]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Mon, 9 Mar 2009 18:34:34 +0000 (11:34 -0700)]
kclient: simplify osd timeouts
Call handle_timeout N seconds after oldest request. In
handle_timeout, reschedule N seconds in future. Reschedule
with new oldest request when oldest requests completes.
Maintain a r_timeout_stamp value for each request.
Sage Weil [Mon, 9 Mar 2009 20:57:22 +0000 (13:57 -0700)]
cosd: fewer osds, with big btrfs devices
Sage Weil [Mon, 9 Mar 2009 20:18:56 +0000 (13:18 -0700)]
initscripts: allow vars with spaces
Sage Weil [Mon, 9 Mar 2009 20:18:05 +0000 (13:18 -0700)]
osdmap: build default crush maps out of mostly straw buckets
Sage Weil [Mon, 9 Mar 2009 19:51:04 +0000 (12:51 -0700)]
confutils: make max line very large
Yehuda Sadeh [Mon, 9 Mar 2009 17:08:51 +0000 (10:08 -0700)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Yehuda Sadeh [Mon, 9 Mar 2009 17:08:29 +0000 (10:08 -0700)]
kclient: add sysfs directory per mds request
Sage Weil [Sat, 7 Mar 2009 00:45:55 +0000 (16:45 -0800)]
filepath: always prepend ino if it's nonzero
That means the only time we don't have an ino perfix is for fully
relative paths.
Sage Weil [Sat, 7 Mar 2009 00:44:37 +0000 (16:44 -0800)]
client: fix request paths
Use filepath, not bare string, whenever we can.
Sage Weil [Sat, 7 Mar 2009 00:26:05 +0000 (16:26 -0800)]
kclient: fix up replay after mds recovery
Do not kick unsafe requests in kick_requests (and in particular, do not
drop the session refs).
Set retry_attempt to MDS_REQUEST_REPLAY to indicate which requests are
replays.
Sage Weil [Sat, 7 Mar 2009 00:25:04 +0000 (16:25 -0800)]
mds: incorporate reconnect cap info after requests replay if ino originally missing
If the client reasserts caps from replayed requests, the inodes don't yet
exists during the reconnect (or even rejoin) stage. So, if we don't find
the inode, keep the reconnect info around. When processing a replayed
request in CInode::encode_inodestat, set wanted/issued appropriately.
This is incomplete. We really need something to ensure we deal with
replayed requests before new requests are handled.. and on a cluster-wide
basis, since requests may involve slave requests to other mds's.
We also should clean up reconnects unclaimed after all replays are
complete. And somehow inform the client when the cap is officially
nonexistent and EBADF.
Sage Weil [Fri, 6 Mar 2009 22:06:19 +0000 (14:06 -0800)]
mds: include path base ino in client reconnect paths
Sage Weil [Fri, 6 Mar 2009 22:04:19 +0000 (14:04 -0800)]
filepath: no more leading / weirdness
Get rid of weird leading '/' iff ino==1 semantic. Clean up interface a bit.
Sage Weil [Fri, 6 Mar 2009 19:51:39 +0000 (11:51 -0800)]
kclient: drop unused redirty code from set_page_dirty
Sage Weil [Fri, 6 Mar 2009 19:31:12 +0000 (11:31 -0800)]
filestore: lower min_sync_interval
We mainly need something non-zero here to ensure we batch some
requests up when we get a stream of requests with 'startsync'.
Yehuda Sadeh [Fri, 6 Mar 2009 18:29:21 +0000 (10:29 -0800)]
kclient: some bookkeeper fixes
Sage Weil [Fri, 6 Mar 2009 18:16:12 +0000 (10:16 -0800)]
kclient: fix bookkeeper ifdef
Sage Weil [Fri, 6 Mar 2009 05:13:22 +0000 (21:13 -0800)]
kclient: patch series descriptions, updates
Sage Weil [Fri, 6 Mar 2009 04:58:01 +0000 (20:58 -0800)]
kclient: docs
Sage Weil [Fri, 6 Mar 2009 04:57:53 +0000 (20:57 -0800)]
kclient: rename ceph_tools -> bookkeeper
Sage Weil [Fri, 6 Mar 2009 02:01:15 +0000 (18:01 -0800)]
kclient: some caps.c cleanup
Sage Weil [Thu, 5 Mar 2009 19:27:41 +0000 (11:27 -0800)]
mon: tolerate stray victory message (restart the election)
No need to core dump.
