os/bluestore: assert shared blob cache cleared on split

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: dump some stats after fsck

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: instrument onode reshard events

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: instrument transaction count

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: instrument big/small writes

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: make blob_t unused helpers use logical length

These were taking min_alloc_size, but this can change
across mounts; better to use the logical blob length
instead (that's what we want anyway!).

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: use block_size for allocator unit

We need to handle objects written during previous mounts
that may have had a smaller min_alloc_size.  Use
block_size, which is a safe lower bound.

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: fix fsck used_block bitmap

This has to be block_size bits because min_alloc_size
can vary over mounts.

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: optimize compress_extent_map

Only examine the range we just wrote to (and to the left
and right).

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: make nid and blobid allocation less racy

We could bump the _max value for a TransContext in it's
prepare state, have it wait for a long time on IO, and
let another txc allocate and commit something with
an id higher than the previous max.

Fix this first by pushing the max ids into the
TransContext where we can deal with them at commit time,
and then making _kv_sync_thread bump the committed
max in a safe way.

Note that this will need to change if/when we do
these commits in parallel.

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: shard extent map

Rewrote much of the persistence of onode metadata.  The
highlights:

 - extents and blobs stored together (the blob with the
   first referencing extent).
 - extents sharded across multiple k/v keys
 - if a blob if referenced from multiple blobs, it's
   stored in the onode key (called a "spanning blob").
 - when we clone a blob we copy the metadata, but mark
   it shared and put (just) the ref_map on the underlying
   blocks in a shared_blob key.  at this point we also
   assign a globally unique id (sbid = shared blob id)
   so the key has a unique name.
 - we instantiate a SharedBlob in memory regardless of
   whether we need to load the ref_map (which is only
   needed for deallocations!).  the BufferSpace is
   attached to this SharedBlob so we get unified caching
   across clones.

Signed-off-by: Sage Weil <sage@redhat.com>

ceph_test_objectstore: test shards for longer

Signed-off-by: Sage Weil <sage@redhat.com>

ceph_test_objectstore: occasional umount/fsck/mount

Signed-off-by: Sage Weil <sage@redhat.com>

ceph_test_objectstore: add SyntheticMatrixSharding

Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #10870 from ceph/wip_build_cls_rbd

cmake: add dependency from ceph_smalliobenchrbd to cls libraries

Reviewed-by: Jason Dillaman <jdillama@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>

Merge pull request #10912 from stiopaa1/journal_journaltimer_moveCRemoveSet

journal: move JournalTrimmer::C_RemoveSet struct

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

Add dependencies from ceph_smalliobenchrbd on cls_rbd, cls_journal,
and cls_lock.

Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>

Merge pull request #10984 from ceph/wip-kfsx-exclusive-lock

test/librbd/fsx: enable exclusive-lock feature in krbd mode

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #10964 from liewegas/wip-bluefs-compaction

os/bluestore/BlueFS: fix async compaction logging bug

Merge pull request #10966 from liewegas/wip-bluefs-runway

os/bluestore/BlueFS: factor unflushed log into runway calculation

Reviewed-by: Varada Kari <varada.kari@sandisk.com>
Reviewed-by: Somnath Roy <somnath.roy@sandisk.com>

Merge pull request #10970 from xiexingguo/xxg-wip-honour-alloweio-flag

os/bluestore: honour allow-eio flag; use global compressor if possible

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #10974 from dillaman/wip-17210

librbd: potential double-unwatch of watch handle upon error

Reviewed-by: Mykola Golub <mgolub@mirantis.com>

Merge pull request #10774 from renhwztetecs/renhw-wip-cephfs-python-umount

libcephfs: add unmount function in cephfs.pyx

Reviewed-by: John Spray <john.spray@redhat.com>
Reviewed-by: Ramana Raja <rraja@redhat.com>

