Whamcloud - gitweb
fs/lustre-release.git
18 years agoenable ext3-extents on mds and for directories.
nikita [Tue, 12 Dec 2006 21:40:02 +0000 (21:40 +0000)]
enable ext3-extents on mds and for directories.

18 years agobranch: b_new_cmd
ericm [Tue, 12 Dec 2006 20:56:05 +0000 (20:56 +0000)]
branch: b_new_cmd
test_bit take bit number not the value.

18 years agobranch: b_new_cmd
ericm [Tue, 12 Dec 2006 18:41:25 +0000 (18:41 +0000)]
branch: b_new_cmd
sanity-lmv: it's unnecessary to obtain krb5 credential since runas id
is not used even.

18 years agoremove unused constants.
nikita [Tue, 12 Dec 2006 12:34:44 +0000 (12:34 +0000)]
remove unused constants.

18 years agoremove stacktrace debugging: we have got numbers after all.
nikita [Tue, 12 Dec 2006 12:34:21 +0000 (12:34 +0000)]
remove stacktrace debugging: we have got numbers after all.

18 years agoBranch:b_new_cmd
wangdi [Tue, 12 Dec 2006 09:18:39 +0000 (09:18 +0000)]
Branch:b_new_cmd
2 typo fixes

18 years agoBranch:b_new_cmd
wangdi [Tue, 12 Dec 2006 05:39:32 +0000 (05:39 +0000)]
Branch:b_new_cmd
clear error msg for cmd3-test18

18 years agobranch: b_new_cmd
ericm [Mon, 11 Dec 2006 20:36:22 +0000 (20:36 +0000)]
branch: b_new_cmd
fix return error value of checksum verification, the upper layer and
recovery depend on the exact error value to decide to reply error rpc.

18 years ago- more fixes in modifying led_held_locks, comment fix;
yury [Mon, 11 Dec 2006 18:03:13 +0000 (18:03 +0000)]
- more fixes in modifying led_held_locks, comment fix;
- more fixes in lov.

18 years ago- rewrite handle in close request
tappro [Mon, 11 Dec 2006 15:19:15 +0000 (15:19 +0000)]
- rewrite handle in close request

18 years agoOne of main bottlenecks on b_new_cmd md server is the necessity to keep
nikita [Sun, 10 Dec 2006 15:17:33 +0000 (15:17 +0000)]
One of main bottlenecks on b_new_cmd md server is the necessity to keep
additional data structures like object index (oi), and fids in directory
entries. To minimize overhead of this on-disk structures should be as compact
as possible.

For directory entries this was achieved by "compressing" fids, but this
approach cannot be applied to oi as is, because here fids are *keys* rather
than records, and iam doesn't support variable sized keys.

Instead of complicating iam with such support, this patch splits object index
into multiple indices with different key size. Currently two indices are
created:

 oi.16 --- with full sized 16 byte fid as a key, and

 oi.5  --- with compact 5 byte representation of "usual" fid as a key.

"Usual fid" is defined as a fid with

        fid_seq(fid) <  0xffffff
        fid_oid(fid) <    0xffff
        fid_ver(fid) ==        0

(which gives 5 bytes of data). It is expected that under usual workload
majority of fids fall into second category.

18 years ago- disable not used label in lmv;
yury [Sun, 10 Dec 2006 12:26:33 +0000 (12:26 +0000)]
- disable not used label in lmv;
- fixed two possible races in lov about checking lov_qos.lq_dirty_rr.

18 years ago- split limit was bumped - increase number of created files in test to cause split.
yury [Sun, 10 Dec 2006 10:41:24 +0000 (10:41 +0000)]
- split limit was bumped - increase number of created files in test to cause split.

18 years ago- yet smaller number of fld lookups in lmv in create paths. FLD lookups number is...
yury [Sun, 10 Dec 2006 10:37:31 +0000 (10:37 +0000)]
- yet smaller number of fld lookups in lmv in create paths. FLD lookups number is decreased by factor 1.25.

18 years ago- small fix with flags.
tappro [Sun, 10 Dec 2006 10:11:04 +0000 (10:11 +0000)]
- small fix with flags.

