os/bluestore/BlueFS: do direct io by default (not buffered)

Somnath is doing this in his testing with good results.

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

os/bluestore: break down onode vs bnode_map size in log

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

Merge pull request #10717 from liewegas/wip-bluefs-async-compaction

os/bluestore/BlueFS: async compaction

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

Merge pull request #10698 from liewegas/wip-bluestore-bluefs-allocation

bluestore: do not use freelist to track bluefs_extents

Merge pull request #10550 from tanghaodong25/in-memory-db

kv: fix some bugs in memdb

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

Merge pull request #10775 from xiexingguo/xxg-wip-clean-storetest

test/store_test: tidy-up SyntheticWorkloadState class

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

test/store_test: avoid dynamic allocation of trans

For the sake of performance.

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

test/store_test: drop redundant transaction of callbacks

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

test/store_test: fix incorrect prompt given by stash callback

"clone finish" belongs to C_SyntheticOnClone instead.

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

kv: fix iterator bug in prev

Signed-off-by: Haodong Tang <haodong.tang@intel.com>

kv: fill 'm_key_value' after 'seek_to_first' or 'seek_to_last'

Signed-off-by: Haodong Tang <haodong.tang@intel.com>

kv: fix iterator invalidation in memdb.

Signed-off-by: Haodong Tang <haodong.tang@intel.com>

kv: delete store after pg destructor is called in OSD shutdown.

Using memdb for bluestore kvbackend, we will hit segfault when we use
'kill' command to shut down osd process. After destructing pg, some
reference to bluestore will be release, but bluestore has been deleted
at this time.

Signed-off-by: Haodong Tang <haodong.tang@intel.com>

Merge pull request #10481 from dillaman/wip-16171

librbd: request exclusive lock if current owner cannot execute op

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

Merge pull request #10737 from weiqiaomiao/wqm-wip-rgw-core

rgw: fix radosgw daemon core when reopen logs

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

Merge pull request #10728 from xiexingguo/xxg-wip-kill-bs-warn

bluestore: initialize csum_order properly

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

Merge pull request #10703 from dillaman/wip-16855

rbd-mirror: improve split-brain detection logic

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

Merge pull request #10756 from linuxbox2/fix_gc_interval

rgw_file: restore local definition of RGWLibFS gc interval
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #10686 from liewegas/wip-bluefs-remove

os/bluestore/BlueFS: do not op_file_update deleted files

Tested-by: Somnath Roy <somnath.roy@sandisk.com>

rgw_file: restore local definition of RGWLibFS gc interval

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

Merge pull request #10755 from SUSE/wip-cmake-use-pie

cmake: Add -pie to CMAKE_EXE_LINKER_FLAGS

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

Merge pull request #10754 from SUSE/wip-cmake-fix-rgw-soversion

cmake: Fix mismatched librgw VERSION / SOVERSION

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

cmake: Add -pie to CMAKE_EXE_LINKER_FLAGS

Without this, rpmlint (on openSUSE Tumbleweed) fails with:

ceph-radosgw.x86_64: E: non-position-independent-executable
(Badness: 10000) /usr/bin/radosgw
This executable must be position independent.  Check that it
is built with -fPIE/-fpie in compiler flags and -pie in linker
flags.

Signed-off-by: Tim Serong <tserong@suse.com>

cmake: Fix mismatched librgw VERSION / SOVERSION

Without this, rpmlint (on openSUSE Tumbleweed) fails with:

librgw2.x86_64: E: shlib-policy-name-error (Badness: 10000) librgw1
Your package contains a single shared library but is not named
after its SONAME.

It seems that the VERSION/SOVERSION mismatch results in the
creation of librgw.so.1 and librgw.so.2.0.0, whereas it should
be librgw.so.2 and librgw.so.2.0.0.

Signed-off-by: Tim Serong <tserong@suse.com>

Merge pull request #10749 from mikulely/fix-rgw-client-cb

rgw: fix flush_read_list() error msg
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #10747 from oliveiradan/master

gcc 6.1.1 complains about missing include: <random>. 4.8.3 does not c…

rgw: fix flush_read_list() error msg

Signed-off-by: Jiaying Ren <jiaying.ren@umcloud.com>

test: gcc 6.1.1 complains about missing include: <random>. 4.8.3 does not complain about it

Signed-off-by: Daniel Oliveira <doliveira@suse.com>

Merge pull request #10264 from yuyuyu101/wip-async-abstract-backend

msg/async: allow other async backend implementations

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

test_msgr: ensure wait for ms_dispatch executed

Otherwise in slow runner, ms_dispatch may not executed when assert

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

msg/async/Event: remove global_centers assert

In tests we allow to reset EventCenter instance in the same CephContext,
so it may let global_centers->centers to set the same position multi times.