Sage Weil [Fri, 6 Mar 2009 00:28:09 +0000 (16:28 -0800)]
kclient: BUG instead of WARN on wanted item on rdcaps list
Sage Weil [Fri, 6 Mar 2009 00:27:47 +0000 (16:27 -0800)]
kclient: cosmetic
Sage Weil [Fri, 6 Mar 2009 00:25:27 +0000 (16:25 -0800)]
kclient: wait for requests to be safe before tearing down dcache
We need to drop dentry refs in kill_sb _before_ kill_anon_super
tears down the dcache. Wait for safe replies for a while, then
drop what's left on the floor.
Sage Weil [Thu, 5 Mar 2009 23:35:39 +0000 (15:35 -0800)]
kclient: don't fall over on !dirty page in invalidatepage
Spit out a warning, in case we end up with writeback accounting
errors. I _think_ it's ok, though.
Yehuda Sadeh [Fri, 6 Mar 2009 00:18:21 +0000 (16:18 -0800)]
kclient: some documentation
Yehuda Sadeh [Fri, 6 Mar 2009 00:14:32 +0000 (16:14 -0800)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Yehuda Sadeh [Fri, 6 Mar 2009 00:14:12 +0000 (16:14 -0800)]
kclient: fix osd timeout handling
Sage Weil [Thu, 5 Mar 2009 23:02:06 +0000 (15:02 -0800)]
kclient: more sanity checks in set_page_dirty
These don't fix anything, just make things break sooner...
Sage Weil [Thu, 5 Mar 2009 22:26:54 +0000 (14:26 -0800)]
kclient: only remember leases on regular dentries
...not for snapdir or snap dentries, which don't have matching
d_release methods.
Sage Weil [Thu, 5 Mar 2009 22:15:55 +0000 (14:15 -0800)]
kclient: fix snap creation/deletion, memory leak
This broke with the new mds request dentry pointers.
Also, clean up kmalloced snap.
Sage Weil [Thu, 5 Mar 2009 22:03:31 +0000 (14:03 -0800)]
kclient: fix snap_realm leak
Separate lookup/create from get (this is safe because we always
hold span_rwsem during these operations).
Yehuda Sadeh [Thu, 5 Mar 2009 21:29:58 +0000 (13:29 -0800)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Yehuda Sadeh [Thu, 5 Mar 2009 21:28:32 +0000 (13:28 -0800)]
kclient: revise osd client timeout mechanism
Sage Weil [Thu, 5 Mar 2009 21:18:45 +0000 (13:18 -0800)]
kclient: simplify invalidatepage; strengthen assertions
Sage Weil [Thu, 5 Mar 2009 19:01:39 +0000 (11:01 -0800)]
kclient: writepages cleanup
Sage Weil [Thu, 5 Mar 2009 18:40:20 +0000 (10:40 -0800)]
synclient: optionally do a sync on object writes; wait for sync, not ack
Sage Weil [Thu, 5 Mar 2009 18:22:03 +0000 (10:22 -0800)]
filestore: enforce min and max sync interval
Ideally the min sync interval should be self-tuning (ala ext3, xfs),
but that can come later.
Sage Weil [Thu, 5 Mar 2009 18:08:31 +0000 (10:08 -0800)]
kclient: dosync in writepages if sync_mode is WB_SYNC_ALL
If we dosync if !pdflush, then we end up doing syncs from
balance_dirty_pages, wreaking havoc on the OSDs.
It's possible the current_is_pdflush() check isn't needed at all..
Sage Weil [Thu, 5 Mar 2009 03:54:37 +0000 (19:54 -0800)]
kclient: fix readpages zeroing bug
Move to the next page after zeroing the first partial page.
Sage Weil [Thu, 5 Mar 2009 00:55:02 +0000 (16:55 -0800)]
kclient: return 0 from readpage on success
Sage Weil [Thu, 5 Mar 2009 00:54:50 +0000 (16:54 -0800)]
kclient: addr debug, cleanup
Sage Weil [Thu, 5 Mar 2009 00:36:26 +0000 (16:36 -0800)]
kclient: fix readpages zeroing
Yehuda Sadeh [Thu, 5 Mar 2009 00:39:18 +0000 (16:39 -0800)]
kclient: reschedule timeout for timeout length if remaining is zero
Yehuda Sadeh [Thu, 5 Mar 2009 00:32:23 +0000 (16:32 -0800)]
kclient: merge conflict
Yehuda Sadeh [Thu, 5 Mar 2009 00:31:27 +0000 (16:31 -0800)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Conflicts:
src/kernel/caps.c
src/kernel/super.h
Yehuda Sadeh [Thu, 5 Mar 2009 00:30:23 +0000 (16:30 -0800)]
kclient: sparse cleanup
Yehuda Sadeh [Thu, 5 Mar 2009 00:09:15 +0000 (16:09 -0800)]
kclient: small cleanup
Sage Weil [Wed, 4 Mar 2009 23:55:32 +0000 (15:55 -0800)]
kclient: fix readpages
do_sync_request returns an error code or bytes read; no need to do
the 0 -> bytes conversion in the caller.
