1.5.1 until such time as the new kernel support has landed.
Andrew Perepechko [Fri, 22 Oct 2010 00:23:19 +0000 (04:23 +0400)]
b=22386 disallow wrong conf_param options
i=Johann Lombardi
i=Nathan Rutman
Jian Yu [Thu, 21 Oct 2010 05:33:16 +0000 (13:33 +0800)]
b=18551 adapt lu_site to new cfs_hash
o=zhen.liang
i=eric.mei
i=mikhail.pershin
yangsheng [Thu, 21 Oct 2010 08:23:42 +0000 (16:23 +0800)]
b=22514 Update RHEL5.5 & OEL5.5 to latest kernel.
--RHEL5 2.6.18-194.17.1.el5.
--OEL5 2.6.18-194.17.1.0.1.el5.
--Switch using 'inkernel' OFED stack.
--Build fixes for ppc64 & ia64.
Eric Mei [Wed, 20 Oct 2010 23:57:12 +0000 (17:57 -0600)]
b=19486 fix a typo in previous patch.
r=di.wang
Alexander.Zarochentsev [Tue, 19 Oct 2010 16:34:07 +0000 (20:34 +0400)]
b=22417 remove obd arg from ptlrpc_start_thread
i=Mikhail.Pershin
i=Andew.Perepechko
Johann Lombardi [Tue, 19 Oct 2010 02:01:36 +0000 (22:01 -0400)]
b=20744 add changelog entry
o=johann
i=brian
Brian J. Murrell [Tue, 19 Oct 2010 02:01:35 +0000 (22:01 -0400)]
b=20744 set arch correctly for ofed-devel
Oh boy, more fallout from our not building i586 kernels for sles.
If we are building lustre servers and NOT rebuilding the Suse ofed
RPM, then we need to look for the i586 ofed-devel package.
i=yangsheng
i=yibin.wang
Vitaly Fertman [Thu, 21 Oct 2010 23:27:19 +0000 (03:27 +0400)]
b=20744 remedial fix for inkernel ofed handling
o=brian
i=yangsheng
i=yibin.wang
It seems that the patch to add inkernel handling for SLES10's OFED introduced a
bug for other inkernel uses. This fixes that bug.
Brian J. Murrell [Tue, 19 Oct 2010 02:01:33 +0000 (22:01 -0400)]
b=20744 hacks for Novell bug 639581
i=mjmac
i=wangyb
Brian J. Murrell [Tue, 19 Oct 2010 02:01:32 +0000 (22:01 -0400)]
b=20744 hacks for Novell bug 578796
i=mjmac
i=wangyb
These hacks are needed to deal with Novell bug 578796.
Brian J. Murrell [Tue, 19 Oct 2010 02:01:31 +0000 (22:01 -0400)]
b=20744 use Novell supplied ofed-devel
Several changes to support using ofed-devel supplied in SLES10 SP3
(currently as PTFs).
Fix a bug with passing arguments to rpm2cpio that should not be passed.
Add a hook for distro specific module building after the kernel[-devel]
build and unpack.
Fix resolve_arch to be able to resolve either the "canonical" arch for
i?86, or the arch that the patchless kernel is expected to be.
Bail out of download_ofed earlier if using "inkernel".
Fix a bug where an rpm2cpio command was not being error checked.
Fix a bug where we were not processing and making available files
in all of the RPMS/i?86 subdirs.
i=mjmac
i=wangyb
Vitaly Fertman [Tue, 19 Oct 2010 23:54:17 +0000 (03:54 +0400)]
Revert "b=21760 Applications hung in direct I/O"
This reverts commit
75b7e5a7b448238e486b8f9bd753cda5f4b4e68b.
Vitaly Fertman [Tue, 19 Oct 2010 23:53:29 +0000 (03:53 +0400)]
Revert "b=19325 adjust waiting extent locks during 1st enqueue"
This reverts commit
5cced5eb96914a5a51a551ed12e92ebfc7834029.
Vitaly Fertman [Tue, 19 Oct 2010 00:09:16 +0000 (04:09 +0400)]
Revert "b=22176 Add .sync_fs super block handler"
This reverts commit
8cad1d968ad4efee46d08aee1e6a6266d4713e84.
