Whamcloud - gitweb
tappro [Mon, 16 Oct 2006 06:58:07 +0000 (06:58 +0000)]
- initiate create_spec
tappro [Mon, 16 Oct 2006 06:38:06 +0000 (06:38 +0000)]
- initialize values in mdt_thread_info
- change log level for debug message
huanghua [Mon, 16 Oct 2006 05:21:42 +0000 (05:21 +0000)]
disable sanity 60a tempprarily, which has some problem to fix.
huanghua [Mon, 16 Oct 2006 05:20:14 +0000 (05:20 +0000)]
info->mti_transno is overwritten in initialization.
yury [Sun, 15 Oct 2006 19:36:47 +0000 (19:36 +0000)]
- possibly (I'm almost sure) fix for FLD lookup err -2. Allocating new seq and setup FLD for it should be atomic. Otherwise two threads allocating fids race and first (which allocated seq) deals with FLD setup and second (which just allocated fid in new seq) quickly sends fid in new seq to server and servers can't find it in FLD, because FLD did not see update yet;
- fix in lmv_placement_policy(). Take into account split dirs for creating child dirs too, not only files.
nikita [Sun, 15 Oct 2006 18:58:19 +0000 (18:58 +0000)]
mdt: another "uninitialized mdt_thread_info" fix.
nikita [Sun, 15 Oct 2006 18:45:53 +0000 (18:45 +0000)]
kill overwritten initialization.
yury [Sun, 15 Oct 2006 17:11:20 +0000 (17:11 +0000)]
- serialize fld reqs on client and server;
- cleanups about server fld client. Now it is created in cmm;
- cleanups in split code;
nikita [Sun, 15 Oct 2006 13:43:20 +0000 (13:43 +0000)]
mdt: do not zero mdt_thread_info
nikita [Sun, 15 Oct 2006 13:41:46 +0000 (13:41 +0000)]
mdd: simplify mdd_txn_param_build(): 0. @op can me used as an array index without scanning. 1. remove mdd_txn_lock: it protects nothing anyway.
yury [Sun, 15 Oct 2006 12:14:15 +0000 (12:14 +0000)]
- cleanups in split, remove obsoletes and unite fids alloc and objs create for slaves.
fanyong [Sun, 15 Oct 2006 11:07:05 +0000 (11:07 +0000)]
Small fix.
yury [Sun, 15 Oct 2006 10:38:31 +0000 (10:38 +0000)]
- split fixes
fanyong [Sun, 15 Oct 2006 10:30:56 +0000 (10:30 +0000)]
Initialize obd_lvfs_ctxt correctly.
Fix "ASSERTION(atomic_read(&new_ctx->pwd->d_count))" for rmtacl.
yury [Sun, 15 Oct 2006 09:52:46 +0000 (09:52 +0000)]
- some cleanups;
- added debug messages to fld
tappro [Sun, 15 Oct 2006 08:43:43 +0000 (08:43 +0000)]
- small fix. skip first mea slot.
tappro [Sun, 15 Oct 2006 08:26:40 +0000 (08:26 +0000)]
- return back old split code but keep new one under #if 0 for now
nikita [Sun, 15 Oct 2006 02:16:04 +0000 (02:16 +0000)]
mdt: mdt_thread_info_init(): oops, last commit was too fast: zeroing is still required.
nikita [Sun, 15 Oct 2006 02:12:33 +0000 (02:12 +0000)]
mdt: mdt_thread_info_init(): do not zero out whole struct mdt_thread_info: that is too expensive.
nikita [Sun, 15 Oct 2006 00:53:21 +0000 (00:53 +0000)]
dt/osd: cleanup of txn credit constants handling.
tappro [Sat, 14 Oct 2006 23:52:58 +0000 (23:52 +0000)]
- don't count self-export as stale export
tappro [Sat, 14 Oct 2006 23:51:55 +0000 (23:51 +0000)]
- pass mdt_obd through process_config parameters to the mdc instead of storing
it in lu_device->lu_obd.
- rename mdc_add_obd, mdc_del_obd to the common nameing format in b_new_cmd
<object>_<action>
nikita [Sat, 14 Oct 2006 23:44:33 +0000 (23:44 +0000)]
lu: adjust hash table size dynamically.
nikita [Sat, 14 Oct 2006 23:42:29 +0000 (23:42 +0000)]
iam: 0. make heavily used functions static inline; 1. add r5 hash and make it defauls.
ericm [Sat, 14 Oct 2006 18:12:01 +0000 (18:12 +0000)]
branch: b_new_cmd
only include gss subdir if gss is enabled.
