Whamcloud - gitweb
alex [Wed, 3 Aug 2005 13:29:38 +0000 (13:29 +0000)]
b=7049
- it's enough to check for block number
alex [Wed, 3 Aug 2005 12:18:42 +0000 (12:18 +0000)]
b=7049
- it's better to unmap buffers holding no truncate sem
note: unmap_underlying_metadata() is commented out to see whether
a race happens at all
alex [Wed, 3 Aug 2005 12:04:15 +0000 (12:04 +0000)]
b=7049
- check for aliases in bd-mapping for writes only
alex [Wed, 3 Aug 2005 10:20:08 +0000 (10:20 +0000)]
b=7049
- let's check whether dirty pages/buffers in bd-mapping
alex [Wed, 3 Aug 2005 07:34:22 +0000 (07:34 +0000)]
b=7237
- missed wake_up() has been returned
- few CDEBUG() have changed their level to D_ERROR to simplify debugging
adilger [Tue, 2 Aug 2005 23:29:58 +0000 (23:29 +0000)]
Merge b1_4_bug3389 from b1_4 (20050729_0312)
Description: Mounting a Lustre file system on a node running as an OST could
lead to deadlocks
Details : OSTs now allocate memory needed to write out data at
startup, instead of when needed, to avoid having to
allocate memory in possibly low memory situations.
Specifically, if the file system is mounted on on OST,
memory pressure could force it to try to write out data,
which it needed to allocate memory to do. Due to the low
memory, it would be unable to do so and the node would
become unresponsive.
b=6514, b=5137
Description: Addition of lconf --service command line option
Details : lconf now accepts a '--service <arg>' option, which is
shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
b=7015
Description: Failover mode is now the default for OSTs.
Details : By default, OSTs will now run in failover mode. To return to
the old behaviour, add '--failout' to the lmc line for OSTs.
b=6101
Description: Health checks are now provided for MDS and OSTs
Details : Additional detailed health check information on MSD and OSTs
is now provided through the procfs health_check value.
b=1693
Description: Disk fragmentation on the OSTs could eventually cause slowdowns
after numerous create/delete cycles
Details : The ext3 inode allocation policy would not allocate new inodes
very well on the OSTs because there are no new directories
being created. Instead we look for groups with free space if
the parent directories are nearly full.
b=4466
Description: Network or server problems during mount may cause partially
mounted clients instead of returning an error.
Details : The config llog parsing code may overwrite the error return
code during mount error handling, returning success instead
of an error.
b=6302
alex [Tue, 2 Aug 2005 19:46:23 +0000 (19:46 +0000)]
b=3984
- a reference on resource must be taked under spinlock
alex [Tue, 2 Aug 2005 18:46:35 +0000 (18:46 +0000)]
b=7039
- concurrent lock cancelling might find inode dead while it's not yet
alex [Tue, 2 Aug 2005 17:34:20 +0000 (17:34 +0000)]
b=7231
- asynchronous commit in obdfilter/2.6
alex [Tue, 2 Aug 2005 15:43:33 +0000 (15:43 +0000)]
b=7231
- wait on bio's completion after commit. to evaluate performance impact
alex [Tue, 2 Aug 2005 09:09:02 +0000 (09:09 +0000)]
b=7039
- stop progress and dump a log for debugging purposes
alex [Sun, 31 Jul 2005 12:43:22 +0000 (12:43 +0000)]
- b_size_on_mds landed on HEAD:
5984 - oops in mdc_set_lock_data
we must drop lookup lock in ll_clear_inode()
7216 - abort due to missed clients
replay already committed (open/close rpcs) requests
even if some clients haven't connected in time and
got evicted
7057 - sanity: test_66: /mnt/lustre/f66 blocks 0 < 8
send close rpc only when all dirty pages are written
(otherwise i_blocks isn't valid)
6817 - don't update attrs. on mds if the client hasn't modified file
using openhandle being closed
ericm [Sun, 31 Jul 2005 04:49:34 +0000 (04:49 +0000)]
land b_hd_remote_acl: support get/set ACL from remote client.
jacob [Fri, 29 Jul 2005 23:54:32 +0000 (23:54 +0000)]
update patches and scripts for SLES 9 SP2
wangdi [Fri, 29 Jul 2005 14:06:12 +0000 (14:06 +0000)]
Branch: b_hd_crypto
minor fixes about crypto api
ericm [Thu, 28 Jul 2005 18:31:21 +0000 (18:31 +0000)]
land minor fixes from b_hd_sec:
* "EA" is used for lustre internal metadata ea. so change user ea's name to
"XATTR", including ACLs.
