Whamcloud - gitweb
wangdi [Wed, 5 Apr 2006 06:20:31 +0000 (06:20 +0000)]
Branch: b_new_cmd
update mdd prototype
yury [Tue, 4 Apr 2006 13:24:23 +0000 (13:24 +0000)]
- added fid_lock()/fid_unlock()
- many changes about using fid. Idea is that we do not want to change whole mds stuff to use new fids but we want to use new fids for new modules.
tappro [Tue, 4 Apr 2006 09:08:52 +0000 (09:08 +0000)]
anoter occurance of ll_fid
tappro [Tue, 4 Apr 2006 08:58:37 +0000 (08:58 +0000)]
use lu_fid instead of ll_fid
wangdi [Tue, 4 Apr 2006 08:27:04 +0000 (08:27 +0000)]
Branch: b_new_cmd
1.move share structure to mdd
2.some minor fixes
tappro [Tue, 4 Apr 2006 07:44:01 +0000 (07:44 +0000)]
move cmm name to the obd.h
tappro [Tue, 4 Apr 2006 07:41:12 +0000 (07:41 +0000)]
add missed makefile stuff
yury [Tue, 4 Apr 2006 06:41:15 +0000 (06:41 +0000)]
- renames of ll_fid (what is expanded to llite_fid) into lu_fid (lustre_fid)
tappro [Mon, 3 Apr 2006 18:29:26 +0000 (18:29 +0000)]
add CMM to makefiles
tappro [Mon, 3 Apr 2006 18:28:25 +0000 (18:28 +0000)]
CMM support
tappro [Mon, 3 Apr 2006 18:27:35 +0000 (18:27 +0000)]
generic cmm device support
tappro [Mon, 3 Apr 2006 18:23:41 +0000 (18:23 +0000)]
move generic md_ stuff from mdt.h to md_object.h. It is extention for
lu_object.h
nikita [Mon, 3 Apr 2006 15:30:35 +0000 (15:30 +0000)]
move LUSTRE_MDT0_NAME into global header: utils need it
nikita [Mon, 3 Apr 2006 15:29:52 +0000 (15:29 +0000)]
mark local declaration "static"
nikita [Mon, 3 Apr 2006 15:29:25 +0000 (15:29 +0000)]
mdt: use new ptlrpc_key interface
nikita [Mon, 3 Apr 2006 15:28:45 +0000 (15:28 +0000)]
kill unused declaration
nikita [Mon, 3 Apr 2006 15:28:18 +0000 (15:28 +0000)]
fix OBD_ALLOC_* macros to (0) avoid multiple evaluation of @ptr; (1) return value
nikita [Mon, 3 Apr 2006 15:27:28 +0000 (15:27 +0000)]
add temporary backpointer from lu_device to obd_device
nikita [Mon, 3 Apr 2006 15:26:43 +0000 (15:26 +0000)]
new ptlrpc interface allowing modules to allocate per-thread values
(conceptually similar to pthread_key_create(3)).
wangdi [Mon, 3 Apr 2006 14:42:30 +0000 (14:42 +0000)]
Branch: b_new_cmd
update mdd proto type of mdd
wangdi [Mon, 3 Apr 2006 10:36:45 +0000 (10:36 +0000)]
Branch: b_new_cmd
update mdd prototype
nikita [Sun, 2 Apr 2006 15:21:44 +0000 (15:21 +0000)]
mdt prototype updates: cleanup of mdt_handle()
wangdi [Sun, 2 Apr 2006 03:01:23 +0000 (03:01 +0000)]
Branch: b_new_cmd
add mdd_handler.c mdd_internal.h for mdd prototype
nikita [Fri, 31 Mar 2006 13:26:50 +0000 (13:26 +0000)]
mdt proto changes:
- instroduce mdt_lock_handle, and add it to mdt_thread_info to reduce stack
consumption;
- fix mdt_mkdir() sample method to create second (child) object.
huanghua [Fri, 31 Mar 2006 09:58:30 +0000 (09:58 +0000)]
comment out all the code and make it to be able to compile
nikita [Fri, 31 Mar 2006 09:45:55 +0000 (09:45 +0000)]
mdt prototype updates
nikita [Fri, 31 Mar 2006 09:45:22 +0000 (09:45 +0000)]
lu_object updates
nikita [Fri, 31 Mar 2006 09:44:44 +0000 (09:44 +0000)]
mds_fid2locked_dentry(): kill unused parameters
huanghua [Fri, 31 Mar 2006 09:27:20 +0000 (09:27 +0000)]
PROTO for Adding MDT in CMD with mountconf.