yury [Sat, 14 Oct 2006 18:09:17 +0000 (18:09 +0000)]
- fixes in split:
- split dir may happen no any MDS - fids of slaves and master should be saved in coherence with mdsnum. Do not save fid of slave on mds0 to mea[2] for instance. This is what LMV needs in some cases and this is what may allow to get rid of doing additional fld_lookups in LMV;
- in cmm_create_slave_objects() pass to cmm_create_remote_obj() last arg as sizeof(*slave_lmv) instead of sizeof(slave_lmv), otherwise edatalen is not correct and mea on slaves get broken;
- in many split functions, look like this: for (i = 1; i < cmm->cmm_tgt_count + 1; i++) is not correct even for formet solution, it makes it possible to overwrite/access data out of array memory;
- in cmm_create_slave_objects() memory leak in case of error (OBD_FREE_PTR(lmv) should be added);
- cleanups, comments.
lsy [Sat, 14 Oct 2006 15:45:52 +0000 (15:45 +0000)]
fixed a logic error in ll_add_capa(), which will make list not sorted.
remove debug code.
wangdi [Sat, 14 Oct 2006 15:43:37 +0000 (15:43 +0000)]
Branch: b_new_cmd
remove obsolote debug warning msg
yury [Sat, 14 Oct 2006 13:03:28 +0000 (13:03 +0000)]
- fixed format in CWARN()
yury [Sat, 14 Oct 2006 12:49:32 +0000 (12:49 +0000)]
- fixed replay of meta-seq alloc whereas super-seq replay is also needed if ctl and srv run on same node.
wangdi [Sat, 14 Oct 2006 11:12:20 +0000 (11:12 +0000)]
Branch: b_new_cmd
1)reset hash end with next segment's hash start, not the next start entry's
hash offset for be compatiable with readdir.
2)set ldp_start with the hash offset, not the start entry's hash
ericm [Fri, 13 Oct 2006 23:57:13 +0000 (23:57 +0000)]
branch: b_new_cmd
from fanyong: re-enable mapping database lookup in lsvcgss.
nikita [Fri, 13 Oct 2006 20:10:30 +0000 (20:10 +0000)]
osd: get rid of /*OBJ-TEMP* hack.
yury [Fri, 13 Oct 2006 19:10:13 +0000 (19:10 +0000)]
- added debug to fld client fucntions to catch wrong activities.
ericm [Fri, 13 Oct 2006 18:14:35 +0000 (18:14 +0000)]
*** empty log message ***
yury [Fri, 13 Oct 2006 18:08:57 +0000 (18:08 +0000)]
- set in cmm mdc devices obd flags from mdt_obd in shutdown case. That should fix oops on canceling locks and make shutdown cleaner.
- show obd_name in fld_client_add_target() as it is more understandable (needed for debug, seems fld targets on diff. nodes may have diff. order what is cause of really bad issues);
nikita [Fri, 13 Oct 2006 16:17:39 +0000 (16:17 +0000)]
cleanup to shrinker environment setup and finalization.
nikita [Fri, 13 Oct 2006 14:59:58 +0000 (14:59 +0000)]
iam: use read-write lock for iam container.
lsy [Fri, 13 Oct 2006 13:47:44 +0000 (13:47 +0000)]
don't hijack any field in mdt_body as return value.
instead don't set OBD_MD_FLOSSCAPA in reply can work.
nikita [Fri, 13 Oct 2006 13:32:32 +0000 (13:32 +0000)]
mdt: move ESERIOUS deifnition to lustre_mdt.h
nikita [Fri, 13 Oct 2006 13:28:20 +0000 (13:28 +0000)]
iam: split conditionally compiled functions properly.
yury [Fri, 13 Oct 2006 13:06:17 +0000 (13:06 +0000)]
- do not issue RPC inside spin lock;
lsy [Fri, 13 Oct 2006 12:59:56 +0000 (12:59 +0000)]
update comment.
lsy [Fri, 13 Oct 2006 12:58:52 +0000 (12:58 +0000)]
revert last commit, transfer a positive value should be better.
lsy [Fri, 13 Oct 2006 12:54:18 +0000 (12:54 +0000)]
use mdt_body->size instead of mdt_body->flags to store errno,
whose type is __u64, thus can transfer negative value between 32 and 64 platform.
lsy [Fri, 13 Oct 2006 12:37:24 +0000 (12:37 +0000)]
for capability renewal, once object not exist, don't return -ENOENT directly,
mdt_renew_capa will handle this error.
tappro [Fri, 13 Oct 2006 12:32:13 +0000 (12:32 +0000)]
- fix missed result code due to wrong GOTO usage
yury [Fri, 13 Oct 2006 11:48:31 +0000 (11:48 +0000)]
- fixed two missed lmv_obj_put() what cause sinability to free kmem cache later.
yury [Fri, 13 Oct 2006 11:11:40 +0000 (11:11 +0000)]
- using __mdd_lookup_locked() in mdd_parent_fid();
- fixes in mdt_renew_capa() about using @body.
