]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Tue, 31 Mar 2009 19:41:59 +0000 (12:41 -0700)]
kclient: make ->write_inode() flush dirty caps
Sage Weil [Tue, 31 Mar 2009 19:41:14 +0000 (12:41 -0700)]
mds: leave simplelock in EXCL if there is a loner
Sage Weil [Tue, 31 Mar 2009 19:34:52 +0000 (12:34 -0700)]
kclient: fix flush_ack cleaned logic
We cleaned the bits we said were dirty. It has nothing to do with
whether we also released them.
Sage Weil [Tue, 31 Mar 2009 19:08:35 +0000 (12:08 -0700)]
kclient: drop old_atime cruft
Sage Weil [Tue, 31 Mar 2009 18:23:24 +0000 (11:23 -0700)]
kclient: fix cap_release accounting
Count num caps we can release, not messages.
Sage Weil [Tue, 31 Mar 2009 18:14:33 +0000 (11:14 -0700)]
kclient: no need to clean out cap_delay_list
Inodes get removed from __remove_cap. And there are no pins, so
we don't interfere with umount.
Sage Weil [Tue, 31 Mar 2009 18:13:56 +0000 (11:13 -0700)]
mds: release caps only via cap_release
Sage Weil [Tue, 31 Mar 2009 18:13:43 +0000 (11:13 -0700)]
kclient: never release via client_cap message
Sage Weil [Tue, 31 Mar 2009 17:47:50 +0000 (10:47 -0700)]
Merge branch 'unstable' into nofilltrace
Conflicts:
src/include/ceph_fs.h
Sage Weil [Mon, 30 Mar 2009 22:11:52 +0000 (15:11 -0700)]
scripts: make mds request checker show the operation
Sage Weil [Mon, 30 Mar 2009 23:41:53 +0000 (16:41 -0700)]
mds: disallow setlayout on non-new file
Sage Weil [Tue, 31 Mar 2009 17:23:53 +0000 (10:23 -0700)]
todo
Sage Weil [Tue, 31 Mar 2009 17:20:17 +0000 (10:20 -0700)]
kclient: release caps in destroy_inode
Sage Weil [Mon, 30 Mar 2009 22:58:54 +0000 (15:58 -0700)]
mds: add bulk MClientCapRelease message
Sage Weil [Mon, 30 Mar 2009 22:15:34 +0000 (15:15 -0700)]
kclient: do not pin cap inodes
Sage Weil [Mon, 30 Mar 2009 22:01:05 +0000 (15:01 -0700)]
kclient: pin dirty caps
Sage Weil [Mon, 30 Mar 2009 21:49:54 +0000 (14:49 -0700)]
kclient: drop cap lru
Sage Weil [Mon, 30 Mar 2009 20:34:58 +0000 (13:34 -0700)]
kclient: only request wanted caps if they aren't already issued
We still probably want something here to avoid updating wanted on
non-auth caps.
Sage Weil [Mon, 30 Mar 2009 20:31:13 +0000 (13:31 -0700)]
caps: fix CEPH_FILE_MODE_RD caps (no WR!)
Sage Weil [Mon, 30 Mar 2009 20:30:56 +0000 (13:30 -0700)]
mds: try choosing loner on encode_inodestat
Sage Weil [Mon, 30 Mar 2009 20:30:39 +0000 (13:30 -0700)]
mds: avoid going to MIX when unnecessary
Sage Weil [Mon, 30 Mar 2009 20:09:30 +0000 (13:09 -0700)]
kclient: do async wanted adjustment on open if we hold any caps
Sage Weil [Mon, 30 Mar 2009 20:12:24 +0000 (13:12 -0700)]
mds: issue FILE_RD+FILE_RDCACHE caps speculatively
This should let us do readdir + open + read with no mds
interaction on the open.
Yehuda Sadeh [Mon, 30 Mar 2009 20:01:22 +0000 (13:01 -0700)]
cosd: dump error when failed to mount, fix logs
Yehuda Sadeh [Mon, 30 Mar 2009 20:00:53 +0000 (13:00 -0700)]
cmon: add mds state changes logclient notifications
Sage Weil [Mon, 30 Mar 2009 19:50:50 +0000 (12:50 -0700)]
kclient: fix touch_cap; release old caps
Sage Weil [Mon, 30 Mar 2009 18:38:12 +0000 (11:38 -0700)]
kclient: make clean_caps lru; explicitly release old caps
Use check_caps for now. Eventually we want something more
efficient.