Till now, it is not compillable!!
nikita [Wed, 29 Mar 2006 18:52:52 +0000 (18:52 +0000)]
mdt prototype changes
nikita [Wed, 29 Mar 2006 18:51:50 +0000 (18:51 +0000)]
add OBD_FAIL_MDS_SET_INFO_{PACK,NET}: having OBD_FAIL's for all mds methods makes mdt code simpler
nikita [Wed, 29 Mar 2006 18:48:37 +0000 (18:48 +0000)]
yank lu_device and lu_device_type inside of obd_device and obd_type resp. This is temporary measure to avoid changing a lot of code at prototype stage
nikita [Wed, 29 Mar 2006 18:47:04 +0000 (18:47 +0000)]
add lfid_eq() function
nikita [Wed, 29 Mar 2006 18:46:32 +0000 (18:46 +0000)]
fix uninitialized variable reported by gcc
nikita [Wed, 29 Mar 2006 18:44:40 +0000 (18:44 +0000)]
fixes for lu_* types
nikita [Wed, 29 Mar 2006 18:41:38 +0000 (18:41 +0000)]
->o_setup() methods always take struct lustre_cfg in buffer, make this explicit in typing
nikita [Tue, 28 Mar 2006 15:56:58 +0000 (15:56 +0000)]
add mds-laeyring prototype
cvs2svn [Mon, 27 Mar 2006 10:33:33 +0000 (10:33 +0000)]
This commit was manufactured by cvs2svn to create branch 'b_new_cmd'.
yujian [Mon, 27 Mar 2006 10:33:32 +0000 (10:33 +0000)]
b=9858
add some new test cases:
-test_20: start mds first
-test_21: start mds last
-test_22: start mds between two osts
-test_23: add a new ost before a client has started
-test_24: add a new ost after a client has started
nathan [Sat, 25 Mar 2006 01:40:05 +0000 (01:40 +0000)]
Branch b1_4_mountconf
b=9858
test changes for mountconf
nathan [Fri, 24 Mar 2006 20:49:19 +0000 (20:49 +0000)]
Branch b1_4_mountconf
b=9858
more cleanup of conf-sanity
nathan [Fri, 24 Mar 2006 00:04:39 +0000 (00:04 +0000)]
Branch b1_4_mountconf
b=9858
stupid strsep caused memory leak,
more changes for conf-sanity
nathan [Thu, 23 Mar 2006 01:34:43 +0000 (01:34 +0000)]
Branch b1_4_mountconf
b=9858
conf-sanity working (at least starting)
nathan [Wed, 22 Mar 2006 22:49:14 +0000 (22:49 +0000)]
Branch b1_4_mountconf
b=9858
sanityN working
nathan [Wed, 22 Mar 2006 20:00:00 +0000 (20:00 +0000)]
Branch b1_4_mountconf
b=9858
replace llmount/llmountcleanup with mountconf.sh
fix sanity
nathan [Tue, 21 Mar 2006 20:34:04 +0000 (20:34 +0000)]
Branch b1_4_mountconf
b=4482
Fix some recovery problems with late-added OSTs. A "late-added" OST is one
which is added to the FS after the MDT is added. Since the config logs are
not re-written, these OSTs get added to the LOV after the MDT starts up,
every time the MDT starts up. We have to fix the mdsize, and re-setup the
llogs whenever we add a late OST.
This fix in particular: during replay on an MDT, we still have to set the
right mdsize, but don't re-setup the llogs until recovery finishes.
(replay-single test 13)
Also check recovery w/o replay: create file, failover MDT, create another file.
nathan [Tue, 21 Mar 2006 20:02:34 +0000 (20:02 +0000)]
Branch b1_4_mountconf
b=9858
move mountconf deltas into test-framework as much as possible
replay-single now works reasonably well
nathan [Tue, 21 Mar 2006 16:53:12 +0000 (16:53 +0000)]
Branch b1_4_mountconf
b=9858
get service name from disk label
yujian [Tue, 21 Mar 2006 11:20:09 +0000 (11:20 +0000)]
b=9853
- verify service IP addresses with failover nodes
yujian [Tue, 21 Mar 2006 11:18:22 +0000 (11:18 +0000)]
b=9853
- scp ha.cf to all the nodes in the failover group, that is, all the nodes
in one failover group have the same ha.cf file
yujian [Tue, 21 Mar 2006 11:16:06 +0000 (11:16 +0000)]
b=9853
- modify the csv samples
- MGMT_NID->MGS_NIDS, NETWORKS->MODULE_OPTS
- --failover->--failnode, --mgmtnid->--mgsnode, --mgmt->--mgs
cliffw [Mon, 20 Mar 2006 23:07:08 +0000 (23:07 +0000)]
This is needed to place the new file in the make dist tarball, rpm build was broken.
