]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Thu, 26 Feb 2009 19:25:37 +0000 (11:25 -0800)]
logger: identify fields with int constants instead of strings
This avoids the crufty hashes and so forth. The only downside is that
_all_ fields must be predeclared. That's probably a good thing anyway.
Logger will assert if the key is invalid or the name hasn't been
registered. Hopefully I haven't missed any.
Sage Weil [Thu, 26 Feb 2009 17:52:20 +0000 (09:52 -0800)]
config: remove ms_hosts
Sage Weil [Thu, 26 Feb 2009 17:52:00 +0000 (09:52 -0800)]
logger: configurable output dir
Sage Weil [Thu, 26 Feb 2009 17:47:42 +0000 (09:47 -0800)]
osd: strip out hard-coded osd device paths
The correct device path should be specified on the command line.
Sage Weil [Thu, 26 Feb 2009 17:23:34 +0000 (09:23 -0800)]
kclient: zero_user_segments doesn't appear until 2.6.25
Sage Weil [Wed, 25 Feb 2009 23:47:41 +0000 (15:47 -0800)]
kclient: more sysfs cleanup, bug fixes
Sage Weil [Wed, 25 Feb 2009 23:07:38 +0000 (15:07 -0800)]
kclient: fix kbuild patch, import script
Sage Weil [Wed, 25 Feb 2009 23:07:24 +0000 (15:07 -0800)]
kclient: move parens
Sage Weil [Wed, 25 Feb 2009 22:40:55 +0000 (14:40 -0800)]
kclient: changes for 2.6.29
Sage Weil [Wed, 25 Feb 2009 22:39:08 +0000 (14:39 -0800)]
kclient: avoid simple_prepare_write()
No longer exported.
Sage Weil [Wed, 25 Feb 2009 22:34:55 +0000 (14:34 -0800)]
kclient: separate Kconfig
Sage Weil [Wed, 25 Feb 2009 21:07:23 +0000 (13:07 -0800)]
todos
Sage Weil [Wed, 25 Feb 2009 20:49:48 +0000 (12:49 -0800)]
kclient: add module author
Sage Weil [Wed, 25 Feb 2009 20:49:18 +0000 (12:49 -0800)]
kclient: dup kobj_sysfs_ops from kobject.c
They're not exported, bleh.
Sage Weil [Wed, 25 Feb 2009 20:38:32 +0000 (12:38 -0800)]
journal: initiate full commit when journal hits 50% full
We want to avoid filling up the journal, particularly since
_start_sync is a no-op with a journal enabled.
Sage Weil [Wed, 25 Feb 2009 20:37:28 +0000 (12:37 -0800)]
filestore: don't initiate sync on _start_sync if there's a journal
Journal will commit immediately anyway.. don't slow the rest
down with a full commit to disk.
Sage Weil [Wed, 25 Feb 2009 20:20:43 +0000 (12:20 -0800)]
osd: don't clobber log output stream
Don't print a null char pointer.
Sage Weil [Wed, 25 Feb 2009 20:19:54 +0000 (12:19 -0800)]
mon: only do big sync if we're flushing lots of states
The rest of the time, fsync them individually.
Sage Weil [Wed, 25 Feb 2009 19:10:43 +0000 (11:10 -0800)]
kclient: include sysfs (and not proc) in kernel client patchset
Sage Weil [Wed, 25 Feb 2009 19:05:55 +0000 (11:05 -0800)]
osd: take journal on command line
./cosd dev -j journaldev ...
Sage Weil [Wed, 25 Feb 2009 18:09:19 +0000 (10:09 -0800)]
kclient: remove /proc/fs/ceph
/sys/fs/ceph now does everything it used to.