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

test_async_networkstack: add networkstack tests

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

msg/async/AsyncConnection: fix _conn_prefix racing when stopped

When the connection is lossy and enter fault, it will dispatch reset event.
If cleanup handler is executed as well as ms_handle_reset call mark_down,
it may exists racing for "cs". cleanup handler will reset "cs" but
_conn_prefix in mark_down will access "cs".

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

msg/async: readd set_priority call

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

msg/async: allow EventCenter::set_owner reentry

If daemonlize we need to respawn event threads, it need to allow set_owner again

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

msg/async/Stack: register fork watcher

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

common/ceph_context: add prefork/postfork wathcher support

Because daemon() will termniate all existing threads, it will make something go
wrong.

So we want to add hook at CephContext, do something before/after fork.

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

msg/async/Stack: disable smart thread spawn now

New async msgr runtime need to spawn threads when binding, but ceph-osd will
call daemon() after binding port. So we need to respawn threads if forked.

Then thread spawn delay will increase complexity for this change and it's
really a simple strategy which help less, we disable auto spawn now.

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

msg/async/PosixStack: increase log level

AsyncMessenger will try to loop the bind port range, so it will produce
some addr inuse errors which is not abnormal.

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

msg/async/Event: remove event wakeup flag

Now only dispatch external event will wakeup event thread(previously
delete_time_event will call wakeup), we only need to use
"external_num_events" to indicate whether we have extra events.

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

msg/async/Event: debug event address when dispatching and executing

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

msg/async/AsyncConnection: support NetworkStack api instead of posix

1. replace sd to ConnectedSocket
2. Replace WorkerPool with Stack
3. Use Stack worker

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

msg/async/net_handler: let set_socket_options return value

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

msg/async: add PosixStack support

Add default posix backend support, it should be the full replacement for
the original AsyncMessenger IO logics.

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

msg/async/Stack: add abstract Stack

Stack is a network IO framework which encapsulates all necessary basic network
interface, then it manages threads to work.

Different network backend like posix, dpdk even RDMA need to inherit Stack
class to implement necessary interfaces. So it will make ease for other
network backend to integrated into ceph. Otherwise, each backend need to
implement the whole Messenger logics like reconnect, policy handle, session
maintain...

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

test: fixed unused function warnings in unittest_rbd_mirror

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

rbd-mirror: prevent syncing to non-primary image after failover

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

rbd-mirror: demote/promote in same cluster results in split-brain

Fixes: http://tracker.ceph.com/issues/16855
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: normalize journal promotion/demotion events

A non-primary image's commit possition won't accurately reflect
the current demotion/promotion chain. Therefore, directly specify
the predecessor for promotion events.

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

librbd: block RPC requests when demoting the image

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

librbd: separate journal::TagData predecessor data into new struct

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

rbd-mirror: include tag tid in bootstrap debug log messages

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

qa/workunits/rbd: demote/promote image on same cluster

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

Merge pull request #10691 from ceph/wip-jlayton-dotdot

Clean up handling of "/.." in ceph client

Merge pull request #10733 from dillaman/wip-16608

rbdmap: specify bash shell interpreter

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

Merge pull request #10735 from dillaman/wip-17030

librbd: always respond to "release lock" request if lock owner

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

Merge pull request #10712 from dillaman/wip-16740

librbd: permit disabling journaling if in corrupt state

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

Merge pull request #10711 from dillaman/wip-16984

librbd: fix possible inconsistent state when disabling mirroring fails

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

Merge pull request #10708 from dillaman/wip-16967

rbd: bench io-size should not be larger than image size

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

Merge pull request #10714 from dillaman/wip-16529

qa/workunits: support filtering cls_rbd unit test cases

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

Merge pull request #9688 from shylesh/wip-16280

ceph-disk: change ownership of initfile to ceph:ceph

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

Merge pull request #10376 from onyb/wip-rbd-pypi

pybind: Rework rbd/setup.py for PyPI

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

rgw: fix radosgw daemon core when reopen logs

the rgw_user_st_syn thread use the pointer meta_mgr in RGWUserStatsCache::sync_all_users() fuction, so we should close the thread before delete the pointer in RGWRados::finalize() function when close storage before reopen_logs.

