librbd: possible race condition leads to use-after-free

The invoke async operation state machine can complete before the
owner lock is released.  Use a stack reference to prevent
use-after-free.

Fixes: http://tracker.ceph.com/issues/15690
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit d9421a25b582e41550526714b71fb55f05abb1de)

Merge pull request #9004 from dillaman/wip-15708

jewel: rbd: help message distinction between commands and aliases

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

Merge pull request #8972 from Abhishekvrshny/wip-15697-jewel

jewel: python: clone operation will fail if config overridden with "rbd default format = 1"

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

Merge pull request #9008 from dillaman/wip-15698

jewel: rbd-mirror: lockdep error during bootstrap

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

librbd: disable automatic refresh of image upon lock message

There is no need to refresh the image upon request/release/acquire
of the exclusive lock.  The next IO or maintenance op will kick off
the refresh.  This is interfering with the refresh state machine
unit test case (since two concurrent refreshes shouldn't be possible).

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit ee617a0ca56d752daf62bfffa59219b2c540a428)

librbd: update_features should release lock if acquired

There is a small window where requests are unblocked and before
the image is shut down where new requests can be accepted.  These
are correctly canceled during the shut down, but it can lead to
warning messages.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 608e09851396a9eeb7cf6dc1c7f687ef7a033601)

librbd: avoid applying refreshed image config within librados callback

There is a potential that a synchronous API call could deadlock a
image refresh.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit ce5c701bc47b0959f8453b6b92dee4804d3b1d75)

rbd-mirror: don't use temporary string for snap operation

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 06faf83bbfab000c8200a9cbe781adb0082b3f7b)

librbd: avoid recursive locking within operation state machine

Fixes: http://tracker.ceph.com/issues/15664
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 91a4890ee78c25391c1548fdacb2b51c46a47415)

Merge pull request #8869 from Abhishekvrshny/wip-15669-jewel

jewel: Errors encountered disabling object-map while flatten is in-progress

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

Merge pull request #8870 from Abhishekvrshny/wip-15668-jewel

jewel: rbd disk-usage CLI command should support calculating full image usage

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

Merge pull request #8868 from Abhishekvrshny/wip-15666-jewel

jewel: rbd CLI to retrieve rbd mirror state for a pool / specific image

Conflicts:
src/tools/rbd_mirror/ImageReplayer.cc: trivial resolution

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

Merge pull request #9046 from dillaman/wip-fix-snapc-validation-jewel

jewel: librbd: put the validation of image snap context earlier

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

librbd: put the validation of image snap context earlier

we'd better validate the snap context right after we got the mutable
metadata before we could go any further

Signed-off-by: runsisi <runsisi@zte.com.cn>
(cherry picked from commit ca8ae66ba42e7a598ad23ed911938589dc288456)

Merge pull request #8867 from Abhishekvrshny/wip-15661-jewel

jewel: Parent image is closed twice if error encountered while opening

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

Merge pull request #8866 from Abhishekvrshny/wip-15660-jewel

jewel: rbd-mirror can crash if start up is interrupted

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

Merge pull request #9044 from dillaman/wip-librbd-bad-header-jewel

jewel: librbd: does not crash if image header is too short

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

librbd: does not crash if image header is too short

if something goes wrong with the object, and returns a chunk shorter
than expected, don't panic

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 7b52183addda83f98c8b1f37d3de20ca0fc4687b)

Merge pull request #8803 from SUSE/wip-15605-jewel

jewel: Potential double free of SetSnapRequest instance

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

Merge pull request #8865 from Abhishekvrshny/wip-15649-jewel

jewel: possible double-free of object map invalidation request upon error

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

Merge pull request #8930 from theanalyst/wip-15707-jewel

jewel: multisite: Issues with Deleting Buckets

Merge pull request #9029 from theanalyst/wip-15680

jewel: rgw: radosgw-admin zone set cuts pool names short if name starts with a period

Merge pull request #9022 from yehudasa/wip-rgw-leak-3-jewel

jewel: rgw: leak fixes

Merge pull request #9021 from yehudasa/wip-rgw-xstorageurl-fix-jewel

jewel: rgw: fix printing wrong X-Storage-Url in Swift's TempAuth.