Merge pull request #10914 from batrick/mds-no-maxmds

mds: remove max_mds config option

Reviewed-by: John Spray <john.spray@redhat.com>

Merge pull request #10922 from ceph/wip-jlayton-statx

First pile of statx patches

test/cephfs: add mount&&unmount test

Signed-off-by: huanwen ren <ren.huanwen@zte.com.cn>

Merge pull request #10981 from SUSE/wip-fix-install-deps

build/ops: fix /etc/os-release parsing in install-deps.sh

Reviewed-by: Jan Fajerski <jfajerski@suse.com>
Reviewed-by: Loic Dachary <ldachary@redhat.com>

build/ops: fix /etc/os-release parsing in install-deps.sh

85a370e35fc42031a7f7e24dea9e50a649c0f309 introduced the DISTRO variable whose
value was obtained by parsing /etc/os-release like this:

DISTRO=$(grep  "^ID=" /etc/os-release | sed "s/ID=//")

This unfortunately picks up the double-quotes, so on a CentOS system DISTRO
will be equal to '"centos"'.

Signed-off-by: Nathan Cutler <ncutler@suse.com>

os/bluestore: use global compressor if possible

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

Merge pull request #10969 from xiexingguo/xxg-wip-fix-cache-logger

os/bluestore: replace store with logger in Cache

Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>

Merge pull request #9592 from trociny/wip-15632

librbd: API methods to directly acquire and release the exclusive lock

Reviewed-by: Jason Dillaman <dillaman@redhat.com>

librbd: potential double-unwatch of watch handle upon error

Fixes: http://tracker.ceph.com/issues/17210
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #9758 from SUSE/wip-btrfs-not-ideal

doc: remove btrfs contradiction

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #10469 from wjwithagen/wip-wjw-compileflag

cmake: supress more warnings

Reviewed-by: Kefu Chai <kchai@redhat.com>

os/bluestore: honour allow-eio flag during read

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore: check iterator before using during fsck

Also drop redundant checking.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore: replace store with logger in Cache

As what we really want is to log cache-related statistics.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore/BlueFS: factor unflushed log into runway calculation

Include the size of the in-memory buffer we haven't yet flushed
into the runway calculation when deciding whether to allocate
more log space.

Reported-by: Somnath Roy <somnath.roy@sandisk.com>
Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #10945 from dillaman/wip-17188

librbd: deadlock when replaying journal during image open

Reviewed-by: Mykola Golub <mgolub@mirantis.com>

SQUASH: client: request all shared caps if ctime was requested

Much like the change_attr, ctime changes can potentially be cached on
clients. Request all shared caps if the want mask specifies the ctime.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

Merge pull request #10746 from ceph/wip-async-default

msg: make async backend default

scripts/bdev_grep: parse bluefs style extents too

(they use + instead of ~)

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore/BlueFS: fix async compaction txn

Our transaction writes are labeled with a seq and uuid
to avoid replaying over garbage.

Two bugs, one real, one potential.

1) The second async compaction transactoin didn't have
its seq and uuid set, so replay always stopped.

2) We were writing two separate transactions, one with
all the new metadata, and the next one with a jump to
the new log offset.  If the first write completed but
it was torn and the second transaction didn't hit disk,
we might see an old transaction with seq == 2 and the
same uuid and replay that instead.

Fix both of these by making the async log txn one single
transaction that jumps directly to the new log offset.

Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #10938 from liewegas/wip-bluestore-csum-error

os/bluestore: fix/improve csum error message

Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>

Merge pull request #10961 from xiexingguo/xxg-wip-bluestore-09-02

os/bluestore: add cache-related stats

Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #10953 from liewegas/wip-bluestore-fixes

os/bluestore: misc fixes

Reviewed- by: Igor Fedotov ifedotov@mirantis.com

test/librbd/fsx: enable exclusive-lock feature in krbd mode

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

SQUASH: mds/client: add routines to copy to/from ceph_mds_request_head_legacy

Signed-off-by: Jeff Layton <jlayton@redhat.com>

os/bluestore: trivial cleanups

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore: don't overwrite type if it does already exist and at desired type

