Merge pull request #12104 from pritha-srivastava/wip_sts_role_rest

rgw: Added code for REST APIs for AWS Roles.

Reviewed-by: Adam C. Emerson <aemerson@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>

Merge pull request #13189 from gaurav36/bug-18326-image-rename-non-rbd-pool

rbd: destination pool should be source pool if it is not specified

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

Merge pull request #12948 from trociny/wip-17020

rbd-mirror: add support for active/passive daemon instances

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

Merge pull request #13013 from wangzhengyong/master

os/bluestore: remove no use parameter in bluestore_blob_t::map_bl

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

Merge pull request #12782 from atheism/rgw-orphan-fix-error-messages

rgw: orphan: fix error messages

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

Merge pull request #13199 from chipitsine/master

mds,mon: Clean issues detected by cppcheck

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

Merge pull request #13203 from ifed01/wip-bluestore-more-dump

os: extend ObjectStore interface to dump store's performance counters

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

Merge pull request #12974 from ukernel/wip-multimds-misc

mds: miscellaneous fixes

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

qa/suites/rbd: add rbd-mirror-ha test

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

qa/workunits/rbd: tweaks to use RBD_MIRROR_USE_EXISTING_CLUSTER locally

This var is mostly used when running rbd_mirror test scripts on
teuthology.  It can be used locally though to speedup re-running the
tests:

Set a test temp directory:

  export RBD_MIRROR_TEMDIR=/tmp/tmp.rbd_mirror

Run the tests the first time with NOCLEANUP flag (the cluster and
daemons are not stopped on finish):

  RBD_MIRROR_NOCLEANUP=1 ../qa/workunits/rbd/rbd_mirror.sh

Now, to re-run the test without restarting the cluster, run cleanup
with USE_EXISTING_CLUSTER flag:

  RBD_MIRROR_USE_EXISTING_CLUSTER=1 \
    ../qa/workunits/rbd/rbd_mirror_ha.sh cleanup

and then run the tests:

  RBD_MIRROR_USE_EXISTING_CLUSTER=1
    ../qa/workunits/rbd/rbd_mirror_ha.sh

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

qa/workunits/rbd: tests for rbd mirror HA

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

qa/workunits/rbd: allow to start several daemons for the same cluster

by optionally specifyning daemon instance after cluster name and
colon, like:

  start_mirror ${cluster}:${instance}

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

rbd-mirror HA: pool replayer should be started/stopped when lock acquired/released

Fixes: http://tracker.ceph.com/issues/17020
Signed-off-by: Mykola Golub <mgolub@mirantis.com>

rbd-mirror HA: create pool locker / leader class

Fixes: http://tracker.ceph.com/issues/17019
Signed-off-by: Mykola Golub <mgolub@mirantis.com>

rbd-mirror HA: add new lock released/acquired and heartbeat messages

Fixes: http://tracker.ceph.com/issues/17018
Signed-off-by: Mykola Golub <mgolub@mirantis.com>

rbd-mirror: add class for mirror status watcher

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

librbd: equality operator for managed_lock::Locker

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

librbd: add create_async_context_callback helper

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

librbd: missing const qualifiers

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

librbd: track in-fly break_lock and get_locker requests in managed lock

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

librbd: pass notifier_id to Watcher handle_notify

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

librbd: allow watcher derived classes access ioctx, oid and work queue

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

rbd: destination pool should be source pool if it is not specified

Currently if user perform image rename operation and user give pool
name as a optional parameter (--pool=<pool_name>) then currently
its taking this optional pool name for source pool and making
destination pool name default pool name.
With this fix if user provide pool name as a optional pool name
parameter then it  will consider both soruce and destination pool
name as optional parameter pool name.

Fixes: http://tracker.ceph.com/issues/18326
Reported-by: МАРК КОРЕНБЕРГ <socketpair@gmail.com>
Signed-off-by: Gaurav Kumar Garg <garg.gaurav52@gmail.com>

Merge pull request #13077 from guzhongyan/fix_wrongly_init

librbd: initialize diff parent overlap to zero

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

Merge pull request #12537 from liewegas/wip-optracker

common/TrackedOp: various cleanups and optimizations

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

Merge pull request #13205 from dachary/wip-crush-documentation

crush: API documentation

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

crush: API documentation

Signed-off-by: Loic Dachary <loic@dachary.org>

Merge pull request #13164 from dzafman/wip-18718

osd: Fix useless MAX(0, unsigned) to prevent out of wack misplaced

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

os/bluestore: add submit latency perf counter.

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

test/fio: add performance counters dump on FIO job completion.

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

os: add dump_perf_counters method to ObjectStore implementations.

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

os/bluestore: fix typos in perf counter public names.

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

test/fio: fix misordered parameter for queue_transaction.