Merge pull request #9020 from yehudasa/wip-15597-jewel

jewel: rgw: fix issue #15597

Merge pull request #9019 from yehudasa/wip-rgw-admin-zonegroup-jewel

jewel: radosgw-admin: add missing --zonegroup-id to usage

Merge pull request #9018 from yehudasa/wip-15626-jewel

rgw: add AWS4 completion support for RGW_OP_SET_BUCKET_WEBSITE

Merge pull request #9017 from yehudasa/wip-15625-jewel

jewel: rgw: segfault at RGWAsyncGetSystemObj

Merge pull request #9016 from yehudasa/wip-rgw-admin-modify-jewel

jewel: radosgw-admin: update usage for zone[group] modify

Merge pull request #9015 from yehudasa/wip-rgw-zonegroup-remove-jewel

jewel: radosgw-admin: zonegroup remove command

rgw: fix suffix in RGWZoneParams::fix_pool_names()

Fixes: http://tracker.ceph.com/issues/15598
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 6f6b0c925994b0369eda390a6b3fd20005e70c9b)

rgw: code cleanup

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit d53ffbfe65611425b08af71e0d24bfe2a0076f1e)

rgw: upgrade default zonegroup and set correct zone[group] id

Fixes: http://tracker.ceph.com/issues/15597
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 27806c71922eef2f4cde360d7ba40dee9d5bd4bd)

rgw: RGWHTTPClient requests can unregister themselves early

No need to wait for req_state to complete anymore.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 91f61d68bf5fc39152d75fbc633f088e17d53d9e)

Conflicts:
src/rgw/rgw_http_client.cc

rgw: move around sync_env.init()

call sync_env.init() even if we're initalized in RGWRemoteDataLog::init()

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 3d8442effc1c175e0b7c5064ed41bc642f6cdcb0)

rgw: rados crs, explicit cleanup

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 1c876eb3873692042c0af3644f8ee3ad15ffdbf5)

rgw: RGWHTTPManager, avoid referring to req_data->client when completing

this is past calling req_data->finish(), client can be destroyed by now

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 7ada4df124a71eae413e7dec0e25264a8deffcad)

rgw: civetweb_callback() refer to store inside lock

pe->store might change when reconfiguring

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 2ca38dab071c7d20a224a4e1dacfd68548fe52dd)

rgw: RGWPeriodPusher, stop http manager explicitly

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 98433e589f2a8b519726ed23b1565c5eb331c68e)

rgw: RGWHTTPManager, can call stop() more than once

destructor calls stop(), but it can be also called explicitly, if caller
wants to ensure requests are not being handled anymore.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit bdce81a2ae00f154420386e107a104cbf100d96d)

rgw: RGWReadRESTResourceCR, fix refcounting

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit d1ec31def71505dc95139fbdd3b49c92fec1d296)

rgw: RGWReadRemoteDataLogShardCR fix destructor

was missing destructor, constructor was broken

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 95162a4c1bf61f7329e363b7efd5686f7e1025f0)

rgw: RGWDataSyncStatusManager, cleanup if failing init

init can be called more than once, ended up with leaking some resources

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 14cd44e0847fd914068f48dd0208ce7c1fe4eecb)

rgw: rest crs, explicitly call cleanup

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit a3a3f4f5883f3fc2bf47a7ebfa538206b7ab51d9)

rgw: more leaks fixes

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit cb797d4fd754b4a33eb34c249da3aabc65a7aba3)

rgw: drop a reference to http op

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 4442d1c19acfc28c2dd4a011365ebebe08750963)

rgw: fix printing wrong X-Storage-Url in Swift's TempAuth.

If the option "rgw_swift_account_in_url" is being set to true,
both user and subuser IDs will be present in the X-Storage-Url
header generated by the implementation of Swift API's TempAuth.
This patch rectifies this behavior and makes that only the user
identifier will be placed there.