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

src/CMakeLists.txt: -Woverloaded-virtual is already included in -Wall on Clang

Signed-off-by: Willem Jan Withagen <wjw@digiware.nl>

Merge pull request #10956 from ceph/wip-cmake

cmake: pass -DINTEL* to gf-complete cflags

Reviewed-by: Loic Dachary <ldachary@redhat.com>

os/bluestore: conditional checking label of bluefs devices

As NVMe devices may not support to do so.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

cmake: pass -DINTEL* to gf-complete cflags

* in addition to the -march=foobar, we also need to pass -DFOOBAR to
  gf-complete sources' cflags, so that the SIMD instructions can be
  generated as expected.
* also extract the compiler support for instruction extensions detections
  into SIMDExt.cmake.
* and only check the compiler support if CMAKE_SYSTEM_PROCESSOR matches
  with the instruction-set/arch to be checked.

Signed-off-by: tone.zhang <tone.zhang@linaro.org>
Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #10934 from tchaikov/wip-kill-warnings

Wip kill warnings

Reviewed-by: Brad Hubbard <bhubbard@redhat.com>

os/bluestore: end scope of std::hex properly

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore: add onode/buffer cache statistics

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

SQUASH: client: querying for stx_version field requires getting all shared caps

Otherwise, someone could potentially query for just CEPH_STATX_VERSION,
and see it as unchanged, even when there are changes buffered up on
other clients.

By doing this, I don't think we'll incur any perf hit in the common use
case which is ganesha querying for all attributes. We are adding Xs
here, but unless there is a lot of xattr activity I don't think that
will generally cause a lot of exclusive cap recalls.

Also, we don't actually need AUTH caps to fetch CEPH_STATX_RDEV, remove
that from statx_to_mask. PIN is sufficient there.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

SQUASH: mds/client: don't get clever with ceph_mds_request_head_legacy encode/decode

As requested by Greg...

Declare a legacy object on the stack, and do an extra copy to/from it.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

SQUASH: client: always take a change_attr update from the server if it's newer

I think this is more correct in the client. If we see a newer
change_attr, then we always want to take note of that fact, even if
we have the right caps.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

SQUASH: Revert "mds: ensure that change_attr reflects metadata changes on clients that hold CAP_FILE_EXCL"

This reverts commit 26ab2fa10041ce34b25aa1535715131a2709a27f.

I don't think this is necessary. We don't necessarily need to increment
the change attribute on every change, as long as we ensure that it gets
bumped iff there were changes since the last time you queried for it.

IOW: it's ok to have a single change_attr change encompass a large set
of changes, as long as you ensure that that it is larger by at least 1
after all of those changes.

In order to look at (and potentially cache) other attributes under that
change_attribute (e.g. owner or mode), you need to recall any outstanding
exclusive caps for those attrs. That causes their change_attrs to be
synched to the largest, which is enough to ensure that it changed in
some way.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

SQUASH: client: request AUTH caps on the directory during create if we'll need to check perms locally

Signed-off-by: Jeff Layton <jlayton@redhat.com>

Merge pull request #10517 from wjwithagen/wip-wjw-freebsd-cmake-execludes

cmake: FreeBSD specific excludes in CMakeLists.txt files

Reviewed-by: Kefu Chai <kchai@redhat.com>

ceph_test_objectstore: fix TryMoveRename test

Put the objects in the right pool.

Signed-off-by: Sage Weil <sage@redhat.com>

ceph_test_objectstore: clean up Rename test a bit

Drop useless remove calls.

Leave in the no-op setattr call since it triggered a
bluestore bug :)

Signed-off-by: Sage Weil <sage@redhat.com>

ceph_test_objectstore: assert umount succeeds

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: make object existence check robust

We may have the OnodeRef defined but the object may have
exists==false.  Move the check out of the lookup block
so that we get the right behavior.

Signed-off-by: Sage Weil <sage@redhat.com>

os/bluestore: use _setattr for SETATTR

This is a bit faster.

Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #10843 from liewegas/wip-osdmap-reencode

messages/MOSDMap: mark as enlighten OSDMap encoder

Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #10950 from xiexingguo/xxg-wip-blue-09-01

os/bluestore: gifting bluefs more carefully

Reviewed-by: Sage Weil <sage@redhat.com>

os/bluestore: don't over-gift if bluefs free is smaller than bluestore_bluefs_min

- If free space of bluefs is smaller than bluestore_bluefs_min, then raise it
  exactly up to bluestore_bluefs_min.

