]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Fri, 31 Oct 2008 17:43:48 +0000 (10:43 -0700)]
osd: report pg osds, osd peers to pgmonitor; include in pg dump
Yehuda Sadeh [Fri, 31 Oct 2008 18:07:23 +0000 (11:07 -0700)]
kclient: keep a pointer to the current snap context in the inode
Yehuda Sadeh [Fri, 31 Oct 2008 17:18:00 +0000 (10:18 -0700)]
kclient: use current snap context if not found
Yehuda Sadeh [Thu, 30 Oct 2008 23:35:23 +0000 (16:35 -0700)]
kclient: don't register a new bdi for the same client
Sage Weil [Thu, 30 Oct 2008 23:21:33 +0000 (16:21 -0700)]
osd: do all recovery operations in dedicated recovery thread
Sage Weil [Thu, 30 Oct 2008 22:50:46 +0000 (15:50 -0700)]
dstart.sh: 2x rep only
Sage Weil [Thu, 30 Oct 2008 22:00:43 +0000 (15:00 -0700)]
dstop.sh: stop crun too
Yehuda Sadeh [Thu, 30 Oct 2008 22:57:44 +0000 (15:57 -0700)]
kclient: override rdcache invalidation time when going down
Sage Weil [Thu, 30 Oct 2008 21:23:03 +0000 (14:23 -0700)]
osd: avoid trimming unless pg is clean; keep recovered_to, complete_to log pointers cleared when not in use
Sage Weil [Thu, 30 Oct 2008 21:19:25 +0000 (14:19 -0700)]
kclient: do LCHOWN, not CHOWN, so that mds does not follow the symlink
Sage Weil [Thu, 30 Oct 2008 20:01:17 +0000 (13:01 -0700)]
kclient: update patch generator
Sage Weil [Thu, 30 Oct 2008 18:48:39 +0000 (11:48 -0700)]
mds: fix bad EINVAL error code in unlink
Sage Weil [Thu, 30 Oct 2008 18:48:20 +0000 (11:48 -0700)]
kclient: note locked new_dir in ceph_rename
Sage Weil [Thu, 30 Oct 2008 18:15:55 +0000 (11:15 -0700)]
mds: fix pre_dirty/mark_dirty order for remote dentries in unlink_local
Sage Weil [Thu, 30 Oct 2008 18:05:25 +0000 (11:05 -0700)]
kclient: use vmalloc for messages larger than one page
Sage Weil [Thu, 30 Oct 2008 18:00:46 +0000 (11:00 -0700)]
kclient: make reconnect allocation a bit more generous
Yehuda Sadeh [Wed, 29 Oct 2008 21:38:53 +0000 (14:38 -0700)]
kclient: remove extra dput on failed rmdir/unlink
Sage Weil [Wed, 29 Oct 2008 21:59:35 +0000 (14:59 -0700)]
osd: initialize curclone
Sage Weil [Wed, 29 Oct 2008 21:59:21 +0000 (14:59 -0700)]
filestore: fix getattrs return value
Sage Weil [Wed, 29 Oct 2008 21:47:22 +0000 (14:47 -0700)]
kclient: fix mds request reference after free bug
Sage Weil [Wed, 29 Oct 2008 21:22:02 +0000 (14:22 -0700)]
Revert "mds: CInode::mark_dirty should then dirty the projected parent inode, if present"
We _shouldn't_ use the projected dentry, only the current dentry, since updates
to this inode may be pipelined with other updates and/or a parent change. If
we are really doing a rename, fix the parent, _then_ dirty the inode.
This reverts commit
89c635d3ee6038daee352e238d00b94094927571 .
Sage Weil [Wed, 29 Oct 2008 21:03:05 +0000 (14:03 -0700)]
mon: lossless communication between monitors
Sage Weil [Wed, 29 Oct 2008 20:58:42 +0000 (13:58 -0700)]
kclient: set r_got_reply _after_ we ensure we have a session
Sage Weil [Wed, 29 Oct 2008 20:50:27 +0000 (13:50 -0700)]
mds: ClientCaps typo
Sage Weil [Wed, 29 Oct 2008 20:50:15 +0000 (13:50 -0700)]
mds: eval FileLock after last wrlock releases
Sage Weil [Wed, 29 Oct 2008 20:48:57 +0000 (13:48 -0700)]
kclient: no cap delay if mdsc->stopping (unmounting)
Sage Weil [Wed, 29 Oct 2008 19:14:27 +0000 (12:14 -0700)]
osd: clean up debug output a bit
Less chatter about pg stat updates to monitor
Sage Weil [Wed, 29 Oct 2008 19:06:37 +0000 (12:06 -0700)]
filestore: don't clobber file on clone_range
Sage Weil [Wed, 29 Oct 2008 19:05:51 +0000 (12:05 -0700)]
osd: initialize clone_overlap on clone
Initially, the clone overlaps completely with the newer item (head).
