kv/MemDB: fix wrong output target and add sanity checks
Mark's Comments:
This passed Jenkins checks.
This passed "ceph_test_objectstore --gtest_filter=*/2".
This PR did not appear to have a significant impact on performance tests.
Reviewed-by: Mark Nelson <mnelson@redhat.com>
os/bluestore: add a boundary check of cache read
Mark's Comments:
This passed Jenkins checks.
This passed "ceph_test_objectstore --gtest_filter=*/2".
This PR did not appear to have a significant impact on performance tests.
Reviewed-by: Mark Nelson <mnelson@redhat.com>
if the realm is deleted while the gateway has a watch, it's disconnected
and watch_restart() will fail. this results in a watch handle of 0,
which leads to a segfault on ~RGWRealmWatcher when we pass it to
unwatch()
this commit cleans up the watch when watch_restart() fails, so we don't
try to unwatch() on destruction
Fixes: http://tracker.ceph.com/issues/16817
Signed-off-by: Casey Bodley <cbodley@redhat.com>
Also change bdev of extent from uint16_t to uint8_t, which matches
the fnode prefer_bdev field better, and is good for efficiency.
Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>
return error code "NoSuchLifecycleConfiguration" if lifecycle configuration does not exist
as AmazonS3 latest API does.
Signed-off-by: weiqiaomiao <wei.qiaomiao@zte.com.cn>
When you execute the command gettorrent of a object, a torrent file will be produced and returned.
The torrent also will be save into a pool named default.rgw.torrent.
If the torrent of a object exists in default.rgw.torrent, it will be returned.
Signed-off-by: zhouruisong <236131368@qq.com>
introduce RBD_SNAP_REMOVE_FLATTEN for flags of Image::snap_remove2()
to auto flatten the children of snapshot which we want to remove.
Signed-off-by: Dongsheng Yang <dongsheng.yang@easystack.cn>
currently, we only have one api for snap_remove, but if we want to
pass more options about snapshot removal, that's impossible.
This patch introduce a new api of snap_remove2 here to solve
this problem.
Signed-off-by: Dongsheng Yang <dongsheng.yang@easystack.cn>
msgr is part of libcommon. and the later is linked into both libcephfs
and librados. and hence into cephfs.so and rados.so. that's why we have
double free.
Fixes: http://tracker.ceph.com/issues/16686
Signed-off-by: Kefu Chai <kchai@redhat.com>
This function assumes that the source path
does not point to a null dentry, and that
path_traverse is therefore setting the
passed inode.
Candidate code path for crash
in http://tracker.ceph.com/issues/16807
Signed-off-by: John Spray <john.spray@redhat.com>
ReplicatedPG::new_repop() returns a pointer to RepGather with two refcounts,
one is held by ReplicatedPG::repop_queue, the other is supposed to be
held by the caller of this function. but it's caller
ReplicatedPG::submit_log_entries() assigns it to a
boost::intrusive_ptr<RepGather>() directly, why by default add_ref() in
its constructor. this makes the refcount 3. that's why we have a leak of
RepGather in `ReplicatedPG::new_repop(ObcLockManager&&,
boost::optional<std::function<void ()>>&&)`.
Fixes: http://tracker.ceph.com/issues/16801
Signed-off-by: Kefu Chai <kchai@redhat.com>
this behaviour matches the autotools. and we should not enable profiling
unless asked to do so.
Fixes: http://tracker.ceph.com/issues/16804
Signed-off-by: Kefu Chai <kchai@redhat.com>