nathan [Fri, 17 Mar 2006 22:30:31 +0000 (22:30 +0000)]
b=8007
update to b_rel_146
nathan [Fri, 17 Mar 2006 09:29:38 +0000 (09:29 +0000)]
file tcp-zero-copy-2.6.5-7.244.patch was added on branch b1_4_mountconf on 2006-04-17 21:30:25 +0000
yury [Fri, 17 Mar 2006 09:29:31 +0000 (09:29 +0000)]
file tcp-zero-copy-2.6.5-7.244.patch was added on branch b_new_cmd on 2006-05-19 11:29:25 +0000
nathan [Fri, 17 Mar 2006 09:29:27 +0000 (09:29 +0000)]
file tcp-zero-copy-2.6.12.6.patch was added on branch b1_4_mountconf on 2006-04-17 21:30:25 +0000
yury [Fri, 17 Mar 2006 09:29:20 +0000 (09:29 +0000)]
file tcp-zero-copy-2.6.12.6.patch was added on branch b_new_cmd on 2006-05-19 11:29:25 +0000
nathan [Thu, 16 Mar 2006 23:15:31 +0000 (23:15 +0000)]
Branch b1_4_mountconf
b=9861
Use common code to parse params from mkfs.lustre and from lctl:
mkfs.lustre --mdt --param="default_stripe_size=1114112" /dev/sda
or
lctl conf_param lustre-MDT0000 default_stripe_size=1114112
Implemented so far:
default_stripe_size
default_stripe_count
default_stripe_offset
default_stripe_pattern
obd_timeout
failnode
nathan [Thu, 16 Mar 2006 00:28:45 +0000 (00:28 +0000)]
Branch b1_4_mountconf
b=8007
Have to set the mds_max_mdsize for a recovery startup as well.
This patch does it "late", after the recovery completes -- this may not be
sufficient for replay.
nathan [Wed, 15 Mar 2006 18:03:51 +0000 (18:03 +0000)]
Branch b1_4_mountconf
b=9861
change client mount command from
mount uml1:/lustre-client
to
mount uml1:/lustre
at Braam's request.
nathan [Tue, 14 Mar 2006 22:35:48 +0000 (22:35 +0000)]
Branch b1_4_mountconf
b=9846
add ability to change failover nodes or stripe params via lctl:
mgs> ./lctl conf_param lustre-OST0000 failnode=3@elan,192.168.0.3@tcp0
any client> cat /proc/fs/lustre/osc/lustre-OST0000-osc/ost_conn_uuid
mgs> ./lctl conf_param lustre-mdtlov default_stripe_size=1114112
mgs> ./lctl conf_param lustre-mdtlov default_stripe_count=2
mgs> ./lctl conf_param lustre-mdtlov default_stripe_offset=3
mgs> ./lctl conf_param lustre-mdtlov default_stripe_pattern=0
cat /proc/fs/lustre/lov/lustre-mdtlov/stripe*
nathan [Tue, 14 Mar 2006 21:48:43 +0000 (21:48 +0000)]
Branch b1_4_mountconf
b=9861
make mgsnids and failover nids part of string parameters
(e.g. mgsnode=<nid_list>), to simplify mkfs/mount, and to provide a
distinction between multiple nids per node, and multiple nodes. String
parsing moves into kernel. Arbitrary params can be added to mkfs.lustre
with --param key=value, hopefully preventing any need to ever change the
lustre_disk_data.