- Pre-check if bluestore_bluefs_min will exceed bluestore_bluefs_max_ratio * bluefs_total,
  so if free space of bluefs is smaller than bluestore_bluefs_min_ratio * bluefs_total,
  we'll still be able to gift.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore: end scope of std::hex properly

So it does not extend to the following output stream and
causes unexpected behaviour.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

vstart.sh: add option to turn on multimds

This makes the necessary `ceph fs` calls to get it running.

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

mds: remove max_mds config option

It is now required that all changes to max_mds use the run-time `ceph fs
set max_mds` command. The rationale for this change is that it is
confusing to have a configuration for max_mds which is only observed at
file system creation.

Fixes: http://tracker.ceph.com/issues/17105
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

librbd: deadlock when replaying journal during image open

Fixes: http://tracker.ceph.com/issues/17188
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: improve image state machine debug log messages

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #10404 from ilc/build_fix

make-dist/ceph.spec.in: Fix srpm build breakage.

Reviewed-by: Ken Dreyer <kdreyer@redhat.com>

Merge pull request #10846 from stiopaa1/mds_mdstable_addConstToMemFun

mds/MDSTable: add const to member functions

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #10781 from stiopaa1/osdc_objectcacher_moveClassesToCC

osdc/ObjectCacher: move C_ReadFinish, C_RetryRead

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #10744 from stiopaa1/osdc_journaler_moveCDelayFlushClass

osdc/Journaler: move C_DelayFlush class to .cc

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #10668 from stiopaa1/mon_mdsmonitor_moveCUpdatedClass

mon/MDSMonitor: move C_Updated class to .cc file

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #10628 from jcsp/wip-16909

mon: tear down standby replays on MDS rank stop

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #10937 from ceph/dang-build

Allow compressor build without YASM

Reviewed-by: Ali Maredia <amaredia@redhat.com>

os/bluestore: improve failed csum debug output

Signed-off-by: Sage Weil <sage@redhat.com>

script/bdev_grep: update to use hex

Signed-off-by: Sage Weil <sage@redhat.com>

Allow compressor build without YASM

The isa-l compressors need YASM to build.  Conditionalize on that.

Signed-off-by: Daniel Gryniewicz <dang@redhat.com>

Merge pull request #10931 from xiexingguo/xxg-wip-fix-isa-l-assert

os/bluestore: get rid off "isa-l" type in ZLibCompressor ctor

Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #10923 from xiexingguo/xxg-wip-bluefs-08-30

os/bluestore/bluefs: use map to track dirty files

Reviewed-by: Sage Weil <sage@redhat.com>

os/bluestore: get rid off "isa-l" type in ZLibCompressor ctor

This is introduced by https://github.com/ceph/ceph/pull/10158 and causes
the following failure:

/home/xxg/ceph-dev/src/os/bluestore/bluestore_types.h: 284: FAILED assert(0 == "invalid compression algorithm")