18 years ago-remove unused values
tappro [Sun, 10 Dec 2006 00:28:04 +0000 (00:28 +0000)]
-remove unused values

18 years ago- fix for -EEXIST issue for open|create during replay. In fact it is not error,
tappro [Sat, 9 Dec 2006 23:56:37 +0000 (23:56 +0000)]
- fix for -EEXIST issue for open|create during replay. In fact it is not error,
  because file can exists and open replay should finish normally. Rename
  mdt_mfd_open() to mdt_finish_open(). New mdt_mfd_open() contains only  mfd
  handling and used for open_replay.
- add transno value from reqmsg to the debug_req output for debug purposes.

18 years agomdd,osd,libiam,fid,lu,mkfs: store fids on disk as variable sized records.
nikita [Sat, 9 Dec 2006 20:51:32 +0000 (20:51 +0000)]
mdd,osd,libiam,fid,lu,mkfs: store fids on disk as variable sized records.

18 years agoiam: handle variable sized records.
nikita [Sat, 9 Dec 2006 20:47:47 +0000 (20:47 +0000)]
iam: handle variable sized records.

18 years ago- cleanups in ll_drop_dentry(), some other cleanups;
yury [Sat, 9 Dec 2006 10:56:07 +0000 (10:56 +0000)]
- cleanups in ll_drop_dentry(), some other cleanups;

- fix from 1_5 branch in ldlm_lock_cancel(). There was race possible when we ask client to return lock back and client does it by itself.

18 years ago2.6-rhel series: put kgdb patch back. Hopefully it will compile on x86_64 now.
nikita [Fri, 8 Dec 2006 12:04:45 +0000 (12:04 +0000)]
2.6-rhel series: put kgdb patch back. Hopefully it will compile on x86_64 now.

18 years agoremove BREAKPOINT() calls.
nikita [Fri, 8 Dec 2006 12:03:04 +0000 (12:03 +0000)]
remove BREAKPOINT() calls.

18 years agokgdb: add trivial include/asm-x86_64/kgdb.h
nikita [Fri, 8 Dec 2006 12:00:52 +0000 (12:00 +0000)]
kgdb: add trivial include/asm-x86_64/kgdb.h

18 years agoremove unused code. capa is not stored in nfs fh any more, since getattr don't check...
lsy [Fri, 8 Dec 2006 09:26:13 +0000 (09:26 +0000)]
remove unused code. capa is not stored in nfs fh any more, since getattr don't check it at all because remote object. see mds_getattr().

18 years agobranch: b_new_cmd
ericm [Fri, 8 Dec 2006 04:00:15 +0000 (04:00 +0000)]
branch: b_new_cmd
switch context immediately when we detect ctx losing on server side, to
avoid infinite loop of resending.

18 years agoupdate driver for sk98 Ethernet controller to enabale more nodes in CNBuild.
huanghua [Fri, 8 Dec 2006 03:37:27 +0000 (03:37 +0000)]
update driver for sk98 Ethernet controller to enabale more nodes in CNBuild.
If we do not use this kind of sk98, no need to apply this patch.

18 years agoRemove the kgdb patch because on x86_64 the gcc command includes the -m32
brian [Fri, 8 Dec 2006 01:21:20 +0000 (01:21 +0000)]
Remove the kgdb patch because on x86_64 the gcc command includes the -m32
flag which tells gcc to define __i386__ and that blows up when asm/kgdb.h
cannot be found.  That exists in asm-i386 but not asm-x86_64.
When/if somebody can figure out why our kernel build is including the
-m32 gcc flag and fixes it we can re-enable the kgdb patch.  This is prob-
ably due to some processor option in the kernel config file.
Is anyone using kgdb with this kernel on this branch anyway?  Maybe devs
in local testing situations?

18 years agoseries file for x86_64 based HP orion cluster.
nikita [Thu, 7 Dec 2006 22:18:29 +0000 (22:18 +0000)]
series file for x86_64 based HP orion cluster.

18 years agobranch: b_new_cmd
ericm [Thu, 7 Dec 2006 20:57:11 +0000 (20:57 +0000)]
branch: b_new_cmd
bump version to 1.9.50

18 years agoproc-sleep: fix possible interrupt deadlock.
nikita [Thu, 7 Dec 2006 20:13:59 +0000 (20:13 +0000)]
proc-sleep: fix possible interrupt deadlock.