Vitaly Fertman [Mon, 18 Oct 2010 22:50:27 +0000 (02:50 +0400)]
b=19813 cfs_hash_cond_del doesn't have dance of addref/decref anymore
o=liang
i=vitaly
i=tappro
Eric Mei [Fri, 15 Oct 2010 03:41:11 +0000 (21:41 -0600)]
b=23728 gss: allow oss authenticate with mgs.
r=di.wang
Eric Mei [Fri, 15 Oct 2010 20:26:18 +0000 (14:26 -0600)]
b=22168 write-append-truncate: retry write when receives EINTR.
r=jian.yu
r=vitaly.fertman
Brian J. Murrell [Fri, 15 Oct 2010 16:28:42 +0000 (12:28 -0400)]
b=23721 download OFED from OFA
Download the OFED distribution directly from OFA to ameliorate
re-distribution complications.
i=yangsheng
i=wangyb
Andrew Perepechko [Tue, 12 Oct 2010 12:33:43 +0000 (16:33 +0400)]
b=20533 set the default max_sectors to the raid5/6 stripe size
a=Kevin Van Maren
i=Jinshan Xiong
i=Atul Vidwansa
Brian J. Murrell [Mon, 18 Oct 2010 18:10:51 +0000 (14:10 -0400)]
package liblustre tests for debian
The liblustre tests were not being packaged into lustre-tests.
Brian J. Murrell [Mon, 18 Oct 2010 18:10:50 +0000 (14:10 -0400)]
need to match both master and b1_8 style tags
Brian J. Murrell [Mon, 18 Oct 2010 18:10:49 +0000 (14:10 -0400)]
explicitly set KVERS; allow all automake packages
Explicitly set KVERS to what it'd default value would be if it's
not set by the caller.
Allow any of the compatible automake* packages in debian/control.
Brian J. Murrell [Mon, 18 Oct 2010 18:10:48 +0000 (14:10 -0400)]
make debs refinement
When looking for the most recent tag, look for only tags matching v*.
That should hopefully weed out any non-versioning tags.
Don't convert debian/changelog commits to patches since the changelog
has to be in place right when dpkg-buildpackage starts.
This does mean that one should NEVER commit to another file in the same
commit as a debian/changelog commit.
Make sure debian/patche{s,d} are clean before starting.
Brian J. Murrell [Mon, 18 Oct 2010 18:10:47 +0000 (14:10 -0400)]
debian/patches and non-GA tags
Make the code to populate debian/patches work better with non-GA tags
such as v1_8_3_53.
Brian J. Murrell [Mon, 18 Oct 2010 18:10:46 +0000 (14:10 -0400)]
move .debs into /debs
Move the built .debs from .. into a /debs subdir of the build tree.
We in fact could almost make debian/rules do this using dh_builddeb's
--destdir argument, however dpkg-genchanges wants to find them in ..
and there is no way to pass dpkg-genchange's -u argument to it from
dpkg-buildpackage. So thus, our hack.
Vitaly Fertman [Fri, 15 Oct 2010 12:37:06 +0000 (16:37 +0400)]
b=18456 fix for commit:
248d9f7fc874d137d327992537a9e7672963e021
drop inexistent patch from ext4 serie
Cliff White [Thu, 14 Oct 2010 15:56:36 +0000 (08:56 -0700)]
b=23645 warning at lookup_one_len() durning mount
i=andrew.perepechko
Change lookup_on_len to ll_lookup_one_len when getting iopen dir
Andrew Perepechko [Thu, 14 Oct 2010 14:50:10 +0000 (18:50 +0400)]
b=23818 sanity_quota test_32 fix
wait until MDS-OST synchronization is done
i=Johann Lombardi
i=ZhiYong Tian
Eric Mei [Thu, 14 Oct 2010 14:40:46 +0000 (08:40 -0600)]
b=19486 add server identifier into lu_seq_range.
For future usage to distinguish between MDT and OST fids.
o=di.wang
r=eric.mei
r=mikhail.pershin
Rahul Deshmukh [Thu, 14 Oct 2010 13:09:04 +0000 (18:39 +0530)]
b=22984 change all references to tune.ldiskfs in lustre to tunefs.ldiskfs
o=girish
i=adilger
i=johann
Rahul Deshmukh [Thu, 14 Oct 2010 13:29:45 +0000 (18:59 +0530)]
b=21135 errors seen running llverfs (full mode) at 94% full with lustre fs with three 16TB OST's
o=andreas
i=girish
i=simmonsja
Rahul Deshmukh [Thu, 14 Oct 2010 13:10:12 +0000 (18:40 +0530)]
b=18456 huge OST mount time
o=andreas
i=alex.zhuravlev
i=girish
Johann Lombardi [Mon, 22 Mar 2010 22:01:55 +0000 (23:01 +0100)]
b=22307 rate limit dlm debug message in ll_inode_from_lock()
i=oleg
i=dmitry
Vitaly Fertman [Thu, 14 Oct 2010 23:14:03 +0000 (03:14 +0400)]
Revert "b=22033 Change default filesystem to ext3"
This reverts commit
f53e89dbe756e9da0ee98c1afe1e3691fa556724.
