Merge pull request #156 from SUSE/wip-bsc1063014

luminous: rgw: Remove assertions in IAM Policy

rgw: Remove assertions in IAM Policy

A couple of them could be triggered by user input.

Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
(cherry picked from commit b3118cabb8060a8cc6a01c4e8264cb18e7b1745a)

Merge pull request #155 from kmroz/ses5

rgw: stop/join TokenCache revoke thread only if started.

rgw: stop/join TokenCache revoke thread only if started.

Thread::join triggers an assert otherwise.

Fixes bsc#1059458
Fixes http://tracker.ceph.com/issues/21666

Signed-off-by: Karol Mroz <kmroz@suse.de>
(cherry picked from commit 26f2da083c7dd21b89c1c1e6c498b14e034364a6)

Merge pull request #154 from SUSE/wip-restore-tasksmax

systemd: restore TasksMax=infinity

Revert "Revert "Drop any systemd imposed process/thread limits""

This reverts commit 725343ebed6ee018118bd05bd237d3c5f089d01c.

Signed-off-by: Nathan Cutler <ncutler@suse.com>
Conflicts: (all trivial resolution)
systemd/ceph-mds@.service
systemd/ceph-mon@.service
systemd/ceph-osd@.service
systemd/ceph-radosgw@.service

Merge pull request #153 from SUSE/wip-sigpipe-porting-luminous

luminous: msg: reimplement sigpipe blocking

Reviewed-by: Joao Eduardo Luis <joao@suse.de>
Reviewed-by: Jan Fajerski <jfajerski@suse.com>

msgr: add a mechanism for Solaris to avoid dying on SIGPIPE

This is fairly clean: we define an RAII object in the Messenger.h on
Solaris, and "declare" it with a macro in the implementations. There's
no code duplication and on Linux it's just entirely compiled out.

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit cba20a95e816aaf9f0971b29b14b0be8c524b59d)

Conflicts:
    src/msg/async/PosixStack.cc - luminous #include "common/simple_spin.h" is
            missing in master (trivial resolution)

Revert "SIGPIPE suppression for platforms without SO_NOSIGPIPE or MSG_NOSIGNAL"

This reverts commit 131deb39769c1187c334ee84f552d3be01f1751b. It added
code that shouldn't have entered the project repo.

Conflicts:
src/msg/async/AsyncConnection.cc
src/msg/async/AsyncConnection.h
src/msg/simple/Pipe.cc
src/msg/simple/Pipe.h

Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 21bdb730932196a48407be0effab8813a78c4b1b)

Merge pull request #152 from theanalyst/wip-luminous-keystone-logs

rgw: defer constructing keystone engine unless url is configured

rgw: defer constructing keystone engine unless url is configured

currently we create a keystone revocation thread even when keystone url
is empty, lets defer the construction of keystone unless the urls are
configured

Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>
(cherry picked from commit 104c2f59011e6b515e3845cce529ee44334e76c0)

Merge remote-tracking branch 'ceph/luminous' into ses5

Merge pull request #17862 from smithfarm/wip-21307-luminous

luminous: mon: Client client.admin marked osd.2 out, after it was down for 1504627577 seconds

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

Merge pull request #17865 from smithfarm/wip-21465-luminous

luminous: OSD metadata 'backend_filestore_dev_node' is unknown even for simple deployment

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

Merge pull request #17936 from liewegas/wip-ruleset-errors-luminous

mon/OSDMonitor: error out if setting ruleset-* ec profile property

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

Merge pull request #17946 from liewegas/wip-mgr-localpool-luminous

luminous: mgr/localpool: module to automagically create localized pools

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

Merge pull request #17998 from liewegas/wip-pr-17978-luminous

rbdmap: fix umount when multiple mounts use the same RBD

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

rbdmap: fix umount when multiple mounts use the same RBD

When a Kubernetes Pod consumes a RBD it is mounted two
times on the same host. When the host shutdown umount will
fail leading to a hung system