18 years agomdt: update to match tmp-debug.
nikita [Thu, 7 Dec 2006 16:00:24 +0000 (16:00 +0000)]
mdt: update to match tmp-debug.

18 years agotmp-debug: use stacktrace_record() instead of dunmping onto console.
nikita [Thu, 7 Dec 2006 15:59:44 +0000 (15:59 +0000)]
tmp-debug: use stacktrace_record() instead of dunmping onto console.

18 years agoproc-sleep: add stacktrace_record() function to record callpaths. Visible at /proc...
nikita [Thu, 7 Dec 2006 15:59:00 +0000 (15:59 +0000)]
proc-sleep: add stacktrace_record() function to record callpaths. Visible at /proc/stacktrace.

18 years agofix wrong print format
tappro [Thu, 7 Dec 2006 15:55:55 +0000 (15:55 +0000)]
fix wrong print format

18 years agomdt: use tmp-debug patch.
nikita [Wed, 6 Dec 2006 18:57:23 +0000 (18:57 +0000)]
mdt: use tmp-debug patch.

18 years agotmp-debug: initial jiffies value maybe non-0.
nikita [Wed, 6 Dec 2006 18:56:51 +0000 (18:56 +0000)]
tmp-debug: initial jiffies value maybe non-0.

18 years agoBranch:b_new_cmd
wangdi [Wed, 6 Dec 2006 07:54:43 +0000 (07:54 +0000)]
Branch:b_new_cmd
choose mds by nid instead by name to avoid mds-mds recovery temporarily for debuging 11/17.
currently, the parent is locked before mds-mds rpc, which may caused some problem when do
mds-mds recovery.

18 years agodebug for recovery
tappro [Wed, 6 Dec 2006 00:46:25 +0000 (00:46 +0000)]
debug for recovery

18 years agoadd temporary kernel debugging to track where __find_get_block() and try_to_wake_up...
nikita [Tue, 5 Dec 2006 22:04:59 +0000 (22:04 +0000)]
add temporary kernel debugging to track where __find_get_block() and try_to_wake_up() are called from. Spams console, beware.

18 years agoiam: change number of bits used by TEA portion of sandwich hash.
nikita [Tue, 5 Dec 2006 22:03:24 +0000 (22:03 +0000)]
iam: change number of bits used by TEA portion of sandwich hash.

18 years ago- small debug and cleanups
tappro [Tue, 5 Dec 2006 16:34:46 +0000 (16:34 +0000)]
- small debug and cleanups

18 years ago- set server_timeout for mds-mds import
tappro [Tue, 5 Dec 2006 16:31:54 +0000 (16:31 +0000)]
- set server_timeout for mds-mds import

18 years agoPort from b1_5, cleanup the client import manually if no connect to mdt/ost.
fanyong [Tue, 5 Dec 2006 15:30:38 +0000 (15:30 +0000)]
Port from b1_5, cleanup the client import manually if no connect to mdt/ost.
(found when client mount failed for permission)

18 years agobranch: b_new_cmd
ericm [Tue, 5 Dec 2006 03:52:27 +0000 (03:52 +0000)]
branch: b_new_cmd
port from b1_8_gss: support MIT Kerberos algorithm arcfour-hmac-md5.

18 years agolmv: add a bit of debugging output.
nikita [Tue, 5 Dec 2006 00:06:14 +0000 (00:06 +0000)]
lmv: add a bit of debugging output.

18 years agoBranch:b_new_cmd
wangdi [Mon, 4 Dec 2006 09:03:57 +0000 (09:03 +0000)]
Branch:b_new_cmd
Disable target_stop_recovery_thread temporarily, because it
might cause dead-lock of target_recovery_thread. For example
1) in recovery, if clients are still not reconnected to the MDS.
2) this MDS failed again, try to stop_recovery_thread here.  obd_abort_recovery is set to 1.
3) MDS disconnect stale exports, obd_abort_recovery will be reset to 0. and check the next
   replay, but currently the MDS already disconnect the exports. and next replay req will
   never reached.  so the recovery-thread will hang there.

