Whamcloud - gitweb
Vitaly Fertman [Thu, 16 Dec 2010 00:19:35 +0000 (03:19 +0300)]
b=24324 no new lock on disconnected export
i=green
i=tappro
check if export is being disconnected and do not add new locks to it
Vitaly Fertman [Thu, 16 Dec 2010 00:19:37 +0000 (03:19 +0300)]
b=24218 fix race on ksock_tx_t::tx_resid
o=Liang
i=isaac
If a connection is closed before ksocknal_transmit() returns to
ksocknal_process_transmit(), then nobody has refcount on conn::ksnc_sock
and all pending ZC requests will be finalized by ksocknal_connsock_decref->
ksocknal_finalize_zcreq, ksocknal_finalize_zcreq will mark not-acked ZC
request as error by setting tx::tx_reside = -1.
This is race because ksocknal_process_transmit() will check tx::tx_resid
right after calling ksocknal_transmit(), and it can get tx->tx_resid != 0
and rc == 0 then hit later LASSERT(rc < 0).
Mikhail Pershin [Wed, 15 Dec 2010 15:07:12 +0000 (18:07 +0300)]
b=24078 replay barrier checks sequence allocation
i=grev
Liang Zhen [Wed, 15 Dec 2010 15:04:30 +0000 (23:04 +0800)]
b=23289 revert patch on 21828
this patch reverted change on bug 21828, because svc::srv_n_active_reqs
is dropped by ptlrpc_server_finish_request() now which is used to be
dropped by last calling of ptlrpc_server_drop_request, so recovery
thread has no refcount on svc::srv_n_active_reqs and we should just
remove these refcount changes.
This should also resovled issue of bug 21140, because logic in
ptlrpc_server_request_get() can be totally screwed up by
ptlrpc_server_active_request_inc/dec.
Andreas Dilger [Tue, 14 Dec 2010 07:33:11 +0000 (00:33 -0700)]
b=23289 don't start more than minimum MDT threads
Because there are many different portals running on the MDS node (MDT,
readpage, setattr, SEQ, SEQ server, FLD, DLM), starting more than the
minimum number of threads can result in thousands of threads running
on large MDS nodes, even if many of those threads are not needed.
Start only the minimum number of threads at mount time, and start more
as required by the load.
i=liang
i=vitaly
Andreas Dilger [Tue, 14 Dec 2010 06:54:49 +0000 (23:54 -0700)]
b=23072 avoid duplicate cfs_size_round declarations
Use HAVE_CFS_SIZE_ROUND to determine if libcfs_private.h or lustre_user.h has
declared the cfs_size_round() helper function. This avoids conflicts if both
of these headers are included in different orders.
Vitaly Fertman [Thu, 16 Dec 2010 00:19:36 +0000 (03:19 +0300)]
b=22376 accmode fix
i=tappro
i=green
accmode is fixed to not let other flags to be set together with MAY_EXEC
test is added
Andreas Dilger [Fri, 17 Dec 2010 00:17:22 +0000 (17:17 -0700)]
b=21244 add man page for ll_decode_filter_fid tool
Add man page for new ll_decode_filter_fid(8) tool.
Add references to lustre(7) man page.
Isaac Huang [Wed, 15 Dec 2010 15:35:21 +0000 (08:35 -0700)]
b=20288 kiblnd_check_conns can deadlock
Avoid dropping global lock in kiblnd_check_conns
i=maxim
i=liang
Andreas Dilger [Wed, 15 Dec 2010 06:44:39 +0000 (23:44 -0700)]
b=16605 verify object group before using it
Avoid trying to access an invalid/unknown group sent from the client.
Don't return -EINVAL, since it is possible that the problem is the OST
code is old and this object actually exists. Instead, return -EBADF
so that we can more easily diagnose this problem.
Mikhail Pershin [Wed, 15 Dec 2010 14:51:41 +0000 (17:51 +0300)]
b=24048 set body->easize in mdc_getattr_pack()
i=johann
i=adilger
This allows client to supply the server with enough info about reply buffers.
Server may use this to allocate buffers and don't guess with max_mdsize
Mikhail Pershin [Wed, 15 Dec 2010 14:50:39 +0000 (17:50 +0300)]
b=24171 check the init() iterator results properly
i=vitaly
i=bzzz
Ricardo M. Correia [Mon, 13 Dec 2010 15:56:14 +0000 (15:56 +0000)]
b=24300 Don't run autogen.sh in the spl and zfs repos.
