Whamcloud - gitweb
adilger [Thu, 5 Aug 2004 23:55:53 +0000 (23:55 +0000)]
Remove global atomic counter for MDS opens, it just slows things down.
adilger [Thu, 5 Aug 2004 23:19:53 +0000 (23:19 +0000)]
Print client recovery info to the console for debugging.
Better locking arount mds_obd_create(), fix error path.
adilger [Thu, 5 Aug 2004 23:12:23 +0000 (23:12 +0000)]
Defensive check to avoid re-freeing export during force cleanup.
adilger [Thu, 5 Aug 2004 23:11:16 +0000 (23:11 +0000)]
Remove duplicate but unused declaration of mds_pack_open_flags().
adilger [Thu, 5 Aug 2004 23:07:30 +0000 (23:07 +0000)]
More verbose debugging for llog replay assertion.
b=4033
alex [Wed, 4 Aug 2004 14:41:38 +0000 (14:41 +0000)]
- we need not that CWARN()s
ericm [Tue, 3 Aug 2004 00:52:49 +0000 (00:52 +0000)]
file lutil.h was initially added on branch b_newsysio.
ericm [Tue, 3 Aug 2004 00:52:47 +0000 (00:52 +0000)]
file lutil.c was initially added on branch b_newsysio.
alex [Mon, 2 Aug 2004 23:39:02 +0000 (23:39 +0000)]
b=4059
- race in lmv_check_connect() has been fixed: few threads might try to
establish LMV's connects to MDCs and some of them might find not-yet-
-connected LMV operational
alex [Mon, 2 Aug 2004 22:18:37 +0000 (22:18 +0000)]
b=4030
r=adilger
- new obd method ->o_connect_post() is called when reverse import
gets initialized
- method ->o_llog_connect() gets obd_export, not obd_device
- MDS passes mdsnum in connect_flags slot
- OST saves last known group in LAST_GROUP file
- OST initializes all known groups upon startup, not on demand
- filter_connect_post() initializes context for given group upon
connection from MDS
- filter_connect() initializes group from connect_flags
- osc_set_info() doesn't forward "mds_conn": no need to do this any more
wangdi [Mon, 2 Aug 2004 14:58:13 +0000 (14:58 +0000)]
1)reorganize the smfs hook ops to make smfs walk a list of hooks ops in hook macro
2)remove clonefs in llite, we do not need them now
3)add snap_dot_info in snap_super_info
alex [Mon, 2 Aug 2004 14:39:03 +0000 (14:39 +0000)]
b=4054
- in RESENT case mds_getattr_name() doesn't take dentry for parent.
so, we have no mea to check mds validness
alex [Sat, 31 Jul 2004 14:56:29 +0000 (14:56 +0000)]
- comment fixed
alex [Sat, 31 Jul 2004 14:49:46 +0000 (14:49 +0000)]
b=4051
r=adilger
- lock inversion between mds_reint_unlink() and mdt_obj_create() fixed:
mdt_obj_create() unlinks object, so it doesn't matter what directory
to create objects in
alex [Sat, 31 Jul 2004 14:43:41 +0000 (14:43 +0000)]
b=4049
r=adilger
- fix lock inversion by removing DLM lock acquiring in mdt_obj_create()
instead just mark request difficult explicitly
alex [Sat, 31 Jul 2004 14:35:46 +0000 (14:35 +0000)]
b=4044
r=adilger
- few clients connecting simultaneously might race in mds_lmv_connect().
just take a sem around code that connects to LMV
alex [Sat, 31 Jul 2004 14:31:03 +0000 (14:31 +0000)]
r=adilger
- possible name collision in mdt_obj_create() fixed
adilger [Fri, 30 Jul 2004 21:53:55 +0000 (21:53 +0000)]
Fix error handling.
adilger [Fri, 30 Jul 2004 21:53:14 +0000 (21:53 +0000)]
Grab a lov llog lock so we ensure the llog records for a single transno
are allocated in the same order they will be replayed in. Concurrent
unlinks might otherwise have misordered llog records and not be replayable.
b=4033
r=andreas (for Tian)
alex [Fri, 30 Jul 2004 20:20:25 +0000 (20:20 +0000)]
- ext3_check_dir_entry shouldn't panic if inum in cross-node dentry
larger than local filesystem may have
rread [Fri, 30 Jul 2004 18:08:56 +0000 (18:08 +0000)]
b=3825
typo
rread [Thu, 29 Jul 2004 20:29:03 +0000 (20:29 +0000)]
Make recovery on clients less verbose and misleading.
rread [Thu, 29 Jul 2004 20:23:26 +0000 (20:23 +0000)]
b=3554
Use cond_resched() ll_pgcache_remove_extent, and make sure we've got
the definition for it.