18 years agoBranch:b_new_cmd
wangdi [Mon, 4 Dec 2006 08:30:03 +0000 (08:30 +0000)]
Branch:b_new_cmd
txn_stop_cb should be clear after recovery thread stop

18 years agoBranch:b_new_cmd
wangdi [Sat, 2 Dec 2006 05:57:03 +0000 (05:57 +0000)]
Branch:b_new_cmd
add debug_info for debuging LASSERT when test17

18 years agobranch: b_new_cmd
ericm [Sat, 2 Dec 2006 01:07:04 +0000 (01:07 +0000)]
branch: b_new_cmd
backport from b1_8_gss:
fix gss-krb5 lucid context export format.

18 years agofile loadgen.c was added on branch b_new_cmd on 2007-02-08 09:58:50 +0000
huanghua [Fri, 1 Dec 2006 23:58:25 +0000 (23:58 +0000)]
file loadgen.c was added on branch b_new_cmd on 2007-02-08 09:58:50 +0000

18 years agoBranch:b_new_cmd
wangdi [Fri, 1 Dec 2006 07:40:45 +0000 (07:40 +0000)]
Branch:b_new_cmd
Init cat log for mds when add ost in recoverying, to make sure all unlink log can
be initialized after config log is processed. This could avoid race beween unlink
and llog init postrecovery, which should be the reason of b11218.
b:11218

18 years agotry smaller bh lru size.
nikita [Thu, 30 Nov 2006 21:11:19 +0000 (21:11 +0000)]
try smaller bh lru size.

18 years ago- call md_clear_open_replay_data() right before freeing och instead of doing
tappro [Thu, 30 Nov 2006 14:15:22 +0000 (14:15 +0000)]
- call md_clear_open_replay_data() right before freeing och instead of doing
  that before ll_done_writing(). This fix prevents possible handle loss when
replay happens between open/close and done_writting.

18 years agosabdwich hash.
nikita [Thu, 30 Nov 2006 10:27:46 +0000 (10:27 +0000)]
sabdwich hash.

18 years agoanother return value fix.
nikita [Thu, 30 Nov 2006 10:18:53 +0000 (10:18 +0000)]
another return value fix.

18 years ago2.6-rhel series: increase BH_LRU_SIZE.
nikita [Wed, 29 Nov 2006 21:22:39 +0000 (21:22 +0000)]
2.6-rhel series: increase BH_LRU_SIZE.

18 years agolog_do_checkpoint(): fix return value.
nikita [Wed, 29 Nov 2006 19:09:54 +0000 (19:09 +0000)]
log_do_checkpoint(): fix return value.

18 years agoremove obsolete debugging.
nikita [Wed, 29 Nov 2006 19:09:12 +0000 (19:09 +0000)]
remove obsolete debugging.

18 years agoFix BUG for mount permission:
fanyong [Wed, 29 Nov 2006 13:57:54 +0000 (13:57 +0000)]
Fix BUG for mount permission:
(1) if client failed to identification, it should not reconnect endlessly.
(2) client should check the connection error and return the value correctly.

18 years agofix the create mode bug
huanghua [Wed, 29 Nov 2006 09:59:34 +0000 (09:59 +0000)]
fix the create mode bug

18 years agoBranch:b_new_cmd
wangdi [Wed, 29 Nov 2006 09:44:34 +0000 (09:44 +0000)]
Branch:b_new_cmd
add evict_client proc for mdt

18 years ago- bug in lustre_msg_clear_flags(). For version 2 it clears ALL flags instead of
tappro [Tue, 28 Nov 2006 22:03:22 +0000 (22:03 +0000)]
- bug in lustre_msg_clear_flags(). For version 2 it clears ALL flags instead of
  passed as parameter. b1_5 is affected too.

18 years agobranch: b_new_cmd
ericm [Tue, 28 Nov 2006 21:23:23 +0000 (21:23 +0000)]
branch: b_new_cmd
port from b1_8_gss: add a gss sanity test for enlarge_reqbuf().

18 years agofix for LASSERT(*start <= hash) bug.
nikita [Tue, 28 Nov 2006 21:07:29 +0000 (21:07 +0000)]
fix for LASSERT(*start <= hash) bug.

18 years agofix a memleak @lustre_md->mea in ll_prep_inode().
huanghua [Tue, 28 Nov 2006 15:24:20 +0000 (15:24 +0000)]
fix a memleak @lustre_md->mea in ll_prep_inode().