Sage Weil [Wed, 29 Oct 2008 18:15:50 +0000 (11:15 -0700)]
osd: skip pg log object when generating backlog
Sage Weil [Wed, 29 Oct 2008 18:15:17 +0000 (11:15 -0700)]
filestore: use strtoull (unsigned) instead of strtoll when parsing object names, etc.
Otherwise we fall over on CEPH_NOSNAP (-2), notably.
Sage Weil [Wed, 29 Oct 2008 18:13:44 +0000 (11:13 -0700)]
buffer: print 'no raw' for 'null' bufferptrs
Aids debugging...
Sage Weil [Wed, 29 Oct 2008 18:13:18 +0000 (11:13 -0700)]
objectstore: do not include 0-length bufferptr in bufferlist if attr not found
Having a "null" bufferptr causes problems in a number of places down the
line. We want an empty bufferlist instead.
Sage Weil [Tue, 28 Oct 2008 23:50:38 +0000 (16:50 -0700)]
vstartnew.sh: clean up out/* even if owned by root
Sage Weil [Tue, 28 Oct 2008 23:50:17 +0000 (16:50 -0700)]
osd: sync on shutdown
There may be write completion callbacks in flight that depend on what
we're about to tear down.
Sage Weil [Tue, 28 Oct 2008 23:49:46 +0000 (16:49 -0700)]
osd: batch pg remove messages, do them all at once
Sage Weil [Tue, 28 Oct 2008 23:48:59 +0000 (16:48 -0700)]
filestore: encode transaction only if journaling it
And do it later. No need to encode it before applying, now.
Sage Weil [Tue, 28 Oct 2008 23:48:21 +0000 (16:48 -0700)]
synclient: mksnapfile to generate a simple multi-snapshot object
writes 0~4M, takes snapshot, then overwrites 1~2M.
Sage Weil [Tue, 28 Oct 2008 23:31:36 +0000 (16:31 -0700)]
mds: generate valid seq value on rmsnap
Sage Weil [Tue, 28 Oct 2008 21:34:58 +0000 (14:34 -0700)]
osd: fix typo on setting complete_to
Sage Weil [Tue, 28 Oct 2008 20:49:32 +0000 (13:49 -0700)]
osd: include fsid in MPGStats
Sage Weil [Tue, 28 Oct 2008 20:47:16 +0000 (13:47 -0700)]
buffer: skip crc calc on zero length bufferptrs
Sage Weil [Tue, 28 Oct 2008 20:59:32 +0000 (13:59 -0700)]
mds: CInode::mark_dirty should then dirty the projected parent inode, if present
Sage Weil [Tue, 28 Oct 2008 20:45:30 +0000 (13:45 -0700)]
mon: ensure snap doesn't already exist in pending set
Sage Weil [Tue, 28 Oct 2008 20:26:22 +0000 (13:26 -0700)]
kclient: comment on paths for mds ops
Sage Weil [Mon, 27 Oct 2008 23:07:34 +0000 (16:07 -0700)]
osd: reduce max_rep to 3 by default
Sage Weil [Mon, 27 Oct 2008 22:22:54 +0000 (15:22 -0700)]
mds: increase max_size based on loner AND other caps
Sage Weil [Mon, 27 Oct 2008 21:02:46 +0000 (14:02 -0700)]
Merge branch 'recovery' into unstable
Sage Weil [Mon, 27 Oct 2008 21:01:56 +0000 (14:01 -0700)]
osd: recovery more efficiently keeps N ops in flight
Sage Weil [Mon, 27 Oct 2008 20:28:43 +0000 (13:28 -0700)]
objectcacher: fix commit callback accounting
Sage Weil [Mon, 27 Oct 2008 18:31:38 +0000 (11:31 -0700)]
kclient: clean out when generating patches
Sage Weil [Mon, 27 Oct 2008 18:53:32 +0000 (11:53 -0700)]
xlist: implement push_front()
Yehuda Sadeh [Mon, 27 Oct 2008 18:49:44 +0000 (11:49 -0700)]
kclient: remove warning due to missing format
Sage Weil [Mon, 27 Oct 2008 17:32:52 +0000 (10:32 -0700)]