Sage Weil [Wed, 4 Mar 2009 23:43:32 +0000 (15:43 -0800)]
kclient: use helper to clearly define which caps belond on rdcaps
Adjust list membership in ceph_add_cap, release, flush_ack, and
ceph_put_fmode paths. Use a spinlock to protect list adjustments
because ceph_put_fmode does not have the session mutex.
Make __cap_is_valid() to check if we are on the rdcaps list or not.
Remove unneeded extra checks in rdcaps trimmer; replace with a
WARN.
Yehuda Sadeh [Wed, 4 Mar 2009 22:54:05 +0000 (14:54 -0800)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Sage Weil [Wed, 4 Mar 2009 22:35:59 +0000 (14:35 -0800)]
kclient: adjust debug output
Sage Weil [Wed, 4 Mar 2009 22:14:22 +0000 (14:14 -0800)]
kclient: be more careful about putting cap on rdcaps in ceph_add_cap()
Namely, remove it if it doesn't belong there anymore. This needs
to happen when we release caps, too...
Yehuda Sadeh [Wed, 4 Mar 2009 22:13:12 +0000 (14:13 -0800)]
kclient: some cleanup
Sage Weil [Wed, 4 Mar 2009 22:13:04 +0000 (14:13 -0800)]
kclient: do not drop rdcaps that are dirty or flushing
This screws up cap writeback.
Yehuda Sadeh [Wed, 4 Mar 2009 22:12:57 +0000 (14:12 -0800)]
kclient: remove unused code
Sage Weil [Wed, 4 Mar 2009 20:08:21 +0000 (12:08 -0800)]
kclient: user zero_user_segment helper
Sage Weil [Wed, 4 Mar 2009 19:59:44 +0000 (11:59 -0800)]
kclient: avoid zero_user_segment on old kernels
zero_user_segment appeared in 2.6.25.
Sage Weil [Wed, 4 Mar 2009 19:51:07 +0000 (11:51 -0800)]
kclient: fix __read_mostly attributes
Sage Weil [Wed, 4 Mar 2009 19:01:46 +0000 (11:01 -0800)]
kclient: zero trailing pages in readpage(s) on short read or ENOENT
Return number of bytes read + zeroed.
Tighten up assertions a bit.
Sage Weil [Wed, 4 Mar 2009 18:57:58 +0000 (10:57 -0800)]
kclient: mark debug vars __read_mostly
Sage Weil [Wed, 4 Mar 2009 18:57:15 +0000 (10:57 -0800)]
initscript: --btrfs flag
Optionally mount btrfs
Sage Weil [Wed, 4 Mar 2009 17:11:32 +0000 (09:11 -0800)]
mds: flush journal on truncate if readers/writers are waiting
Sage Weil [Wed, 4 Mar 2009 16:51:14 +0000 (08:51 -0800)]
kclient: include ceph_ver.h in kernel patch series, package
Yehuda Sadeh [Wed, 4 Mar 2009 18:34:27 +0000 (10:34 -0800)]
kclient: remove /sys/fs/ceph when doing cleanup
Sage Weil [Wed, 4 Mar 2009 00:59:11 +0000 (16:59 -0800)]
mds: logger typo
Sage Weil [Wed, 4 Mar 2009 00:40:38 +0000 (16:40 -0800)]
cosd: fixes
Sage Weil [Tue, 3 Mar 2009 19:01:24 +0000 (11:01 -0800)]
mkcephfs: bit more verbose
Sage Weil [Wed, 4 Mar 2009 00:45:58 +0000 (16:45 -0800)]
configure: remove some silly checks
Sage Weil [Wed, 4 Mar 2009 00:45:20 +0000 (16:45 -0800)]
makefile: include make_version in dist tarball
Sage Weil [Wed, 4 Mar 2009 00:35:49 +0000 (16:35 -0800)]
todo
Sage Weil [Wed, 4 Mar 2009 00:35:16 +0000 (16:35 -0800)]
mds: fix logger key id; call validate() for all logtypes
Sage Weil [Wed, 4 Mar 2009 00:02:10 +0000 (16:02 -0800)]
config: no more .ceph_monmap; parse cluster.conf for mon addrs
Look for cluster.conf in /etc/ceph/cluster.conf, then
./cluster.conf. Make vstart.sh generate a cluster.conf.
No more .ceph_monmap default.. only look for monmap file is
specified on command line.