Fixes: http://tracker.ceph.com/issues/15667
Signed-off-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>
(cherry picked from commit 88f4c67d822388d2038454a25ead7e9a11f404b5)

radosgw-admin: add missing --zonegroup-id to usage

also clarified that --rgw-zone takes the zone by name

Signed-off-by: Casey Bodley <cbodley@redhat.com>
Reported-by: John Wilkins <jowilkin@redhat.com>
(cherry picked from commit 97fad754565b262529073e1b9f8d1799bfdfe0de)

rgw: don't allow any concurrent sync requests on the same key

We used to allow concurrent requests, as long as these had different op,
which wasn't correct.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit edea6d58dd25995bcc1ed4fc5be6f72ce4a6835a)

rgw: collect children after waiting for them

Otherwise we're going to busy loop there.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 806fb85d18653910863ba6b8cb56b8db1a0fa3d2)

rgw: don't pass object context into async coroutines

Fixes: http://tracker.ceph.com/issues/15625
The async read data/attrs coroutines may outlive the callers. Instead
of introducing a complicated refcounting scheme, just don't pass it in.
Anyway, it was not thread safe, and the benefit of using it is not
clear.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit b225d1758057efc30721aad25447d54fa36e4cb2)

rgw: check for status >= 0

Fixes: #15565
check only verified that status was > 0, so we ended not updating mtime for
the target object. This resulted in a continuous sync of objects.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 255f19d9ef964536f2db17f70749f489330262a6)

rgw: don't try to delete object if does not exist

If we know that the object does not exist, no need to remove it.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 8d59a30d56b5f197a293f7fba6ebb1a630a33a05)

rgw, cls/rgw: store removed object mtime in bi log

Store the mtime of the object that was removed in the bucket index log.
This will be used in data sync.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 80d64049937e1930ea8273fdad76a51493a5e14a)

rgw: add AWS4 completion support for RGW_OP_SET_BUCKET_WEBSITE

Fixes: http://tracker.ceph.com/issues/15626
Signed-off-by: Javier M. Mellid <jmunhoz@igalia.com>
(cherry picked from commit 05373354e4725c3fc675df652fbf97da26160bcb)

radosgw-admin: update usage for zone[group] modify

the help strings indicated that only --master could be changed

Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit 410d432d11541a29c3fd4608a070cecf8dc2766e)

test/rgw: add test_zonegroup_remove

Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit e2b27c7266f0ace372dbf94e062598fd7a575795)

test/rgw: index zones by name instead of insertion order

Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit 6327ade12bf0255c3b35bd6407f1dd2bcc6936fa)

radosgw-admin: add missing 'zonegroup remove'

Fixes: http://tracker.ceph.com/issues/15684
Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit db0fa48ed5958d37928f893acd44cd66b6b9b990)

rgw: RGWZoneGroup::remove_zone() takes zone id

Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit 3a451511fb38a5f1cd83ca3610c91141e7279f63)

Merge pull request #9006 from athanatos/wip-15655-jewel

OSDMonitor: avoid underflow in reweight-by-utilization if max_change=1

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

rbd:make a distinction of help message between ''rbd snap rollback" and "rbd snap revert"

Fixes: https://tracker.ceph.com/issues/15521
Signed-off-by: Yongqiang He <he.yongqiang@h3c.com>
(cherry picked from commit 1bf26509c6efd06a8facc0e45ab42255592ca74d)

osdc/Objecter: fix race condition for sortbitwise flag detection

During list_objects/list_nobjects(), we are checking osdmap's
sortbitwise flag setting without holding rwlock in hand, which
is a potential race condition.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>
(cherry picked from commit 7fb6be706be2454cc2a831df29808997f0aa5ab3)

librbd: default clone operation to image format 2

Image format 1 doesn't support clones, so unless the user
overrode the setting, default to format 2.