Yehuda Sadeh [Mon, 30 Mar 2009 18:16:30 +0000 (11:16 -0700)]
cmon: less verbose
Sage Weil [Mon, 30 Mar 2009 18:04:58 +0000 (11:04 -0700)]
kclient: drop rdcaps (cap expiration) in favor of clean_caps
This simplifies locking, too, since we have a less complication
definition of clean vs dirty (i.e. one that doesn't depend on
wanted and requires put_fmode weirdness).
Sage Weil [Mon, 30 Mar 2009 18:03:37 +0000 (11:03 -0700)]
kclient: drop cap renewal
Sage Weil [Mon, 30 Mar 2009 17:10:05 +0000 (10:10 -0700)]
kclient: keep track of which cap is authoritative
Yehuda Sadeh [Sat, 28 Mar 2009 01:10:15 +0000 (18:10 -0700)]
osd: add logclient state changes
Sage Weil [Fri, 27 Mar 2009 22:56:17 +0000 (15:56 -0700)]
kclient: drop unused USE_CAP_MDS
Sage Weil [Fri, 27 Mar 2009 22:56:08 +0000 (15:56 -0700)]
kclient: remove unused dentry in ceph_open
Sage Weil [Fri, 27 Mar 2009 21:58:34 +0000 (14:58 -0700)]
more cap notes
Sage Weil [Fri, 27 Mar 2009 21:46:37 +0000 (14:46 -0700)]
kclient: take CAP_PIN refs on requests
Sage Weil [Fri, 27 Mar 2009 21:35:24 +0000 (14:35 -0700)]
cap todos
Sage Weil [Fri, 27 Mar 2009 21:35:06 +0000 (14:35 -0700)]
kclient: i_pin_ref
Ability to carry refs on CAP_PIN.
Sage Weil [Fri, 27 Mar 2009 20:52:59 +0000 (13:52 -0700)]
kclient: fix deep mount refcounting on s_root
Sage Weil [Fri, 27 Mar 2009 20:39:51 +0000 (13:39 -0700)]
mon: avoid dup umount log entires on client unmount
Sage Weil [Fri, 27 Mar 2009 20:23:50 +0000 (13:23 -0700)]
mon: clean up client mount/unmount messages a bit
Sage Weil [Fri, 27 Mar 2009 20:23:34 +0000 (13:23 -0700)]
mon: use separate message type for log ack
Fixes logging from monitor itself.
Sage Weil [Fri, 27 Mar 2009 17:53:00 +0000 (10:53 -0700)]
mds: fix rename common parent checks
Rewrite common parent logic in terms of base inodes, since the
traces may initially be empty, and we may be renaming items in the
root directory.
Also look at _projected_ parents, not current parents.
Sage Weil [Fri, 27 Mar 2009 17:33:57 +0000 (10:33 -0700)]
mds: expose projected inode to loners, even if they don't have EXCL yet
If they are a (new) loner, they soon will!
Sage Weil [Fri, 27 Mar 2009 19:54:01 +0000 (12:54 -0700)]
kclient: handle mds replies with no trace.
Passes posix test suite.
Yehuda Sadeh [Fri, 27 Mar 2009 18:51:00 +0000 (11:51 -0700)]
cmon: add logclient, mount/unmount messsages
Sage Weil [Fri, 27 Mar 2009 17:53:00 +0000 (10:53 -0700)]
mds: fix rename common parent checks
Rewrite common parent logic in terms of base inodes, since the
traces may initially be empty, and we may be renaming items in the
root directory.
Also look at _projected_ parents, not current parents.
Sage Weil [Fri, 27 Mar 2009 17:34:23 +0000 (10:34 -0700)]
kclient: clean up
Sage Weil [Fri, 27 Mar 2009 17:34:15 +0000 (10:34 -0700)]
kclient: d_move only on rename
Sage Weil [Fri, 27 Mar 2009 17:33:57 +0000 (10:33 -0700)]
mds: expose projected inode to loners, even if they don't have EXCL yet
If they are a (new) loner, they soon will!
