]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Yehuda Sadeh [Thu, 10 Jun 2010 23:01:34 +0000 (16:01 -0700)]
rgw: put_obj uses write_full
Yehuda Sadeh [Thu, 10 Jun 2010 21:42:11 +0000 (14:42 -0700)]
testradospp: add a test for librados trunc
Yehuda Sadeh [Thu, 10 Jun 2010 21:41:53 +0000 (14:41 -0700)]
librados: add trunc operation
Yehuda Sadeh [Thu, 10 Jun 2010 21:12:20 +0000 (14:12 -0700)]
testradospp: update test, use write_full
Yehuda Sadeh [Thu, 10 Jun 2010 21:04:07 +0000 (14:04 -0700)]
librados: implement write_full
Yehuda Sadeh [Thu, 10 Jun 2010 18:54:23 +0000 (11:54 -0700)]
osd: fix null dereference in dout
Sage Weil [Thu, 10 Jun 2010 17:17:21 +0000 (10:17 -0700)]
mds: be extra careful
Probably implied by the refs == 0, but be explicit anyway.
Sage Weil [Thu, 10 Jun 2010 17:17:03 +0000 (10:17 -0700)]
mds: fix another instance of bad stray trimming code to use proper helper
Greg Farnum [Thu, 10 Jun 2010 16:10:34 +0000 (09:10 -0700)]
mds: remove erroneous return statement
Greg Farnum [Thu, 10 Jun 2010 16:05:29 +0000 (09:05 -0700)]
mds: a checker function like _is_dir_nonempty should only report results.
Remove the call to reply_request and fix up the callers that need it.
Sage Weil [Thu, 10 Jun 2010 05:17:22 +0000 (22:17 -0700)]
mds: handle dup anchorclient ACKs gracefully
One recovery scenario sees crisscrossing 'agree' and 'commit':
C->S --> commit1
S->C --> agree
C<-S <== agree
C->S --> commit2 .. client resends commit!
S<-C <== commit1
S->C --> ack .. server resends ack!
S<-C <== commit2
S->C --> ack .. server resends ack!
C<-S <== commit1
client journals ack
C<-S <== commit2
client should ignore dup ack ***
*** but doesn't, because the 'remove from committing list' bit above was
never in the code, even as far back as v0.4 (just the comment). Instead,
the map was getting fixed up in the _logged_ack() completion. Move it
up here instead, where it belongs!
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Wed, 9 Jun 2010 21:43:32 +0000 (14:43 -0700)]
mds: fix export caps
Don't mangle sessionmap version by incrementing twice; pv is returned by
prepare_force_open_sessions().
Sage Weil [Wed, 9 Jun 2010 21:06:27 +0000 (14:06 -0700)]
mds: allow import caps pin from multiple replicas at once
e.g. on dentry unlink, replicas send their caps to the auth. There may
be multiple replicas.
Yehuda Sadeh [Wed, 9 Jun 2010 20:49:54 +0000 (13:49 -0700)]
osd: return EINVAL when snapc is invalid and trying to write
Sage Weil [Wed, 9 Jun 2010 20:30:30 +0000 (13:30 -0700)]
osd: don't clobber racing pg_info on active pg
Wido saw a pg go active, but an activate log+info update crossed paths with
a pg_notify info, and the primary overwrote it's updated shiny new info
with the stale old info from the replica. Don't do that. It causes
problems down the line. In this case, we got
osd/OSD.cc: In function 'void OSD::generate_backlog(PG*)':
osd/OSD.cc:3863: FAILED assert(!pg->is_active())
1: (ThreadPool::worker()+0x28f) [0x5b08ff]
2: (ThreadPool::WorkThread::entry()+0xd) [0x4edb8d]
3: (Thread::_entry_func(void*)+0xa) [0x46892a]
4: (()+0x69ca) [0x7f889ff249ca]
5: (clone()+0x6d) [0x7f889f1446cd]
on the replica because it was active but the primary was restarting peering
due to the bad info.
Sage Weil [Wed, 9 Jun 2010 19:29:03 +0000 (12:29 -0700)]
osd: refine SnapContext::is_valid()
Sage Weil [Wed, 9 Jun 2010 19:27:55 +0000 (12:27 -0700)]
osd: implement SnapContext::is_valid()
Sage Weil [Wed, 9 Jun 2010 17:56:17 +0000 (10:56 -0700)]
init-ceph: don't fail if we fail to umount
It may not be mounted to begin with.