Vitaly Fertman [Tue, 12 Oct 2010 23:47:46 +0000 (03:47 +0400)]
Revert "b=23090 Wirecheck.c is still missing some structs"
This reverts commit
56444fda88f0a2a252963bfd1e57372f20e794b2.
Elena Gryaznova [Tue, 12 Oct 2010 20:25:11 +0000 (00:25 +0400)]
b=22157 conf-sanity test_5b defect
i=Andrew.Perepechko
test_5b cleanup for mgs/mds not combined;
test_5* changes to use error instead of echo && return
new test_5f
Elena Gryaznova [Mon, 11 Oct 2010 14:13:40 +0000 (18:13 +0400)]
bug=23542 recovery-small: test_27 is not time bounded
i=Brian.Murrell
Brian J. Murrell [Fri, 8 Oct 2010 14:15:32 +0000 (10:15 -0400)]
b=23685 don't define unused variables
When CONFIG_SYSCTL is not defined in the kernel config, a couple of
variables are left unused, yet still defined.
i=Dmitry Zogin
i=Andrew Perepechko
Vladimir Saveliev [Fri, 8 Oct 2010 10:36:26 +0000 (14:36 +0400)]
b=22766 cascading_rw: take lmm_stripe_count returned by ioctl(LL_IOC_LOV_SETSTRIPE)
i=adilger
yangsheng [Fri, 8 Oct 2010 09:54:42 +0000 (17:54 +0800)]
b=22375 Add walk_stack callback for dump_trace.
o=sebastien.buisson@bull.net
i=adilger
i=yangsheng
Vladimir Saveliev [Fri, 8 Oct 2010 09:06:31 +0000 (13:06 +0400)]
b=22942 check attribute flags correctly
i=vitaly
LiuYing [Fri, 8 Oct 2010 02:48:14 +0000 (10:48 +0800)]
b=21420 Add mds/mgs stats to HEAD
1)remove useless counter from mds and move some definitions from mds to mdt;
2)move LPROCFS_MD_OP_INITs from lprocfs_alloc_md_stats() to
lprocfs_init_mps_stats(), which is needed by this stats;
3)increase mdt counter for each type operation
i=andreas
i=wangdi
Nicolas Williams [Thu, 7 Oct 2010 22:24:46 +0000 (17:24 -0500)]
b=23090 Wirecheck.c is still missing some structs
i=andreas.dilger
i=robert.read
Terry Rutledge [Fri, 8 Oct 2010 21:27:20 +0000 (14:27 -0700)]
Updated for 2.1.0 build 05.
Vitaly Fertman [Tue, 28 Sep 2010 21:52:03 +0000 (01:52 +0400)]
b=19325 adjust waiting extent locks during 1st enqueue
o=bobijam
i=vitaly
i=green
Re-landing. Adjust locks' extents on their first enqueue, so that at the time
they get granted, there is no need for another pass through the queues since
they are already shaped into the proper forms.
Rahul Deshmukh [Tue, 5 Oct 2010 10:28:50 +0000 (15:58 +0530)]
b=22032 race between create lproc and read lproc fix
There is window between lproc create and private data init, where a lproc read
can find data null. this patch fixes it by making it atomic.
i=alexander.zarochentsev
i=rahul
Elena Gryaznova [Tue, 5 Oct 2010 13:06:23 +0000 (17:06 +0400)]
b=23537 do_rpc_nodes does not propagate NAME to remote nodes causing rpc.sh to fail abruptly
o=Chris.Horn <hornc@cray.com>
i=grev
Cliff White [Tue, 5 Oct 2010 18:08:32 +0000 (11:08 -0700)]
b=22755 more shrink grant fixes
i=di.wang
i=oleg.drokin
- filter_preprw_write() should check for OBD_FL_SHRINK_GRANT in o_flags instead
of o_valid;
- The grant shrink code should not use o_flags w/o setting OBD_MD_FLFLAGS since
o_flags can be legitimately reset if OBD_MD_FLFLAGS (e.g. it is done in the
checksum code);
- OBD_FL_MMAP is added to wiretest & wirecheck
Mikhail Pershin [Wed, 6 Oct 2010 12:56:41 +0000 (16:56 +0400)]
b=16509 port recovery ending tests to master
i=grev
Elena Gryaznova [Thu, 7 Oct 2010 11:37:21 +0000 (15:37 +0400)]
b=22546 set quota parameters on all OSS servers.