nathan [Tue, 14 Mar 2006 19:24:59 +0000 (19:24 +0000)]
Branch b1_4_mountconf
b=9858
remove llrmount
nathan [Mon, 6 Mar 2006 19:10:01 +0000 (19:10 +0000)]
file tcp-rto_proc-2.6.9.patch was added on branch b1_4_mountconf on 2006-04-05 01:14:57 +0000
yury [Mon, 6 Mar 2006 19:10:00 +0000 (19:10 +0000)]
file tcp-rto_proc-2.6.9.patch was added on branch b_new_cmd on 2006-04-14 07:37:09 +0000
nathan [Fri, 3 Mar 2006 05:50:14 +0000 (05:50 +0000)]
file lustre_idl.h was added on branch b1_4_mountconf on 2006-04-26 18:45:46 +0000
yury [Fri, 3 Mar 2006 05:50:13 +0000 (05:50 +0000)]
file lustre_idl.h was added on branch b_new_cmd on 2006-05-19 11:29:23 +0000
nathan [Fri, 3 Mar 2006 00:28:12 +0000 (00:28 +0000)]
Branch b1_4_mountconf
b=4482
if there are no osc's, we don't know how big to make the page size
nathan [Thu, 2 Mar 2006 21:55:23 +0000 (21:55 +0000)]
Branch b1_4_mountconf
b=9861
add method to describe multiple failover nodes each with multiple nids to mkfs.lustre.
nathan [Thu, 2 Mar 2006 13:40:01 +0000 (13:40 +0000)]
file lustre_types.h was added on branch b1_4_mountconf on 2006-04-26 18:45:44 +0000
yury [Thu, 2 Mar 2006 13:40:00 +0000 (13:40 +0000)]
file lustre_types.h was added on branch b_new_cmd on 2006-05-19 11:29:21 +0000
nathan [Thu, 2 Mar 2006 13:39:59 +0000 (13:39 +0000)]
file lustre_types.h was added on branch b1_4_mountconf on 2006-04-26 18:45:42 +0000
yury [Thu, 2 Mar 2006 13:39:58 +0000 (13:39 +0000)]
file lustre_types.h was added on branch b_new_cmd on 2006-05-19 11:29:19 +0000
nathan [Wed, 1 Mar 2006 20:49:35 +0000 (20:49 +0000)]
Branch b1_4_mountconf
b=8007
allow old clients to reconnect to a newly-restarted MGS
(plus some minor cleanups)
nathan [Tue, 28 Feb 2006 22:35:35 +0000 (22:35 +0000)]
Branch b1_4_mountconf
b=8007
Check early for double-mount
nathan [Tue, 28 Feb 2006 21:45:52 +0000 (21:45 +0000)]
Branch b1_4_mountconf
b=8007
use random uuid for MGC so that MGS doesn't deny reconnection after parition.
nathan [Tue, 28 Feb 2006 01:13:27 +0000 (01:13 +0000)]
Branch b1_4_mountconf
b=9861
plumb in "add new failover nids"
nathan [Mon, 27 Feb 2006 22:31:06 +0000 (22:31 +0000)]
file tcp-zero-copy-2.6.9-rhel4.patch was added on branch b1_4_mountconf on 2006-04-05 01:14:57 +0000
yury [Mon, 27 Feb 2006 22:31:05 +0000 (22:31 +0000)]
file tcp-zero-copy-2.6.9-rhel4.patch was added on branch b_new_cmd on 2006-04-14 07:37:09 +0000
nathan [Mon, 27 Feb 2006 22:16:41 +0000 (22:16 +0000)]
Branch b1_4_mountconf
b=9846
add global lctl set_timeout
nathan [Mon, 27 Feb 2006 21:43:02 +0000 (21:43 +0000)]
file lustre_ver.h.in was added on branch b1_4_mountconf on 2006-04-26 18:45:41 +0000
yury [Mon, 27 Feb 2006 21:43:01 +0000 (21:43 +0000)]
file lustre_ver.h.in was added on branch b_new_cmd on 2006-05-19 11:29:18 +0000
nathan [Mon, 27 Feb 2006 19:07:04 +0000 (19:07 +0000)]
Branch b1_4_mountconf
b=8007
change sense of umount -f for servers:
umount = failover (next mount will go into recovery)
umount -f = force (non-failover, try to clean up)
nathan [Mon, 27 Feb 2006 16:50:52 +0000 (16:50 +0000)]
Branch b1_4_mountconf
b=8007
Use force/failover flags in mgc shutdown to prevent waiting around
lsy [Mon, 27 Feb 2006 09:57:33 +0000 (09:57 +0000)]
file lustre_capa.h was added on branch b_new_cmd on 2006-09-27 06:34:32 +0000
wangdi [Sat, 25 Feb 2006 19:16:28 +0000 (19:16 +0000)]
file ext3-htree-comments.patch was added on branch b_new_cmd on 2006-04-23 15:33:32 +0000
nathan [Fri, 24 Feb 2006 23:17:51 +0000 (23:17 +0000)]
Branch b1_4_mountconf
b=8214
replay-single updates (and fixes for problems found thereby)
nathan [Fri, 24 Feb 2006 18:33:36 +0000 (18:33 +0000)]
Branch b1_4_mountconf
b=8007
Kick the mgs import a little harder. If an mgs/mdt was restarted on a node
with a live mgc, the mgc wouldn't reconnect to the newly restarted
mgs. Throwing the ptlrpc_invalidate_import in there gives it a little more
incentive to reconnect - the mgs mount still gets EIO the first time, but
if you try it again it works. I need help from somebody with more
knowledge of all these ptlrpc_ fns.