The SPL and ZFS repositories already come with the configure script ready to
run, therefore it's not necessary to run autogen.sh on those directories.
Furthermore, in SPL version 0.5.0 this was causing a version conflict when
regenerating the files with the autotools/libtool that come with RHEL5.
Elena Gryaznova [Mon, 13 Dec 2010 17:37:49 +0000 (20:37 +0300)]
b=24266 increase replay-single test_70d dbench duration for HARD failure mode
i=Mikhail.Pershin
yangsheng [Mon, 13 Dec 2010 18:56:38 +0000 (02:56 +0800)]
b=24037 Using cfs_curproc_umask() instead of reference directly.
i=vitaly
Vitaly Fertman [Tue, 14 Dec 2010 23:07:51 +0000 (02:07 +0300)]
b=22755 Don't consume grant twice on recoverable resend
o=johann
i=green
i=tappro
Vitaly Fertman [Tue, 14 Dec 2010 23:00:07 +0000 (02:00 +0300)]
b=22556 should check # of remained RPCs before aborting
o=liang
i=isaac
lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only when the
transaction is timeout, so if we got "end_session" to interrupt waiting
on transaction, then we can hit ASSERTION(crpc->crp_stamp != 0).
Vladimir Saveliev [Mon, 13 Dec 2010 19:51:45 +0000 (22:51 +0300)]
b=23003 sanity test is added, checks how we work with timestamps before Jan 1 1970
i=grev
minhdiep [Mon, 13 Dec 2010 17:01:05 +0000 (10:01 -0700)]
b=24270 need to mkdir mntpt before mount
o=Minh
i=Elena
yangsheng [Mon, 13 Dec 2010 19:05:33 +0000 (03:05 +0800)]
b=24198 set strictatime to default.
RELATIME set to default in upstream kernel. We set strictatime when
relatime or noatime not given explicitly to keep update atime.
i=adilger
i=vitaly
Vladimir Saveliev [Thu, 9 Dec 2010 20:09:39 +0000 (23:09 +0300)]
b=24175 avoid reporting 'No such file or directory' error
i=rahul.deshmukh
Andreas Dilger [Tue, 14 Dec 2010 08:21:45 +0000 (01:21 -0700)]
b=24037 avoid conflicting DCACHE_LUSTRE_INVALID
In newer kernels our definition of DCACHE_LUSTRE_INVALID needs
to be changed to avoid conflictint with other DCACHE flags.
Use a high value to avoid conflict for a long time, maybe until
we can actually get rid of the need for it entirely.
i=sheng.yang
i=vitaly.fertman
Brian Behlendorf [Thu, 2 Dec 2010 20:20:57 +0000 (12:20 -0800)]
b=24037 Include backing-dev.h header
With ldiskfs disabled we fail to include the backing-dev.h header
which is required for the lustre_sb_info structure. Presumably,
this was not as issue previously because it was included through
one of the ldiskfs headers. The backing-dev.h header exists all
the way back to 2.6.0 kernels so I have included it unconditionally
in the linux/linux-fs.h header. This header is included as part
of libcfs.h which is in turn included by lustre_disk.h.
In file included from
/home/behlendo/src/git/lustre-kdmu/lustre/ofd/ofd_internal.h:12, from
/home/behlendo/src/git/lustre-kdmu/lustre/ofd/ofd_io.c:44:
/home/behlendo/src/git/lustre-kdmu/lustre/include/lustre_disk.h:444:
error: field ‘bdi’ has incomplete type
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
a=brian
i=adilger
i=yangsheng
Andreas Dilger [Tue, 14 Dec 2010 07:37:39 +0000 (00:37 -0700)]
b=23821 Limit bio_alloc() to BIO_MAX_PAGES iovecs.
Fix logic error when patch was originally landed from b=9945.
i=alex.zhuravlev
i=kwestneat
Rahul Deshmukh [Tue, 14 Dec 2010 06:19:09 +0000 (11:49 +0530)]
b=22176 Add .sync_fs super block handler
i=oleg
i=ericm
This patch adds .sync_fs super block handler i.e.
on issuing sync command, the dirty loi's are marked
urgent and synced.