o=James.Simmons <simmonsja@ornl.gov>
i=Brian.Murrell
i=grev
i=landen
Rahul Deshmukh [Thu, 7 Oct 2010 12:01:46 +0000 (17:31 +0530)]
b=22176 Add .sync_fs super block handler
Adding sync_fs functionality by adding .sync_fs to super
block handler. It basically mark all loi's as urgent
(at an instant of time) & move it to separate list and
drain them.
i=oleg
i=eric.mei
Rahul Deshmukh [Thu, 7 Oct 2010 12:06:46 +0000 (17:36 +0530)]
b=19427 Correct lmm_object_id
o=adilger
i=wangdi
i=tapro
i=pravin
i=rahul
Fix boolean logic for limiting the maximum object ID created within a
single group (sequence). For legacy IDIF objects this is 2^48 objects,
since we consume 2^16 SEQ numbers for the high bits of the object ID),
but for all other groups it is only 2^32 objects that can fit into f_oid.
Clean up some duplicate #defines that are no longer needed now that
we declare all of the FID_SEQ values in enum fid_seq in lustre_idl.h.
Rahul Deshmukh [Thu, 7 Oct 2010 12:50:21 +0000 (18:20 +0530)]
b=21898 mdd_max_lmm_get unlikely->likely fix
o=liang
i=andreas
i=rread
Rahul Deshmukh [Thu, 7 Oct 2010 13:01:04 +0000 (18:31 +0530)]
b=22860 lnet_msg_alloc does double memset to zero
i=andreas
i=liang
Vitaly Fertman [Mon, 4 Oct 2010 21:00:34 +0000 (01:00 +0400)]
b=20687 umount symbolic link device
i=vitaly
Elena Gryaznova [Mon, 4 Oct 2010 20:15:38 +0000 (00:15 +0400)]
b=21927 t-f: use the global variables for facets mount points
i=Andreas.Dilger
Andrew Perepechko [Mon, 4 Oct 2010 19:43:29 +0000 (23:43 +0400)]
b=18728 allow queued write syscall termination by a signal
i=Alexander Zarochentsev
i=Vitaly Fertman
Mikhail Pershin [Sun, 3 Oct 2010 10:49:22 +0000 (14:49 +0400)]
b=20101 lfs ost enhancements
a=james.simmons
i=andreas.dilger
i=brian.murrell
Mikhail Pershin [Sun, 3 Oct 2010 09:14:43 +0000 (13:14 +0400)]
b=17485 use $SINGLEMDS, fix /proc access in tests
i=grev
Dmitry Zogin [Thu, 30 Sep 2010 13:09:40 +0000 (09:09 -0400)]
b=20146 Metadata performance degradation from 1.6.5 vs 1.8.0.190
Don't update ctime if we are really removing the inode.
i=mikhail.pershin
i=andreas.dilger
Andrew Perepechko [Thu, 30 Sep 2010 12:32:46 +0000 (16:32 +0400)]
b=22107 a fix for the warning message in filter_destroy_internal
filter_destroy_internal is entered with inode references from
lookup and from the resource used by the extent lock, so the
check is changed to i_count != 2.
i=Vitaly Fertman
i=Johann Lombardi
Alexander.Zarochentsev [Wed, 29 Sep 2010 15:03:06 +0000 (19:03 +0400)]
b=23098 Check for slow i_alloc_sem acquiring
i=alex.zhuravlev
i=vitaly.fertman
Andrew Perepechko [Wed, 29 Sep 2010 08:59:53 +0000 (12:59 +0400)]
b=23216 a fix for a possible memory leak in lov_setup
i=Vitaly Fertman
i=Mikhail Pershin
Andrew Perepechko [Wed, 29 Sep 2010 08:59:52 +0000 (12:59 +0400)]
b=23216 a fix for a memory leak in ldiskfs_mb_load_buddy
i=Alex Zhuravlev
i=Johann Lombardi
Maxim Patlasov [Tue, 28 Sep 2010 10:07:59 +0000 (14:07 +0400)]
b=16842 __l_wait_event should sleep using TASK_INT
i=green
i=andreas
All threads sleeping in non-interruptible way are accounted in "load average".
This is not desirable for lustre threads sleeping in __l_wait_event.
Traditionally, only threads doing I/O are allowed to sleep in
non-interruptible way.
Elena Gryaznova [Wed, 29 Sep 2010 14:58:31 +0000 (18:58 +0400)]
b=20407 replay-ost-single: do not skip for HARD mode and mixed_ost_devs
i=Brian.Murrell
Vladimir Saveliev [Wed, 29 Sep 2010 21:27:14 +0000 (01:27 +0400)]
b=22828 fix unnormalized Linux symbols
introduced by
https://bugzilla.lustre.org/attachment.cgi?id=30855