Yehuda Sadeh [Thu, 26 Mar 2009 23:49:17 +0000 (16:49 -0700)]
kclient: add a "show_bookkeeper" entry to debugfs
Sage Weil [Thu, 26 Mar 2009 23:42:27 +0000 (16:42 -0700)]
mds: return open dn to client when it wants it
Sage Weil [Thu, 26 Mar 2009 23:20:05 +0000 (16:20 -0700)]
partial cleanup of snap related ops. still not really working.
Yehuda Sadeh [Thu, 26 Mar 2009 22:57:13 +0000 (15:57 -0700)]
common: fix buffer::list::read_file, add buffer.cc
Yehuda Sadeh [Thu, 26 Mar 2009 22:35:44 +0000 (15:35 -0700)]
conf: ressurect the g_conf init_stuff
Yehuda Sadeh [Thu, 26 Mar 2009 21:43:37 +0000 (14:43 -0700)]
kclient: cleanup some sparse warnings
Yehuda Sadeh [Thu, 26 Mar 2009 21:35:31 +0000 (14:35 -0700)]
kclient: replace crc32c_le with crc32c
Sage Weil [Thu, 26 Mar 2009 20:58:42 +0000 (13:58 -0700)]
kclient: set s_root, fix misc refcounting
Sage Weil [Thu, 26 Mar 2009 20:53:33 +0000 (13:53 -0700)]
mds: include diri in readdir result
Sage Weil [Thu, 26 Mar 2009 19:52:43 +0000 (12:52 -0700)]
kclient: make mds requests relative to inode or single dentry
Sage Weil [Thu, 26 Mar 2009 18:52:17 +0000 (11:52 -0700)]
mds: simplify reply trace
We have either or both of:
dir inode + dirfrag + dentry
target inode
And that's it.
Clean up the server reply interface a bit.
Sage Weil [Thu, 26 Mar 2009 18:49:00 +0000 (11:49 -0700)]
filepath: leave off / if bare inode (no relative part)
Sage Weil [Wed, 25 Mar 2009 23:41:43 +0000 (16:41 -0700)]
kclient: simplify fill_trace, mount
Don't set sb->s_root.
Sage Weil [Tue, 24 Mar 2009 23:36:10 +0000 (16:36 -0700)]
train wreck
Sage Weil [Tue, 24 Mar 2009 21:34:37 +0000 (14:34 -0700)]
mds: specify number of dentries want in mds request
Let the client inform the MDS whether it wants dentry linkage info
(i.e., whether it holds the directory i_mutex).
Sage Weil [Wed, 25 Mar 2009 23:48:00 +0000 (16:48 -0700)]
todo
Sage Weil [Wed, 25 Mar 2009 22:37:19 +0000 (15:37 -0700)]
config: remove unused init_g_conf()
Sage Weil [Wed, 25 Mar 2009 22:36:39 +0000 (15:36 -0700)]
buffer: make read_file check read(2) result code
Sage Weil [Wed, 25 Mar 2009 23:28:17 +0000 (16:28 -0700)]
kclient: kill bad r_locked_dir in getattr
Sage Weil [Wed, 25 Mar 2009 23:27:18 +0000 (16:27 -0700)]
mds: fix unlink stray snaprealm reference
Stray dir doesn't get a snaprealm currently.
Sage Weil [Wed, 25 Mar 2009 22:31:16 +0000 (15:31 -0700)]
mds: make root dir default to 0755
Sage Weil [Wed, 25 Mar 2009 22:31:03 +0000 (15:31 -0700)]
config: make log sym dir default to log dir
Sage Weil [Wed, 25 Mar 2009 21:22:53 +0000 (14:22 -0700)]
config: open conf file readonly
Sage Weil [Wed, 25 Mar 2009 19:19:36 +0000 (12:19 -0700)]
todo
Sage Weil [Wed, 25 Mar 2009 18:41:48 +0000 (11:41 -0700)]
mds: fix auth_pin vs xlocks
Look at lock->is_stable() to determine whether auth_pin should be
taken or dropped in xlock_start() and eval_gather().
Sage Weil [Wed, 25 Mar 2009 18:34:26 +0000 (11:34 -0700)]
kclient: fix r_locked_dir to match VFS locking
Make sure r_locked_dir properly matches what the VFS is actually
doing.