Elena Gryaznova [Wed, 8 Dec 2010 19:17:53 +0000 (22:17 +0300)]
b=24239 use SAMPLE_FILE instead of termcap
i=Vitaly.Fertman
Vitaly Fertman [Fri, 10 Dec 2010 12:53:44 +0000 (15:53 +0300)]
b=24230 always allow to handle normal request for liblustre
o=liang
i=vitaly
i=zam
as liblustre is 1-threaded we should not prevert normal requests
from being processed if only 1 thread is left
Landen [Fri, 10 Dec 2010 16:47:38 +0000 (00:47 +0800)]
b=24055 a patch to detect if quota is turned on properly
i=johann
i=panda
Elena Gryaznova [Wed, 8 Dec 2010 13:35:29 +0000 (16:35 +0300)]
b=24039 lfs setstripe --pool broken
i=Andreas.Dilger
i=Vladimir.Saveliev
fix llapi_search_fsname() to handle relative pathname
yangsheng [Wed, 8 Dec 2010 18:50:03 +0000 (02:50 +0800)]
b=24037 Remove iopen patch.
i=adilger
i=johann
Elena Gryaznova [Wed, 8 Dec 2010 18:41:08 +0000 (21:41 +0300)]
b=23961 fix for setup with several network interfaces
i=Brian.Murrell
i=Andrew.Perepechko
- metadata-updates fix for setup when several interfaces are UP on host;
hostname could be assigned to IP which is different from lnet network used,
the hostname-s of NODES_TO_USE are now stored in HOSTS
- new SHUTDOWN_ATTEMPTS: the tunable number of attepts to shutdown node
- shutdown_node_hard () fix: do not call "power off" each time,
wait that the node is not pingable before the next "power off" attempt
- unused check_port() is removed
Vitaly Fertman [Fri, 10 Dec 2010 13:01:42 +0000 (16:01 +0300)]
b=23848 mdc_clear_open_replay_data() ASSERTION(mod != LP_POISON && mod != NULL)
it is possible to have no md_open_data on close for some recovery cases.
also avoid ELC for recovery tests which rely on lost blocking AST rpc.
Cliff White [Fri, 10 Dec 2010 07:13:45 +0000 (23:13 -0800)]
b=23827 fix brw_stats issue
i=johann.lombardi
i=oleg.drokin
Correct issue with brw_stats - fix typo
Vitaly Fertman [Fri, 10 Dec 2010 23:14:33 +0000 (02:14 +0300)]
b=23820 Handle unsent requests with rq_net_err in ptlrpc_check_set()
i=johann
i=green
ptlrpc_check_set()) ASSERTION(req->rq_phase == RQ_PHASE_BULK) failed
hongchao.zhang [Fri, 10 Dec 2010 12:40:25 +0000 (20:40 +0800)]
b=23781 fix obdo leak issue
in ll_set_attr_raw, if ll_extent_lock call fails, the newly allocated
obdo has not been freed before return
i=johann
i=johann
i=dmitry.zoguine
Andrew Perepechko [Fri, 10 Dec 2010 17:36:13 +0000 (20:36 +0300)]
b=23701 uninline part of ptlrpc_main to reduce stack usage
i=Oleg Drokin
i=Alexander Zarochentsev
Zhiqi Tao [Thu, 2 Dec 2010 05:50:28 +0000 (13:50 +0800)]
b=23193 Land ISER performance patch for Sundance
The purpose of this patch is to make ISER support 1MB RDMAs instead of splitting 2 512K RDMAs and increase the maximum number of ISCSI commands.
i=alexg
Cliff White [Mon, 6 Dec 2010 18:29:32 +0000 (10:29 -0800)]
b=23277 Increase kernel log buffer size to 256k
i=ricardo.correia
i=brian
Change CONFIG_LOG_BUF_SHIFT to 18 (256k) for ia64/x84_64
Vitaly Fertman [Tue, 7 Dec 2010 22:00:13 +0000 (01:00 +0300)]
b=21804 protect rpc by refcount in ptlrpc_server_handle_req_in()
i=johann
i=adilger
ptlrpc_hpreq_init() inserts rpc to the export list and by the time
of ptlrpc_server_request_add() it could be already handled and
released. To not lose request in between, take an extra reference
on the request.
Matt Wu [Tue, 7 Dec 2010 11:25:54 +0000 (19:25 +0800)]
b=24243 Invalid WNC intent lock makes server LBUG
i=andreas.dilger
i=oleg.drokin
do not grant PDO locks for non-dir objects, but return -ENOTDIR instead.