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

Merge pull request #13193 from dillaman/wip-18738

journal: stop processing removal after error

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

Merge pull request #13069 from SUSE/wip-hammer-retirement

doc: add 0.94.10 and hammer EOL to releases.rst

Reviewed-by: Abhishek Lekshmanan <abhishek@suse.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>

Clean issues detected by cppcheck

[src/mds/MDLog.cc:1172] -> [src/mds/MDLog.cc:1173]: (warning) Either the condition 'sle!=NULL' is redundant or there is possible null pointer dereference: sle.
[src/mon/MDSMonitor.cc:2622] -> [src/mon/MDSMonitor.cc:2625]: (warning) Either the condition 'mds_map!=nullptr' is redundant or there is possible null pointer dereference: mds_map.

Signed-off-by: Ilya Shipitsin <chipitsine@gmail.com>

Merge pull request #13151 from smithfarm/wip-ceph-release-notes

tools: ceph-release-notes: ignore low-numbered PRs

Merge pull request #13190 from Adirl/reduce_num_buffers

msg/async/rdma: reduce number of rdma rx/tx buffers

Reviewed-by: Haomai Wang <haomai@xsky.com>

Merge pull request #13066 from liewegas/wip-aio-read

os/bluestore: use aio for reads

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

Merge pull request #13071 from liewegas/wip-omap-getkeys-rados-test

qa/suites/rados/basic: set low omap limit for rgw workload

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

Merge pull request #13150 from tchaikov/wip-mon-cleanup

mon: post-jewel cleanups

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

Merge pull request #12966 from dillaman/wip-18436

osdc: cache should ignore error bhs during trim

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

Merge pull request #12800 from jcsp/wip-vstart-qasuite

Improve vstart_runner to (optionally) create its own cluster

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

Merge pull request #12817 from liupan1111/wip-list-snapshot-time

librbd: support to list snapshot timestamp

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

doc: update actual firefly and estimated hammer retirement

According to the "Understanding the release cycle" section, firefly was
actually retired when jewel was released (April 2016), and hammer will be
retired when luminous is released (estimated in May 2017).

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

doc: add 0.94.10 to releases.rst

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

Merge pull request #13098 from wido/doc-osd-greace

doc: Document that osd_heartbeat_grace applies to MON and OSD

Reviewed-by: Joao Eduardo Luis <joao@suse.de>
Reviewed-by: Kefu Chai <kchai@redhat.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>

journal: stop processing removal after error

Fixed: http://tracker.ceph.com/issues/18738
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #13141 from dillaman/wip-18653

librbd: introduce new constants for tracking max block name prefix

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

librbd: introduce new constants for tracking max block name prefix

Fixes: http://tracker.ceph.com/issues/18653
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #13188 from liewegas/wip-rest-fs-name

qa/workunits/rest: use unique pool names for cephfs test

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

test: regolden formatted-output.t for timestamp metadata of rbd snapshot

Signed-off-by: Pan Liu <pan.liu@istuary.com>

test: Add unit test for getting rbd snapshot timestamp info

Signed-off-by: Pan Liu <pan.liu@istuary.com>

librbd: support to list snapshot time stamp

Fixes: http://tracker.ceph.com/issues/808
Signed-off-by: Pan Liu <pan.liu@istuary.com>

librbd: Add timestamp metadata when creating rbd snapshot

Signed-off-by: Pan Liu <pan.liu@istuary.com>

Merge pull request #12646 from SirishaGuduru/swift-api-auth-doc-change

doc: Fixes radosgw-admin ex: in swift auth section

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

qa/workunits/rest: use unique pool names for cephfs test

Using cephfs_[meta]data collides with the pools that teuthology
already creates if an mds is defined.

This became a (noticeable) problem with 052c3d3f681017d56b5b0ee5cf6f65bffc952a4c

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

Merge pull request #13054 from oritwas/wip-rgw-period-update-crash

rgw: fix period update crash

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

msg/async/rdma: reduce number of rdma rx/tx buffers

issue: none

Change-Id: Ied523fec3210af152508b102c48aff5633343ab4
Signed-off-by: Adir Lev <adirl@mellanox.com>

Merge pull request #13186 from Adirl/fix_rdma_compile

cmake: fix broken RDMA compilation after merge PR #12878

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

Merge pull request #13003 from Jing-Scott/dev/swift-container-acl-info

rgw: swift: the http referer acl in swift API should be shown

Reviewed-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>

Merge pull request #13005 from Jing-Scott/dev/swift-http-url-parse

rgw: swift: The http referer should be parsed to compare in swift API

Reviewed-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>

Merge pull request #12874 from Jing-Scott/dev/swift-ref

rgw: the swift container acl should support field .ref

Reviewed-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>

cmake: fix broken RDMA compilation after merge PR #12878

issue: 965984

Change-Id: I2e401acf204069d3907001d276c58d9fb4cd0d58
Signed-off-by: Oren Duer <oren@mellanox.com>

doc: Document that osd_heartbeat_grace applies to MON and OSD

This value has to be set in both the [mon] and [osd] section since
both daemons read this value.