Signed-off-by: Alexandre Marangone <amarango@redhat.com>
(cherry picked from commit 40825daecedb2a3481021e4d36a367c339eb9b62)

Merge pull request #17953 from ceph/wip-ceph-ansible-bport

tests - ceph-ansible backports

use ovh instead of vps

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 56d1511bdfaee5a2a0d85bd606348632bfcbf09e)
Signed-off-by: Yuri Weinstein <yweinste@redhat.com>

12.2.1

Merge pull request #17970 from idryomov/wip-pr-17346-luminous

luminous: qa/suites/krbd: rbd_xfstests job overhaul

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

qa/suites/krbd: new rbd_xfstests job definition

ext4 seems to be a better choice for our purposes -- less test churn,
rather small and reliable exclude list.

All excluded tests but generic/050 fail with no krbd in the mix, most
have popped up on linux-ext4 list at least once.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
(cherry picked from commit 6647c65ebe0688283cc8b9da9f638a4d6ffb75de)

qa/tasks/rbd.xfstests: take exclude list from yaml

Different filesystems (and further, different configurations of the
same filesystem) need different exclude lists.  Hard coding the list in
a wrapper script is inflexible.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
(cherry picked from commit 01914be5a3b35479037054293b85d96596cb870b)

Conflicts:
qa/run_xfstests_krbd.sh [ trivial, caused by commit
  3aae5ca6fda8 ("scripts: fix bash path in shebangs") ]

qa/run_xfstests.sh: quit building xfstests on test nodes

xfstests is a pain to build on trusty, xenial and centos7 with a single
script.  It is also very sensitive to dependencies, which again need to
be managed on all those distros -- different sets of supported commands
and switches, some versions have known bugs, etc.

Download a pre-built, statically linked tarball and use it instead.
The tarball was generated using xfstests-bld by Ted Ts'o, with a number
of tweaks by myself (mostly concerning the build environment).

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
(cherry picked from commit 1a982fea53c30fb246c2ab28822af6a69165ffea)

qa/run_xfstests.sh: drop *_MKFS_OPTIONS variables

AFAICT ./check doesn't query EXT4_MKFS_OPTIONS or BTRFS_MKFS_OPTIONS,
We don't need anything special for xfs, so remove all of them to avoid
confusion.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
(cherry picked from commit 7a0e45046d556c6624e0225ff006c31d8dbae2a8)

qa/run_xfstests.sh: kill FS_TYPE variable

./check expects FSTYP, so use that throughout.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
(cherry picked from commit ca99f1e8406c14e5899d393ab0bb503061c47b9b)

doc/mgr: add "local pool" plugin to toc

Signed-off-by: Kefu Chai <kchai@redhat.com>
(cherry picked from commit 0b831e7886daad5836ca3cb58c01c123f99a1cde)

qa/tests: skip test that use public bucket landsat-pds from AWS

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit e11a1f773d8795dd9bd9819aaf229c6598e22ca6)

qa/tests: update required ceph-ansbile vars

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 184d5e6bad46451b3a8c5787f56148566bb42f20)

qa/tests: Fix ceph-ansible upstream vars

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 077904a0b5dffe9d64feade94cf30ffc92f1c0e5)

Fix get_system_type failure due to invalid remote name

recent changes caused the remote name to be invalid, fix the
arg passed to get_system_type

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 9fe58d5ed6bd2338271a9e003f51d80c2e440e23)

Stop the mgr on node which is not client, this will ensure
the client.0's mgr is active mgr for workunit to work.