Eric Mei [Mon, 6 Dec 2010 22:50:07 +0000 (15:50 -0700)]
b=20563 fix fid_flatten32().
a=andreas.dilger
r=eric.mei
r=di.wang
r=vitaly.fertman
Eric Mei [Mon, 6 Dec 2010 22:49:54 +0000 (15:49 -0700)]
b=20563 fix fid_flatten().
r=andreas.dilger
r=di.wang
Terry Rutledge [Sat, 4 Dec 2010 05:19:03 +0000 (21:19 -0800)]
Needed to create a new tag for 2.1.0 build 09. For some reason the
previous tag was not being propagated out, so v2_0_58_0 is being
created for this build.
Terry Rutledge [Sat, 4 Dec 2010 05:14:57 +0000 (21:14 -0800)]
Updated for 2.1.0 build 09.
Vitaly Fertman [Fri, 3 Dec 2010 19:49:06 +0000 (22:49 +0300)]
Revert "b=22965 enable lockless truncate"
This reverts commit
e768e65ce59b6294813b538f308049ae2cbe74a0.
Vitaly Fertman [Fri, 3 Dec 2010 18:48:36 +0000 (21:48 +0300)]
b=24267 wirecheck fix for OBD_FL_MMAP
i=adilger
i=vitaly
yangsheng [Fri, 3 Dec 2010 14:33:54 +0000 (22:33 +0800)]
b=24214 Discard preallocation blocks after failed allocated.
i=adilger
i=girish
Elena Gryaznova [Thu, 2 Dec 2010 11:35:49 +0000 (14:35 +0300)]
b=24118 test_70b rundbench load failed
i=Brian.Murrell
i=Jian.Yu
- give rundbench a chance to start before the dbench load check
- new check_for_process () and killall_process () to check/kill any defined progs
instead of "dbench" only
- fix 70a, 70b to mount the clients on MOUNT instead of DIR
LiuYing [Fri, 3 Dec 2010 02:03:30 +0000 (10:03 +0800)]
b=24097 move upcall_cache from lvfs to libcfs
part of params_tree platform-independent code
i=andreas
i=alexg
yangsheng [Thu, 2 Dec 2010 19:02:46 +0000 (03:02 +0800)]
b=24037 unlock new inode.
2.6.32 kernel return a locked inode after new_inode. So we need unlock it.
Jian Yu [Thu, 2 Dec 2010 13:45:46 +0000 (21:45 +0800)]
b=23289 set S_NOCMTIME in OSD and remove some mark_inode_dirty
o=Liang Zhen
i=andreas.dilger
Jian Yu [Wed, 1 Dec 2010 12:16:21 +0000 (20:16 +0800)]
b=23289 new API: cfs_waitq_add_exclusive_head
With this patch, we can reduce total number of active threads because
waitq is a LIFO list for exclusive waiting.
o=Liang Zhen
i=andreas.dilger
i=eric.mei
yangsheng [Fri, 3 Dec 2010 14:24:49 +0000 (22:24 +0800)]
b=24037 Changes of 2.6.32 kernel.
--request_queue new member and new macro;
--quota interface name changes;
--capa_* relate changes;
i=adilger
i=vitaly
yangsheng [Fri, 3 Dec 2010 14:24:17 +0000 (22:24 +0800)]
b=19706 porting patch for master.
Since 2.6.32, kernel add pde_user in proc_dir_entry to protect
directory entry from deleted. So we don't need extra lock in
lprocfs_remove().
i=hongchao.zhang
i=johann
Isaac Huang [Mon, 29 Nov 2010 07:08:24 +0000 (00:08 -0700)]
b=23575 O2iblnd credit deadlock regression
This fixed a regression of bug 14425.
i=liang
Jian Yu [Mon, 29 Nov 2010 13:37:56 +0000 (21:37 +0800)]
b=23289 cleanup assertion on atomic
Remove assertion on atomic from addref functions, which is less helpful,
also, add macros to reduce atomic_read() in assertion.
o=Liang Zhen
i=andreas.dilger
i=mikhail.pershin
i=vitaly.fertman
elena.gryaznova@oracle.com [Fri, 26 Nov 2010 22:31:09 +0000 (01:31 +0300)]
b=24228 fix test duration check to be more accurate
i=vitaly
Wang Yibin [Wed, 24 Nov 2010 03:06:02 +0000 (11:06 +0800)]
b=24202 Fix broken PATHS in autoconf file.