* remove a unnecessary param from md_getattr().
* various more fixes.
phil [Wed, 27 Jul 2005 23:48:35 +0000 (23:48 +0000)]
add missing entry for bug 6993
alex [Wed, 27 Jul 2005 18:54:34 +0000 (18:54 +0000)]
b=7200
- protect lock->l_resource from concurrent ldlm_lock_change_resource()
adilger [Wed, 27 Jul 2005 18:18:31 +0000 (18:18 +0000)]
Branch b1_4_bug3389
Fix portals_nid2str() to print something reasonable for Cray NALs.
b=6902
r=bogl
alex [Wed, 27 Jul 2005 17:20:02 +0000 (17:20 +0000)]
b=5210
r=phil(actual patch)
A workaround for the crash in ll_unhash_aliases on 2.6.
The 2.6 lookup code is now lockless (thanks to the RCU machinations), which
allows it to race with the unhashing in ll_unhash_aliases. The 2.6 __d_drop
doesn't leave the list head ->pprev in a safe state, so the next pass through
this loop (which tries to re-hash it after we lose the dput race), would crash.
For the record, it looks like the kernel's own d_prune_aliases is broken in
the very same way, except that instead of crashing immediately, it just leaves
the d_hash pointers poisoned for someone else to crash on.
wangdi [Wed, 27 Jul 2005 09:28:29 +0000 (09:28 +0000)]
Branch:b_hd_crypto
update crypto api code
alex [Sat, 23 Jul 2005 20:55:29 +0000 (20:55 +0000)]
b=7017
- atomic_dec_and_test() with subsequent spin_lock() against the list
is racy. atomic_dec_and_lock() must be used. otherwise other cpu
can hit the race window, increment refcount, drop it again and we
get to the situation when both cpus are freeing the structure
ericm [Fri, 22 Jul 2005 17:23:24 +0000 (17:23 +0000)]
some env (cnbuild) can't build without explicitly include
errno.h
ericm [Thu, 21 Jul 2005 17:17:36 +0000 (17:17 +0000)]
land b_hd_pag: rudiment support for PAG.
brian [Thu, 21 Jul 2005 14:01:53 +0000 (14:01 +0000)]
Copyright is now License in RPM spec files.
jacob [Wed, 20 Jul 2005 21:24:18 +0000 (21:24 +0000)]
quiet
brian [Wed, 20 Jul 2005 18:06:59 +0000 (18:06 +0000)]
Should allow the uml_switch to be used if wanted.
wangdi [Wed, 20 Jul 2005 12:50:30 +0000 (12:50 +0000)]
Branch: HEAD
remove redundant set_md in mds_reint_create and cmobd_reint_create
wangdi [Wed, 20 Jul 2005 12:34:28 +0000 (12:34 +0000)]
Branch: HEAD
add setxattr kml post hook for reint set md to master mds.
wangdi [Tue, 19 Jul 2005 12:56:14 +0000 (12:56 +0000)]
Branch: HEAD
should reset mcd_last_close_xid to 0 when reconnect.
b:6102
r:alex
alex [Tue, 19 Jul 2005 11:35:38 +0000 (11:35 +0000)]
- need couple routines to be accessible from modules for size-on-mds feature
wangdi [Sat, 16 Jul 2005 07:24:10 +0000 (07:24 +0000)]
Branch: HEAD
add kernel-2.6.10-suse-opteron.config for opteron arch
jacob [Fri, 15 Jul 2005 20:40:56 +0000 (20:40 +0000)]
i stand corrected
jacob [Fri, 15 Jul 2005 20:06:06 +0000 (20:06 +0000)]
this is the fix i intended
mjmac [Fri, 15 Jul 2005 19:52:29 +0000 (19:52 +0000)]
r=adilger,nic
stage() should return 0 if $STAGEDIR is not set, not 1.