Sage Weil [Wed, 9 Jun 2010 05:34:57 +0000 (22:34 -0700)]
osd: add --dump-pg-log command
Sage Weil [Wed, 9 Jun 2010 04:36:56 +0000 (21:36 -0700)]
osd: clear newly_removed_snaps on osdmap update if unchanged
We only want to apply _newly_ removed snaps once, or else we try to trim
the same snaps multiple times, and crash like so
./include/interval_set.h: In function 'void interval_set<T>::insert(T, T) [with T = snapid_t]':
./include/interval_set.h:202: FAILED assert(0)
1: (interval_set<snapid_t>::insert(snapid_t, snapid_t)+0x12c) [0x6b1728]
2: (interval_set<snapid_t>::insert(snapid_t)+0x2f) [0x6b195d]
3: (ReplicatedPG::snap_trimmer()+0x1c02) [0x66d5d6]
4: (OSD::SnapTrimWQ::_process(PG*)+0x24) [0x6dc2ac]
5: (ThreadPool::WorkQueue<PG>::_void_process(void*)+0x28) [0x6fa28a]
6: (ThreadPool::worker()+0x23a) [0x7f57a4]
7: (ThreadPool::WorkThread::entry()+0x19) [0x73e9b1]
8: (Thread::_entry_func(void*)+0x20) [0x6508a4]
9: /lib/libpthread.so.0 [0x7fa2707dc73a]
10: (clone()+0x6d) [0x7fa26fa0669d]
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Tue, 8 Jun 2010 23:43:29 +0000 (16:43 -0700)]
mds: use helper to send message to client; fix send to null connection
Sometimes session->connection is NULL; use session->inst in that case.
Greg Farnum [Tue, 8 Jun 2010 23:33:59 +0000 (16:33 -0700)]
mds: remove erroneous bracket
Greg Farnum [Mon, 7 Jun 2010 17:04:33 +0000 (10:04 -0700)]
add checks for being a snapshot root to dir_is_nonempty
Sage Weil [Tue, 8 Jun 2010 22:36:55 +0000 (15:36 -0700)]
throttle: allow take(0)
Sage Weil [Tue, 8 Jun 2010 21:46:30 +0000 (14:46 -0700)]
mds: fix stale lease trimming xlist iterator abuse
Greg Farnum [Tue, 8 Jun 2010 19:10:47 +0000 (12:10 -0700)]
mon: fix memory-leaked messages
Greg Farnum [Mon, 7 Jun 2010 21:57:47 +0000 (14:57 -0700)]
buffer: fix padding distances
Greg Farnum [Mon, 7 Jun 2010 19:01:34 +0000 (12:01 -0700)]
osd: init auid to CEPH_AUTH_UID_DEFAULT in case authorizer doesn't set it.
We should probably also require the authorizer to set it for us.
Sage Weil [Tue, 8 Jun 2010 20:38:15 +0000 (13:38 -0700)]
mds: scan stray dir, eval strays on mds startup
Sage Weil [Tue, 8 Jun 2010 16:42:40 +0000 (09:42 -0700)]
mon: make mon lease clock check protocol change backward compatible
Sage Weil [Tue, 8 Jun 2010 05:18:59 +0000 (22:18 -0700)]
qa: add untar_snap_rm.sh
Sage Weil [Tue, 8 Jun 2010 05:05:08 +0000 (22:05 -0700)]
osd: print rollback osd_op nicely
Sage Weil [Mon, 7 Jun 2010 23:03:37 +0000 (16:03 -0700)]
mds: wire Connection to Session when Session already exists on connect
Sage Weil [Mon, 7 Jun 2010 23:03:14 +0000 (16:03 -0700)]
mds: funnel mds->client messages through single Session* helper
Simplify callers where possible.
Sage Weil [Mon, 7 Jun 2010 22:42:19 +0000 (15:42 -0700)]
mon: simplify clock drift checks
Ignore lease sent vs lease_ack receive times bc multiple lease msgs may
be in flight and the ack may be from a previous one. This was causing
spurious
[WRN] : lease_ack from follower sent at time(10.06.07_15:07:11.441391), before lease extend was sent (10.06.07_15:07:11.826340)! Clocks not synchronized.
messages.
