Whamcloud - gitweb
Andrew Perepechko [Wed, 17 Mar 2010 07:00:49 +0000 (10:00 +0300)]
b=20101 lfs getstripe -d test for sanity 27w
o=James Simmons (ORNL)
i=Andreas Dilger
i=Robert Read
Nathan Rutman [Fri, 12 Mar 2010 22:25:53 +0000 (14:25 -0800)]
b=20878 update for commit
5b268907
Nathan Rutman [Fri, 12 Mar 2010 17:46:01 +0000 (09:46 -0800)]
b=20878 change kernelcomms from netlink to pipes
i=alexg
i=emoly.liu
for portability to Solaris, change kernel-userspace comm
mechanism from NetLink to pipes
Wang Di [Fri, 12 Mar 2010 17:46:01 +0000 (09:46 -0800)]
b=16721 fix rename pdif lock, when source and target in the same directory.
i=tappro
i=ericm
Robert Read [Thu, 11 Mar 2010 05:49:04 +0000 (21:49 -0800)]
b=22097 mount.lustre fails to pass some options to mount()
i=yangsheng
i=dmitry
Brian J. Murrell [Tue, 9 Mar 2010 19:26:11 +0000 (11:26 -0800)]
b=17952 allow relative pathnames
This patch allows one to give relative pathnames to --with-linux and
friends.
i=adilger
i=dmitry.zogin
Andrew Perepechko [Tue, 9 Mar 2010 19:26:11 +0000 (11:26 -0800)]
b=21669 cast the le64 to %llu in all ext*mmp patches
i=Andreas Dilger
i=Andrew Perepechko
a bunch of fixes for extN_warnings in ldiskfs to avoid compilation warnings
Elena Gryaznova [Tue, 9 Mar 2010 19:26:11 +0000 (11:26 -0800)]
b=22169 t-f:start_client_loads () wait the background threads to start
i=Robert.Read
Elena Gryaznova [Tue, 9 Mar 2010 19:26:11 +0000 (11:26 -0800)]
b=22169 t-f cleanup: new do_nodev (), do_nodesv () functions
i=Robert.Read
Landen [Tue, 9 Mar 2010 19:26:11 +0000 (11:26 -0800)]
b=22030 don't reveal unnecessary message to users
i=adilger
i=robert.read
Fan Yong [Tue, 9 Mar 2010 19:26:10 +0000 (11:26 -0800)]
b=21986 allow to apply OPEN_LOCK even when file created
Allow to apply OPEN_LOCK even when file created.
i=robert.read
i=oleg.drokin
Dmitry Zogin [Tue, 9 Mar 2010 19:26:10 +0000 (11:26 -0800)]
b=21900 ost-pools test_25: FAIL
Make ost-pools test_25 more robust
i=manoj.joseph
Dmitry Zogin [Tue, 9 Mar 2010 19:26:10 +0000 (11:26 -0800)]
b=22127 lustre 1.8.2 lfs permissions Patch corrects cfs_curproc_euid() logic.
o=bschubert@ddn.com
i=oleg.drokin
i=johann
Manoj Joseph [Tue, 9 Mar 2010 19:26:10 +0000 (11:26 -0800)]
b=20837 Reduce the output of test_14
Fill up the OST by createing 9 files of size OST0_SIZE/10.
i=nathan.rutman
i=grev
Fan Yong [Tue, 9 Mar 2010 19:25:59 +0000 (11:25 -0800)]
b=15962 statahead should not alias dentry with inode, which should be done by VFS layer operation with parent dir's i_mutex held
1) statahead should not alias dentry with inode, which should be done by VFS layer operation with parent dir's i_mutex held
2) code cleanup
i=tom.wang
i=jinshan.xiong
Terry Rutledge [Thu, 4 Mar 2010 20:42:31 +0000 (13:42 -0700)]
Updated version string for build 38.
Brian J. Murrell [Thu, 4 Mar 2010 16:58:05 +0000 (08:58 -0800)]
b=20383 fix errant m4 "dnl" usage
Some dnl() usage seems to have been causing some errors in the resulting
configure script.
o=tappro
i=brian
i=adilger
Eric Mei [Thu, 4 Mar 2010 01:04:44 +0000 (17:04 -0800)]
b=22074 block if grant is available or maybe available soon
even if there's no optimal-sized RPC can be sent, don't fall back to
synchronous write if there's still grant available.
r=adilger
r=wangdi
Eric Mei [Thu, 4 Mar 2010 01:04:36 +0000 (17:04 -0800)]
b=21106 retry page_gang_lookup.
