Whamcloud - gitweb
tappro [Mon, 6 Nov 2006 14:20:56 +0000 (14:20 +0000)]
- finish reqp if -ERESTART received, add debug
nikita [Mon, 6 Nov 2006 11:21:29 +0000 (11:21 +0000)]
osd: fix ipd handling for iterators.
wangdi [Mon, 6 Nov 2006 10:36:49 +0000 (10:36 +0000)]
Branch: b_new_cmd
add debug_info for 11123
fanyong [Mon, 6 Nov 2006 07:19:06 +0000 (07:19 +0000)]
Check WRITE permission of parent when create.
yury [Sun, 5 Nov 2006 14:12:23 +0000 (14:12 +0000)]
- optimization for open. In case mdt_reint_open() sees that child object does not exist and should be created, tell mdd do not lookup child object again in sanity check.
tappro [Sun, 5 Nov 2006 12:37:04 +0000 (12:37 +0000)]
correct block/inode ratio for OST, don't use defaults
tappro [Sun, 5 Nov 2006 12:31:13 +0000 (12:31 +0000)]
- enable 4G of memory usage, disable preemt
nikita [Sat, 4 Nov 2006 23:24:09 +0000 (23:24 +0000)]
osd/mdd: new (weaker) locking for ->do_attr_{g,e}set(), see message to colibri@ for details.
nikita [Sat, 4 Nov 2006 21:14:47 +0000 (21:14 +0000)]
mdd: little cleanup.
nikita [Sat, 4 Nov 2006 21:14:11 +0000 (21:14 +0000)]
0. iam: pub r5 hash back.
1. iam/osd: fix a race that caused 11150.
yury [Sat, 4 Nov 2006 15:33:38 +0000 (15:33 +0000)]
- remove debug stuff for b11150
tappro [Sat, 4 Nov 2006 15:30:35 +0000 (15:30 +0000)]
- make 16K bytes/inode ratio only for really big devices > 100Gb. This is
workaround against -ENOSPC on OST due to lack of inodes
yury [Sat, 4 Nov 2006 08:27:23 +0000 (08:27 +0000)]
- cleanups and fixes in fid/lproc* stuff;
- in mdd_dir_page_build() added check for wrong hash.
- in mdd_dir_page_build() fixed using not swabbed fid.
huanghua [Sat, 4 Nov 2006 07:31:39 +0000 (07:31 +0000)]
(1) check whether reply buffer has been packed;
(2) shrink reply buffer if needed;
(3) call ptlrpc_lprocfs_unregister_obd() while cleanup.
huanghua [Sat, 4 Nov 2006 06:08:00 +0000 (06:08 +0000)]
fix a typo
ericm [Fri, 3 Nov 2006 21:06:07 +0000 (21:06 +0000)]
branch: b_new_cmd
after negotiation handling might invalidate the whole negotiation, do
proper cleanup, and client don't repeatedly try (bug found by fanyong).
nikita [Fri, 3 Nov 2006 18:30:47 +0000 (18:30 +0000)]
ptlrpc: put back into ll_rpc_opcode[] symbolic rpc names lost during 1.5 merge.
nikita [Fri, 3 Nov 2006 18:30:03 +0000 (18:30 +0000)]
mdt: export ptlrpc stats fro mdt services.
nikita [Fri, 3 Nov 2006 18:28:32 +0000 (18:28 +0000)]
seq: fix (paper over, in fact) NULL pointer dereference during read of /proc/fs/lustre/seq/cli-ctl-* files.
tappro [Fri, 3 Nov 2006 16:15:21 +0000 (16:15 +0000)]
revert back changes about MAX_HASH_SIZE usage during hash_segment calculations.
yury [Fri, 3 Nov 2006 15:36:06 +0000 (15:36 +0000)]
- cleanups in mdd about update CTIME etc., for all operaitons. Removed dead code unbder if 0
tappro [Fri, 3 Nov 2006 12:39:07 +0000 (12:39 +0000)]
- more fixes with hash_segment calculation. make them similar during split and
at client after that.