alex [Thu, 29 Jul 2004 09:07:47 +0000 (09:07 +0000)]
b=3986
r=adilger
- we can't initialize few groups in parallel because
filter_read_group_internal() reallocates bunch of
global-per-obd structures. so, here is simple patch
to serialize it. indeed, this isn's very scalable
to hundreds of MDSs, but I think it's enough for a
while. tested on mountain and my home cluster
alex [Wed, 28 Jul 2004 21:25:47 +0000 (21:25 +0000)]
- just to avoid gcc warning
alex [Wed, 28 Jul 2004 21:22:30 +0000 (21:22 +0000)]
b=3825
r=robert
landed on HEAD
alex [Wed, 28 Jul 2004 19:17:32 +0000 (19:17 +0000)]
b=4019
r=adilger
- obd_connect() gets one more arg: connect flags
- client_connect_import() puts this flag in imp_connect_flags
- ptlrpc_connect_import() sends this flag to peer
- mds_lmv_connect() passes OBD_OPT_MDS_CONNECTION when connects to MDS
- mds_connect() looks for OBD_OPT_MDS_CONNECTION and treats
connection with this flag as MDS-to-MDS and doesn't try
to connect to own LMV
this way connection to LMV gets delayed and MDS reconnects to LMV
after recovery abort
eeb [Wed, 28 Jul 2004 18:09:53 +0000 (18:09 +0000)]
file small_scatterlist-2.4.21-chaos.patch was initially added on branch b1_4_eeb_perf.
eeb [Wed, 28 Jul 2004 18:09:52 +0000 (18:09 +0000)]
file blkdev_tunables-2.4.21-chaos.patch was initially added on branch b1_4_eeb_perf.
alex [Tue, 27 Jul 2004 15:37:43 +0000 (15:37 +0000)]
- lmv.sh config generator may add several OST nodes to setup as specified
in OSTCOUNT environment variable
alex [Tue, 27 Jul 2004 15:35:06 +0000 (15:35 +0000)]
b=3999
- IT_UNLINK doesn't use 4rd buffer
alex [Tue, 27 Jul 2004 12:38:12 +0000 (12:38 +0000)]
b=3550
r=adilger
- replay of unlink. landed for testing on china cluster
alex [Tue, 27 Jul 2004 12:32:15 +0000 (12:32 +0000)]
- don't show error message in replay case
alex [Tue, 27 Jul 2004 12:17:06 +0000 (12:17 +0000)]
b=3995
- mdc_enqueue() saved wrong easize for IT_OPEN replay
alex [Mon, 26 Jul 2004 22:50:10 +0000 (22:50 +0000)]
b=3983
brown paper bug have been fixed
rread [Mon, 26 Jul 2004 16:20:25 +0000 (16:20 +0000)]
b=3550
Preemptive LASSERT to make sure replay creates the correct
inode here.
wangdi [Mon, 26 Jul 2004 13:03:28 +0000 (13:03 +0000)]
Update snapfs: 1) some fix on clonefs read in lustre
alex [Mon, 26 Jul 2004 11:20:43 +0000 (11:20 +0000)]
b=3983
- adds a bit more balanced hash that takes all chars into account
- stores hash type in mea
- converts old non-marked mea to new one with magic
alex [Mon, 26 Jul 2004 11:14:15 +0000 (11:14 +0000)]
- more info in test 1c
alex [Mon, 26 Jul 2004 10:59:34 +0000 (10:59 +0000)]
- mds_reint_unlink() shouldn't kill slaves if master directory can't be unlinked
alex [Sat, 24 Jul 2004 12:10:35 +0000 (12:10 +0000)]
b=3874
- this BKL collides with jbd a lot. remove it.
alex [Sat, 24 Jul 2004 12:06:50 +0000 (12:06 +0000)]
b=3874
- avoid needless fsfilt_get_md() and taking i_sem on a non-splittable parent
rread [Fri, 23 Jul 2004 18:06:05 +0000 (18:06 +0000)]
b=3633,3934
debugging
rread [Thu, 22 Jul 2004 19:45:34 +0000 (19:45 +0000)]
b=3899
Handle server recovery in a new thread, instead of tying up request
processing threads. Centralizes the recovery management, including the
abort recovery checking.
rread [Thu, 22 Jul 2004 19:33:03 +0000 (19:33 +0000)]
b=3934
Add some more debugging.
rread [Thu, 22 Jul 2004 17:43:21 +0000 (17:43 +0000)]
b=3892
Move the transno update so it's done after we check for the -ENOTCONN
reply.
rread [Thu, 22 Jul 2004 16:30:26 +0000 (16:30 +0000)]
b=2525
I've seen the "server went back in time" message while testing HEAD,
so this will fix that. Patch already on 1.4.