There should be a space between paths.
i=sheng.yang
i=brian.murrell
Landen [Thu, 25 Nov 2010 09:03:37 +0000 (17:03 +0800)]
b=24188 rehash patch makes oops in quota_search_lqs
Problem is: we should do "cfs_hash_getref(qctxt->lqc_lqs_hash)"
when we search and create a new quota hash entry.
This patch does some optimization for lquota code so that lquota
can take one less reference for lqs.
i=landen
i=vitaly.fertman
Andrew Perepechko [Tue, 23 Nov 2010 23:38:37 +0000 (02:38 +0300)]
b=24051 sanity test_76 fix
Due to per-cpu slab caches, "active" stat from /proc/slabinfo
is not always accurate.
i=ZhiYong Tian
Elena Gryaznova [Fri, 26 Nov 2010 08:08:35 +0000 (11:08 +0300)]
b=24020 lustre doesn't start with ext4 based ldiskfs.
i=Andrew.Perepechko
load jbd2 module in load modules
Vladimir Saveliev [Fri, 26 Nov 2010 06:51:07 +0000 (09:51 +0300)]
b=23996 grep lfs getstripe output for pools correctly
i=zam
Aleksandr Guzovskiy [Wed, 24 Nov 2010 23:01:26 +0000 (18:01 -0500)]
b=23842 some issues with recent changes to build/nn-final-symbol-list.txt
i=adilger
yangsheng [Wed, 24 Nov 2010 14:09:01 +0000 (22:09 +0800)]
b=19808 fix d_obtain_alias() misuse due to compat macro.
o=Adreas.Dilger
i=yangsheng
i=Oleg.Drokin
ll_iget_for_nfs() get a change from d_alloc_anon() to d_obtain_alias().
The former would NOT release the inode reference if there was an error, while
the latter does. Hide this detail in the d_obtain_alias() compat macro, for
kernels that do not have this function, and remove it from the error handling
code at the caller, where it would drop the inode reference twice in rare error
cases on all kernels WITH d_obtain_alias() support (RHEL5 and newer).
Andreas Dilger [Wed, 24 Nov 2010 23:34:33 +0000 (16:34 -0700)]
b=15599 remove OBD_CONNECT_LAYOUTLOCK from supported bits
Until the rest of the layout lock support is landed, it doesn't
make sense for the MDT to advertize the OBD_CONNECT_LAYOUTLOCK flag
at connection time.
Remove the flag from MDS_CONNECT_SUPPORTED until the actual
MDS_INODEBITS_LAYOUT ibits lock is landed and functional.
Vitaly Fertman [Wed, 24 Nov 2010 01:30:44 +0000 (04:30 +0300)]
b=22598
disable thandle debug usage statistics
Vitaly Fertman [Tue, 23 Nov 2010 21:44:50 +0000 (00:44 +0300)]
b=22965 enable lockless truncate
i=zam
i=adilger
Vitaly Fertman [Tue, 23 Nov 2010 21:44:49 +0000 (00:44 +0300)]
b=23170 lockless enqueue lbug
lockless enqueue still could be called if first enqueue attempt set
lockless ops but found a conflict and therefore tried to enqueue again
later when the confict was resolved.
i=green
i=ericm
Andreas Dilger [Tue, 23 Nov 2010 21:17:32 +0000 (14:17 -0700)]
b=14836 Fail if "lfs setstripe" has positional parameters
Since 1.6.1 it has been possible to use named options -c/-i/-s to
specify the layout parameters for "lfs setstripe".
Since 1.8.0 the use of the old positional parameters has generated
a warning that they are deprecated.
The code was set to fail building at 2.1.0.0, so this needs to be
resolved before moving to a release build. Instead of simply removing
the conditional code, change "lfs setstripe" to return an error if
positional parameters are used. That would avoid the confusion of
creating files with names of the old {size}, {offset}, and {count}
parameters, instead of returning a meaningful error to the user.
Jian Yu [Sat, 20 Nov 2010 12:15:18 +0000 (20:15 +0800)]
b=24022 don't return NULL in lprocfs_alloc_stats to satisfy userspace error checker
o=Liang Zhen
i=vitaly.fertman
Eric Mei [Mon, 22 Nov 2010 15:47:59 +0000 (08:47 -0700)]
b=23994 hide EMFILE error in ptlrpc_start_threads().
i=vitaly.fertman
yangsheng [Mon, 22 Nov 2010 16:01:50 +0000 (00:01 +0800)]
b=23985 Fix upstream jbd2 issue.