- fix cmr_link() issue. use correct place to store fid during lookup
nikita [Fri, 3 Nov 2006 12:32:31 +0000 (12:32 +0000)]
cmm: replace some dynamic allocations with context key values.
tappro [Fri, 3 Nov 2006 12:02:13 +0000 (12:02 +0000)]
avoid loosing last hashes
fanyong [Fri, 3 Nov 2006 10:55:23 +0000 (10:55 +0000)]
Fix ASSERTION for local client when identity_upcall is disabled.
fanyong [Fri, 3 Nov 2006 07:35:51 +0000 (07:35 +0000)]
sanity check for user authenticate.
nikita [Thu, 2 Nov 2006 21:43:46 +0000 (21:43 +0000)]
mdd: initialize mdd->mdd_txn_cb.dtc_linkage to avoid crash in error handling path (bug found by Elena).
ericm [Thu, 2 Nov 2006 21:29:20 +0000 (21:29 +0000)]
branch: b_new_cmd
fix assertion: enc pages might be allocated by previous failed bulk rpc.
tappro [Thu, 2 Nov 2006 17:16:40 +0000 (17:16 +0000)]
- fix split issue with create file on wrong mds. Check for split is needed
always in cml_create, even after cmm_split_try()
- fix issue with cross-ref link to existent name, in that case the extra ref
occur on inode. Additional lookup() in cmr_link() prevents such situation.
nikita [Thu, 2 Nov 2006 16:22:49 +0000 (16:22 +0000)]
sync with last changes to iam_lvar.c.
nikita [Thu, 2 Nov 2006 16:15:59 +0000 (16:15 +0000)]
procsleep: simplify stack-walker, making it more robust.
wangdi [Thu, 2 Nov 2006 15:47:54 +0000 (15:47 +0000)]
Branch: b_new_cmd
try another hash method to check whether 11150 is related with iam
fanyong [Thu, 2 Nov 2006 13:44:24 +0000 (13:44 +0000)]
Small fix, user authentication first.
yury [Thu, 2 Nov 2006 12:48:43 +0000 (12:48 +0000)]
- fixes in split about le/cpu byte order, cleanups;
- added dump page in split in case of error in remove page code.
fanyong [Thu, 2 Nov 2006 08:24:16 +0000 (08:24 +0000)]
(1) For setattr / setxattr, change ctime.
(2) For link / unlink, change child's ctime, change parent's ctime and mtime.
(3) For create / delete, change parent ctime and mtime.
(4) For rename, change source's and target's ctime, change ctime and mtime of
source's parent and target's parent.
huanghua [Thu, 2 Nov 2006 05:24:13 +0000 (05:24 +0000)]
fix a typo which makes cross link() fail.
huanghua [Thu, 2 Nov 2006 04:06:56 +0000 (04:06 +0000)]
(1) add some test case for link();
(2) add ONLY=MOUNT for sanity and sanity-lmv to only mount Lustre.
tappro [Thu, 2 Nov 2006 01:13:38 +0000 (01:13 +0000)]
- simplify handling of same fids issue during link. mdt_reint_link() returns
error immediately, mdd_link_sanity_check() check the source is dir at first,
therefore src cannot be the same as tgt.
tappro [Wed, 1 Nov 2006 23:20:28 +0000 (23:20 +0000)]
- make clo_split enum type
- fix is_subdir() -EREMOTE issues
- avoid usage of mdt_body mode field to store the result
nikita [Wed, 1 Nov 2006 23:02:56 +0000 (23:02 +0000)]
osd: remove unused fields.
nikita [Wed, 1 Nov 2006 23:02:00 +0000 (23:02 +0000)]
increase mdt thread pool size limit.