yury [Fri, 24 Feb 2006 13:15:15 +0000 (13:15 +0000)]
file ext3-filterdata-2.6.15.patch was added on branch b_new_cmd on 2006-08-20 13:12:43 +0000
nathan [Thu, 23 Feb 2006 23:22:29 +0000 (23:22 +0000)]
Branch b1_4_mountconf
b=9858 regression test mods
b=8007 need refcount around config_llog_data for arbitrary cleanup
nathan [Thu, 23 Feb 2006 00:44:31 +0000 (00:44 +0000)]
Branch b1_4_mountconf
b=9858
-sanity test fixes
-brief nap before updating to new configs
yury [Wed, 22 Feb 2006 22:54:19 +0000 (22:54 +0000)]
file ext3-wantedi-2.6.15.patch was added on branch b_new_cmd on 2006-08-20 13:12:43 +0000
yury [Wed, 22 Feb 2006 22:54:13 +0000 (22:54 +0000)]
file ext3-extents-2.6.15.patch was added on branch b_new_cmd on 2006-08-20 13:12:43 +0000
nathan [Wed, 22 Feb 2006 18:58:37 +0000 (18:58 +0000)]
Branch b1_4_mountconf
b=8007
get rid of unneeded lustre_mgs.h
nathan [Wed, 22 Feb 2006 00:04:26 +0000 (00:04 +0000)]
Branch b1_4_mountconf
b=8007
Hunt down some memory leaks:
1. There are problems with the llog_finish/init pair in mds_lov_add_ost.
Firstly, they must be atomic as a pair.
Secondly, the setup sets up more than the cleanup cleans up (see
note in llog_setup), so we hackishly fix that.
2. Make sure we don't exit the umount until after we've dropped all our
references to the mount. Class_cleanup isn't synchronous, and we can't
exit the umount and risk our sb getting destroyed until we're really done.
So added a polling check. Also, move the MDS/OSS cleanup check until after
the poller, since I'm just looking at the class refcount.
nathan [Tue, 21 Feb 2006 23:43:54 +0000 (23:43 +0000)]
Branch b1_4_mountconf
b=8007
going after the memory leaks:
1. problems with the llog_finish / reinitialize in mds_lov_add_ost -
the pair must be atomic in case someone else is adding
the cleanup doesn't clean up as much as the setup sets up. This is
falloout
nathan [Mon, 20 Feb 2006 21:47:39 +0000 (21:47 +0000)]
Branch b1_4_mountconf
b=9863 - record timestamp in config log markers for future 9863 use
b=8192 - don't match existing locks in mgc_enqueue
b=9846 - fix client lov name for lctl conf_param
nathan [Mon, 20 Feb 2006 18:47:20 +0000 (18:47 +0000)]
Branch b1_4_mountconf
b=9846
Add lctl conf_param path via ioctl to MGS to set LCFG_PARAMs in logs.
Specifically, this allows things like:
./lctl conf_param lustre-mdtlov default_stripe_size=1055000
to be added to the MDT and client logs.
nathan [Fri, 17 Feb 2006 00:51:21 +0000 (00:51 +0000)]
Branch b1_4_mountconf
b=9860
upgrading from 1.4.6 - don't copy the old client log, just the old mdt log.
Regen client log as osts/mdt re-register after upgrade.
This means that new clients won't know about all the old osc's that the mdt
knows about, so don't start them until after all servers are upgraded.
But this also means no frankenstein hybrid old/new client log, and we don't
need to know the client name.
Use the old uuids parsed out of last_rcvd for the upgrading servers.
We can go back to 1.4.6 with no action required.
nathan [Thu, 16 Feb 2006 19:14:55 +0000 (19:14 +0000)]
Branch b1_4_mountconf
b=9860
namespace cleanup, new flags for upgrading from 146