The jbd2_stats_proc_init() was placed on wrong location in
jbd2_journal_init_dev(). This may cause /proc/fs/jdb2/<dev>/*
cannot be created when using external journal device.
i=adilger
i-johann
Elena Gryaznova [Fri, 19 Nov 2010 18:02:00 +0000 (21:02 +0300)]
b=23535 sgpdd-survey.sh should check for sg_map
i=Brian.Murrell
check that iokit sgpdd-survey and sg_map are installed
Brian J. Murrell [Fri, 19 Nov 2010 17:22:14 +0000 (01:22 +0800)]
b=23533 ldiskfs to depend on latest e2fsprogs
ldiskfs should depend on the latest e2fsprogs package so that users
are required to keep their e2fsprogs RPMs up to date.
i=mjmac
i=adilger
Vitaly Fertman [Tue, 23 Nov 2010 21:03:08 +0000 (00:03 +0300)]
b=23460 remove expensive memory check
i=green
remove cfs_mem_in_cache check when checking validity of pages and locks
Vitaly Fertman [Tue, 23 Nov 2010 21:03:09 +0000 (00:03 +0300)]
b=23460 make page state changes tracking optinal
i=green
The stats are unused by anybody, yet they take away several percents of read
speed. Let's make them optional.
Jian Yu [Tue, 23 Nov 2010 05:24:11 +0000 (13:24 +0800)]
b=23289 no need to always start transaction for attr_set(atime) on close
To reduce unnecessary transaction overhead on file closing.
o=Liang Zhen
i=mikhail.pershin
i=di.wang
minhdiep [Mon, 22 Nov 2010 20:29:15 +0000 (13:29 -0700)]
b=21525 debug enhancements + minor comment updates
a=Issac
i=Liang
i=Maxim
- when a message is dropped, the lnet counters should be increased;
- /proc/sys/lnet/peers should contain some data for debugging, like
the last_alive time stamp.
Liang Zhen [Fri, 19 Nov 2010 14:44:02 +0000 (22:44 +0800)]
b=18551 adapt ldlm_namespace to new cfs_hash
Use new cfs_hash to implement ldlm_namespace
o=Liang Zhen
i=Andreas Dilger
i=Oleg Drokin
i=Vitaly Fertman
Terry Rutledge [Fri, 19 Nov 2010 21:33:04 +0000 (13:33 -0800)]
Updated for 2.1.0 build 08.
minhdiep [Thu, 18 Nov 2010 22:04:54 +0000 (15:04 -0700)]
b=24170 checking len of debug string before matching
i=vitaly
i=rahul
We need to check the len before matching the string
to prevent something like net matches neterror
Jian Yu [Wed, 17 Nov 2010 02:15:26 +0000 (10:15 +0800)]
b=23289 grained lock for ptlrpc service
- cacheline optimization for struct ptlrpc_service
- split ptlrpc_service::srv_lock to three locks:
. ptlrpc_service::srv_lock
serialize operations on rqbd and requests queued on srv_req_in_queue
. ptlrpc_service::srv_rq_lock
serialized operations on active requests
. ptlrpc_service::srv_rs_lock
serialize operations on RS
- cleanup logic in ptlrpc_main()
- remove unused atomic srv_outstanding_replies
o=liang
i=andreas.dilger
i=mikhail.pershin
Isaac Huang [Wed, 17 Nov 2010 14:14:46 +0000 (07:14 -0700)]
b=20153 add IB bonding failover support to o2iblnd
O2iblnd changes to support failover events from an IB
bonding IPoIB interface. Mostly to recreate device
specific resources, e.g. listener CMID.
i=isaac
i=liang
Mikhail Pershin [Thu, 18 Nov 2010 07:14:58 +0000 (10:14 +0300)]
b=19890 fix mgs conf for mds and client for nosvc option
i=zam
i=ericm
Brian J. Murrell [Mon, 15 Nov 2010 21:15:29 +0000 (05:15 +0800)]
b=24087 reverse order of $LINUX{,_OBJ}/include
It is important that /usr/src/linux-...-obj/include is searched for
includes before /usr/src/linux-.../include so that the inclusion of
"include/linux/autoconf.h" picks up the one for the kernel we are
trying to build against, and not the one for the currently running
kernel, which is what is in /usr/src/linux-.../ copy is.