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit b1fd615e9f507d4243744b34887366086c1890ac)

Test ceph-mgr RESTful api

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit e5b5a1b8c8d0e750d29fe41a16252dcb0adf449f)

Add additional variables as required by ceph-ansible for upstream
installation.

possible revert if those are made default
issue: https://github.com/ceph/ceph-ansible/issues/1834

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit e963bfd6ca0ea1525e37dc35bf3309548c7a3831)

Rename folders to fix task order

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 9cc00c5c1a6e99323a4655aad5c093b8d87609f6)

use bluestore with dmcrypt option

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 1041c803f1f8f7be0edd0e78df352152fa3c4e6f)

Add dmcrypt option

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit f6de5d9f9e2a3666116311c146a8f2a45f3b8dbb)

Separate the main task from options

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 60d00e0eadda3a7a7cfbac2acdc81a4d55a27570)

Catchup with recent changes with ceph-ansible

Adds osd_scenario and ceph_stable_release variables

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 0395b844884a63a2cb7480dcc40762df6915b403)

Add workaround for http://tracker.ceph.com/issues/20950

mgr bootstrap key differs on disk, rewrite the new key
using auth get.

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 5c43c5972ddb04724c8659a1b8d71cda1e0acdfb)

Add kraken to luminous upgrade case

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 71ea4987e0d55cc49e90dfbaf9722769cc892f3f)

Add kernel tar/untar and systemd task

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 526d4aef1cb73ffa6f73e1353dc06eed6cccb9f9)

Restart services after upgrade

This seems to be an issue and should be fixed in our systemd scripts,
need to discuss more on this.

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 2116f76a1f1fbb32ca5844c1d051e22b1b657e0f)

since the roles are mapped inside ceph-deploy, store the roles that
are mapped and use the new mapped role for upgrades during later
stage.

eg: mon.a is mapped to mon.mira002 during install, store this mapping
and durig upgrade map it back to appropriate name to find the hostname
with that role

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 8a2b9a08095dc97e0e440cc900c650d5efb55a4f)

Add doc string for missing ceph-deploy-branch

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit e7a0a4f87f2fbea0c9e27341c952245f1ad34ab8)

use mon.a as ceph-admin node, the previous get_first_mon returns
a sorted(mons)[0] as first mon.

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 407ce2613d37252e48c2a648b94b065ab070ec46)

use elif instead of else to handle cases for jewel install

jewel needs neither filestore or bluestore as an option, so provide none
when running with jewel branch.

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit f3ec7cabf66c76d712f77a4dfefe96f374507982)

Jewel to luminous upgrade scenario using ceph-deploy

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 9c85c76a90dcf46ec2bcc6840f4fc57ed86583b5)

Add upgrade functionality using ceph-deploy

This is to test for customer like upgrade scenarios and to find
any issues that may be related to systemd, packaging etc

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit 9a73127c16d3395934d003024964ed1373d81083)

Fix old flake8 issues

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit b409c3d163194b0b65e26b0fbc93def8e17a10b4)

Add option to skip mgr install for old release

Signed-off-by: Vasu Kulkarni <vasu@redhat.com>
(cherry picked from commit a9be0c387ed2ae09227dad833c8647af110bd2de)

Merge pull request #17945 from batrick/i21540

luminous: qa whitelist fixes

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

Merge pull request #17821 from smithfarm/wip-21472-luminous

luminous: tests: kcephfs: ignorable MDS cache too large warning

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>

qa/workunits/mgr/test_localpool: simple test for localpool mode

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 6383fa5b3004a606c15efd2be1b7ce5d57f05086)

pybind/mgr/localpool: module to automagically create localized pools

By default, this will create a pool per rack, 3x replication, with a host
failure domain.  Those parameters can be customized via mgr config-key
options.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 1a0f42b70a4c9fa68dc47f2f521d0f1e8f5bb220)

qa: whitelist expected MDS_CACHE_OVERSIZED

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 71f0066f6ec32be18abc0c220a49e203ac3b9786)

Conflicts:
   qa/suites/kcephfs/recovery/tasks/client-limits.yaml - fix
      already applied by some other commit

qa: whitelist expected rstat warning

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit c53d091e5110011866aa85495ff4e5abe0442d71)

