If the mds is in clientreplay, a session close
request needs to be delayed until it reaches
active. Otherwise, the session state gets set to
'closing', and the replay requests get dropped on the
floor.
Fixes#4564.
Signed-off-by: Sam Lang <sam.lang@inktank.com>
If the mds is standby, when a segment is trimmed, we need
to clear the backtrace updates list to avoid the following
assertion when the segment is deleted.
./include/elist.h: 92: FAILED assert(_head.empty())
ceph version 0.59-478-g8befbca (8befbca77a)
(MDLog::standby_trim_segments()+0xce5) [0x6ccec5]
(MDS::C_MDS_StandbyReplayRestartFinish::finish(int)+0x39) [0x4e86b9]
(Journaler::_finish_reprobe(int, unsigned long, Context*)+0x190)
[0x6d3210]
(Filer::_probed(Filer::Probe*, object_t const&, unsigned long,
utime_t)+0x558) [0x704a88]
(Objecter::C_Stat::finish(int)+0xc0) [0x705900]
(Objecter::handle_osd_op_reply(MOSDOpReply*)+0xe38) [0x6f1df8]
(MDS::handle_core_message(Message*)+0xae8) [0x4dc318]
(MDS::_dispatch(Message*)+0x2f) [0x4dc4df]
(MDS::ms_dispatch(Message*)+0x1db) [0x4ddf7b]
(DispatchQueue::entry()+0x341) [0x81f561]
(DispatchQueue::DispatchThread::entry()+0xd) [0x79c6ad]
(()+0x7e9a) [0x7f346bb9ee9a]
(clone()+0x6d) [0x7f346a3574bd]
Fixes#4539.
Signed-off-by: Sam Lang <sam.lang@inktank.com>
Otherwise, we update the stat.stat structure, but not the
stat.invalid_stats part. This will result in a recently
split primary propogating the invalid stats but not the
invalid marker. Sending the whole pg_stat_t structure
also mirrors MOSDSubOp.
Fixes: #4557
Backport: bobtail
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Always increment the iterator when adding old pools
to the backtrace. This fixes a bug on files where
the layout had been set to a different pool and then
back to the same pool, causing continuous looping in
the build_backtrace() function.
Fixes#4537.
Signed-off-by: Sam Lang <sam.lang@inktank.com>
Fixes: #4508
Backport: bobtail
On certain bucket index operations we didn't check whether
the bucket was a system bucket, which caused the operations
to fail. This triggered an error message on bucket removal
operations.
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
On rename, only set the other inode if the
lookup for the destination succeeds, otherwise we hit
a segv in set_other_inode().
Fixes#4517.
Signed-off-by: Sam Lang <sam.lang@inktank.com>
Tested-by: Noah Watkins <jayhawk@cs.ucsc.edu>
Introduced by fc80c1dc6e,
the client should _not_ fail if the lookup for the
destination path on rename returns ENOENT.
The previous code also did not check that the lookup
returned ENOENT or success. We add the check and fail
if we get any other errors.
Fixes#4517.
Signed-off-by: Sam Lang <sam.lang@inktank.com>
We should clobber configuration and log data, but *not* user data. Leave
/var/lib/ceph alone.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Gary Lowell <gary.lowell@inktank.com>
This was a hack for dev purposes ages ago; remove it. The predictable
filename is a security issue.
CVE-2013-1882
Reported-by: Michael Scherer <misc@zarb.org>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
The predictable file name is a security problem.
CVE-2013-1882
Reported-by: Michael Scherer <misc@zarb.org>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
The predictable file name is a security problem.
CVE-2013-1882
Reported-by: Michael Scherer <misc@zarb.org>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
This block of code tries to limit the number of aios in flight by waiting
for the amount of data to be written to grow relative to a function of the
number of aios. Strictly speaking, the condition we are waiting for is a
function of both aio_num and the write queue, but we are only woken by
changes in aio_num, and were (in rare cases) waiting when aio_num == 0 and
there was no possibility of being woken.
Fix this by verifying that aio_num > 0, and restructuring the loop to
recheck that condition on each wakeup.
Fixes: #4079
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
By not deleting and setting NULL the session's auth_handler, we could
hit a scenario in which we'd end up dispatching a previously-wait-listed
auth message and we wouldn't start its auth session.
This only happened when increasing max_global_id via Paxos (in which case
we would wait-list the message) and would only be noticeable when running
with cephx disabled.
Fixes: #4519
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
Being unable to run a ::statfs() may be a symptom of something bigger.
We want to cleanly shutdown the monitor ASAP if such thing happens.
Fixes: #4509
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
Caused by 3bd48cbbad
feature 4207 implementation
Signed-off-by: David Zafman <david.zafman@inktank.com>
Reviewed-by: Gary Lowell <gary.lowell@inktank.com>
(cherry picked from commit e1e2d5d217)
Older guests may not send flushes properly (i.e. never), so if this is
enabled, rbd_cache=true is safe for them transparently.
Disable by default, since it will unnecessarily slow down newer guest
boot, and prevent writeback caching for things that don't need to send
flushes, like the command line tool.
Refs: #3817
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This ensures that the paxos state is not active when the PaxosService
restart() methods run right afterwards, and that EAGAIN waiters will get
requeued appropriately.
Signed-off-by: Sage Weil <sage@inktank.com>
If 7aec13f749 we started passing non-zero
return values to these completions; now we have to deal with them
accordingly.
RetryMessage behaves just like the Monitor variant.
Propose and Committed update state but otherwise ignore non-zero
return values.
Signed-off-by: Sage Weil <sage@inktank.com>