fanyong [Fri, 13 Oct 2006 10:33:42 +0000 (10:33 +0000)]
Process cross-mkdir with SGID mode.
nikita [Fri, 13 Oct 2006 10:28:15 +0000 (10:28 +0000)]
iam: use config.h switches, defined by ./configure
huanghua [Fri, 13 Oct 2006 10:22:24 +0000 (10:22 +0000)]
save a fid copy before freeing the original request.
tappro [Fri, 13 Oct 2006 09:59:02 +0000 (09:59 +0000)]
- issue with GOTO;
- make double checks for rc > 0 as one
huanghua [Fri, 13 Oct 2006 09:43:28 +0000 (09:43 +0000)]
follow {parent, child} lock order
huanghua [Fri, 13 Oct 2006 09:42:51 +0000 (09:42 +0000)]
remove lock recursion.
huanghua [Fri, 13 Oct 2006 09:42:23 +0000 (09:42 +0000)]
added debug code
huanghua [Fri, 13 Oct 2006 09:41:52 +0000 (09:41 +0000)]
fix typo in RETURN
lsy [Fri, 13 Oct 2006 07:02:59 +0000 (07:02 +0000)]
assert capa buffer which should not be NULL.
wangdi [Fri, 13 Oct 2006 06:53:11 +0000 (06:53 +0000)]
Branch: b_new_cmd
rename obsolete is_parent check in mdd_rt_sanity_check
lsy [Fri, 13 Oct 2006 06:52:20 +0000 (06:52 +0000)]
once do_capa_get failed, it will return errno instead of NULL, add check for it.
wangdi [Fri, 13 Oct 2006 06:00:04 +0000 (06:00 +0000)]
Branch: b_new_cmd
miss lmv_obj_put
wangdi [Fri, 13 Oct 2006 05:14:01 +0000 (05:14 +0000)]
Branch: b_new_cmd
check filename before pack name in mdc_getattr_name, since filename might be NULL
when handle cross-ref or split object
huanghua [Fri, 13 Oct 2006 04:25:03 +0000 (04:25 +0000)]
fix a typo
huanghua [Fri, 13 Oct 2006 03:53:56 +0000 (03:53 +0000)]
using cross-ref lock for link source object
ericm [Fri, 13 Oct 2006 03:10:24 +0000 (03:10 +0000)]
branch: b_new_cmd
from tappro: a better version of size_roundup_power2().
ericm [Fri, 13 Oct 2006 01:54:03 +0000 (01:54 +0000)]
branch: b_new_cmd
second part of b11012: support enlarge_reqbuf() in gss authenticate
and privacy mode.
nikita [Thu, 12 Oct 2006 23:40:07 +0000 (23:40 +0000)]
lu_site_init(): remove duplicate initialization
nikita [Thu, 12 Oct 2006 23:29:32 +0000 (23:29 +0000)]
lu_object: add lu_object cache shrinking.
nikita [Thu, 12 Oct 2006 20:18:02 +0000 (20:18 +0000)]
ptlrpc: mark branches in hot path as unlikely()
nikita [Thu, 12 Oct 2006 20:16:03 +0000 (20:16 +0000)]
iam: further cpu consumtion fixes
nikita [Thu, 12 Oct 2006 20:05:47 +0000 (20:05 +0000)]
lu_object: cpu consumption reduction: increase hash table size. (Should be automatically tuned, in fact.)
nikita [Thu, 12 Oct 2006 20:05:02 +0000 (20:05 +0000)]
ptlrpc: mark branches in hot path as unlikely()
yury [Thu, 12 Oct 2006 18:37:46 +0000 (18:37 +0000)]
- do not allocate super-seq in controller assign time. Do that later when that is really needed.
- save defult value on controller in strt time.
yury [Thu, 12 Oct 2006 18:23:05 +0000 (18:23 +0000)]
- added ldiskfs debug configure keys.
ericm [Thu, 12 Oct 2006 17:59:27 +0000 (17:59 +0000)]
branch: b_new_cmd
- first part of b11012: mdc_realloc_openmsg() should not manipulate
request buffer directly, should be done in sptlrpc layer. implement
enlarge_msg() in null and plain policy.
- restore default to plain for rpcs to MDTs
fanyong [Thu, 12 Oct 2006 16:08:22 +0000 (16:08 +0000)]
Roll back for the change of server_put_mount.
Use server_get_mount_2 instead of server_get_mount.
lsy [Thu, 12 Oct 2006 15:57:27 +0000 (15:57 +0000)]
give friendly error message if crypto failed.
lsy [Thu, 12 Oct 2006 15:55:05 +0000 (15:55 +0000)]
ll_revalid_it will pack fid2, while its capability might expire yet,
but actually mds will use its name to locate object, so this message
might not indicate an error.
lsy [Thu, 12 Oct 2006 15:35:35 +0000 (15:35 +0000)]
* all capabilities of opened inode should always renew because many file
operations do from file->f_dentry->d_inode directly.