yury [Wed, 1 Nov 2006 19:44:13 +0000 (19:44 +0000)]
- fixes in split about using correct byte order;
- in cmm_split_try() init la_size used in split info messages;
- cleanups and debug in llite close thread related stuff;
- fixes in mdd about link() operation. Fixed deadlock for case when tgt and src is same object. Fixed lost error code after insert name which is possibly existing. Cleanups;
- in mdt_reint_open() use MDT_CROSS_LOCK for opened file in case it is located on remote MDT (by huanghua);
- in mdt_attr_set() fixed wrong goto to label out;
- cleanups in mdt_reint_link(), mdt_reint_unlink(), mdt_reint_rename() (by huanghua);
- fixed lock mode in md_reint_link(). It should LCK_EX instead of LCK_PW;
- fixed deadlock in mdt_reint_link() in case src and dst are same object;
- use MDT_CROSS_LOCK for src in mdt_reint_link() as it maybe located on remote MDT (by huanghua);
- used MDT_CROSS_LOCK for in mdt_reint_rename() as name may be located on remote MDT (by huanghua);
- fixed -ENODATA case in mdt_getxattr_pack_reply(). It should cause err_serious() with errors in console. Cleanups;
wangdi [Wed, 1 Nov 2006 16:40:02 +0000 (16:40 +0000)]
Branch b_new_cmd
check -EREMOTE before return
fanyong [Wed, 1 Nov 2006 09:39:50 +0000 (09:39 +0000)]
Do not call mdt_handle_idmap here, it will be called from mdt_req_handle again.
fanyong [Wed, 1 Nov 2006 04:05:41 +0000 (04:05 +0000)]
Do not permit remote reaml node to claim to be local.
lsy [Wed, 1 Nov 2006 03:48:44 +0000 (03:48 +0000)]
capa code cleanup.
wangdi [Wed, 1 Nov 2006 03:41:19 +0000 (03:41 +0000)]
Branch: b_new_cmd
Only when writing page succeed, then we check whether the inode should be
put the close list, since when page writing failed, it will be redirtied and
writing again later.
fanyong [Wed, 1 Nov 2006 02:33:20 +0000 (02:33 +0000)]
Add proc interface for distinguishing clinet type -- local / remote.
tappro [Tue, 31 Oct 2006 21:02:11 +0000 (21:02 +0000)]
- assert on wrong transno during replay
tappro [Tue, 31 Oct 2006 20:58:13 +0000 (20:58 +0000)]
remove extra debug messages
tappro [Tue, 31 Oct 2006 20:52:04 +0000 (20:52 +0000)]
take into account initial_conn value during reconnect
tappro [Tue, 31 Oct 2006 19:47:33 +0000 (19:47 +0000)]
split optimization to avoid often attr_get() calling
nikita [Tue, 31 Oct 2006 18:26:34 +0000 (18:26 +0000)]
iam: 0. add optional calls to schedule() to improve pdirops testing. 1. fix locking in split_index_node().
nikita [Tue, 31 Oct 2006 18:23:07 +0000 (18:23 +0000)]
osd: remove oi lock: it is no longer needed now that we have iam-pdirops.
yury [Tue, 31 Oct 2006 18:01:19 +0000 (18:01 +0000)]
- added dir size it was split on to message about split;
- in ll_queue_done_writing() replace assert by check-and-doing nothing if inode is already on close thread queue. Seems it maybe called two times for same inode. If that is not correct Vitaly will fix that later. Tests pass.
yury [Tue, 31 Oct 2006 15:28:16 +0000 (15:28 +0000)]
- fix in comment;
- use -EALRADY instead of -1 in ll_close_next_lli to point out that thread should be stopped.
yury [Tue, 31 Oct 2006 14:47:13 +0000 (14:47 +0000)]
- cleanups in mdt_lock_pdo_mode(), fix in error message;
- fixed typo in comment.
yury [Tue, 31 Oct 2006 12:53:28 +0000 (12:53 +0000)]
- fixed bug with possible concurrent split;
- comments, cleanups.
yury [Tue, 31 Oct 2006 11:21:11 +0000 (11:21 +0000)]
- added missed rc = 0 in mdt_getattr_name_lock() for resent case.
yury [Tue, 31 Oct 2006 10:07:14 +0000 (10:07 +0000)]
- protect export's held locks list by spin lock while adding there fixed up lock in mdt_intent_lock_replace();
- some cleanups.
fanyong [Tue, 31 Oct 2006 08:50:45 +0000 (08:50 +0000)]
Support flush specific user identity info.
fanyong [Tue, 31 Oct 2006 07:25:29 +0000 (07:25 +0000)]
(1) drop some comment.