Fixes: http://tracker.ceph.com/issues/17036
Signed-off-by: weiqiaomiao <wei.qiaomiao@zte.com.cn>

librbd: always respond to "release lock" request if lock owner

Fixes: http://tracker.ceph.com/issues/17030
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #10709 from cgwalters/doc-use-https-for-repos

doc/install: Use https:// for download.ceph.com

Reviewed-by: Dan Mick <dmick@redhat.com>

doc: added rbdmap man page to RBD restructured index

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

rbdmap: specify bash shell interpreter

Fixes: http://tracker.ceph.com/issues/16608
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #10596 from cbodley/wip-rgw-period-puller

rgw multisite: fixes for period puller

Merge pull request #10330 from rzarzynski/wip-rgw-objdeler-fixes

rgw: multiple fixes for Swift's object expiration
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #10619 from chengweiv5/fix-doc

doc: fix rados/configuration/osd-config-ref.rst

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

Merge pull request #10138 from Yan-waller/yj-wip-rgwrados-0705

rgw: fix the error return variable in log message and cleanups

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

Merge pull request #10729 from SUSE/wip-rnotes-script-regex

script: improve ceph-release-notes regex

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

Merge pull request #10705 from chengweiv5/doc-image-format

doc: format 2 now is the default image format

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

script: improve ceph-release-notes regex

Tolerate multiple spaces after ':', remove duplicate "fs", add "log" and
"msgr".

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

librbd: fix possible memory leak when image open fails

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

bluestore: initialize csum_order properly

To kill the following warning:

In file included from /home/jenkins-build/build/workspace/ceph-pull-requests/src/os/bluestore/BlueStore.h:39:0,
                 from /home/jenkins-build/build/workspace/ceph-pull-requests/src/os/bluestore/BlueStore.cc:22:
/home/jenkins-build/build/workspace/ceph-pull-requests/src/os/bluestore/bluestore_types.h: In member function ‘int BlueStore::_do_alloc_write(BlueStore::TransContext*, BlueStore::WriteContext*)’:
/home/jenkins-build/build/workspace/ceph-pull-requests/src/os/bluestore/bluestore_types.h:558:29: warning: ‘csum_order’ may be used uninitialized in this function [-Wmaybe-uninitialized]
     csum_chunk_order = order;
                             ^
/home/jenkins-build/build/workspace/ceph-pull-requests/src/os/bluestore/BlueStore.cc:6145:14: note: ‘csum_order’ was declared here
     unsigned csum_order;

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

Merge pull request #10118 from weiqiaomiao/wqm-wip-bucketcreate

rgw: for the create_bucket api, if the input creation_time is zero, we should set it to 'now"
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #10431 from ifed01/wip-bluestore-fix-csum-disable2

os/bluestore: proper handling for csum enable/disable settings

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

Merge pull request #10719 from stiopaa1/mon_pgmonitor_moveC_StatsStr

mon/PGMonitor: move C_Stats struct to cc file

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

unittest_bluefs: make flush_3 test device larger

I filled it up on my nvme device.

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

os/bluestore/BlueFS: Cleanup unused interfaces

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

os/bluestore/BlueFS: unit tests for log compaction

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

os/bluestore/BlueFS: Helper functions for the unit test framework

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

os/bluestore/BlueFS: Add sync and async compaction

Patch contains implementation for sync and async
log compaction, deafulted to async compaction.

Signed-off-by: Sage Weil <sage@redhat.com>
Signed-off-by: Varada Kari <varada.kari@sandisk.com>

os/bluestore/BlueFS: use hex

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

os/bluestore/BlueFS: assert we never extend log at flush

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

os/bluestore/BlueFS: init in header at decl

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

os/bluestore/BlueFS: interface to get bluefs space usage

Returns total bytes consumed by BlueFS across devs

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

os/bluestore/BlueFS: move metadata dump (compaction) into a helper

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

os/bluestore/BlueFS: allow log position to when doing op_jump

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

os/bluestore/BlueFS: add OP_JUMP

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

Merge pull request #10208 from BlaXpirit/cmake-py3

Include Python 3 bindings into the cmake build and make packages for them

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

client: add a testcase for "/.." handling

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

client: don't use special faked-up inode for /..

The CEPH_INO_DOTDOT thing is quite strange. Under most OS (Linux
included), the parent of the root is itself. IOW, at the root, '.' and
'..' refer to the same inode.

Change the ceph client to do the same, as this allows users to get
valid stat info for '..', as well as elimnating some special-casing.

Also in several places, we're checking dn_set.empty as an indicator
of being the root. While that is true for the root, it's also true
for unlinked directories.

This patch has treats them the same. An unlinked directory will
be reparented to itself, effectively acting as a root of its own.

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

mon/PGMonitor: move C_Stats struct to cc file

Signed-off-by: Michal Jarzabek <stiopa@gmail.com>

os/bluestore/BlueFS: _replay(false) to dump log file

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

os/bluestore/BlueFS: _should_compact_log

Signed-off-by: Varada Kari <varada.kari@sandisk.com>

Merge pull request #10685 from liewegas/wip-bluefs-extent

os/bluestore/bluefs_types: fix extent operator<<

Merge pull request #10676 from Yan-waller/yj-wip-rgwdocmultisite

rgw/doc: fix indent

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

qa/workunits: support filtering cls_rbd unit test cases

Fixes: http://tracker.ceph.com/issues/16529
Signed-off-by: Jason Dillaman <dillaman@redhat.com>