wangdi [Fri, 15 Jul 2005 07:11:18 +0000 (07:11 +0000)]
Branch: HEAD
add fid and group in kml for cmobd test
jacob [Fri, 15 Jul 2005 04:34:12 +0000 (04:34 +0000)]
move AC_SUBST into lustre-build.m4 (b=7126)
jacob [Thu, 14 Jul 2005 23:04:14 +0000 (23:04 +0000)]
add --stage option to lbuild
jacob [Thu, 14 Jul 2005 23:00:40 +0000 (23:00 +0000)]
revert a broken change from an earlier fix
adilger [Thu, 14 Jul 2005 22:53:45 +0000 (22:53 +0000)]
Branch b1_4
Use a wildcard for /usr/share/lustre.
jacob [Thu, 14 Jul 2005 22:25:09 +0000 (22:25 +0000)]
just remove unused macro as i can't get it to work right anyway
jacob [Thu, 14 Jul 2005 22:02:43 +0000 (22:02 +0000)]
my mind fell out of m4 mode for a moment
jacob [Thu, 14 Jul 2005 21:50:07 +0000 (21:50 +0000)]
restripe snmp to b1_4 so that andreas will add the right globs later
jacob [Thu, 14 Jul 2005 21:47:27 +0000 (21:47 +0000)]
remove harmless error message when snmp is not present
jacob [Thu, 14 Jul 2005 21:10:04 +0000 (21:10 +0000)]
add support for the snmp module.
jacob [Thu, 14 Jul 2005 21:00:40 +0000 (21:00 +0000)]
This commit was generated by cvs2svn to compensate for changes in r49775,
which included commits to RCS files with non-trunk default branches.
jacob [Thu, 14 Jul 2005 21:00:40 +0000 (21:00 +0000)]
Initial import of lustre-snmp module.
jacob [Thu, 14 Jul 2005 17:11:21 +0000 (17:11 +0000)]
oops
jacob [Thu, 14 Jul 2005 17:02:25 +0000 (17:02 +0000)]
- skip modules lacking a tag
- print a summary of failed updates/checkouts at the end
- also return an error status in this case
jacob [Thu, 14 Jul 2005 16:23:21 +0000 (16:23 +0000)]
sort correctly
brian [Thu, 14 Jul 2005 14:25:34 +0000 (14:25 +0000)]
Revert Tom's commit. It breaks recovery.
r=alex
r=umka
wangdi [Thu, 14 Jul 2005 12:46:48 +0000 (12:46 +0000)]
Branch HEAD
add fid and group in kml log
ericm [Thu, 14 Jul 2005 03:05:21 +0000 (03:05 +0000)]
spin_is_locked() always return false in UP kernel.
ericm [Wed, 13 Jul 2005 22:22:17 +0000 (22:22 +0000)]
minor: verbose on sec register.
ericm [Wed, 13 Jul 2005 22:08:59 +0000 (22:08 +0000)]
conf-sanity: don't start/stop lsvcgssd/lgssd frequently.
ericm [Wed, 13 Jul 2005 21:53:00 +0000 (21:53 +0000)]
supply nettype to conf-sanity 5d
r=6327
yury [Wed, 13 Jul 2005 16:23:33 +0000 (16:23 +0000)]
- changes in cmobd.sh - there should not be cache LMV
- fixes in lmc
alex [Wed, 13 Jul 2005 11:54:30 +0000 (11:54 +0000)]
- remove out of date warning
liangzhen [Wed, 13 Jul 2005 10:20:05 +0000 (10:20 +0000)]
Spinlock twice.
ericm [Wed, 13 Jul 2005 00:59:04 +0000 (00:59 +0000)]
land b_hd_sec: perm/acl authorization for remote users.
yury [Tue, 12 Jul 2005 13:05:52 +0000 (13:05 +0000)]
- many cleanups with case like foo (). Replaced by foo()
- in cmobd fixed init of EA size for LOV with multiple OSTs as master. As there is no another way MD layer is getting asked for correct lovdesc, as it is needed in some cases. Thus, added handler for "lovdesc" nfo key to all MD devices.For instance, cmobd asks LMV for lov desc, it asks first MDS and MDS has it from LOV anyway.
- fixed possible LMV obd corruption in cmobd_init_ea_size() where it unconditionaly accessed u.cli things abd modified them whereas here was LMV.
- removed obd_getready() as it is not needed anymore. It was needed earlier for CMOBD to get LMV ready before asking rootid, etc. Currently alternative way is found. lmv_get_info() is just taking care of LMV targets readiness it self.