Sage Weil [Wed, 25 Feb 2009 18:06:03 +0000 (10:06 -0800)]
kclient: add other maps to sysfs
Sage Weil [Wed, 25 Feb 2009 17:29:56 +0000 (09:29 -0800)]
kclient: add debug levels, mask to /sys/fs/ceph
Sage Weil [Wed, 25 Feb 2009 05:14:45 +0000 (21:14 -0800)]
kclient: some sysfs cleanup
Sage Weil [Wed, 25 Feb 2009 00:58:29 +0000 (16:58 -0800)]
dstop: remove cosd hostname prefix
Sage Weil [Wed, 25 Feb 2009 00:51:09 +0000 (16:51 -0800)]
todos
Sage Weil [Wed, 25 Feb 2009 00:51:04 +0000 (16:51 -0800)]
dstart: remove cosd hostname prefix
Sage Weil [Wed, 25 Feb 2009 00:37:55 +0000 (16:37 -0800)]
kclient: consider issued AND implemented AND dirty when updating inode
We do not want to update the local inode field(s) if we
* still hold EXCL
* are having EXCL revoked (not issued, but still implemented)
* are writing back EXCL (not issued or implemented, but dirty)
Sage Weil [Wed, 25 Feb 2009 00:33:46 +0000 (16:33 -0800)]
mds: flush cap update if affected lock is unstable
Sage Weil [Tue, 24 Feb 2009 22:30:12 +0000 (14:30 -0800)]
debian: include new binaries, scripts in ceph-common
Sage Weil [Tue, 24 Feb 2009 22:30:00 +0000 (14:30 -0800)]
Makefile: mount.ceph goes in usr/sbin
Sage Weil [Tue, 24 Feb 2009 22:21:25 +0000 (14:21 -0800)]
debian: pull version from configure.ac
Sage Weil [Tue, 24 Feb 2009 22:13:37 +0000 (14:13 -0800)]
debian: specify arch on command line
Sage Weil [Tue, 24 Feb 2009 22:08:19 +0000 (14:08 -0800)]
makefile: add missing files for kernel module
Sage Weil [Tue, 24 Feb 2009 21:57:16 +0000 (13:57 -0800)]
debian: script to build and upload debs to repo
Sage Weil [Tue, 24 Feb 2009 21:15:53 +0000 (13:15 -0800)]
make unstable v0.7
Sage Weil [Tue, 24 Feb 2009 20:33:21 +0000 (12:33 -0800)]
common: uninit var
Sage Weil [Tue, 24 Feb 2009 20:29:28 +0000 (12:29 -0800)]
debian: cleanup, changelog fix
Sage Weil [Tue, 24 Feb 2009 20:22:16 +0000 (12:22 -0800)]
kclient: typo
Sage Weil [Tue, 24 Feb 2009 20:21:45 +0000 (12:21 -0800)]
mds: xlock filelock during truncate
This allows client to see projected truncation before it commits.
Sage Weil [Tue, 24 Feb 2009 19:55:58 +0000 (11:55 -0800)]
mds: look at active (not projected) inode when issuing truncation
We are interested in applying the truncation that just committed, not one
that is projected.
Sage Weil [Tue, 24 Feb 2009 19:55:23 +0000 (11:55 -0800)]
mds: only do truncation process when file size goes down
For file size increase, no special action is required...
Sage Weil [Tue, 24 Feb 2009 19:52:56 +0000 (11:52 -0800)]
mds: limit propagation of full auth_pin count up hierarchy
We only care if there are _some_ auth pins, not the exact count. So only
propagation whether there are any pings from dir -> inode, to avoid walking
all the way up the hierarchy for each pin/unpin event.
Sage Weil [Mon, 23 Feb 2009 21:46:46 +0000 (13:46 -0800)]
kclient: don't remove from clients list if we're not on it
Sage Weil [Mon, 23 Feb 2009 18:18:46 +0000 (10:18 -0800)]
Makefile: link libcommon, lockdep last
Link libcommon.a last for everything.
Include lockdep last in libcommon so to initialize first, shutdown
last, so that g_lockdep is initialized to zero for all Mutexes.
Put config.cc next to last so that the atomic_t thread counter is
initialized before all threads are created, shutdown after all
threads are destroyed.
Yehuda Sadeh [Fri, 20 Feb 2009 22:49:44 +0000 (14:49 -0800)]
objecter: shouldn't swap bufferlist buffers
Yehuda Sadeh [Thu, 19 Feb 2009 23:42:56 +0000 (15:42 -0800)]
common: remove warning
Yehuda Sadeh [Thu, 19 Feb 2009 23:41:36 +0000 (15:41 -0800)]
common: virtual destructor for class Timer
Sage Weil [Thu, 19 Feb 2009 23:25:21 +0000 (15:25 -0800)]
todo
Yehuda Sadeh [Fri, 13 Feb 2009 21:28:24 +0000 (13:28 -0800)]
SafeTimer: getting rid of global timer
Sage Weil [Thu, 19 Feb 2009 23:22:12 +0000 (15:22 -0800)]
kclient: only unregister after both acks
Partially reverts behavior of
b732c0ddb8a990a97b98606a6d9a387e5a21e3e5
We need the request to stay in the request_tree so that when the
safe arrives we can look it up (duh). Also, we need to keep a
reference to the request during that period...