Fixes: http://tracker.ceph.com/issues/15685
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit f24c70f54343aa6698af8ac1f0774282b0659968)

librbd: add rbd_image_options_is_set helper method to API

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 2633b045e0b57827cc10c2e7707bd5a5e344e59a)

rgw_op: pass delete bucket op to master zone first

Basically first ask the master to process a delete bucket request before
we process it in the zone locally, otherwise we reach a situation where
we drop the bucket locally and master will still maintain an index

Fixes: http://tracker.ceph.com/issues/15540
Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>
(cherry picked from commit da6b44a6e47399493adae53ad3c36a029f1e2a23)

rgw: add errno entry for -ENOTEMPTY

currently if master returns an -ENOTEMTPY, we return a -EIO as we dont
have an entry for this

Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>
(cherry picked from commit 40c3bfef8cc3bce7d15fbb0e135c1a909ee59368)

fix deb package /etc/default/ceph location

commit 7384a14f243519547a61534d22e21e6069aae016 have introduce
the /etc/default/ceph file at a wrong location : /etc/default/ceph/ceph

(cherry picked from commit 13804fafd4c33068d8ed6d0480b594daec0f9341)

mon/OSDMonitor: improve reweight_by_utilization() logic

By calling reweight_by_utilization() method, we are aiming at an evener result
of utilization among all osds. To achieve this, we shall decrease weights of
osds which are currently overloaded, and try to increase weights of osds which
are currently underloaded when it is possible.
However, we can't do this all at a time in order to avoid a massive pg migrations
between osds. Thus we introduce a max_osds limit to smooth the progress.

The problem here is that we have sorted the utilization of all osds in a descending
manner and we always try to decrease the weights of the most overloaded osds
since they are most likely to encounter a nearfull/full transition soon, but
we won't increase the weights from the most underloaded(least utilized by contrast)
at the same time, which I think is not quite reasonable.

Actually, the best thing would probably be to iterate over teh low and high osds
in parallel, and do the ones that are furthest from the average first.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>
(cherry picked from commit e7a32534ebc9e27f955ff2d7a8d1db511383301e)

mon/OSDMonitor: avoid potential expensive grace calculation

The grace calculation during check_failure() is now very complicated
and time-consuming. Therefore we shall skip this when it is possible.

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>
(cherry picked from commit 3557903d5d57642179b2ae137bedc389974b1956)

mon/OSDMonitor: max_osds must be > 0 for reweight-by-*

Include a few test cases.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 84bd0dce18f6dd0c35d7e6cfee91ce8c5a0f4db7)

Merge pull request #8787 from xiaoxichen/wip-15631

jewel: CephFSVolumeClient should isolate volumes by RADOS namespace

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

OSDMonitor: avoid underflow in reweight-by-utilization if max_change=1

Fixes: http://tracker.ceph.com/issues/15655
Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit c121bc51a5a877a49094cfbd3bd16af1f3ae5090)

rbd: disk usage now includes all snapshots by default

Added a new "--from-snap" optional to restrict the calculation to
specific subsets of snapshots and the HEAD revision.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 1ccdcb5b6c1cfd176a86df4f115a88accc81b4d0)

librbd: block RPC requests while updating features

Disabling the journal and object map require acquiring the exclusive
lock locally.  We don't want to start executing long-running ops
for this quick process.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit db7aaff0f6d7a57e2b36d22b609f915f9b2b3893)

librbd: return -ESHUTDOWN when a refresh is requested on a closed image

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit ed132a1eb10e6b2eb4bc465658eb0379daca3d67)

librbd: synchronous block_writes should return possible error result

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 88e76cb3c9a74b4718ad39c827f2194666f5be2a)

librbd: refresh image before executing maint op

If a maint op is interrupted and restarted, it's possible that the
image will need to be refreshed.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 9cf6821bd9f58c181720a5c8156cec5ab33d35d4)

librbd: reduce log level when lock owner not detected

This can happen under normal operating conditions and the
lock request is automatically retried.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 5a4893b7a7e50cc399696e1d0cb4d4f8785e723d)

librbd: reduce log level for interrupted maint ops

Fixes: http://tracker.ceph.com/issues/15572
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 1617328402345c68bf5d54828da6d687e51ef42b)

qa/workunits/rbd: test mirror status in pool directory

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit f15cd51a1ddbd4079a8218bc2446a4bf68f95107)