(2) small fix for lock using.
(3) replace kernel flag with LUSTRE flag.
wangdi [Tue, 31 Oct 2006 06:03:00 +0000 (06:03 +0000)]
Branch:b_new_cmd
disable 27m in sanity.sh, since create_precreate has been increase,
so we need touch tons of file to fill the precreate cache, so disable
this test temporarily
ericm [Tue, 31 Oct 2006 00:58:40 +0000 (00:58 +0000)]
branch: b_new_cmd
minor message fix for gss server daemon.
yury [Mon, 30 Oct 2006 18:24:44 +0000 (18:24 +0000)]
- added assert in mdt_is_subdir(). It only should be called against local objects;
- cleanups in cmr_xxx methods. No need to use RETURN() without ENTRY, just return is ok;
- in mdt_reint_rename() use namelen.
nikita [Mon, 30 Oct 2006 17:36:13 +0000 (17:36 +0000)]
ldlm and its users: do not pass struct ldlm_res_id by value, pass a pointer instead. This reduces stack and cpu consumption. Also constify for a good measure. Also remove what looks like obsolete code in ldlm_cli_enqueue_local(): dead assignment to local variable.
nikita [Mon, 30 Oct 2006 17:30:32 +0000 (17:30 +0000)]
lmv_readpage(): suppress compiler warning for user-level compilation.
nikita [Mon, 30 Oct 2006 17:29:50 +0000 (17:29 +0000)]
0. increase OST_MIN_PRECREATE; 1. move fid_res_name_eq() to lustre_fid.h and use offset constants in it.
yury [Mon, 30 Oct 2006 16:19:24 +0000 (16:19 +0000)]
- use usual MDT name in mdd procfs stuff instead of mdd-0;
- make mdc/mdc_is_subdir() static to not confuse it with same name in cmm.
wangdi [Mon, 30 Oct 2006 15:44:13 +0000 (15:44 +0000)]
Branch:b_new_cmd
fix the len bug in cmm_split_entries, the original method has some problem
when entries are not full in the page
fanyong [Mon, 30 Oct 2006 10:31:28 +0000 (10:31 +0000)]
Small fix.
yury [Mon, 30 Oct 2006 10:21:05 +0000 (10:21 +0000)]
- fixed small typo in cmm_split_remove_page()
yury [Mon, 30 Oct 2006 10:07:45 +0000 (10:07 +0000)]
- handle empty page in cmm_split_remove_page() at Tom's suggestion;
- fixes about le_to_cpu in mdt_write_dir_page()
wangdi [Mon, 30 Oct 2006 09:54:23 +0000 (09:54 +0000)]
Branch: b_new_cmd
not root dir, it should be NOT expect splitted.
yury [Mon, 30 Oct 2006 09:01:43 +0000 (09:01 +0000)]
- cleanups in split - function names are changed to make split understandable not only for author;
- in cml_create() check if name came to correct MDT only if cmm_split_try() returned -EALREADY, that is dir is split. In other cases that is not needed and only slows things down;
- comments, allocating new fid and creating object for slaves is united to one function.
fanyong [Mon, 30 Oct 2006 03:27:22 +0000 (03:27 +0000)]
Since have called mdd_la_get before and held lock yet,
call __mdd_permission_internal directly.
fanyong [Mon, 30 Oct 2006 02:31:53 +0000 (02:31 +0000)]
Local client should not use "lfs getfacl / lfs setfacl".