Merge pull request #17922 from batrick/kcephfs-backports

luminous: qa: kcephfs whitelist fixes

Reviewed-by: Yan, Zheng <zyan@redhat.com>

mon/OSDMonitor: error out if setting ruleset-* ec profile property

We change ruleset -> crush back in dc7a2aaf7a34b1e6af0c7b79dc44a69974c1da23.
If someone tries to use the old property, error out early, instead of
silently not doing the thing they thought they told us to do.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 6376d75eda648789b5e316a1ac2883708db7128e)

Merge pull request #17855 from smithfarm/wip-21490-luminous

luminous: tests: test_rebuild_simple_altpool triggers MDS assertion

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

qa: get config only on running MDS

Fixes: http://tracker.ceph.com/issues/21466
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 8a535d9c72965c79692dccc6ff418b46fb86c3eb)

qa: whitelist cache too large in client-limits

Fixes: http://tracker.ceph.com/issues/21510
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 87039eb30a25158a192ab7099fa81fbb9ce02953)

qa: whitelist failover log warnings

Fixes: http://tracker.ceph.com/issues/21509
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 37a7737712f36422b5f035a0e02bab25ae2da890)

qa: ignore evicted client in client-recovery

Fixes: http://tracker.ceph.com/issues/21508
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 8ecef57bb10a83587f02ce087b657d303ae7c589)

Merge PR #17878 into luminous

* refs/remotes/upstream/pull/17878/head:
qa/cephfs: Fix error in test_filtered_df

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Amit Kumar <amitkuma@redhat.com>

Merge PR #17877 into luminous

* refs/remotes/upstream/pull/17877/head:
client: Delete onsafe
client: add mountedness check inside client_lock

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

Merge PR #17870 into luminous

* refs/remotes/upstream/pull/17870/head:
mds: reset Formatter on error

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Amit Kumar <amitkuma@redhat.com>

Merge PR #17854 into luminous

* refs/remotes/upstream/pull/17854/head:
mds: void sending cap import message when inode is frozen
client: fix message order check in handle_cap_export()

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

Merge PR #17852 into luminous

* refs/remotes/upstream/pull/17852/head:
mds: check ongoing catter-gather process before capping log

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

Merge PR #17822 into luminous

* refs/remotes/upstream/pull/17822/head:
qa/cephfs: kill mount if it gets evicted by mds
qa/cephfs: fix test_evict_client

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Zheng Yan <zyan@redhat.com>

Merge pull request #17817 from smithfarm/wip-21464-luminous

luminous: qa: ignorable MDS_READ_ONLY warning

Reviewed-by: Yan, Zheng <ukernel@gmail.com>
Reviewed-by: Yuri Weinstein <yweins@redhat.com>

Merge pull request #17879 from smithfarm/wip-21449-luminous

luminous: tests: test_misc creates metadata pool with dummy object resulting in WRN: POOL_APP_NOT_ENABLED

Reviewed-by: Amit Kumar amitkuma@redhat.com
Reviewed-by: Yan, Zheng <ukernel@gmail.com>

Merge pull request #17886 from smithfarm/wip-21486-luminous

luminous: tests: [kcephfs] test_client_pin times out waiting for dentry release from kernel

Reviewed-by: Amit Kumar amitkuma@redhat.com
Reviewed-by: Yan, Zheng <ukernel@gmail.com>

Merge pull request #17887 from smithfarm/wip-21487-luminous

luminous: cephfs: MDS rank add/remove log messages say wrong number of ranks

Reviewed-by: Amit Kumar amitkuma@redhat.com
Reviewed-by: Yan, Zheng <ukernel@gmail.com>

Merge pull request #17888 from smithfarm/wip-21488-luminous

luminous: tests: failures from pjd fstest

Reviewed-by: Amit Kumar amitkuma@redhat.com
Reviewed-by: Yan, Zheng <ukernel@gmail.com>

Merge remote-tracking branch 'ceph/luminous' into ses5

Merge pull request #17812 from liewegas/wip-21410-luminous

Revert "osd/OSDMap: allow bidirectional swap of pg-upmap-items"

Reviewed-by: Josh Durgin <jdurgin@redhat.com>
Reviewed-by: xie xingguo <xie.xingguo@zte.com.cn>

ceph: do link/rename semantic checks after srcdn is readable

For hard link, source inode must not be directory. For rename,
types of source/destination inodes must match. If srcdn is replica
and we do these checks while it's not readble, it's possible that
wrong source inode is used in these checks.

