Whamcloud - gitweb
alex [Mon, 29 Aug 2005 10:28:16 +0000 (10:28 +0000)]
- dump log if OSC recovery takes too long
alex [Mon, 29 Aug 2005 09:49:28 +0000 (09:49 +0000)]
r=tappro
- couple unused vars have been removed
wangdi [Mon, 29 Aug 2005 07:22:19 +0000 (07:22 +0000)]
Branch: HEAD
revert back the fix for b7379, it is not right.
alex [Sun, 28 Aug 2005 17:57:15 +0000 (17:57 +0000)]
- fill io epoch in MDS_CLOSE only if file is writtable
alex [Sun, 28 Aug 2005 13:16:48 +0000 (13:16 +0000)]
- set and check io epoch only for writtable files
alex [Sun, 28 Aug 2005 13:14:33 +0000 (13:14 +0000)]
b=7412
r=ericm
- keep FLSIZE for non-regular files
wangdi [Sun, 28 Aug 2005 02:56:22 +0000 (02:56 +0000)]
Branch: HEAD
a tmp fix for b7379.
lsy [Sat, 27 Aug 2005 10:15:23 +0000 (10:15 +0000)]
a minor reorganization of capability code.
some cleanup.
tappro [Fri, 26 Aug 2005 12:59:06 +0000 (12:59 +0000)]
b=7390
previous fix was wrong
tappro [Fri, 26 Aug 2005 11:48:04 +0000 (11:48 +0000)]
b=7390
- enable audit on every child in dir while setting audit on that dir.
tappro [Fri, 26 Aug 2005 11:44:26 +0000 (11:44 +0000)]
- fix case when inode can be NULL.
tappro [Fri, 26 Aug 2005 11:42:29 +0000 (11:42 +0000)]
b=7390
- add special handler for auditing access failures
yury [Fri, 26 Aug 2005 08:53:03 +0000 (08:53 +0000)]
- all stats gatherd in OSC debugging time (cache loading, etc.) moved to proc. Added new proc file ./cache_stats showing them.
- some RPCs related new stats added to ./rpc_stats
wangdi [Fri, 26 Aug 2005 03:45:05 +0000 (03:45 +0000)]
branch: HEAD
some updates in sanity-crypto.sh
lsy [Thu, 25 Aug 2005 14:55:33 +0000 (14:55 +0000)]
shouldn't touch this.
lsy [Thu, 25 Aug 2005 14:52:52 +0000 (14:52 +0000)]
fixes:
* previous implementation doesn't support multi-open/truncate by different
users on the same inode.
* update DLD.
* cleanup.
tappro [Thu, 25 Aug 2005 12:58:02 +0000 (12:58 +0000)]
fix for namelen used for audit purposes.
tappro [Thu, 25 Aug 2005 09:41:12 +0000 (09:41 +0000)]
b=7390
- move reint auditing into mds_audit
- auth audit should use child inode if it exists to recognize correct audit setting
tappro [Thu, 25 Aug 2005 09:36:28 +0000 (09:36 +0000)]
b=7390
audit on file didn't work on unlink due to loss of audit setting. Now audit mask is saved in smfs_inode_info also.
wangdi [Thu, 25 Aug 2005 07:07:27 +0000 (07:07 +0000)]
Branch: HEAD
b7148 typo fix
yury [Thu, 25 Aug 2005 06:34:25 +0000 (06:34 +0000)]
- returned back grant related changes moved from 1_4 as they seem cause issues with #45
wangdi [Thu, 25 Aug 2005 06:28:16 +0000 (06:28 +0000)]
Branch: HEAD
tmp hack fix for b7148, now we use confobd to write config log, so mds_config_version
is useless in config log name defination. Remove it temporarily.
b7148
alex [Wed, 24 Aug 2005 21:49:24 +0000 (21:49 +0000)]
b=7178
r=tappro
- avoid possibility to FLSIZE being sent back to client w/o checking
alex [Wed, 24 Aug 2005 21:31:01 +0000 (21:31 +0000)]
b=7178
- let's try to catch the case when late MDS_CLOSE updates attrs. from
out-of-date i/o epoch
nikita [Wed, 24 Aug 2005 19:27:03 +0000 (19:27 +0000)]
file random-reads.c was initially added on branch b1_4.
yury [Wed, 24 Aug 2005 14:05:51 +0000 (14:05 +0000)]
- returned back OSC dirty size.