18 years ago(1) Fix setxid permission check order according to the upcall result packed order.
fanyong [Tue, 28 Nov 2006 13:13:48 +0000 (13:13 +0000)]
(1) Fix setxid permission check order according to the upcall result packed order.
(2) More detail comment.

18 years ago(1)Add more rules for setxid.conf, permit to enable/disable setxid permission explicitly.
fanyong [Tue, 28 Nov 2006 09:39:47 +0000 (09:39 +0000)]
(1)Add more rules for setxid.conf, permit to enable/disable setxid permission explicitly.
(2)Fix bug in cmd3-53 for local client setgrp permission out-of-control.
(3)More detail comment.

18 years ago- make split size 128K to ensure that no split happens in recovery but still happens...
yury [Tue, 28 Nov 2006 07:58:05 +0000 (07:58 +0000)]
- make split size 128K to ensure that no split happens in recovery but still happens in perf. tests.

18 years ago- fix recovery issue with SOM. The setattr/done_writting use mfd on server but
tappro [Mon, 27 Nov 2006 19:46:51 +0000 (19:46 +0000)]
- fix recovery issue with SOM. The setattr/done_writting use mfd on server but
  can't find the correct one during replay because mfd handle is changed.
Current fix add old_handle field in mfd so during replay it is possible to find
it comparing old handle.

18 years agoBranch b_new_cmd
vitaly [Mon, 27 Nov 2006 17:44:01 +0000 (17:44 +0000)]
Branch b_new_cmd
b=22564

cancel all the locks on umount before shutting down the close thread
to flush all the dirty cache and send needed Size-on-MDS updates on MDS.

18 years agoBranch b_new_cmd
vitaly [Mon, 27 Nov 2006 14:48:53 +0000 (14:48 +0000)]
Branch b_new_cmd
b=22564,11103

check if unhashed and rehash under the dcache lock.
print epoch to be closed properly.

18 years agoBranch:b_new_cmd
wangdi [Mon, 27 Nov 2006 08:58:05 +0000 (08:58 +0000)]
Branch:b_new_cmd
remove useless debug info after 11123 is fixed

18 years agoFix test script for local acl operation.
fanyong [Mon, 27 Nov 2006 05:40:09 +0000 (05:40 +0000)]
Fix test script for local acl operation.

18 years agoBranch:b_new_cmd
wangdi [Sun, 26 Nov 2006 04:41:20 +0000 (04:41 +0000)]
Branch:b_new_cmd
if ost failed is detected in lfs find, only return 0 to make tranverse continue
for cmd3-test18

18 years ago- some cleanups in lmv.
yury [Sat, 25 Nov 2006 13:08:44 +0000 (13:08 +0000)]
- some cleanups in lmv.

18 years agobranch: b_new_cmd
ericm [Fri, 24 Nov 2006 18:15:11 +0000 (18:15 +0000)]
branch: b_new_cmd
the part missing from last commit: don't taint passed-in obj until we
know everything is fine.

18 years agolmv: 0. more fid correctness checks, 1. fix wrong mea_hash_segment calculation.
nikita [Fri, 24 Nov 2006 17:27:15 +0000 (17:27 +0000)]
lmv: 0. more fid correctness checks, 1. fix wrong mea_hash_segment calculation.

18 years agothis fix a bug in Lustre over NFS. But this flag is temparory as the patch said.
huanghua [Fri, 24 Nov 2006 13:48:29 +0000 (13:48 +0000)]
this fix a bug in Lustre over NFS. But this flag is temparory as the patch said.

18 years ago- cleanups in lmv:
yury [Fri, 24 Nov 2006 13:29:37 +0000 (13:29 +0000)]
- cleanups in lmv:
  - removed not needed @body initializatiion;
  - added assert that body field in reply should be swabbed in mdc before using it in lmv.

18 years agoBranch: b_new_cmd
wangdi [Fri, 24 Nov 2006 12:39:18 +0000 (12:39 +0000)]
Branch: b_new_cmd
We should try to retrieve the MEA from MDS anyway in getattr

18 years agoAdd ASSERT for filename validity check.
fanyong [Fri, 24 Nov 2006 09:49:29 +0000 (09:49 +0000)]
Add ASSERT for filename validity check.