- fixed layering violations and possible oopses in cmobd data code where master export was unconditionaly expected as LOV and accessed as LOV what could be wrong in the case of using OSC. OSC using with no LOV is deprecated, but in principle is possible.
- in LMV module many functions are added by static specificator as they are not used outside of their object files.
yury [Tue, 12 Jul 2005 06:14:49 +0000 (06:14 +0000)]
- fixed possible access to of bounds in cmobd
yury [Mon, 11 Jul 2005 17:18:36 +0000 (17:18 +0000)]
- cleanups in cmobd and others:
- change LUSTRE_MDC_NAME (and others) to form like OBD_MDC_DEVICENAME to be coherent with
OBD_LMV_DEVICENAME and others. Because some OBDs used one form and others used another one.
- removed different guesses of OBD types from name of some functions in cmobd. Say
cmobd_find_master() now is called cmobd_find_obd() as all it does is finds obd
by uuid and may be used not only for finding master. Also cmobd_lmv_obd() is renamed to
cmobd_md_obd() etc. Improved function (a bit smaller of code) wich looks for master.
- in memory leaks detector do not remove memory track in case of freeing memory chunk different
of size. marking it instead as such to show in final log of leaks with special hint.
- fixes in lconf. COBD and CMOBD should use LMV and LOV devices with diffrent names to avoid
issues in the case when both they run on the same node (like in test 45) and try to setup
and clanup the same device twise (from lconf).
tappro [Mon, 11 Jul 2005 16:41:50 +0000 (16:41 +0000)]
- plugin handler should pass EXIT call in any case.
alex [Mon, 11 Jul 2005 11:43:43 +0000 (11:43 +0000)]
b=6817
r=niu
- memleak in lov_match() fixed: it didn't handle LDLM_FL_TEST_LOCK well
alex [Mon, 11 Jul 2005 11:28:09 +0000 (11:28 +0000)]
- memleak in mds_create_objects():
we should free ids array if an error is returned
alex [Mon, 11 Jul 2005 11:05:19 +0000 (11:05 +0000)]
- dump log if memory leak observed
wangchao [Mon, 11 Jul 2005 04:50:14 +0000 (04:50 +0000)]
move the initiation of fo_denylist/fo_denylist_lock from filter_setup to filter_common_setup
alex [Sun, 10 Jul 2005 23:10:17 +0000 (23:10 +0000)]
b=3984
- b_ldlm_newlocking landed. we need it to pass some CMD2 performance tests
NOTE: the new locking rules introduced by the patch are still experimental!
alex [Sun, 10 Jul 2005 23:05:06 +0000 (23:05 +0000)]
- drop reference for open replay data in mdc_commit_open() instead of
mdc_commit_close() as in some cases we may have no close at all
alex [Sun, 10 Jul 2005 22:12:32 +0000 (22:12 +0000)]
- out of date warning has been removed
alex [Sun, 10 Jul 2005 22:07:43 +0000 (22:07 +0000)]
- make 512-bytes inode size by default: with new EAs (fid, parent) we just
can't fit 256-bytes inodes
alex [Sun, 10 Jul 2005 20:38:42 +0000 (20:38 +0000)]
- convert CWARN() to CDEBUG()
alex [Sun, 10 Jul 2005 19:53:09 +0000 (19:53 +0000)]
- we don't need deadlock-monitor in general series
alex [Sun, 10 Jul 2005 19:52:22 +0000 (19:52 +0000)]
b=6817
- b_size_on_mds landed (prototype for CMD2 w/o recovery support on OST)
alex [Sun, 10 Jul 2005 18:36:24 +0000 (18:36 +0000)]
- minor debug level change to have only DEBUG_REQ with D_RPCTRACE
alex [Sun, 10 Jul 2005 18:31:06 +0000 (18:31 +0000)]
- update enqueue time upon lock conversion otherwise we get too long
enqueue wait time warning
ericm [Sun, 10 Jul 2005 05:17:51 +0000 (05:17 +0000)]
mds_open() better to check permission even it's open request for device
files, otherwise client vfs will check that and failed siliently which
cause request leak. (client's interaction with vfs seems not perfect --
can't handle all error cases)
ericm [Sat, 9 Jul 2005 19:33:55 +0000 (19:33 +0000)]
try to be a little bit more nice to remote use: as long as it be mapped
to an local user, the supp groups of target usr will take part in fs
access checking.