If client cached a huge number of page, cl_lock_page_out() may hog CPU
for too long time without releasing, especially with high debug grade.
This patch break this process with voluntary resched.
r=jay
r=wangdi
Rahul Deshmukh [Thu, 4 Mar 2010 01:03:31 +0000 (17:03 -0800)]
b=21500 2.6.31-fc12 patchless client support
i=adilger
i=brian
i=sheng.yang
Rahul Deshmukh [Thu, 4 Mar 2010 01:03:21 +0000 (17:03 -0800)]
b=20492 allow llverdev/llverfs to handle media errors better
i=adilger
i=rread
Rahul Deshmukh [Thu, 4 Mar 2010 01:03:09 +0000 (17:03 -0800)]
b=20738 FC11 support in Yala is needed
Added support for fc11 patchless client build on lbat
i=brain
Elena Gryaznova [Thu, 4 Mar 2010 01:02:49 +0000 (17:02 -0800)]
b=21255 parallel-scale statahead test fix
use mpi for create/delete files instead of createmany and rm
i=Vladimir.Saveliev
i=Andrew.Perepechko
Andrew Perepechko [Thu, 4 Mar 2010 01:02:11 +0000 (17:02 -0800)]
b=21919 Do not resend quota_ctl requests
Do not resend quota_ctl requests infinitely.
i=Johann Lombardi
i=ZhiYong Tian
Manoj Joseph [Thu, 4 Mar 2010 01:01:30 +0000 (17:01 -0800)]
b=21066 ost-pools test_14 should not assert that files are from a specific OST
Round-robin allocation test should not assert that files are allocated
in strict round-robin fashion.
i=nathan.rutman
i=grev
Robert Read [Wed, 24 Feb 2010 03:02:00 +0000 (19:02 -0800)]
b=22225 Pad all test descriptions with "=" characters.
Add banner() function so we can log other messages in the same format.
i=nico
Brian J. Murrell [Wed, 3 Mar 2010 01:32:48 +0000 (17:32 -0800)]
b=21380 make dist seems to exclude the "darwin" bits
Include all of the darwin bits in the distribution tarball created with
make dist.
i=adilger
i=rread
Mike Tappro [Wed, 3 Mar 2010 01:32:45 +0000 (17:32 -0800)]
b=21485 Keep in-memory client data until export destroy
Client data on disk is removed upon disconnect but in-memory data is needed
while export is still used and must be kept until its destroy.
i=adilger
i=rread
Robert Read [Tue, 2 Mar 2010 02:08:23 +0000 (18:08 -0800)]
Revert "b=21485 Keep in-memory client data until export destroy"
This reverts commit
64d7c036a5429a9ce5435aea96e30935b1cef44a.
Causes compile error:
.../lustre/include/lu_target.h:59: error: 'LR_CLIENT_BITMAP_SIZE' undeclared here (not in a function)
Mike Tappro [Tue, 2 Mar 2010 01:57:44 +0000 (17:57 -0800)]
b=21485 Keep in-memory client data until export destroy
Client data on disk is removed upon disconnect but in-memory data is needed
while export is still used and must be kept until its destroy.
i=adilger
i=rread
Alexander.Zarochentsev [Tue, 2 Mar 2010 01:57:39 +0000 (17:57 -0800)]
b=21988 serialize mdt and obd state bit access
Use atomic bit functions to access mdt state in mdt_allow_cli(). Otherwise bit
updates can be lost in case of concurrent modifications. Protect obd_no_conn =
0 by the obd_task_processing_lock spinlock as in the rest of the code
excluding setup.
i=tappro
i=robert.read
Elena Gryaznova [Tue, 2 Mar 2010 01:57:34 +0000 (17:57 -0800)]
b=21948 skip parallel grouplock test for NFSCLIENT mode
i=Johann.Lombardi
Vitaly Fertman [Tue, 2 Mar 2010 01:57:26 +0000 (17:57 -0800)]
b=20814 enable frame pointers for rhel5 kernels
re-land the patch, it was occasionally dropped
o=rread
i=vitaly
i=eric
Eric Mei [Tue, 2 Mar 2010 01:57:23 +0000 (17:57 -0800)]
b=20731 add handling of statfs ioctl in mdc.
when cluster is upgraded from 1.8, client don't have LMV up and running,
so statfs ioctl will have to handled by mdc directly.