yury [Sun, 29 Oct 2006 20:05:14 +0000 (20:05 +0000)]
- disable slow test
yury [Sun, 29 Oct 2006 19:48:28 +0000 (19:48 +0000)]
- more fixes in split, now 24o works and split is "counted" as bug free ;) at least I do not know any of them.
yury [Sun, 29 Oct 2006 19:03:49 +0000 (19:03 +0000)]
- fixes in split. Tom! Your changes in cmm_remove_entries() were wrong;
yury [Sun, 29 Oct 2006 14:56:53 +0000 (14:56 +0000)]
- removed debug stuff.
fanyong [Sun, 29 Oct 2006 14:19:19 +0000 (14:19 +0000)]
(1) "lfs getfacl" should return base file mode anyway.
(2) make "lfs setfacl" more like local client setfacl.
fanyong [Sun, 29 Oct 2006 13:12:19 +0000 (13:12 +0000)]
For remote client, return directly when lfs getfacl / lfs setfacl
if server do not support ACL.
fanyong [Sun, 29 Oct 2006 11:53:51 +0000 (11:53 +0000)]
Drop unnecessary warning message.
wangdi [Sun, 29 Oct 2006 11:07:02 +0000 (11:07 +0000)]
Branch:b_new_cmd
add group num when set attr to OST
wangdi [Sun, 29 Oct 2006 07:11:22 +0000 (07:11 +0000)]
Branch:b_new_cmd
cleanup in cmm split, after pdo added.
huanghua [Sat, 28 Oct 2006 15:37:19 +0000 (15:37 +0000)]
some code cleanup in split.
nikita [Sat, 28 Oct 2006 13:31:56 +0000 (13:31 +0000)]
procsleep patch: on statistics reset, allocate info if necessary.
yury [Sat, 28 Oct 2006 11:28:05 +0000 (11:28 +0000)]
- fixed ma_valid in cmm_mdsnum_check();
- fixes in cmm_try_to_split();
- added error messages to cmm_try_to_split();
- cache mdsnum for lmv slaves so that we need not to lookup them all the time. This reduces number of fld cache lookups from ~46000 to ~12000 on sanity 24o.
ericm [Sat, 28 Oct 2006 04:25:03 +0000 (04:25 +0000)]
branch: b_new_cmd
- replace spinlock with rwlock for policy table protection.
- fix concurrent load ptlrpc_gss module.
- minor message adjustment.
nikita [Fri, 27 Oct 2006 23:48:55 +0000 (23:48 +0000)]
kernel_patches: add statistics clearing for /proc/sleep.
nikita [Fri, 27 Oct 2006 23:48:18 +0000 (23:48 +0000)]
iam: add -EAGAIN handling for index split.
tappro [Fri, 27 Oct 2006 21:38:53 +0000 (21:38 +0000)]
- filter MDS_DONE_WRITING during recovery
yury [Fri, 27 Oct 2006 17:56:59 +0000 (17:56 +0000)]
- removed obsolete debug stuff.
huanghua [Fri, 27 Oct 2006 16:14:47 +0000 (16:14 +0000)]
more efficient unlinkmany
wangdi [Fri, 27 Oct 2006 10:22:50 +0000 (10:22 +0000)]
Branch:b_new_cmd
add debug info for b11123
huanghua [Fri, 27 Oct 2006 08:59:39 +0000 (08:59 +0000)]
Be sure not to take PDO lock for non-existing or remote directory.
yury [Fri, 27 Oct 2006 07:11:39 +0000 (07:11 +0000)]
- drop not directory dentries in ll_drop_dentry(). Let's see if it helps with OOM on clients.
wangdi [Fri, 27 Oct 2006 06:10:24 +0000 (06:10 +0000)]
Branch: b_new_cmd
add this ldlm debug info temporarily
wangdi [Fri, 27 Oct 2006 06:06:28 +0000 (06:06 +0000)]
Branch: b_new_cmd
add gr in these 2 places