mds: make loner unitialized warning go away
Sage Weil [Mon, 27 Oct 2008 18:23:06 +0000 (11:23 -0700)]
kclient: bleh
Sage Weil [Mon, 27 Oct 2008 18:20:10 +0000 (11:20 -0700)]
kclient: include documentation in patch queue
Sage Weil [Mon, 27 Oct 2008 17:28:18 +0000 (10:28 -0700)]
filestore: be less noisy
Sage Weil [Mon, 27 Oct 2008 17:03:30 +0000 (10:03 -0700)]
Merge commit 'origin/unstable' into recovery
Sage Weil [Mon, 27 Oct 2008 03:09:42 +0000 (20:09 -0700)]
kclient: typo
Sage Weil [Mon, 27 Oct 2008 03:07:02 +0000 (20:07 -0700)]
mds: fix filelock loner modes
Rewrote Filelock handling of "loner" mode, which gives a single client access
to do cached/bufferd read/write on a file. To make this sane, CInode::loner_cap
identifies the distinguished client (if any), and the FileLock caps methods
provides issued/allowed/wanted values for loner and other clients. A new
lock state is added to track revocation of RDCACHE from lock->loner on the
non-loner clients.
Overall, the new code is much cleaner and should be less susceptible to
odd corner cases...
Sage Weil [Fri, 24 Oct 2008 23:47:19 +0000 (16:47 -0700)]
mds: fix rfile/rsubdirs accounting vs linkunlink
Yehuda Sadeh [Fri, 24 Oct 2008 22:53:51 +0000 (15:53 -0700)]
kclient: xattr rsubdirs fix
Sage Weil [Fri, 24 Oct 2008 22:29:52 +0000 (15:29 -0700)]
kclient: take i_lock in flush_snaps
Yehuda Sadeh [Fri, 24 Oct 2008 21:07:29 +0000 (14:07 -0700)]
kclient: get code up to date with latest linux kernel
Sage Weil [Fri, 24 Oct 2008 20:56:45 +0000 (13:56 -0700)]
kclient: flush/release all caps back to mds before closing a session
This allows us to handle races that can occur if an inode moves between
MDSs. It also ensures that _all_ dirty cap data is safely flushed before
we throw it out.
To make this work, we drop the PIN reference on the subdirectory that
we mounted. Since VFS doesn't tell us when each vfsmount goes away,
we can't remove them as we go. And we're more or less screwed anyway
if the directory gets renamed (currently, at least), so there's not
much point in having this reference in the first place.
Sage Weil [Fri, 24 Oct 2008 18:44:11 +0000 (11:44 -0700)]
crush: return 1.0 from get_offload if osd is not in the map
Sage Weil [Fri, 24 Oct 2008 18:30:06 +0000 (11:30 -0700)]
kclient: remove realm from parent's child list on removal; avoid letting ref count hit 0 during split
Yehuda Sadeh [Fri, 24 Oct 2008 18:09:12 +0000 (11:09 -0700)]
kclient: remove some sparse warnings
Sage Weil [Fri, 24 Oct 2008 18:16:56 +0000 (11:16 -0700)]
kclient: fix cap flush, and drop ref when acked by mds
Yehuda Sadeh [Fri, 24 Oct 2008 17:47:09 +0000 (10:47 -0700)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Sage Weil [Fri, 24 Oct 2008 17:57:39 +0000 (10:57 -0700)]
kclient: finish_cap_snap fixes
Sage Weil [Fri, 24 Oct 2008 17:47:52 +0000 (10:47 -0700)]
kclient: queue finished cap_snaps and flush them after snap_rwsem is dropped
This avoids some lock ordering problems. It's also just smarter.. we don't
want to send all those messages while holding snap_rwsem.