r=adilger
r=pravin
Elena Gryaznova [Thu, 25 Feb 2010 08:33:27 +0000 (00:33 -0800)]
b=21968 conf-sanity changes for config mgs and mds are not combined
i=Nathan.Rutman
includes t-f:combined_mgs_mds ()
Alexander.Zarochentev [Thu, 25 Feb 2010 08:32:58 +0000 (00:32 -0800)]
b=17660 remove unnecessary spinlock wrappings
Read access to obd_recovering is atomic and doesn't need to be protected
by a spinlock.
i=tappro
i=adilger
Fan Yong [Thu, 25 Feb 2010 08:32:12 +0000 (00:32 -0800)]
b=16680 diagnostic patch to verify lustre read-only device mechanism
Diagnostic patch to check whether lustre read-only device mechanism works well or not.
i=alex.zhuravlev
i=rahul.deshmukh
Jian Yu [Thu, 25 Feb 2010 08:31:58 +0000 (00:31 -0800)]
b=21097 quote the passing on of it's own arguments through $@
Fix do_nodes() in acc-sm test-framework to quote the passing on
of it's own arguments through $@ to do_node().
i=jian.yu
i=edg
Robert Read [Tue, 23 Feb 2010 00:00:43 +0000 (16:00 -0800)]
Revert "b=20989 Don't search for inode in nfs get_dentry"
This reverts commit
06889c56aec37cead923dbbd636d31dba0a46950.
This caused a performance regression on NFS.
Wang Di [Mon, 22 Feb 2010 22:21:32 +0000 (14:21 -0800)]
b=20984 cleanup md_op_data and add getstripe -M
i=adiger
i=robert
i=vitaly
Elena Gryaznova [Mon, 22 Feb 2010 22:21:27 +0000 (14:21 -0800)]
b=21932 disable some tests on NFSCLIENT
Disable simul, write_disjoint, write_append_truncate for NFSCLIENT mode.
Setstripe on lustre client (nfsserver) instead of nfs client for ior test.
i=Oleg.Drokin
Maxim Patlasov [Mon, 22 Feb 2010 22:20:54 +0000 (14:20 -0800)]
b=21853 define cfs_*_nested for CONFIG_DEBUG_LOCK_ALLOC=y
Post-nn fix: define cfs_mutex_lock_nested and other libcfs nested lock prims as linux native nested prims when CONFIG_DEBUG_LOCK_ALLOC is defined.
i=rread
i=alex
Maxim Patlasov [Mon, 22 Feb 2010 22:20:43 +0000 (14:20 -0800)]
b=21911 fix for double release of ibc_lock in o2iblnd
Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra reference to conn before calling kiblnd_post_tx_locked() to avoid scenario when conn disappears inside kiblnd_post_tx_locked().
i=liang
Eric Mei [Mon, 22 Feb 2010 22:20:21 +0000 (14:20 -0800)]
b=21835 fix mdsrate script.
Fix mdsrate test script to also account in inodes numbers used by
test directories.
r=robert.read
Terry Rutledge [Fri, 19 Feb 2010 17:30:48 +0000 (10:30 -0700)]
Updated version for buld 37.
Mikhail Pershin [Thu, 18 Feb 2010 16:26:54 +0000 (19:26 +0300)]
Revert "b=21501 flock in process hangs on eviction, does not fail"
This reverts commit
7e750a8d9f9b87b7a3a608454c194d2fe7b69704.
It produces new bug 22111.
Mikhail Pershin [Wed, 17 Feb 2010 08:07:47 +0000 (11:07 +0300)]
b=21718 refer to osc_io only if lock was held.
i=wangdi
i=ericm
Mike Tappro [Wed, 17 Feb 2010 07:45:43 +0000 (10:45 +0300)]
b=20997 Reset imp_vbr_fail flag upon eviction
Flag was reset when recovery ends but recovery can be aborted and this flag
stands after reconnection and cause false eviction later
i=rread
i=zam
Oleg Drokin [Wed, 17 Feb 2010 07:01:03 +0000 (10:01 +0300)]
b=21501 flock in process hangs on eviction, does not fail
Wake up waiters when failed lock gets completion AST due to eviction.
i=adilger
i=bzzz
Vladimir Saveliev [Tue, 16 Feb 2010 16:30:28 +0000 (19:30 +0300)]
b=21982 long file truncate problem
fix (use loff_t for truncate size instead of size_t) and unit test
i=jinshan.xiong
i=adilger
Andrew Perepechko [Sat, 13 Feb 2010 00:27:14 +0000 (03:27 +0300)]
b=21147 several fixes and enhancements for quota
1) allow consecutive quotaons
2) a locking fix
3) call build_lqs from a single point in the code
i=Johann Lombardi
i=ZhiYong Tian
Vitaly Fertman [Fri, 12 Feb 2010 15:45:05 +0000 (18:45 +0300)]
b=16195 handle request size correctly in ptlrpc_add_rqs_to_pool()
i=adilger
i=johann
Robert Read [Fri, 12 Feb 2010 17:27:23 +0000 (09:27 -0800)]
b=21881 Fix trap in rpc.sh.