18 years agobranch: b_new_cmd
ericm [Fri, 24 Nov 2006 04:36:51 +0000 (04:36 +0000)]
branch: b_new_cmd
back port from b1_8_gss:
 - don't taint passed-in obj until we know everything is fine.
 - multi-process client gss daemon, to prevent one unresponsive server
   block other context negotiation. nfs-utils patch not updated though.
 - fix regression of exporting context of des3-hmac-sha1 on old kerberos.

18 years agoosd: disable back expensive fid_is_local() assertion.
nikita [Thu, 23 Nov 2006 18:45:22 +0000 (18:45 +0000)]
osd: disable back expensive fid_is_local() assertion.

18 years agolmv/iam: new "scalable" readdir.
nikita [Thu, 23 Nov 2006 18:44:45 +0000 (18:44 +0000)]
lmv/iam: new "scalable" readdir.

18 years agofid: factor common function out.
nikita [Thu, 23 Nov 2006 18:06:59 +0000 (18:06 +0000)]
fid: factor common function out.

18 years agoliblustre: add hash_long() from linux/hash.c.
nikita [Thu, 23 Nov 2006 18:05:40 +0000 (18:05 +0000)]
liblustre: add hash_long() from linux/hash.c.

18 years agoDo NOT count the terminating '\0' of filename for length.
fanyong [Thu, 23 Nov 2006 14:01:52 +0000 (14:01 +0000)]
Do NOT count the terminating '\0' of filename for length.

18 years agoremote operation should set MDS_CROSS_REF flag.
huanghua [Thu, 23 Nov 2006 10:09:12 +0000 (10:09 +0000)]
remote operation should set MDS_CROSS_REF flag.

18 years agoMore sanity check for filename length.
fanyong [Thu, 23 Nov 2006 06:39:59 +0000 (06:39 +0000)]
More sanity check for filename length.

18 years agofix a typo and clear out the fid.
huanghua [Thu, 23 Nov 2006 05:17:28 +0000 (05:17 +0000)]
fix a typo and clear out the fid.

18 years agomore precise debugging code.
huanghua [Thu, 23 Nov 2006 02:20:44 +0000 (02:20 +0000)]
more precise debugging code.

18 years agoBranch: b_new_cmd
wangdi [Wed, 22 Nov 2006 16:34:15 +0000 (16:34 +0000)]
Branch: b_new_cmd
miss kunmap for those reset offset page. found by umka

18 years agoMore sanity check for nlink of file.
fanyong [Wed, 22 Nov 2006 14:55:39 +0000 (14:55 +0000)]
More sanity check for nlink of file.

18 years agojbd-checkpoint-on-commit: fix locking.
nikita [Wed, 22 Nov 2006 14:43:48 +0000 (14:43 +0000)]
jbd-checkpoint-on-commit: fix locking.

18 years ago- fix recovery issue. dropping obd_recovering flag before notify() cause loosing
tappro [Wed, 22 Nov 2006 13:38:51 +0000 (13:38 +0000)]
- fix recovery issue. dropping obd_recovering flag before notify() cause loosing
  the objects on OST due to orphan destroying. The obd_recovering should be
  zeroed after notify()

18 years ago- make sure there is transno for close/done_write() to keep them in replay queue
tappro [Wed, 22 Nov 2006 13:36:06 +0000 (13:36 +0000)]
- make sure there is transno for close/done_write() to keep them in replay queue

18 years ago- count -EAGAIN during close() as valid answer
tappro [Wed, 22 Nov 2006 13:22:55 +0000 (13:22 +0000)]
- count -EAGAIN during close() as valid answer

18 years ago- fix objid update call. It should be done before transaction stop. Separate it
tappro [Wed, 22 Nov 2006 13:17:03 +0000 (13:17 +0000)]
- fix objid update call. It should be done before transaction stop. Separate it
  from freeing lov_create data.
- encapsulate checking no_lov_create flag inside mdd_lov_create

18 years agoadd debug code to catch b11241
huanghua [Wed, 22 Nov 2006 13:06:18 +0000 (13:06 +0000)]
add debug code to catch b11241

18 years agolet export Lustre over NFS work.
huanghua [Wed, 22 Nov 2006 07:37:30 +0000 (07:37 +0000)]
let export Lustre over NFS work.
basic test works now. More tests needed.