Sage Weil [Thu, 19 Feb 2009 22:47:03 +0000 (14:47 -0800)]
mds: reverse rename argument order
Order change introduced by
3dc43aa8e6606be15695d565715b6a7a1b2d38ea
Sage Weil [Thu, 19 Feb 2009 22:45:37 +0000 (14:45 -0800)]
client: reverse rename argument order
Sage Weil [Thu, 19 Feb 2009 22:38:33 +0000 (14:38 -0800)]
kclient: add missing r_wait list_head init
Sage Weil [Thu, 19 Feb 2009 22:30:16 +0000 (14:30 -0800)]
kclient: initialize request list_heads
Sage Weil [Thu, 19 Feb 2009 22:23:38 +0000 (14:23 -0800)]
kclient: cast path
path[12] will either point to freshly allocated paths (that we
later kfree) or a provided const path.
Sage Weil [Thu, 19 Feb 2009 21:59:15 +0000 (13:59 -0800)]
kclient: unregister mds requests on unsafe reply
Fresh vs unsafe mds request replay is handled separately. So,
unregister requests on first reply. This avoids a stray dentry
reference if we unmount before getting a final safe (although
eventually those should be explicitly flushed as well).
Sage Weil [Thu, 19 Feb 2009 20:10:40 +0000 (12:10 -0800)]
kclient: clean up some mdsc locking
Do not drop mds mutex in __do_request().
Sage Weil [Thu, 19 Feb 2009 21:08:27 +0000 (13:08 -0800)]
kclient: fix path memory leak
Sage Weil [Thu, 19 Feb 2009 21:10:40 +0000 (13:10 -0800)]
kclient: async mds requestion completion callbacks
Sage Weil [Thu, 19 Feb 2009 00:35:45 +0000 (16:35 -0800)]
kclient: fix fh_to_dentry; move build_path to mds_client.c
Sage Weil [Thu, 19 Feb 2009 00:19:39 +0000 (16:19 -0800)]
kclient: mds requests in terms of dentries, not paths
Generate mds request paths when request is sent, so that it reflects
the current state of mds capabilities. Eventually we can then make
the relative paths based on the current capabilities for that mds..
Sage Weil [Wed, 18 Feb 2009 22:13:49 +0000 (14:13 -0800)]
kclient: async mds requests
Restructure the mds client to handle mds request asynchronously, so
that we can handle requests that are not managed by a blocking
thread.
Yehuda Sadeh [Thu, 19 Feb 2009 19:53:11 +0000 (11:53 -0800)]
kclient: remove extra log output
Yehuda Sadeh [Thu, 19 Feb 2009 17:43:28 +0000 (09:43 -0800)]
kclient: sync writeout when not pdflush
Sage Weil [Thu, 19 Feb 2009 19:01:51 +0000 (11:01 -0800)]
mds: only flush journal when we need to
Force a flush every tick just in case.
Sage Weil [Wed, 18 Feb 2009 22:20:05 +0000 (14:20 -0800)]
kclient: set 'sync' flag for fsync
This will ensure that fsynced data is committed immediately by the
OSDs.