It is sufficient to just check for messages received from the future. To
avoid cruftiness trying to do that when the only stamp is the lease
timeout, add a sent_timestamp to the message and use that instead. This
simplifies things quite a bit, at the expense of not being backward
compatible.
Sage Weil [Mon, 7 Jun 2010 22:04:37 +0000 (15:04 -0700)]
monc: behave in ms_handle_reset if cur_mon is < 0
Sage Weil [Mon, 7 Jun 2010 22:03:28 +0000 (15:03 -0700)]
msgr: don't throttle.get 0
Sage Weil [Mon, 7 Jun 2010 22:00:23 +0000 (15:00 -0700)]
throttle: allow put(0)
Still returns a consistent value for the count.
Sage Weil [Mon, 7 Jun 2010 21:59:16 +0000 (14:59 -0700)]
msgr: don't thottle.put 0
Sage Weil [Mon, 7 Jun 2010 21:47:56 +0000 (14:47 -0700)]
Merge remote branch 'origin/msgr' into unstable
Sage Weil [Mon, 7 Jun 2010 19:05:55 +0000 (12:05 -0700)]
mds: use cap on head if there is none on the snapped inode
This is needed, in particular, when we're flushing snap data on an inode
that already got COWed.
Sage Weil [Mon, 7 Jun 2010 18:40:32 +0000 (11:40 -0700)]
osd: use low-level helper getting obc in sub_op_push
find_object_context does all sorts of stuff we don't need here: we know
which object the context is for. Just set it up.
Greg Farnum [Mon, 7 Jun 2010 12:55:05 +0000 (05:55 -0700)]
throtle: add asserts on max and change parameters where appropriate
Greg Farnum [Mon, 7 Jun 2010 12:54:47 +0000 (05:54 -0700)]
throttle: fix assert count to actually use count
Sage Weil [Mon, 7 Jun 2010 05:14:50 +0000 (22:14 -0700)]
crypto: don't clean up EVP table on every decrypt()
Don't think that's appropriate? And certainly doesn't happen for the
encrypt() case.
Sage Weil [Mon, 7 Jun 2010 05:14:21 +0000 (22:14 -0700)]
crypto: don't leak memory in CryptoAES::encrypt()
Sage Weil [Mon, 7 Jun 2010 05:04:29 +0000 (22:04 -0700)]
mon: don't leak MAuth
Greg Farnum [Sat, 5 Jun 2010 00:01:32 +0000 (17:01 -0700)]
throttle: use signed counters and assert that count never drops below 0
Greg Farnum [Fri, 4 Jun 2010 23:19:00 +0000 (16:19 -0700)]
msgr: Fix uses of get_[data, payload, middle] to use throttling-aware functions.
Greg Farnum [Fri, 4 Jun 2010 22:01:30 +0000 (15:01 -0700)]
msgr: put throttler usage on Message destruct
Greg Farnum [Fri, 4 Jun 2010 21:02:56 +0000 (14:02 -0700)]
osd: fix compile issues
Greg Farnum [Fri, 4 Jun 2010 21:02:40 +0000 (14:02 -0700)]
msgr: switch to get/set functions for Message:throttler
Greg Farnum [Fri, 4 Jun 2010 20:37:25 +0000 (13:37 -0700)]
osd: add osd_client_message_size_cap option to config; default 500MB
And change the name in cosd to be that
Sage Weil [Fri, 4 Jun 2010 23:31:42 +0000 (16:31 -0700)]
objectcacher: add verify_stats() debugging helper
Sage Weil [Fri, 4 Jun 2010 23:31:08 +0000 (16:31 -0700)]
objectcacher: fix stat accounting when resizing bufferheads
Must keep stats in mind when adjusting bufferheads!
Sage Weil [Fri, 4 Jun 2010 23:10:44 +0000 (16:10 -0700)]
objectcacher: cleanup formatting
Sage Weil [Fri, 4 Jun 2010 23:05:55 +0000 (16:05 -0700)]
objectcacher: fix use of invalid iterator in map_write()
The p points to bh, which is removed by merge_left. Move it back to final,
so we can advance to the new next a few lines down.