- enabled filter_tally_write() to have brw_stats on OST
wangdi [Wed, 24 Aug 2005 09:45:12 +0000 (09:45 +0000)]
Branch: HEAD
b7375
when decrypt page, should decrypt the whole page array, not only the first
page
alex [Wed, 24 Aug 2005 04:36:29 +0000 (04:36 +0000)]
b=7356
- we should allow processing for request being resent during request
replay stage. but we may do that only when all clients are connected
(or after evicting clients that haven't connect in time)
wangdi [Wed, 24 Aug 2005 03:33:33 +0000 (03:33 +0000)]
Branch: HEAD
add lmv_precleanup for cleanup llog for lmv
tappro [Tue, 23 Aug 2005 15:11:58 +0000 (15:11 +0000)]
b=7391
lctl fix for setattr
remove unused symlink opcode
tappro [Tue, 23 Aug 2005 15:08:25 +0000 (15:08 +0000)]
remove unused variable
tappro [Tue, 23 Aug 2005 08:40:34 +0000 (08:40 +0000)]
b=7214
- fix duplicate name in open/stat auditing. Name is needed only if lustre id cannot be saved.
lsy [Tue, 23 Aug 2005 06:26:36 +0000 (06:26 +0000)]
set fsuid upon readpage.
ericm [Tue, 23 Aug 2005 03:48:48 +0000 (03:48 +0000)]
add "blacklist" support into OSS capabilities.
pjkirner [Mon, 22 Aug 2005 23:47:06 +0000 (23:47 +0000)]
Add LNET CVS Repository Support
ericm [Mon, 22 Aug 2005 21:13:47 +0000 (21:13 +0000)]
remove some temporary debug msgs.
yury [Mon, 22 Aug 2005 16:54:08 +0000 (16:54 +0000)]
few changes for debug purposes in OSC:
- increased number in flight requests by factor 2. Thus twise is increased cl_dirty_max (in fact cache size)
- disabled check for in flight requests in osc_check_rpcs() - number of in flight RPCs is limited anyway by ->cl_dirty_max
yury [Mon, 22 Aug 2005 14:46:58 +0000 (14:46 +0000)]
- added CERROR LBUG in filter_grant_check() from 1_4
yury [Mon, 22 Aug 2005 14:28:25 +0000 (14:28 +0000)]
- added missed fix to osc_exit_cache() from 1_4. It updates lost_grant mor carefuly what is important for cases when blocksize is not equal to page size.
- fixed warning about init of var in llog test stuff.
yury [Mon, 22 Aug 2005 11:57:40 +0000 (11:57 +0000)]
- added more debug in osc. Currently two ideas are working for checks why osc does not give desired band:
(1) cache size is choosen kind of cumbersome way and not enough, what makes writing thread waiting for cache in osc_enter_cache(). Cache size curremtly does not depend on how good wire is able to send RPCs to remote node what seems to be an artificial limit in some cases. If this is so, the fix would be to adjust cache size depending on much it is needed. This however does not explain why second subtest suffers from slowdown whereas first one does not.
(2) due to some reasons (limited cache or limited grant from OST) many async RPCs are falled back to synchronous IO what makes substential slowdown.
Added debug on real cluster should show what is going on there. Idea that issue on server has a little chance to be true, as server uses direct IO and there is almost not ability to be slowed down due to some cache issues, etc.
eeb [Mon, 22 Aug 2005 11:42:01 +0000 (11:42 +0000)]
* simplified gmnal thread startup/shutdown
tappro [Mon, 22 Aug 2005 11:21:59 +0000 (11:21 +0000)]
while sanity test60 uninitialized llog_cookie in llog_test_4() can invoke assertion in llog_cat_current_log()
eeb [Mon, 22 Aug 2005 10:58:22 +0000 (10:58 +0000)]
* gmnal: removed spinlock round GM API calls that don't need to be serialised
(they'd been added "blindly" previously)
* gmnal netregression ran for 36 hours.
yury [Mon, 22 Aug 2005 09:48:30 +0000 (09:48 +0000)]
- fixes about cobd_switch() which caused issues with cleanup:
- avoid double connections to choosen export due to using lctl on umounted client (1st connection) and after that mount the client (2nd connection). Avoid double EA init along with double connection.
- disconnect correct export on cobd_disconnect()
- avoid possible oops on double cobd_cleanup() due to freed cmobd->write_srv, but not zerod out after that.
- in cobd_switch() set ->cache_on field into new value only if COBD_CONNECT or COBD_SWITCH bias is specified.
- in LMV and LOV in lconf cleanup LMV and LOV first and after that cleanup all their targets. This is because LMV, etc. cleanup needs targets OBD attached, with correct obd_name, etc. This fixes prevents oops in disconnect time whrn using test45 configuration.
lsy [Mon, 22 Aug 2005 08:31:00 +0000 (08:31 +0000)]
fixes:
* pack fid for direct io.
* seperate verify_fid from verify_capa to fully support truncate.