Sage Weil [Wed, 18 Feb 2009 19:53:51 +0000 (11:53 -0800)]
script: fix check_osd_request_latency
Sage Weil [Wed, 18 Feb 2009 19:47:23 +0000 (11:47 -0800)]
script: check_osd_request_latency
Yehuda Sadeh [Wed, 18 Feb 2009 19:07:19 +0000 (11:07 -0800)]
conf: can turn on debug logs in startup.conf
Yehuda Sadeh [Tue, 17 Feb 2009 21:26:24 +0000 (13:26 -0800)]
conf: add debug_ms option
Sage Weil [Tue, 17 Feb 2009 20:15:39 +0000 (12:15 -0800)]
mds: don't rdlock stat fields that client has EXCL cap for
yehudasa [Tue, 17 Feb 2009 18:57:48 +0000 (10:57 -0800)]
add ceph_init to distribution
Yehuda Sadeh [Tue, 17 Feb 2009 18:45:34 +0000 (10:45 -0800)]
kclient: backport changes to kernel 2.6.25
Yehuda Sadeh [Sat, 14 Feb 2009 01:09:12 +0000 (17:09 -0800)]
kclient: fix mdsc->snap_empty cleanup
Yehuda Sadeh [Sat, 14 Feb 2009 01:06:22 +0000 (17:06 -0800)]
kclient: remove client from client list before destorying
Yehuda Sadeh [Fri, 13 Feb 2009 23:02:51 +0000 (15:02 -0800)]
kclient: fix proc cleanup
Yehuda Sadeh [Fri, 13 Feb 2009 00:47:56 +0000 (16:47 -0800)]
assert: some more assert_warn
Yehuda Sadeh [Thu, 12 Feb 2009 23:05:12 +0000 (15:05 -0800)]
cosd: change some assert to assert_warn and clean exit
Yehuda Sadeh [Thu, 12 Feb 2009 20:43:17 +0000 (12:43 -0800)]
kclient: show some meaningful info in client proc entry
Yehuda Sadeh [Thu, 12 Feb 2009 19:05:45 +0000 (11:05 -0800)]
kclient: per client proc data file
Yehuda Sadeh [Wed, 11 Feb 2009 00:21:20 +0000 (16:21 -0800)]
kclient: added a proc function to read client data
Yehuda Sadeh [Wed, 11 Feb 2009 00:08:02 +0000 (16:08 -0800)]
kclient: proc subdir for each ceph client
Yehuda Sadeh [Tue, 10 Feb 2009 23:42:51 +0000 (15:42 -0800)]
kclient: keep global list of ceph clients
Yehuda Sadeh [Tue, 10 Feb 2009 00:29:43 +0000 (16:29 -0800)]
assert: warn on assertion if requested for current thread
Yehuda Sadeh [Mon, 9 Feb 2009 23:46:02 +0000 (15:46 -0800)]
added a common initialization function
Yehuda Sadeh [Mon, 9 Feb 2009 22:59:07 +0000 (14:59 -0800)]
tls: addeed a thread local storage infrastructure
(cherry picked from commit
1b6b128f4e2aaab016575f48f2ec50d4abe002c9 )
Yehuda Sadeh [Sat, 31 Jan 2009 00:24:41 +0000 (16:24 -0800)]
osd: fix log level
Sage Weil [Fri, 30 Jan 2009 23:13:43 +0000 (15:13 -0800)]
mds: make sure we eval_stray when inode was replicated
Also, drop dn to end of lru if unlinked and last cap.
Sage Weil [Fri, 30 Jan 2009 23:04:25 +0000 (15:04 -0800)]
mds: remove traces of CDir::committed_version_equivalent
Don't think it's coming back
Sage Weil [Fri, 30 Jan 2009 23:03:21 +0000 (15:03 -0800)]
kclient: release CAP_LINK_RDCACHE before unlink
Sage Weil [Fri, 30 Jan 2009 23:01:03 +0000 (15:01 -0800)]
vstart: fix call to stop.sh
Sage Weil [Fri, 30 Jan 2009 23:00:53 +0000 (15:00 -0800)]
todos
Sage Weil [Fri, 30 Jan 2009 23:00:45 +0000 (15:00 -0800)]
osd: print ops args correctly
Sage Weil [Fri, 30 Jan 2009 23:00:34 +0000 (15:00 -0800)]
mds: use generic helper when removing a cap
This way we make sure to do all the right checks on the locks,
mds_caps, and to call eval_stray if the inode is unlinked.
Sage Weil [Fri, 30 Jan 2009 22:59:24 +0000 (14:59 -0800)]
mds: fix try_remove_unlinked_dn
Clear dentry NEW flag when we commit it.. pretty sure that was the
old bug. And we can't cheat the directory commit because the fnode
also gets updated.
Sage Weil [Fri, 30 Jan 2009 21:51:49 +0000 (13:51 -0800)]
mds: fix purge of 0-length inode
Sage Weil [Fri, 30 Jan 2009 21:15:20 +0000 (13:15 -0800)]
mds: remove unused inode purge fields
Yehuda Sadeh [Fri, 30 Jan 2009 23:56:36 +0000 (15:56 -0800)]
ceph_init: fix initalization
Yehuda Sadeh [Fri, 30 Jan 2009 23:48:56 +0000 (15:48 -0800)]
ceph_init: added missing file