Sage Weil [Fri, 4 Jun 2010 21:26:05 +0000 (14:26 -0700)]
objectcacher: match states before merging in map_write
The caller is going to set us to dirty, so we don't care what state we
have, so long as the left and right bits we're merging match all is ok.
Yehuda Sadeh [Fri, 4 Jun 2010 23:20:35 +0000 (16:20 -0700)]
osd: fix rollback when head points at the rolled back snapshot
Greg Farnum [Fri, 4 Jun 2010 20:23:57 +0000 (13:23 -0700)]
msg: remove copy_payload and copy_data functions; change set to use throttler
Sage Weil [Fri, 4 Jun 2010 20:10:28 +0000 (13:10 -0700)]
Merge branch 'rbd' into unstable
Sage Weil [Fri, 4 Jun 2010 20:09:42 +0000 (13:09 -0700)]
osd: clean up rollback debug output
Sage Weil [Fri, 4 Jun 2010 19:54:25 +0000 (12:54 -0700)]
uclient: handle inode with no caps from mds
This happens when you readdir and some inodes are in a different snaprealm.
Greg Farnum [Fri, 4 Jun 2010 19:57:59 +0000 (12:57 -0700)]
osd: filter_xattrs on a rollback op
Greg Farnum [Fri, 4 Jun 2010 19:55:27 +0000 (12:55 -0700)]
osd: fix naughty iterator usage after invalidating it
Greg Farnum [Fri, 4 Jun 2010 19:19:06 +0000 (12:19 -0700)]
osd: _make_clone now properly duplicates xattrs
Greg Farnum [Fri, 4 Jun 2010 19:49:20 +0000 (12:49 -0700)]
osd: add filter_xattrs function to remove non-user xattrs from a map of them
Greg Farnum [Fri, 4 Jun 2010 19:04:08 +0000 (12:04 -0700)]
progress
Sage Weil [Fri, 4 Jun 2010 18:07:09 +0000 (11:07 -0700)]
mds: fix straydn->first part deux
9ed0c30ecf6611193db52e1facc1f46b37f04bc4 forgot to remove the old code.
Greg Farnum [Fri, 4 Jun 2010 01:22:24 +0000 (18:22 -0700)]
debugging output
Greg Farnum [Fri, 4 Jun 2010 01:22:18 +0000 (18:22 -0700)]
rados: print out pool instead of object
Sage Weil [Fri, 4 Jun 2010 00:32:39 +0000 (17:32 -0700)]
mds: only purge dentries with no extra refs (besides dirty)
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Fri, 4 Jun 2010 00:30:41 +0000 (17:30 -0700)]
mds: set straydn first to match inode on unlink
Sage Weil [Fri, 4 Jun 2010 00:26:11 +0000 (17:26 -0700)]
mds: don't export stray (~mdsfoo/stray), and ignore in balancer
We _must_ keep mdsdir and stray on local mds for normal operations.
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Thu, 3 Jun 2010 23:55:14 +0000 (16:55 -0700)]
mds: make discover work for multiversion inodes (e.g. dirs)
If we don't have the specific snap, look up the head and see if it's
multiversion.
This doesn't give us a "range" lookup like we get with dentries because
the inode_map is a hash, not a map. However, we shouldn't need it,
because we always have a specific snapped inode we're looking for (because
it is refered to by a dentry) or we are looking at a multiversion head.
Sage Weil [Thu, 3 Jun 2010 23:19:36 +0000 (16:19 -0700)]
mds: fix CDir::take_sub_waiting vs dnwaiter pin
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Thu, 3 Jun 2010 23:09:23 +0000 (16:09 -0700)]
mds: kill open_foreign_stray; but open remote mdsdirs instead
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Thu, 3 Jun 2010 22:00:44 +0000 (15:00 -0700)]
mds: fix cap clone logic to look at matching first, not last
The cap->client_follows is set to follows+1 by flushsnap, since the real
follows value isn't convenient. But it is enough to know that it is more
than the old version's follows, so do that.