ericm [Sat, 9 Jul 2005 16:04:16 +0000 (16:04 +0000)]
adjust gss timeout valud, don't use fixed value.
wangdi [Sat, 9 Jul 2005 03:19:27 +0000 (03:19 +0000)]
Branch HEAD
fix a typo in sanity-cmobd.sh
yury [Fri, 8 Jul 2005 15:23:42 +0000 (15:23 +0000)]
- putting file fid into o_inline for cases when crow object may be created. It will be used later for storing into OSS object EA what is needed for quota.
tappro [Fri, 8 Jul 2005 13:57:25 +0000 (13:57 +0000)]
b=6285
- remove few debug and LBUG()
yury [Fri, 8 Jul 2005 10:35:04 +0000 (10:35 +0000)]
- cleanups in cobd and cmobd
- moved cmobd's level in lconf to be under mountpoint. This prevents oops in local config when both cobd and cmobd are used and refer to the same master device.
yury [Fri, 8 Jul 2005 07:29:45 +0000 (07:29 +0000)]
- using push_ctxt() and pop_ctxt() when setting uid/gid in filter_create_object() at Alex's suggest.
jacob [Thu, 7 Jul 2005 21:43:41 +0000 (21:43 +0000)]
- add things to /etc/modules.conf | /etc/modprobe.conf
- tweaks to init scripts for SLES 9
buffalo-pull [Thu, 7 Jul 2005 18:04:09 +0000 (18:04 +0000)]
uid == gid is no longer a valid assumption.
ericm [Thu, 7 Jul 2005 17:41:19 +0000 (17:41 +0000)]
land b_hd_sec on HEAD. various security fixes.
alex [Thu, 7 Jul 2005 16:29:19 +0000 (16:29 +0000)]
b=7039
- additional debug info: dump backtrace
alex [Thu, 7 Jul 2005 14:27:05 +0000 (14:27 +0000)]
- minor typo
alex [Thu, 7 Jul 2005 14:23:22 +0000 (14:23 +0000)]
b=6285
- sometimes mds_getattr_lock() is asked to refresh attributes by fid.
it doesn't take LOOKUP lock, thus has no protected from concurrent
unlink. this means mds_getattr_lock() can return already unlinked
(and being removed) inode. to avoid this we check i_nlink right
after we got UPDATE lock and return -ENOENT if nlink is zero
alex [Thu, 7 Jul 2005 13:38:55 +0000 (13:38 +0000)]
b=6913
- retract extents lock from clients even if OSS hasn't created object yet
alex [Thu, 7 Jul 2005 13:03:01 +0000 (13:03 +0000)]
- minor comment about openlock and test 24n from sanity.sh
tappro [Thu, 7 Jul 2005 12:21:56 +0000 (12:21 +0000)]
b=6285
- more debug info was added
tappro [Thu, 7 Jul 2005 08:33:07 +0000 (08:33 +0000)]
b=6285
- add LBUG() to get kernel-logs in case of this issue
yury [Thu, 7 Jul 2005 07:57:07 +0000 (07:57 +0000)]
- better to make vfs_create() to set correct uid and gid for new inode, as dirtying inode in ext3 is too expensive. Thanks to Andreas.
alex [Wed, 6 Jul 2005 21:53:09 +0000 (21:53 +0000)]
b=7023
- export can be failed asynchronously with regard to processing request.
this case cause open rpc to be replied to zero transno. zero transno
means no reference for request and open request gets freed after
ll_file_open() though mdc_open_data still has a reference on the
request. and this causes oops when mdc_close() tries to access the
request. the workaround is to take additional reference on the request
for mdc_open_data
alex [Wed, 6 Jul 2005 21:06:19 +0000 (21:06 +0000)]
- disable test 24n: it fails because of disabled opencache. current
mds_open() doesn't take lookup lock and client can't maintain
dcache in coherence
alex [Wed, 6 Jul 2005 20:13:41 +0000 (20:13 +0000)]
- make recovery-small.sh working correct with local configuration (1 mds)
yury [Wed, 6 Jul 2005 16:49:20 +0000 (16:49 +0000)]
- passing full options page to do_kern_mount() in confobd and smfs too like it is done in MDS and OST