i=wangyb
i=cliff
Brian J. Murrell [Mon, 15 Nov 2010 21:15:28 +0000 (05:15 +0800)]
b=24095 don't really need configure --with-linux
Don't really need to have run configure --with-linux for version-tag.pl
to work.
i=wangyb
i=cliff
Brian J. Murrell [Mon, 15 Nov 2010 21:15:27 +0000 (05:15 +0800)]
b=24093 not all build files/scripts being distributed
Some files that need to be are not being included in the tarball
when make dist is being run.
i=cliff
i=minh
Brian J. Murrell [Wed, 17 Nov 2010 18:27:19 +0000 (02:27 +0800)]
b=24091 find_linux_rpms utility
Write a utility to find the the kernel "devel" package needed.
Restructure some code to facilitate reuse of existing code to
use in the utility.
i=cliff
i=minh
Brian J. Murrell [Wed, 17 Nov 2010 18:27:18 +0000 (02:27 +0800)]
b=24090 distro and target autodetection
Automatically detect the target and distro. It's in fact a little
crazy that these are two separate options. They should be merged.
i=cliff
i=minh
Brian J. Murrell [Wed, 17 Nov 2010 18:27:17 +0000 (02:27 +0800)]
b=24089 avoid reuse cache storage collisions
This is a hack. Basically, all nodes do their build and the last one
to store in the cache wins.
A better solution would be for nodes to announce their intention to
build for a given hash and any other nodes that will do the same wait
for the first node to finish and store it. Subsequent nodes then just
use the first node's stored result.
i=wangyb
i=cliff
LiuYing [Tue, 16 Nov 2010 01:38:14 +0000 (09:38 +0800)]
b=24117 params_tree code for YangtzeI
part of params_tree code for YangtzeI
-add safe cfs_{v}snprintf
-move linux proc related stuff to params_tree.h
i=andreas
i=alexg
Jian Yu [Tue, 16 Nov 2010 15:38:24 +0000 (23:38 +0800)]
b=24017 fix the performance issue of lu_cache_shrink
- increase busy counter in lu_obj_hop_get, so that we don't have to hack out
this in llite
- make some changes to cfs_hash:
. do not need to return anything by cfs_hash_get/put, they are not used by
anyone, cfs_hash_object also can do the same thing
. pass cfs_hash to cfs_hash_get/put
o=liang
i=vitaly.fertman
yangsheng [Tue, 16 Nov 2010 06:04:24 +0000 (14:04 +0800)]
b=24003 Shows stripe_count: 65535 but should be '-1'.
i=johann
Elena Gryaznova [Tue, 16 Nov 2010 11:18:17 +0000 (14:18 +0300)]
b=23954 MGS device has stopped when we try to start the second mgs
i=Mikhail.Pershin
add test_24b to ALWAYS_EXCEPT list for configuration mgs/mds are not combined
Jian Yu [Tue, 16 Nov 2010 01:21:43 +0000 (09:21 +0800)]
b=23289 remove global atomic from lustre handle
o=liang
i=andreas.dilger
i=mikhail.pershin
Jian Yu [Tue, 16 Nov 2010 01:11:37 +0000 (09:11 +0800)]
b=23289 less global lock & atomic for ldlm pool
o=liang
i=oleg.drokin
i=andreas.dilger
Elena Gryaznova [Tue, 16 Nov 2010 10:53:03 +0000 (13:53 +0300)]
b=22544 delete module_setup.sh
i=Brian.Murrell
yangsheng [Tue, 16 Nov 2010 05:57:45 +0000 (13:57 +0800)]
b=20744 Update to latest SLES10 SP3 kernel 2.6.16.60-0.69.1(patchless client).
i=brian
Elena Gryaznova [Tue, 16 Nov 2010 11:30:43 +0000 (14:30 +0300)]
b=18717 make "lfs check" output consistent on stdout
o=Andreas.Dilger
i=Brian.Murrell
i=Elena.Gryaznova
Brian J. Murrell [Thu, 11 Nov 2010 19:14:56 +0000 (03:14 +0800)]
b=24086 separate kernel and userspace __u64 checking
Even if we don't configure lustre --with-linux= we still need to determine
what type __u64 is for userspace otherwise the LP*64 macros won't be
properly defined.
i=whitebear
i=panda