Fixes: http://tracker.ceph.com/issues/21383
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 46962b253563a867707e7c5d7887abf2060cc4d7)

mon/MDSMonitor: fix clog msgs on MDS rank add/rm

Because these were generated before the actual
change, the get_num_in values did not make sense.

Fixes: http://tracker.ceph.com/issues/21421
Signed-off-by: John Spray <john.spray@redhat.com>
(cherry picked from commit 511439b515414ae93d791f54a187e44d5fdffa7b)

client: set client_try_dentry_invalidate to false by default

By default, ceph-fuse uses side effect of 'dentry invalidation' to
trim kernel dcache if it runs on kernel < 3.18. The implemention of
kernel function d_invalidate() changed in 3.18 kernel, the method no
longer works for upstream kernel >= 3.18.

RHEL 3.10 kernel includes backport of patches that change implemention
of d_invalidate(). So checking kernel version to decide if 'dentry
invalidation' method works is unreliable.

Fixes: http://tracker.ceph.com/issues/21423
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 69adaabface27880dd6c8dbfdeeb06cf3f3d346a)

qa/tasks/cephfs: Whitelist POOL_APP_NOT_ENABLED for test_misc

test_misc verifies that ceph fs new will not create a filesystem
on a pool that already contains objects. As part of the test, it
inserts a dummy object into a pool and then attempts to use it for
CephFS. This triggers POOL_APP_NOT_ENABLED. Setting the application
metadata for the pool (and having ceph fs new fail because of the
existing metadata) would then exercise a different failure case.

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit eba01c48ca8b3113b1faf353990533a5ae39232c)

qa/cephfs: Fix error in test_filtered_df

ceph df accounts for pool size, so there is no need to do it in the test.

Fixes: http://tracker.ceph.com/issues/21381
Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit b059cb6290d81df07bd8ec4e1ec9ef0be24892a2)

client: Delete onsafe

Variable "onsafe" going out of scope leaks the storage it points to. This fixes the Coverity
Scan CID 1417473.

Signed-off-by: Jos Collin <jcollin@redhat.com>
(cherry picked from commit e24a91e9189587c800ef5709c78f1bc0e0ac8419)

client: add mountedness check inside client_lock

Currently we check for mountedness in the high level wrappers, but those
checks are lockless. It's possible to have a call that races with
ceph_unmount(). It could pass one of the is_mounted() checks in the
wrapper, and then block on the client_lock while the unmount is actually
running. Eventually it picks up and runs after the unmount returns, with
questionable results -- possibly even a crash in some cases.

For now, we can explain this away with a simple admonition that
applications should ensure that no calls are running when ceph_unmount
is called. In the future though, we may need to forcibly shut down the
mount when certain events occur (not returning a lease or delegation in
time, for instance).

Sprinkle in a bunch of "unmounting" checks after taking the client_lock,
and simply have the functions return errors (or sensible values in some
cases) when the Client is being downed. With that, we ensure that this
sort of race can't occur, even when the unmount is not being driven by
userland. Note too that in some places I've replaced assertions in the
code with error returns, as that's nicer behavior for libraries.

Note that this can't replace the ->is_mounted() checks in the lockless
wrappers as those are needed to determine whether the client pointer in
the ceph_mount_info is still valid. The admonition not to allow
ceph_unmount to race with other calls is therefore still necessary.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
(cherry picked from commit efca340596ef4da2b254ff1c64ec4c462d7b95a5)

mds: reset Formatter on error

This prevents partial results from concatenating with the error message.