alex [Thu, 22 Jul 2004 15:27:42 +0000 (15:27 +0000)]
b=3920
bits from 3599 landed to help debugging
alex [Thu, 22 Jul 2004 15:24:25 +0000 (15:24 +0000)]
- PTLDEBUG var to control debug set
alex [Thu, 22 Jul 2004 09:19:49 +0000 (09:19 +0000)]
b=2185
- ext3-inode-reuse is part of vanilla-2.4.24 series
- MDS doesn't take lock to prevent inode recreation
(this improves performance a bit)
alex [Thu, 22 Jul 2004 09:08:10 +0000 (09:08 +0000)]
- ost_brw_write() should return an error rather than uninitialized npages
rread [Thu, 22 Jul 2004 06:36:02 +0000 (06:36 +0000)]
b=3940
Corrected. OBD_MD_MDS should only be set when the directory was
created on a different node.
wangchao [Thu, 22 Jul 2004 06:07:20 +0000 (06:07 +0000)]
b=3362
r=braam
update lctl.lyx
wangchao [Thu, 22 Jul 2004 06:04:07 +0000 (06:04 +0000)]
b=3399
r=braam
update lmc.lyx and lconf.lyx
rread [Thu, 22 Jul 2004 06:02:01 +0000 (06:02 +0000)]
b=3940
revert while i figure out what i did wrong.
rread [Thu, 22 Jul 2004 05:25:49 +0000 (05:25 +0000)]
b=3940
Also reconstruct what we can in the body, which is hopefully enough.
wangdi [Wed, 21 Jul 2004 18:32:29 +0000 (18:32 +0000)]
add uml_clone.sh for clonefs test
wangdi [Wed, 21 Jul 2004 18:28:32 +0000 (18:28 +0000)]
1)add clonefs read in lustre
2)add clientfsoption in lmc and lconf
3)some fix in cobd for cache miss handler
jacob [Wed, 21 Jul 2004 00:32:57 +0000 (00:32 +0000)]
file lustrefs was initially added on branch b1_2_smallfix.
jacob [Wed, 21 Jul 2004 00:26:04 +0000 (00:26 +0000)]
file rhel-2.4.target was initially added on branch b1_2_smallfix.
jacob [Wed, 21 Jul 2004 00:26:03 +0000 (00:26 +0000)]
file x86-fpu-crash.patch was initially added on branch b1_2_smallfix.
jacob [Wed, 21 Jul 2004 00:26:02 +0000 (00:26 +0000)]
file qlogic-suse-2.4.21-2.patch was initially added on branch b1_2_smallfix.
jacob [Wed, 21 Jul 2004 00:26:01 +0000 (00:26 +0000)]
file kernel-2.4.21-rhel-2.4-i686.config was initially added on branch b1_2_smallfix.
jacob [Wed, 21 Jul 2004 00:26:00 +0000 (00:26 +0000)]
file kernel-2.4.21-rhel-2.4-i686-smp.config was initially added on branch b1_2_smallfix.
alex [Tue, 20 Jul 2004 18:01:05 +0000 (18:01 +0000)]
b=3896
oscc_internal_create() increase pre-creation window smoothly
rread [Tue, 20 Jul 2004 16:55:07 +0000 (16:55 +0000)]
b=3869
r=bzzz
Reconstruct create reply for CROSS_REF directories from the dentry.
rread [Tue, 20 Jul 2004 15:44:20 +0000 (15:44 +0000)]
b=3869
use the correct bug#
rread [Tue, 20 Jul 2004 06:29:01 +0000 (06:29 +0000)]
b=3892
more diagnostics.
rread [Tue, 20 Jul 2004 04:54:12 +0000 (04:54 +0000)]
b=3488
more debugging
rread [Mon, 19 Jul 2004 22:55:28 +0000 (22:55 +0000)]
b=3887
renmae 2nd ptlrcd "ptlrpcd-recov"
rread [Mon, 19 Jul 2004 22:49:58 +0000 (22:49 +0000)]
b=3886
A small races exists between receiving a reply and timeout. This diff
prevents ptlrpc_expire_one_request from timing out a req that has
already received the reply.
rread [Mon, 19 Jul 2004 22:38:40 +0000 (22:38 +0000)]
b=3885
THe hack on b1_2 branch to call set_info("mds_conn") on a specific OSC
instead of all of them was removed on HEAD, and this broke
mds_lov_synchronize. Instead of introducing a new hack to do this,
for now this changes m_l_s to send the set_info() to all the connected
OSTs. This should be harmless.