Yehuda Sadeh [Fri, 24 Oct 2008 17:47:07 +0000 (10:47 -0700)]
kclient: cleanup some sparse messages
Yehuda Sadeh [Fri, 24 Oct 2008 17:33:24 +0000 (10:33 -0700)]
kclient: free snap_rwsem at ceph_handle_snap
Yehuda Sadeh [Fri, 24 Oct 2008 17:31:08 +0000 (10:31 -0700)]
kclient: remove the use of snap_rwsem at ceph_mdsc_close_sessions()
Yehuda Sadeh [Fri, 24 Oct 2008 17:30:21 +0000 (10:30 -0700)]
kclient: fix __ceph_flush_snaps use of 'follows'
Yehuda Sadeh [Thu, 23 Oct 2008 19:48:42 +0000 (12:48 -0700)]
kclient: drop reply mutex, add a flag instead
Yehuda Sadeh [Thu, 23 Oct 2008 18:49:27 +0000 (11:49 -0700)]
kclient: use a new reply mutex, fixes locks dependency problem
Sage Weil [Thu, 23 Oct 2008 03:36:52 +0000 (20:36 -0700)]
kclient: fix allocation flags for osd client radix tree
Sage Weil [Thu, 23 Oct 2008 02:17:21 +0000 (19:17 -0700)]
kclient: try to release RDCACHE if _no_ caps are wanted
As Yehuda pointed out, we aren't really concerned about invalidating pages and
releasing RDCACHE if we are still open for write. So only bother releasing
RDCACHE when we have _no_ more interest in the file. If there is a conflict,
the MDS will just revoke.
Sage Weil [Thu, 23 Oct 2008 00:16:06 +0000 (17:16 -0700)]
kclient: invalidate pages if RDCACHE used but not wanted
Sage Weil [Wed, 22 Oct 2008 23:40:44 +0000 (16:40 -0700)]
kclient: changing around the rdcache invalidation stuff
Sage Weil [Wed, 22 Oct 2008 22:58:36 +0000 (15:58 -0700)]
kclient: fix messenger tmp var name
Yehuda Sadeh [Wed, 22 Oct 2008 22:32:42 +0000 (15:32 -0700)]
kclient: cleanup
Yehuda Sadeh [Wed, 22 Oct 2008 22:12:12 +0000 (15:12 -0700)]
kclient: revocation of rdcache handling
Sage Weil [Wed, 22 Oct 2008 22:26:50 +0000 (15:26 -0700)]
kclient: more patch queue futzing
Sage Weil [Wed, 22 Oct 2008 22:26:04 +0000 (15:26 -0700)]
crush: comments, cleanup
Sage Weil [Wed, 22 Oct 2008 19:35:43 +0000 (12:35 -0700)]
kclient: flush cap_snaps on session close
Sage Weil [Wed, 22 Oct 2008 17:42:15 +0000 (10:42 -0700)]
kclient: build patches too in import script
Sage Weil [Wed, 22 Oct 2008 17:54:44 +0000 (10:54 -0700)]
kclient: remove remaining #include/#ifdef cruft from ceph_fs.h
Sage Weil [Wed, 22 Oct 2008 05:14:06 +0000 (22:14 -0700)]
kclient: patch queue commit comments
Sage Weil [Wed, 22 Oct 2008 04:42:48 +0000 (21:42 -0700)]
clean up ceph_fs.h a bit
Sage Weil [Tue, 21 Oct 2008 23:57:27 +0000 (16:57 -0700)]
kclient: more patch queue fun
Sage Weil [Wed, 22 Oct 2008 00:04:45 +0000 (17:04 -0700)]
kclient: adjust import queue
Sage Weil [Tue, 21 Oct 2008 23:27:48 +0000 (16:27 -0700)]
kclient: allow monitor port to be specified with the ip(s) during mount
Sage Weil [Tue, 21 Oct 2008 22:08:46 +0000 (15:08 -0700)]
kclient: include Documentation/filesystems/ceph.txt in patch series
Yehuda Sadeh [Tue, 21 Oct 2008 22:16:15 +0000 (15:16 -0700)]
kclient: some changes for 2.6.28
Yehuda Sadeh [Tue, 21 Oct 2008 00:00:12 +0000 (17:00 -0700)]
kclient: added crush.h for the git import tool
Yehuda Sadeh [Mon, 20 Oct 2008 23:45:30 +0000 (16:45 -0700)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Sage Weil [Mon, 20 Oct 2008 23:55:49 +0000 (16:55 -0700)]
kclient: add dput to getattr for do_lookup splice case
If do_lookup returns a dentry, it has a new ref on it. Make sure we
dput() that, since we don't actually need it. (In contrast, if there is no
splice, do_lookup does not dget, as the caller has a ref on the dentry it
provided in the first place.)