Fixes: http://tracker.ceph.com/issues/21252
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 2b2b3d2035a700a65c3c469eedae69ad31023cf6)

os/filestore: fix device/partition metadata detection

The UUID thing (a) relies on partition labels to work, which isn't
always true (and won't be true for ceph-volume going forward), and
(b) reportedly doesn't work anyway.  The fd-based helper works
just fine (even for vstart).

Fixes: http://tracker.ceph.com/issues/20944
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit e62862be0b68a17763e19ed8453436c457294505)

mon: fix `osd out` clog message

This was printing the absolute time instead of the period.

Fixes: http://tracker.ceph.com/issues/21249
Signed-off-by: John Spray <john.spray@redhat.com>
(cherry picked from commit f07480fec729d66f4cdc19a1c3bde3d185df1289)

qa/cephfs: Refactor alternate pool test

Remove the alternate pool recovery test from test_data_scan. Newer
commits will place the test in its own file.

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit 47318f8ac4391148a1394846dd77c8452b7cf5ae)

qa/cephfs: test CephFS recovery pools

Test recovering metadata in to a separate RADOS pool with
cephfs_data_scan and friends.

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit 6af2ae80d327e564ad17f2caeded9c8142becf29)

qa/cephfs: support CephFS recovery pools

Add support for testing recovery of CephFS metadata into an alternate
RADOS pool, useful as a disaster recovery mechanism that avoids
modifying the metadata in-place.

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit 8f9a25202093339afb5308051d354d3ae79c6b2d)

qa/cephfs: Allow deferred fs creation

Permit Filesystem objects to be created and settings modified before
calling Filesystem.create().

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit 5fafc03cb9e6d5f39ce494375ea28c89b896e2f5)

qa/ceph_test_case: support CephFS recovery pools

Add support for testing recovery of CephFS metadata into an alternate
RADOS pool, useful as a disaster recovery mechanism that avoids
modifying the metadata in-place.

Signed-off-by: Douglas Fuller <dfuller@redhat.com>
(cherry picked from commit c85562c94a80b8a18975b8d0ee6a7fbd932cf024)

mds: void sending cap import message when inode is frozen

To export an inode to other mds, mds need to:

- Freeze the inode (stop issuing caps to clients)
- Flush client sessions (ensure client have received all cap messages)
- Send cap export message

These steps guarantee that clients receive cap import/export messages
in proper order (In the case that inode gets exported servel times
within a short time)

When inode is frozen, mds may have already flushed client sessions.
So mds shouldn't send cap import messages.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 48a9e695e58ac15f1e494977e0db7a164bb2fe98)

client: fix message order check in handle_cap_export()

If importer mds' cap already exists, but cap ID mismatches, client
should have received corresponding import message (the imported caps
got released later). Because cap ID does not change as long as client
holds the caps.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 3c31256a4addcc4eca30c62109e33f77224258c2)

mds: check ongoing catter-gather process before capping log

When deactivating mds, MDLog::trim() may start scatter-gather
process on mdsdir inode. Locker::scatter_writebehind() submits
log entry. So mds should make sure there is no scatter-gather
before capping log.

Fixes: http://tracker.ceph.com/issues/21467
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>

Merge pull request #17829 from jdurgin/wip-21428-luminous

osd/PrimaryLogPG: request osdmap update in the right block

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

osd/PrimaryLogPG: request osdmap update in the right block

Fixes: http://tracker.ceph.com/issues/21428
Signed-off-by: Josh Durgin <jdurgin@redhat.com>
(cherry picked from commit afc6624f768ea3c6e2d155122797db5cce8836f7)

qa/cephfs: kill mount if it gets evicted by mds

otherwise, teardown() hange at umount

Fixes: http://tracker.ceph.com/issues/21275
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 98d86a075242f0c4576814db15cc3f8fd8e06017)

qa/cephfs: fix test_evict_client

executing mount_a.kill() twice, then executing mount_b.kill_cleanup()
twice do not make sense.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 8433ced84733da86d545ccacc06ddea3a0282071)