Sage Weil [Tue, 24 Mar 2009 23:46:01 +0000 (16:46 -0700)]
debian: no restart on upgrade, no start
Yehuda Sadeh [Tue, 24 Mar 2009 23:44:07 +0000 (16:44 -0700)]
conf: compare mon addr in monmap and in conf, warn
Sage Weil [Tue, 24 Mar 2009 23:40:39 +0000 (16:40 -0700)]
todo
Sage Weil [Tue, 24 Mar 2009 22:12:50 +0000 (15:12 -0700)]
mds: avoid including caps in readdir items if snaprealm differs
Since the reply snaprealm starts at the _dir_ inode, we want to
exclude caps for any nested items if the realm differs. Otherwise
the client is screwed...
Sage Weil [Tue, 24 Mar 2009 22:11:27 +0000 (15:11 -0700)]
mds: no snaprealm for stray dir is needed
Sage Weil [Tue, 24 Mar 2009 21:53:52 +0000 (14:53 -0700)]
initscript: don't fail if $btrfs_path already mounted
Sage Weil [Tue, 24 Mar 2009 21:02:35 +0000 (14:02 -0700)]
osd: separate CEPH_OSD_OP_* and FLAGS_* namespaces
Ops and op flags are separate namespaces; fix the naming.
Sage Weil [Tue, 24 Mar 2009 20:48:49 +0000 (13:48 -0700)]
todo
Sage Weil [Tue, 24 Mar 2009 20:47:56 +0000 (13:47 -0700)]
kclient: print reassert_version in /debug/ceph/.../osdc
If it's there, it means a sync write was acked but not yet
committed.
Sage Weil [Tue, 24 Mar 2009 20:13:52 +0000 (13:13 -0700)]
buffer: check posix_memalign result code
Yehuda Sadeh [Tue, 24 Mar 2009 20:35:43 +0000 (13:35 -0700)]
kclient: clean osd read requests after receiving reply
Sage Weil [Tue, 24 Mar 2009 19:55:01 +0000 (12:55 -0700)]
initscripts: make do_cmd properly bail when command fails
And print what command it was that failed.
Yehuda Sadeh [Tue, 24 Mar 2009 19:47:11 +0000 (12:47 -0700)]
conf: exit if -c specified and conf file not found
Sage Weil [Tue, 24 Mar 2009 19:33:11 +0000 (12:33 -0700)]
initscript: fix .conf push to other nodes
Sage Weil [Tue, 24 Mar 2009 19:28:18 +0000 (12:28 -0700)]
todo
Sage Weil [Tue, 24 Mar 2009 19:27:12 +0000 (12:27 -0700)]
mds: creation, subtree map tweaks
Don't include mydir if it doesn't exist just yet, etc.
Yehuda Sadeh [Tue, 24 Mar 2009 19:23:23 +0000 (12:23 -0700)]
kclient: fix default timeout values
Yehuda Sadeh [Tue, 24 Mar 2009 19:20:35 +0000 (12:20 -0700)]
kclient: add configurable caps_delay option
Yehuda Sadeh [Mon, 23 Mar 2009 23:43:07 +0000 (16:43 -0700)]
kclient: less verbose
Sage Weil [Tue, 24 Mar 2009 19:15:38 +0000 (12:15 -0700)]
mkcephfs: fix monmap/osdmap copying
Sage Weil [Tue, 24 Mar 2009 18:56:59 +0000 (11:56 -0700)]
mds: fix fs creation, replay
Include mydir in subtree map, and journal it on mkfs. Invent
root inode during replay as needed (it will get rejoined).
Sage Weil [Tue, 24 Mar 2009 18:18:52 +0000 (11:18 -0700)]
kclient: locally open file based on what MDS thinks we want, not what we still want
As long as we have our interest in the caps still registered with
the MDS, we can do a local open. Checking what we currently still
want is only helpful if the same file still is open by another
process or something. Now, we can do a local open on a close +
wait + reopen, as long as we haven't done a cap release yet.
Sage Weil [Tue, 24 Mar 2009 18:17:16 +0000 (11:17 -0700)]
kclient: properly maintain mds_wanted value
Keep track of what wanted caps we have registered with the mds.
Sage Weil [Tue, 24 Mar 2009 18:14:48 +0000 (11:14 -0700)]
mds: put pending (not added) caps in reply caps
Otherwise the client thinks it had fewer caps than it has.