Nicolas Williams [Thu, 11 Feb 2010 21:26:24 +0000 (15:26 -0600)]
b=21881 mdt_num_threads tuning
Fixes to mdt_handler.c so that it honors mdt_num_threads in the same manner as
ost_handler.c does for oss_num_threads. Enhancements to the test framework and
new tests (53a and b, to check for thread tunable behavior) for conf-sanity are
included.
i=adilger@sun.com
i=robert.read@sun.com
hongchao.zhang [Thu, 11 Feb 2010 20:41:06 +0000 (12:41 -0800)]
b=19069 save nid string into buffer before creating dir
in lprocfs_exp_setup, save the nid string gotten from
libcfs_nid2str into a temporary buffer before calling
lprocfs_register to create proc directory, then to
avoid the race on the internal nid string buffer.
i=johann@sun.com
i=zhen.liang@sun.com
Vitaly Fertman [Thu, 11 Feb 2010 20:39:36 +0000 (12:39 -0800)]
b=1028 drop SOM dependency on quota
once SOM attributes are moved to EA, there is no need to disabled quota when SOM is enabled
i=fanyong
i=tappro
Vitaly Fertman [Thu, 11 Feb 2010 20:39:33 +0000 (12:39 -0800)]
b=21882 handle SLV==1 on client side
Initialize ldlm pool SLV to 0 on client side to handle SLV==1 obtained from server correctly
i=reen
i=adilger
Vitaly Fertman [Thu, 11 Feb 2010 20:39:33 +0000 (12:39 -0800)]
b=21882 lru resize SLV can get stuck
calculate SLV with a greater precision to not lose small changes due to interger math truncation;
round up SLV only if the amount of granted locks less than the limit to not get stuck with this SLV
i-adilger
i=green
Vitaly Fertman [Thu, 11 Feb 2010 20:39:25 +0000 (12:39 -0800)]
b=15977 analyse only consistent part of the log
due to debug log buffer overflow, old messages are lost;
as each CPU has its own buffers which overflows at different time,
it may happen that FREE is lost for still existing ALLOC;
make leak_finder to analyze the log since the oldest overflow time.
i=rread
i=green
Robert Read [Thu, 11 Feb 2010 05:07:17 +0000 (21:07 -0800)]
b=20057 Add the timestamp back to the test headers.
a=manoj
i=rread
i=grev
yangsheng [Tue, 9 Feb 2010 21:06:02 +0000 (05:06 +0800)]
b=21411 Improvement for AT to handle timeout more accurate.
i=nathan
i=tappro
Landen [Tue, 9 Feb 2010 05:23:49 +0000 (13:23 +0800)]
b=16909 Simplify MDT/OST service start message
i=nathan
i=adilger
Dmitry Zogin [Tue, 9 Feb 2010 02:56:16 +0000 (21:56 -0500)]
b=21900 ost-pools test_25: FAIL: /mnt/lustre/d0.ost-pools/d25/file1 not allocated from OSTs 0.
Modify ost-pools test_25 to wait for MDS-OST connection to re-establish.
i=johann
Eric Mei [Mon, 8 Feb 2010 23:57:43 +0000 (16:57 -0700)]
b=21278 fix linkea definition and unaligned access.
link_ea_entry->lee_reclen should be defined as unsigned char instead
of char which might get sign expansion during conversion.
fix unaligned memory access to link_ea_entry->lee_parent_fid.
i=nathan
i=rahul
Eric Mei [Mon, 8 Feb 2010 23:57:23 +0000 (16:57 -0700)]
b=21278 fix a memory leak in mdd_links_get().
i=nathan
i=rahul
Wang Di [Mon, 8 Feb 2010 23:52:21 +0000 (18:52 -0500)]
b=17197 stride read-ahead fix
Eric Mei [Mon, 8 Feb 2010 23:24:46 +0000 (16:24 -0700)]
b=16526 don't send raw inode flags on the wire.
inode flags should be convert to platform-neutral MDS_*_FL before
being sent on the wire. Also removed unused flags MDS_BFLAG_EXT_FLAGS.
i=adilger
Robert Read [Wed, 10 Feb 2010 18:47:19 +0000 (10:47 -0800)]
b=21049 Fix a missed change from original patch.