* all dir capabilities should always renew because ll_revalid_it needs
parent dir capability.
* filter update capability key more intelligently.
* rename ll_i2mdscapa -> ll_mdscapa_get, ll_lookup_oss_capa -> ll_osscapa_get.
* other code cleanup.
nikita [Thu, 12 Oct 2006 14:15:23 +0000 (14:15 +0000)]
iam,cmm,llite: cleanup handling of corner cases in readdir, especially for split directories. Get rid of -ERANGE and -E2BIG errors.
yury [Thu, 12 Oct 2006 14:11:48 +0000 (14:11 +0000)]
- zero out used locks in MDT before finishing req handling;
- fixes in seq-mgr recovery (meta and super);
- cleanups, more debug and comments, fixes in comments.
fanyong [Thu, 12 Oct 2006 13:32:17 +0000 (13:32 +0000)]
(1) Do not call server_deregister_mount unless ref conut is zero.
(2) Enable acl under default mode.
tappro [Thu, 12 Oct 2006 11:58:34 +0000 (11:58 +0000)]
- serialize done_writting with close requests
- use 'close' fields in last_rcvd also for done_writting
yury [Thu, 12 Oct 2006 10:31:49 +0000 (10:31 +0000)]
- using find in sanity-lmv again.
wangdi [Thu, 12 Oct 2006 08:47:07 +0000 (08:47 +0000)]
Branch: b_new_cmd
ma_lmm_size is maxisam of lov md size initially, which should be update
in req2attr update
tappro [Thu, 12 Oct 2006 08:03:58 +0000 (08:03 +0000)]
- put some asserts for debug
- remove unneeded debug message
wangdi [Thu, 12 Oct 2006 07:49:54 +0000 (07:49 +0000)]
Branch: b_new_cmd
Log:
oops, typo again, sorry :(
wangdi [Thu, 12 Oct 2006 07:47:06 +0000 (07:47 +0000)]
Branch: b_new_cmd
Log:
fix typo in the last commit
wangdi [Thu, 12 Oct 2006 07:40:01 +0000 (07:40 +0000)]
correct bit size for computing hash in client
huanghua [Thu, 12 Oct 2006 07:30:29 +0000 (07:30 +0000)]
fix return value for mdt_is_subdir().
huanghua [Thu, 12 Oct 2006 02:28:57 +0000 (02:28 +0000)]
passing valid argument
ericm [Wed, 11 Oct 2006 18:07:30 +0000 (18:07 +0000)]
branch: b_new_cmd
hack version of b11058.
wangdi [Wed, 11 Oct 2006 16:08:05 +0000 (16:08 +0000)]
Branch: b_new_cmd
use ma in cmm_thread_info, not allocate MA
tappro [Wed, 11 Oct 2006 13:22:37 +0000 (13:22 +0000)]
- fix the checking stale objects in mdt_getattr_name_lock
yury [Wed, 11 Oct 2006 11:22:10 +0000 (11:22 +0000)]
- remove check for obd_configured from target_handle_connect(), this should be fixed some another way.
yury [Wed, 11 Oct 2006 10:24:51 +0000 (10:24 +0000)]
- stop recovery thare in obdfilter before freeing namespace;
- cleanups in error messages and comments;
- more verbosity to class_config_notify_end();
- set obd->obd_configured = 0 in class_setup() before doing anything;
- do not pre-allocate seq-ranges in mdc. It usually will not do that asap
as remote target is not yet configured. It will be allocated later in very
first opeartion time, when you fid is needed;
- zero out lock handle in mdt_intent_lock_replace() to not confuse mdt_info_thread_fini();
- do not set obd->obd_configured if log processing finishes with error;
- do not allow gss connection if obd is not configured;
- disable FLD relay test 0c for multi MDS config.
fanyong [Wed, 11 Oct 2006 08:48:12 +0000 (08:48 +0000)]
Drop security ctxt when umount, for avoiding ldlm_revoke_export_locks
accessing freed resource.
wangdi [Wed, 11 Oct 2006 08:08:51 +0000 (08:08 +0000)]
Branch: b_new_cmd
fix the problem in reset hash segment
ericm [Wed, 11 Oct 2006 04:14:22 +0000 (04:14 +0000)]
branch: b_new_cmd
mdc_realloc_openmsg() is violating rules of sptlrpc, LBUG it for this
moment. real fixes shold come asap.
tappro [Wed, 11 Oct 2006 00:26:32 +0000 (00:26 +0000)]
remove '\n' from the end of the debug message