rbd-mirror: in replay status store number of entries behind master

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit e4229d1d4b163f1003e68fef1db63f796393ba23)

librbd: API to retrieve image mirroring status

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit a2045d61874f9e418b5b3c3136af4482603c8e7f)

test: MockJournaler: add get_cached_client method

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit 7633407ab584ce8c7777a525554773c2a358ce5f)

rbd: CLI to retrieve rbd mirror state for a pool / specific image

Fixes: #15144
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit fe5207278d884464c05d6808245a0c4df03c441c)

rbd-mirror: store replay status in mirroring object

Fixes: #14420
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit 52b2fe14cfc932b2aa53ce76a3150ce87ce4d377)

test: cls_rbd: methods to store/retrieve mirroring status

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit 8aefad59813489cee112f46b3fe10331bb039b57)

cls::rbd:: methods to store/retrieve mirroring status

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit 17dd40df4f4201cf0c0c6d58b1728177c5ee726b)

objclass: add method to list watchers

Signed-off-by: Mykola Golub <mgolub@mirantis.com>
(cherry picked from commit 44a31d2b770ef016271f9285447c601921b9c838)

Merge pull request #8843 from SUSE/wip-msd-typo-fix-jewel

systemd: fix typo in preset file

Reviewed-by: Boris Ranto <branto@redhat.com>

librbd: propagate any image refresh errors back to caller

Previously, the saved error code was only propagated when an image
flush was required.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 105ef8a24aa6ac8719f68c5a9366129ddb41ee88)

librbd: avoid second close attempt of invalid parent image

Fixes: http://tracker.ceph.com/issues/15574
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit e91e7f8d9ce10f0d19c48facfb6318d6b49798e8)

rbd-mirror: interrupting image replay startup can cause crash

Properly shut down the journaler if a shut down has been requested
while start up is in-progress.

Fixes: http://tracker.ceph.com/issues/15630
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 3a71052a5b14af34ed81f6ee77f862f576a7a5cc)

librbd: fix potential assertion for object map invalidation

if update_flags fails we should not go any further, or the async request
will be removed from m_image_ctx.async_requests twice

Signed-off-by: runsisi <runsisi@zte.com.cn>
(cherry picked from commit dd0eb719a50e9fb24eca5a840c2e8adb53be41b2)

systemd: fix typo in preset file

Signed-off-by: Nathan Cutler <ncutler@suse.com>
(cherry picked from commit 80be4a8cbf1d42232ff61494d035e7abd90c2c14)

systemd: fix typo in preset file

Signed-off-by: Nathan Cutler <ncutler@suse.com>
(cherry picked from commit 80be4a8cbf1d42232ff61494d035e7abd90c2c14)

Merge pull request #8801 from SUSE/wip-15633-jewel

jewel: ceph-{mds,mon,osd} packages need scriptlets with systemd code

Reviewed-by: Nathan Cutler <ncutler@suse.com>
Reviewed-by: Boris Ranto <branto@redhat.com>

librbd: fix potential double free of SetSnapRequest instance

if image feature EXCLUSIVE_LOCK is not enabled we should not try to
initialize the exclusive lock, or we may end with two async Contexts
to finish the same SetSnapRequest instance

Fixes: http://tracker.ceph.com/issues/15571
Signed-off-by: runsisi <runsisi@zte.com.cn>
(cherry picked from commit 0a6eaac12fe29762c3dd29068f9427840cfea100)

rpm: Add rpm scripts for ceph-rbd-mirror

We are currently missing the systemd scripts for the new ceph-rbd-mirror
daemons. This patch introduces them.

Signed-off-by: Boris Ranto <branto@redhat.com>
(cherry picked from commit 14652f32165f72efde569a939179fc26584ab155)

rpm: Start all the targets in %post

We need to have the sub-targets active for the ceph.target to be able
to propagate its calls to the services. If the sub-target is inactive,
the main target won't propagate the stop/restart calls.

Signed-off-by: Boris Ranto <branto@redhat.com>
(cherry picked from commit 73f457986753b504fd088c1fa064608aa9d340c9)