Yehuda Sadeh [Thu, 3 Jun 2010 23:44:31 +0000 (16:44 -0700)]
libatomic: fix assert.h compilation
Greg Farnum [Thu, 3 Jun 2010 23:40:32 +0000 (16:40 -0700)]
msgr: add Throttle pointer to Policy
Greg Farnum [Thu, 3 Jun 2010 23:20:47 +0000 (16:20 -0700)]
Merge branch 'unstable' into msgr
Greg Farnum [Thu, 3 Jun 2010 20:54:24 +0000 (13:54 -0700)]
osd: make sure we don't return EAGAIN to client
Sage Weil [Thu, 3 Jun 2010 21:14:04 +0000 (14:14 -0700)]
mds: open past snap parents at end of rejoin phase
We really need past parents open before we go active or else anything
that needs to build a snap context will fail.
Sage Weil [Thu, 3 Jun 2010 20:48:10 +0000 (13:48 -0700)]
mdsmap: show individual mds states in summary
Sage Weil [Thu, 3 Jun 2010 20:26:39 +0000 (13:26 -0700)]
osd: improve snap_trimmer debug output
Sage Weil [Thu, 3 Jun 2010 20:24:48 +0000 (13:24 -0700)]
mds: another cap_exports message/mdcache encoding fix
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Thu, 3 Jun 2010 20:08:16 +0000 (13:08 -0700)]
mds: only adjust dn->first on lock msg if !multiversion
The multiversion dn->first references a range of inode versions; don't
drag it forward. Fixes
38cb2403c043e6676b563197d086edeb11b71ddf .
Signed-off-by: Sage Weil <sage@newdream.net>
Sage Weil [Thu, 3 Jun 2010 19:03:23 +0000 (12:03 -0700)]
mds: more fix cap_exports typing
Sage Weil [Thu, 3 Jun 2010 18:59:50 +0000 (11:59 -0700)]
mds: fix scatter_nudge infinite loop
Sage Weil [Thu, 3 Jun 2010 18:08:00 +0000 (11:08 -0700)]
mds: fix ESessions type
Sage Weil [Thu, 3 Jun 2010 18:04:05 +0000 (11:04 -0700)]
mds: drag in->first forward with straydn in handle_dentry_unlink
Sage Weil [Thu, 3 Jun 2010 17:38:56 +0000 (10:38 -0700)]
mds: fix anchorclient dup lookups, again
Sage Weil [Thu, 3 Jun 2010 17:17:37 +0000 (10:17 -0700)]
mds: only log successful requests as completed
Sage Weil [Thu, 3 Jun 2010 17:09:19 +0000 (10:09 -0700)]
mds: anchor dir on mksnap
CC Lien [Thu, 3 Jun 2010 16:45:10 +0000 (09:45 -0700)]
mkcephfs: error when creating journal file in a directory that differs from OSD data dir
mkcephfs creates osd data directory automatically, but it doesn't create a
directory for the osd journal file.
When you have a journal file in a directory that differs from the osd data
directory in your configuration, like:
osd data = /osd/osd$id
osd journal = /journal/osd$id
You will receive a "mount failed to open journal /journal/osd0/journal: No
such file or directory" error when doing mkcephfs
Signed-off-by: CC Lien <cc_lien@tcloudcomputing.com>
Sage Weil [Thu, 3 Jun 2010 16:40:57 +0000 (09:40 -0700)]
mds: fix mismatched cap_exports type between msg and MDCache
The types need to match because they are encoded/decoded interchangeably.
See MMDSCacheRejoin::decode() and MDCache::rejoin_send_rejoins().
Sage Weil [Thu, 3 Jun 2010 16:33:27 +0000 (09:33 -0700)]
mds: fix trim_unlinked iterator badness
We may remove the next inode in the map. Queue up unlinked roots first,
which we know remove_inode_recursive() won't reach, and iterate over those.
Sage Weil [Thu, 3 Jun 2010 16:28:15 +0000 (09:28 -0700)]
mds: define MDS_REF_SET in unstable
Sage Weil [Thu, 3 Jun 2010 16:27:56 +0000 (09:27 -0700)]
mds: clear dirtyscattered in remove_inode()
Sage Weil [Thu, 3 Jun 2010 16:17:13 +0000 (09:17 -0700)]
mds: allow dup lookups in anchorclient
It's not practical for callers to avoid dups, particularly since they may
be unaware of each other. And it's trivial to support it here.