alex [Mon, 19 Jul 2004 11:57:30 +0000 (11:57 +0000)]
b=3874
- with CMD each MDS has own group on OST, so group != 0 always. this is
why filter_should_precreate() did return 1 always, so precreation path
didn't take a place. I've comment out check for group out for a while
alex [Mon, 19 Jul 2004 09:45:03 +0000 (09:45 +0000)]
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
alex [Sun, 18 Jul 2004 12:32:35 +0000 (12:32 +0000)]
- minor fixes to make the patch compatible with quadrics patches
alex [Sun, 18 Jul 2004 12:29:24 +0000 (12:29 +0000)]
- more fixes
alex [Sun, 18 Jul 2004 12:21:22 +0000 (12:21 +0000)]
- minor fixes to the patch. previous version couldn't be popped cleanly
alex [Sun, 18 Jul 2004 08:36:01 +0000 (08:36 +0000)]
- teached to catch double-free messy
alex [Sun, 18 Jul 2004 08:24:45 +0000 (08:24 +0000)]
b=3772
- race in dynlock_unlock() fixed: several threads may find refcount=0 and
try to free lock member at the same time. thanks to Andreas!
alex [Sun, 18 Jul 2004 05:41:57 +0000 (05:41 +0000)]
- if MDS aborts recovery it disconnects all the exports and lmv_disconnect()
is called. the last one didn't drop lmv->connected to 0 and this prevent
subsequent lmv_connect() to work properly
rread [Sun, 18 Jul 2004 00:29:37 +0000 (00:29 +0000)]
b=3869,1742
These are the remaining fixes from 1742 that are needed to allow a
client to reconnect during recover. The request that triggered 3869
on the last run was actually a RESENT request that was sent before
recovery had finished, which this patch will fix.
wangdi [Sat, 17 Jul 2004 15:26:39 +0000 (15:26 +0000)]
1 add snap read in fsfilt_getpage
2 some fix for regular file cow
alex [Sat, 17 Jul 2004 12:39:41 +0000 (12:39 +0000)]
b=3869
- more debug info
rread [Sat, 17 Jul 2004 00:57:13 +0000 (00:57 +0000)]
b=1451
Don't allow client to reconnect if an RPC is already in progress for
that client. Added an extra counter for this, because the export
refcount is also held by locks and who knows what else.
green [Fri, 16 Jul 2004 18:57:47 +0000 (18:57 +0000)]
file uml-2.6.7-01-bb2.patch was initially added on branch b1_4_smallfix.
wangdi [Fri, 16 Jul 2004 17:41:10 +0000 (17:41 +0000)]
some fix for obdfilter cow and snap ext3 patch
alex [Fri, 16 Jul 2004 17:08:04 +0000 (17:08 +0000)]
- vmlist needs to be exported for modules support
alex [Fri, 16 Jul 2004 12:40:54 +0000 (12:40 +0000)]
- slab-use-after-free debug tool added to vanilla-2.4.24 series to debug 3772
- dynlocks use own slab: slab-use-after-free may be enabled for only dynlocks
alex [Fri, 16 Jul 2004 08:07:25 +0000 (08:07 +0000)]
b=3772
- more debug checks and printks
rread [Fri, 16 Jul 2004 00:27:46 +0000 (00:27 +0000)]
b=3843
r=bzzz
Initialize req before passing it to down.
wangdi [Thu, 15 Jul 2004 17:59:18 +0000 (17:59 +0000)]
1)add more snap cow hook for dir ops.
2)add snap cow for obdfilter
3)some fix in smfs_cow
alex [Thu, 15 Jul 2004 14:53:37 +0000 (14:53 +0000)]
b=3772
- debug info for corrupted lock case
rread [Wed, 14 Jul 2004 20:37:55 +0000 (20:37 +0000)]
b=3844
Add debug LASSERT to ldlm_replay_locks()
alex [Wed, 14 Jul 2004 15:04:46 +0000 (15:04 +0000)]
b=3824
- new inode flag EXT3_EXTRA_ISIZE to mark inodes having i_extra_isize
initialized properly
alex [Wed, 14 Jul 2004 14:04:27 +0000 (14:04 +0000)]
b=3829
- mds_copy_unlink_reply() shouldn't be called if md_unlink() returns an error
zab [Wed, 14 Jul 2004 00:12:09 +0000 (00:12 +0000)]
file Makefile.mk was initially added on branch b1_4_smallfix_iib.
zab [Wed, 14 Jul 2004 00:12:08 +0000 (00:12 +0000)]
file autoMakefile.am was initially added on branch b1_4_smallfix_iib.
zab [Wed, 14 Jul 2004 00:12:07 +0000 (00:12 +0000)]
file Makefile.in was initially added on branch b1_4_smallfix_iib.
zab [Wed, 14 Jul 2004 00:12:06 +0000 (00:12 +0000)]
file .cvsignore was initially added on branch b1_4_smallfix_iib.
rread [Tue, 13 Jul 2004 01:57:19 +0000 (01:57 +0000)]
b=3405
Use lmv by default for insanity.sh.