Signed-off-by: Wido den Hollander <wido@42on.com>

Merge pull request #13142 from Adirl/remove_compilation_warning

msg/async/rdma: Remove compilation warning

Reviewed-by: Haomai Wang <haomai@xsky.com>

osd: Fix useless MAX(0, unsigned) to prevent out of wack misplaced

Still want to prevent strange results if for some reason num_objects
is less than osd_missing.

Fixes: http://tracker.ceph.com/issues/18718
Signed-off-by: David Zafman <dzafman@redhat.com>

Merge pull request #13057 from wido/issue-18635

systemd: Restart Mon after 10s in case of failure

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

Merge pull request #13140 from liewegas/wip-bluestore-statfs

os/bluestore: fix statfs to not include DB partition in free space

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

Merge pull request #13135 from liewegas/wip-kerneldevice-debug

os/bluestore/KernelDevice: fix debug message

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

Merge pull request #13124 from ukernel/wip-18675

client: remove request from session->requests when handling forward

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

Merge pull request #13095 from ukernel/wip-18660

mds: skip fragment space check for replayed request

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

Merge pull request #13052 from ukernel/wip-16768

mds: don't modify inode that is not projected

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

Merge pull request #12973 from ukernel/wip-18578

mds: fix incorrect assertion in Server::_dir_is_nonempty()

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

Merge pull request #12825 from stiopaa1/r11124_refuseToDoFsNew

mon/MDSMonitor.cc:refuse fs new on pools with obj

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

Merge pull request #12489 from gregsfortytwo/wip-fix-rename-bounds

client: fix the cross-quota rename boundary check conditions

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

Merge pull request #12440 from runsisi/wip-fix-fp-exception

include/fs_types: fix unsigned integer overflow

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

common/TrackedOp: log at level 6

This is above the default log level.  I have yet to encounter a situation
where the optracker output was helpful during a crash, and this saves
a significant amount of cpu time (~2-3% in my test).

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

common/TrackedOp: some cleanup

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

common/TrackedOp: manage current in parent class

No reason for child to muck with this!

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

common/TrackedOp: cache op description

Only calculate it once.  Invalidate from OSD after a full decode (when it
changes).

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

common/TrackedOp: take option time arg

Fix the callers to use this instead of calling into Tracker directly.

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

osd/OpRequest: drop useless helpers

No callers!

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

common/TrackedOp: use preallocated vector for events

This avoids an allocation per mark_event() call.

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

common/TrackedOp: fix formatter output

Spaces not allowed.  Lowercase by convention.

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

common/TrackedOp: add const char* and string variants of event

This avoids the full std::string (copy) for most events.  Wrap it up
into an Event class to hide some of the guts.

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

common/TrackedOp: drop unneeded _mark_event; pass c string

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

osd: use intrusive list for session op queue

We can avoid these list<> allocations in the fast dispatch path.

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

common/TrackedOp: add get() and put()

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

common/TrackedOp: make TrackedOp an intrusive::list::list_base_hook<>

This lets us put these in efficient lists.  Note that callers must take
care to bump the ref count manually!

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

common/TrackedOp: xlist -> intrusive::list

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

common/TrackedOp: use explicit ref count, intrusive_ptr

This is more efficient and more explicit than shared_ptr.  It will also
allow us to put these in intrusive lists.

Note that before we used the shared_ptr OnDelete property only for the
live ops and left it off for the history.  Here we rename is_tracked to
state and explicit note whether we are untracked, live, or history.

Also, now that we #include OpRequest in osd_types.h, we have to include
the .cc file (and ~OpRequest definition) in libcommon to avoid a link
error on all the unit tests etc.

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

mds/Mutation: make Mutation a TrackedOp

This seems silly now, but is needed in order to switch from shared_ptr to
intrusive_ptr.  The TrackOp is the refcounted thing, and we want to be
able to cast beween MutationRefs and MDRequestRefs, so... we need to make
sure the refcounting is done via a common parent, TrackedOp.

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

messages/MLock: drop unnecessary #include

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

os/bluestore: only use aio for read if there are >1 blobs

If we have a single blob to read it is not worth the context switch.

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

os/bluestore/BlockDevice: allow sync read to accumulate on bl

Do not clear the target bl; instead, append to it.  This makes our
behavior consistent with aio_read, which does the same.

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

os/bluestore: do read io via aio in parallel

Dispatch all blob reads in parallel via aio.

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

os/bluestore/KernelDevice: avoid possible use-after-free for ioc->priv

If aio_wake() triggers destruction, ioc->priv might be a
use-after-free (this is the case for the BlueStore read path).

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

os/bluestore/NVMEDevice: implement aio_read

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

os/bluestore/BlockDevice: add aio_read API

NVMEDevice not implemented yet.

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