 ceph version v11.0.0-1941-g47c8a21 (47c8a214356983f0539f24fe1c5724b3bbae0406)
 1: (ceph::__ceph_assert_fail(char const*, char const*, int, char const*)+0x85) [0x7f5668c1ad45]
 2: (BlueStore::_do_alloc_write(BlueStore::TransContext*, BlueStore::WriteContext*)+0x1a65) [0x7f5668b19d55]
 3: (BlueStore::_do_write(BlueStore::TransContext*, boost::intrusive_ptr<BlueStore::Collection>&, boost::intrusive_ptr<BlueStore::Onode>, unsigned long, unsigned long, ceph::buffer::list&, unsigned int)+0x4ac) [0x7f5668b3b3fc]
 4: (BlueStore::_write(BlueStore::TransContext*, boost::intrusive_ptr<BlueStore::Collection>&, boost::intrusive_ptr<BlueStore::Onode>&, unsigned long, unsigned long, ceph::buffer::list&, unsigned int)+0x10f) [0x7f5668b3bf2f]
 5: (BlueStore::_txc_add_transaction(BlueStore::TransContext*, ObjectStore::Transaction*)+0x1067) [0x7f5668b405f7]
 6: (BlueStore::queue_transactions(ObjectStore::Sequencer*, std::vector<ObjectStore::Transaction, std::allocator<ObjectStore::Transaction> >&, std::shared_ptr<TrackedOp>, ThreadPool::TPHandle*)+0x382) [0x7f5668b42432]
 7: (SyntheticWorkloadState::write()+0x3a5) [0x7f5668a2ace5]
 8: (doSyntheticTest(boost::scoped_ptr<ObjectStore>&, unsigned long, unsigned long, unsigned long)+0xb9d) [0x7f56689dfa7d]
 9: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x29d) [0x7f56689dfffd]
 10: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x9e) [0x7f56689dfdfe]
 11: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x9e) [0x7f56689dfdfe]
 12: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x9e) [0x7f56689dfdfe]
 13: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x9e) [0x7f56689dfdfe]
 14: (do_matrix_choose(char const* (*) [10], int, int, int, boost::scoped_ptr<ObjectStore>&)+0x9e) [0x7f56689dfdfe]
 15: (do_matrix(char const* (*) [10], boost::scoped_ptr<ObjectStore>&)+0x317) [0x7f56689e0347]
 16: (()+0x297f3c) [0x7f56689e0f3c]
 17: (StoreTest_SyntheticMatrixCompressionAlgorithm_Test::TestBody()+0x118) [0x7f56689e1078]
 18: (void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*)+0x43) [0x7f5668d673e3]
 19: (testing::Test::Run()+0xb7) [0x7f5668d5a127]
 20: (testing::TestInfo::Run()+0x9e) [0x7f5668d5a1ce]
 21: (testing::TestCase::Run()+0xa5) [0x7f5668d5a2d5]
 22: (testing::internal::UnitTestImpl::RunAllTests()+0x278) [0x7f5668d5a5b8]
 23: (testing::UnitTest::Run()+0x68) [0x7f5668d5a888]
 24: (main()+0x2f4) [0x7f566895be84]
 25: (__libc_start_main()+0xf5) [0x7f5663e5ab15]
 26: (()+0x265ba5) [0x7f56689aeba5]
 NOTE: a copy of the executable, or `objdump -rdS <executable>` is needed to interpret this.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

test/perf_local: fix -Wmaybe-uninitialized warnings

Signed-off-by: Kefu Chai <kchai@redhat.com>

test,pybind: silence -Wsign-compare warnings

Signed-off-by: Kefu Chai <kchai@redhat.com>

os/CMakeLists.txt: undubble kv.cc usage.

 - kv.cc is both in bluestore and in kstore.
   But atm. both are the same
 - Bluestore is not build on FreeBSD (no AIO)

So use the kstore one to be the one to build against.

Signed-off-by: Willem Jan Withagen <wjw@digiware.nl>

os/bluestore/bluefs: drop redundant eof check for read_random()

Because we have done the sanity check and fixed the length
at the entrance, so we don't have to check it again,
especially in the while loop, which is less efficient.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore/bluefs: update mtime of inode in unified path

To reduce code redundancy.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

os/bluestore/bluefs: use map to track dirty files

So dirty files are arranged by dirty_seq and then
those of same dirty_seq are grouped into list and
as a result we can clean up dirty files much faster
during _flush_and_sync_log().

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

libcephfs: add unmount function in cephfs.pyx

we need unmount function when change the mount point

Signed-off-by: huanwen ren <ren.huanwen@zte.com.cn>

Merge pull request #10158 from Ved-vampir/zlib_isal_extension

Isa-l extention for zlib compression plugin

Merge pull request #10858 from atheism/rgw_getattr-return

rgw: remove duplicated calls to getattr

verified: f23

Merge pull request #10927 from liewegas/wip-0.94.9

doc/release-notes: v0.94.9

Reviewed-by: Alfredo Deza <adeza@redhat.com>

doc/release-notes: v0.94.9

Signed-off-by: Sage Weil <sage@redhat.com>