Sage Weil [Tue, 3 Mar 2009 23:21:34 +0000 (15:21 -0800)]
todo: cas notes
Sage Weil [Tue, 3 Mar 2009 23:21:19 +0000 (15:21 -0800)]
mds: small cleanups
Sage Weil [Tue, 3 Mar 2009 23:20:27 +0000 (15:20 -0800)]
mds: fix file_eval
Be more aggressive about shifting filelock state to something more
useful. Mainly, do NOT stop everything if there are wrlocks! This
causes all kinds of annoying stalls waiting for the log to flush.
Yehuda Sadeh [Wed, 4 Mar 2009 00:07:10 +0000 (16:07 -0800)]
ceph_ver: write log message whenever a new log is opened
Yehuda Sadeh [Wed, 4 Mar 2009 00:03:57 +0000 (16:03 -0800)]
ceph_ver: add ceph_ver.h to list of .h files
Yehuda Sadeh [Wed, 4 Mar 2009 00:01:48 +0000 (16:01 -0800)]
kclient: write ceph git version when module loads
Yehuda Sadeh [Tue, 3 Mar 2009 23:46:06 +0000 (15:46 -0800)]
ceph_ver: add unconditional rule
Yehuda Sadeh [Tue, 3 Mar 2009 23:37:02 +0000 (15:37 -0800)]
ceph_ver: uncomment important line
Yehuda Sadeh [Tue, 3 Mar 2009 23:32:20 +0000 (15:32 -0800)]
ceph_ver: show the git sha1 version in the logs
Yehuda Sadeh [Tue, 3 Mar 2009 23:23:12 +0000 (15:23 -0800)]
Merge branch 'unstable' of ssh://ceph.newdream.net/git/ceph into unstable
Yehuda Sadeh [Tue, 3 Mar 2009 22:35:56 +0000 (14:35 -0800)]
kclient: wakeup wq waiting on caps on write
Sage Weil [Tue, 3 Mar 2009 22:27:15 +0000 (14:27 -0800)]
kclient: some fmode cleanup
BUG instead of WARN. Rename field.
Sage Weil [Tue, 3 Mar 2009 22:26:28 +0000 (14:26 -0800)]
kclient: fix fmode ref count buf
We weren't taking an fmode reference on the root dir. This would
cause a WARN if our root caps were allowed to expire and we had
to ropen them. Duplicated wonky r_fmode conditional from later
fill_inode() call in fill_trace().
Sage Weil [Tue, 3 Mar 2009 20:28:11 +0000 (12:28 -0800)]
kclient: wake up i_cap_wq in ceph_add_cap
We could a thread blocked on caps when we are issued a new cap in
another thread. They need to be woken up.
Sage Weil [Tue, 3 Mar 2009 20:27:31 +0000 (12:27 -0800)]
kclient: check d_inode before clearing its COMPLETE flag.
Sage Weil [Tue, 3 Mar 2009 20:25:58 +0000 (12:25 -0800)]
mds: fix try_remove_unlinked_dn
Just because a dir is_complete() doesn't mean that we can safely
do away with a null dentry. What matters is if the dentry existed
before the last commit.. and having everything in our cache isn't
relevant at all there. If we are complete, a fetch() would be a
no-op, but since we can clear complete at any time, that's not
helpful.
Sage Weil [Tue, 3 Mar 2009 19:43:47 +0000 (11:43 -0800)]
kclient: do not return new null dentry from lookup
The VFS doesn't appear to like it when lookup returns a replacement
dentry that is null... it BUGs on unmount with a ref count that is
still 1.
So, if we get ENOENT and no trace, return ERR_PTR(-ENOENT) instead.
Sage Weil [Tue, 3 Mar 2009 19:33:42 +0000 (11:33 -0800)]
cfuse: fix startup
Yehuda Sadeh [Tue, 3 Mar 2009 18:43:35 +0000 (10:43 -0800)]
conf: some cleanup
Yehuda Sadeh [Tue, 3 Mar 2009 18:41:21 +0000 (10:41 -0800)]
conf: some more fixes (again)
Yehuda Sadeh [Tue, 3 Mar 2009 18:40:15 +0000 (10:40 -0800)]
conf: fix typo
Sage Weil [Tue, 3 Mar 2009 18:42:10 +0000 (10:42 -0800)]
mds: clean up auth_pin nesting code a bit
Sage Weil [Tue, 3 Mar 2009 18:32:50 +0000 (10:32 -0800)]
mkcephfs: take optional crush map on command line
Yehuda Sadeh [Tue, 3 Mar 2009 18:37:29 +0000 (10:37 -0800)]
conf: lockdep again
Yehuda Sadeh [Tue, 3 Mar 2009 18:17:13 +0000 (10:17 -0800)]
conf: fix some other errors
Yehuda Sadeh [Tue, 3 Mar 2009 18:14:17 +0000 (10:14 -0800)]
conf: the return of --lockdep, other fixes