* debug message cleanup.
tappro [Mon, 22 Aug 2005 07:41:59 +0000 (07:41 +0000)]
sanity test58 was failed due to structures changing. Setting proper values did fix that.
tappro [Sun, 21 Aug 2005 22:56:12 +0000 (22:56 +0000)]
b=7354
smfs didn't update backfs file pos with smfs file pos value before operation. This were critical for llog processing in some situations.
wangdi [Sun, 21 Aug 2005 06:32:18 +0000 (06:32 +0000)]
Branch: HEAD
comment out llog_init in lmv_add_mdc, because we already do llog_init in mdc_setup
wangdi [Sat, 20 Aug 2005 19:37:38 +0000 (19:37 +0000)]
Branch HEAD
land add mds code to HEAD
ericm [Sat, 20 Aug 2005 19:08:25 +0000 (19:08 +0000)]
don't let ctx obj share storage with request buffer, which might be
released in case of recovery, request be saved and delayed handed.
yury [Sat, 20 Aug 2005 16:46:00 +0000 (16:46 +0000)]
- more debug stuff in osc. Added av. pipe loading, max pipe loading, etc.
- some names cleanup in gns and capa threads related stuff to make all llite
threads (gns, capa, close) coherent in names to each other.
ericm [Fri, 19 Aug 2005 19:57:22 +0000 (19:57 +0000)]
add some debug message.
wangdi [Fri, 19 Aug 2005 17:55:54 +0000 (17:55 +0000)]
Branch: HEAD
b7374,7375 offset should be mapped inside page. when crypt and decrypt
eeb [Fri, 19 Aug 2005 17:11:42 +0000 (17:11 +0000)]
* cleaned up startup/shutdown handling
* queue rx descriptors for handling by thread directly
tappro [Fri, 19 Aug 2005 10:31:51 +0000 (10:31 +0000)]
- temporary disable type check
niu [Fri, 19 Aug 2005 09:36:15 +0000 (09:36 +0000)]
use inode->i_mode instead of fid type, and comment out the assertions, because
type in lustre id are not always correct.(client don't update type in lli_id;
id for cross-ref inode never store mode)
niu [Fri, 19 Aug 2005 08:50:23 +0000 (08:50 +0000)]
pack lustre id to oss in setattr on open
adilger [Fri, 19 Aug 2005 08:40:32 +0000 (08:40 +0000)]
file lustre-version-revert_suse.patch was initially added on branch b1_4.
adilger [Fri, 19 Aug 2005 07:36:02 +0000 (07:36 +0000)]
Branch HEAD
Quiet compiler warnings on x86_64.
lsy [Fri, 19 Aug 2005 07:29:33 +0000 (07:29 +0000)]
several fixes: expiry timer adjusted.
add support for ost-authorization-revocation and fid verification.
yury [Fri, 19 Aug 2005 06:54:36 +0000 (06:54 +0000)]
- added more debug client side pipe related stuff
tappro [Thu, 18 Aug 2005 22:54:10 +0000 (22:54 +0000)]
b=7214
- turn off audit for setattr/unlink on OSS
- add few checks for stored lustre_id
tappro [Thu, 18 Aug 2005 22:51:20 +0000 (22:51 +0000)]
b=7360
- wrong lustre_id was used in audit logging.
jacob [Thu, 18 Aug 2005 20:18:39 +0000 (20:18 +0000)]
fix possible double-version with suse kernels
eeb [Thu, 18 Aug 2005 17:43:26 +0000 (17:43 +0000)]
* GMNAL
- cleaned up structs
- removed buggy RDMA handling (previous version is tagged
GM_PRE_REMOVE_BUGGY_RDMA for reference)
- fixed memory leak on teardown
alex [Thu, 18 Aug 2005 16:02:49 +0000 (16:02 +0000)]
- disable true sync back as it makes recovery more complex (I_LOCK)
alex [Thu, 18 Aug 2005 14:34:49 +0000 (14:34 +0000)]
- dump log about long syncing in worse cases only
alex [Thu, 18 Aug 2005 10:42:25 +0000 (10:42 +0000)]
b=7357
- check lop body as well
alex [Thu, 18 Aug 2005 10:36:44 +0000 (10:36 +0000)]
b=7357
- couple LASSERTs to catch use-after-free
wangdi [Thu, 18 Aug 2005 10:26:15 +0000 (10:26 +0000)]
Branch: HEAD
fix memory leak after land ost add/del
alex [Thu, 18 Aug 2005 10:16:44 +0000 (10:16 +0000)]
b=7356
- show error message always to simplify bug catching
alex [Thu, 18 Aug 2005 09:56:57 +0000 (09:56 +0000)]
- let's try to serialize all allocations and see how does that help
niu [Thu, 18 Aug 2005 08:19:09 +0000 (08:19 +0000)]
- fix typo of chkconnect in lmv_set_info.