Liang Zhen [Wed, 10 Feb 2010 13:30:31 +0000 (05:30 -0800)]
b=21486 Unlink truncated reply buffer
i=eric.mei
i=eeb
Liang Zhen [Wed, 10 Feb 2010 13:30:10 +0000 (05:30 -0800)]
b=21396 Don't hog cpu in connd.
i=he.huang
i=maxim.patlasov
Girish Shilamkar [Fri, 5 Feb 2010 15:04:15 +0000 (20:34 +0530)]
b=21564 Print mmp_check_interval in kmmpd
The patch not only prints mmp_check_interval but also makes it possible
to abort mount operation in case it takes too long.
i=adilger
Alexander.Zarochentev [Fri, 5 Feb 2010 07:25:08 +0000 (10:25 +0300)]
b=18030 deadlock fix
Start the transaction earlier in llog_lvfs_destroy to get transaction start
and inode mutex lock nested properly.
i=zam
i=adilger
Nathan Rutman [Wed, 3 Feb 2010 20:33:30 +0000 (12:33 -0800)]
b=21738 conf-sanity 42 should set conf_param on the MGS
i=rread
Jay Xiong [Thu, 4 Feb 2010 04:17:05 +0000 (12:17 +0800)]
b=21551 holding ->lli_size_sem to update inode size
- Fixed i_size_write in vvp_io_commit_write, it has to be called under ->lli_size_sem
- clenaup the code.
i=wangdi
i=green
Robert Read [Thu, 4 Feb 2010 23:55:21 +0000 (15:55 -0800)]
Prepare for Build 36.
Nathan Rutman [Wed, 3 Feb 2010 20:09:37 +0000 (12:09 -0800)]
b=21924 create test dir before starting changelogs
i=rread
A test-level directory (d0.lustre-rsync-test) was recently added to the $tdir path.
This directory is created before the tests are run. But lustre-rsync requires the
source and target systems to be identical before the changelog is started; the
presence of this directory in the source, missing in the target, causes the
failure seen here. This patch simply creates $tdir on the source and target
before each test is run, insuring a coherent starting state.
Alexander.Zarochentev [Wed, 3 Feb 2010 18:01:03 +0000 (21:01 +0300)]
b=21558 improve test 21b COS=0 check
In test 21b, when testing that the dependent renames are not replied if COS=0,
there is still a window when transactions may be written to disk before the
mds device is set R/O. To avoid such a rare test failure, the check is
repeated several times.
i=tappro
i=rober.read
Alexander.Zarochentev [Wed, 3 Feb 2010 18:01:02 +0000 (21:01 +0300)]
b=21264 workaround dd bus error
A buggy coreutils/gettext combination workaround. Suppressing dd xfer
statistic makes dd do not call gnu gettext library and avoid crashing.
i=robert.read
Dmitry Zogin [Wed, 3 Feb 2010 14:41:27 +0000 (09:41 -0500)]
b=21259 "lfs check" is only allowed for root.
Code cleanup around obd_class_*() functions and sanity test for non-root lfs check
i=adilger
i=andrew.perepechko
Alexander.Zarochentev [Wed, 3 Feb 2010 09:00:47 +0000 (12:00 +0300)]
b=21879 fix hr->hr_index overload
i=eric.barton
i=zhen.liang
Jay Xiong [Wed, 3 Feb 2010 03:13:22 +0000 (11:13 +0800)]
b=21355 fix initiailzation of recovery thread
Recovery thread calls cfs_daemonize() to detach the thread context, this causes all recovery threads share a common fs_struct: init_fs. We should call cfs_daemonize_ctxt() instead.
i=wangdi
i=tappro
yangsheng [Tue, 2 Feb 2010 12:40:24 +0000 (20:40 +0800)]
b=21632 Kernel update for RHEL5.4 2.6.18-164.11.1.el5.
i=johann
Fan Yong [Tue, 2 Feb 2010 10:54:58 +0000 (18:54 +0800)]
b=21433 trace the processing for "no quota" flag on client-side
Trace the processing for "no quota" flag on client-side.
i=robert
i=landen
Rahul Deshmukh [Tue, 2 Feb 2010 08:44:41 +0000 (14:14 +0530)]
b=19742 fixed the fiemap functionality
Fixed the fiemap functionality and enable sanity test_130*.
i=adilger
i=johann
Robert Read [Wed, 3 Feb 2010 08:15:15 +0000 (00:15 -0800)]
b=21890 Remove bash -TE flags from acceptance-small.sh
These flags is not needed since test-framework sets the error trap in all the scripts.
Oleg Drokin [Tue, 2 Feb 2010 05:20:23 +0000 (00:20 -0500)]
b=20989 Don't search for inode in nfs get_dentry
Getting inode separately without some sort of locking enabled a situation where
several threads tried to populate same inode with different lsm data at the same
time leading to lsm pointer change (and leak).
i=johann
i=bzzz
Mikhail Pershin [Mon, 1 Feb 2010 19:27:24 +0000 (11:27 -0800)]
b=21499 check stale object before version checks
i=wangdi
i=fanyong
Andrew Perepechko [Mon, 1 Feb 2010 19:27:18 +0000 (11:27 -0800)]
b=21489 fix several write+utimes race conditions
Disable mtime updates on write and serialize fmd_mactime_xid checks in order to
avoid certain write(2)+utimes(2) race conditions on OSS
i=Johann Lombardi
i=Vitaly Fertman
Landen [Fri, 29 Jan 2010 07:04:54 +0000 (15:04 +0800)]
b=20970 need add an additional barrier for write_disjoint
i=grev
i=rread
Landen [Fri, 29 Jan 2010 07:04:53 +0000 (15:04 +0800)]
b=21846 add a hs_key entry in lqs hash operations
When lqs rehashes, it will call its hs_key method. If it isn't
defined, an assert will be triggered. The patch just add it
for lqs hash.
a=johann
i=landen
i=adilger
Brian Behlendorf [Fri, 29 Jan 2010 01:11:46 +0000 (17:11 -0800)]
b=21882 lru-resize comments
Unrelated to the actual bug, this commit just cleanups up various
spelling mistakes in the comments.
Rahul Deshmukh [Thu, 28 Jan 2010 11:36:39 +0000 (17:06 +0530)]
b=19808 2.6.29-fc11 patchless client support changes.
Final changes (except build related changes) for adding FC11
patchless client support.
i=alexey.lyashkov
i=oleg.drokin
Alexander.Zarochentev [Wed, 27 Jan 2010 15:01:12 +0000 (18:01 +0300)]
b=19813 add missing lprocfs_exp_cleanup calls
add missing lprocfs_exp_cleanup calls to
error handling paths in {mdt,filter,mgs}_connect.
i=tappro
i=andrew.perepechko
Robert Read [Sat, 23 Jan 2010 00:29:27 +0000 (16:29 -0800)]
b=18690 Enable rehashing and increase max hash table sizes.
i=eric.mei
Elena Gryaznova [Tue, 26 Jan 2010 14:10:59 +0000 (17:10 +0300)]
b=19387 integrate LST into acc-sm
new acc-sm test suite: lnet-selftest
i=Maxim.Patlasov
i=He.Huang
Nicolas Williams [Mon, 25 Jan 2010 22:09:10 +0000 (16:09 -0600)]
b=21490 1.8<->2.0 interop: in 2.0 server and 1.8 client env. read 1. 8 file hung
The fix for bug 18631 introduced an interop bug for 1.8.x clients with 2.0 OSTs.
The problem was that a zero-length RMF_RCS was included in the OST_BRW replies
for READs, which caused the RPC header to have eight more bytes for storing the
length of that zero-length buffer. The fix is to split RQF_OST_BRW into two,
one for READs and one for WRITEs, with RMF_RCS present only in the reply for
OST_BRW_WRITE.
i=robert.read@sun.com
i=tom.wang@sun.com
Nathan Rutman [Mon, 25 Jan 2010 20:01:21 +0000 (12:01 -0800)]
b=21746 compare full filename when writeconf'ing
When erasing MGS logs, we could confuse fsname e.g. test with test2 and erase both sets of logs.
i=breitz
i=brian
Robert Read [Mon, 25 Jan 2010 19:36:28 +0000 (11:36 -0800)]
b=21889 Rename lustre-resync-test
Fix Makefile.
Elena Gryaznova [Mon, 25 Jan 2010 11:42:11 +0000 (14:42 +0300)]
b=20918 improve log warning
i=Brian
Elena Gryaznova [Mon, 25 Jan 2010 11:28:36 +0000 (14:28 +0300)]
b=20866 DEPS assignment needs quotes
o=Brian
i=grev