- add MDS_PARSE_ID check.
wangdi [Thu, 18 Aug 2005 07:54:27 +0000 (07:54 +0000)]
Branch: HEAD
serval fix about ost add/del, after landing, which will fix
some memory leak
alex [Thu, 18 Aug 2005 06:22:38 +0000 (06:22 +0000)]
b=7346
- debug addon: track how long iget() and ll_writepages() take
wangdi [Thu, 18 Aug 2005 04:54:08 +0000 (04:54 +0000)]
Branch: HEAD
fix bug 7348, when chmod, should recompute mac
eeb [Wed, 17 Aug 2005 18:35:20 +0000 (18:35 +0000)]
* running on 2 GM nodes @ HP
wangdi [Wed, 17 Aug 2005 16:21:04 +0000 (16:21 +0000)]
Branch: HEAD
revert back ptldebug after merge.
alex [Wed, 17 Aug 2005 16:07:18 +0000 (16:07 +0000)]
- minor fixes against gcc's warnings
alex [Wed, 17 Aug 2005 15:38:31 +0000 (15:38 +0000)]
- initialize namesize var to fix gcc's warning
alex [Wed, 17 Aug 2005 15:37:17 +0000 (15:37 +0000)]
- minor message correction
wangdi [Wed, 17 Aug 2005 15:25:24 +0000 (15:25 +0000)]
Branch: HEAD
the inflight page should also be teardown when unmount
alex [Wed, 17 Aug 2005 14:41:55 +0000 (14:41 +0000)]
b=7346
- debug addon
wangdi [Wed, 17 Aug 2005 13:00:24 +0000 (13:00 +0000)]
Branch: HEAD
for inactive tgts, we should clear l_ast_data too, thanks bzzz remind
yury [Wed, 17 Aug 2005 12:45:22 +0000 (12:45 +0000)]
- more debug info of pipe feeding on client
wangdi [Wed, 17 Aug 2005 09:45:52 +0000 (09:45 +0000)]
Branch: HEAD
revert LBUG back in target_recovery_thread
wangdi [Wed, 17 Aug 2005 09:32:06 +0000 (09:32 +0000)]
Branch: HEAD
land ost add/del to HEAD
adilger [Wed, 17 Aug 2005 09:06:39 +0000 (09:06 +0000)]
file kjournald_affinity.patch was initially added on branch b1_4.
alex [Wed, 17 Aug 2005 08:30:52 +0000 (08:30 +0000)]
b=7268
- additional debug info
yury [Tue, 16 Aug 2005 15:13:55 +0000 (15:13 +0000)]
- returned back not needed changes committed unintentionaly.
yury [Tue, 16 Aug 2005 14:55:10 +0000 (14:55 +0000)]
- small cleanups in cobd
- added a patch for debugging write gaps (av. gap time and sum gaps time)
ericm [Tue, 16 Aug 2005 02:16:57 +0000 (02:16 +0000)]
add some debug message
alex [Mon, 15 Aug 2005 22:00:00 +0000 (22:00 +0000)]
- drop dchild to NULL in order to prevent further access at cleanup
alex [Mon, 15 Aug 2005 21:56:46 +0000 (21:56 +0000)]
- pass connect flag claiming llite is a client
adilger [Mon, 15 Aug 2005 21:02:18 +0000 (21:02 +0000)]
Branch b1_4
Allow tests to optionally be built with cray portals.
b=7117
r=bogl
yury [Mon, 15 Aug 2005 10:04:48 +0000 (10:04 +0000)]
- removed not needed filter_group_set_fs_flags()
- small cleanups, more detailed error messages, fied error codes in ost_rec_create_pack(),
fixed mixed declations and non declation stuff in osc_get_info()
wangdi [Mon, 15 Aug 2005 08:18:26 +0000 (08:18 +0000)]
Branch: HEAD
1) add mks type in sanity-crypto.sh
2) fix decrypt problem in mks type
wangdi [Sun, 14 Aug 2005 17:55:12 +0000 (17:55 +0000)]
Branch: HEAD
should reset the mem to 0, if not get the indicated next block.
which should fix the audit log oops.
alex [Sun, 14 Aug 2005 14:44:59 +0000 (14:44 +0000)]
- per-inode wait queue to wait for I/O completion
- never leave non-handled close works (in close thread)
alex [Sun, 14 Aug 2005 14:40:50 +0000 (14:40 +0000)]
- as we don't use Writeback bit in 2.6, we need other way for sync I/O.
let's do that in ll_writepages(). otherwise, igrab() from I/O completion
can find inode being freed