mon/MgrMonitor: send mgr beacons to leader

The leader has the last_beacon map that matters; make sure all
beacons go there.

At the same time drop the somewhat redundant checks in
preprocess_beaon and just send *everything* to prepare_beacon,
which is smart enough to do no update if there is none.

Fixes: http://tracker.ceph.com/issues/19502
Signed-off-by: Sage Weil <sage@redhat.com>

mgr/Mgr: unlock lock while blocking for metadata load

Fixes: http://tracker.ceph.com/issues/19516
Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #14115 from liewegas/wip-bluestore-metadata

os/bluestore: implement collect_metadata

Merge pull request #14304 from rchagam/18576_bug_fix

tracing: fix segv

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

Merge pull request #14328 from liewegas/wip-mgrmap-features

mgr: fix several init/re-init bugs

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

Merge pull request #14333 from liewegas/wip-bluestore-misc

os/bluestore: misc fixes

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

Merge pull request #14210 from liewegas/wip-osd-dup

os: allow offline conversion of filestore -> bluestore (or anything else)

Reviewed-by: David Zafman <dzafman@redhat.com>

Merge pull request #12175 from dongbula/fix-slavezonegroup-cannot-enable-bucket-versioning

rgw: fix slave zonegroup cannot enable the bucket versioning

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

Merge pull request #14342 from theanalyst/cleanup/rgw/rgw-lc-inc

rgw_lc: drop a bunch of unused headers

Merge pull request #13622 from zhangsw/feature-lifecycle-multipart

rgw: add support for multipart upload expiration.

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

Merge pull request #14337 from smithfarm/wip-19492

common: support s390 and unknown architectures in spin-wait loop

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

Merge pull request #14339 from haklein/master

add RGW ldap auth documentation

common: support s390 and unknown architectures in spin-wait loop

This fixes the s390 build which was broken by 34481768cbce0991bfba1d511dad81893a9004a9

It also reinstates the previous architecture-independent for-loop
implementation as a catch-all for "unknown" (not-explicitly-supported)
architectures.

Fixes: http://tracker.ceph.com/issues/19492
Signed-off-by: Nathan Cutler <ncutler@suse.com>

Merge pull request #14240 from dillaman/wip-rbd-mirror-peer-uuids

rbd-mirror: pool watcher should track mirror uuid

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

Merge pull request #14293 from badone/wip-ceph-disk-list-selinux

ceph-disk: Populate mount options when running "list"

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

Merge pull request #14032 from joscollin/wip-18511-returntype

librbd: changed the return type of ImageRequestWQ::discard()

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

Merge pull request #13957 from wangzhengyong/import

rbd: error out if import image format failed

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

Merge pull request #13836 from gaurav36/bug-14535-rbd-snap-pool-name-validatation

rbd: validate pool and snap name optionals

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

Merge pull request #14282 from wjwithagen/wip-wjw-freebsd-ceph-fuse

build: let FreeBSD build ceph-fuse

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

Merge pull request #14308 from badone/SyntheticClient_random_walk_warning

client/SyntheticClient.cc: Fix warning in random_walk

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

Merge pull request #10146 from yonghengdexin735/wip-zzz-filestore-write

os/filestore: handle error returned from write_fd()

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

Merge pull request #13815 from theanalyst/cleanup/rgw/rest-read-input

rgw: don't read all user input for a few param requests

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

mon/MgrMonitor: make debug prefix useful

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

ceph_test_objectstore: error via derr too

This helps establish timing relative to the log.

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

os/bluestore: increase aio queue depth

We've seen some reports of the aio EAGAIN backoff in logs; use a larger
aio ring buffer to reduce the risk of this.

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

os/bluestore: reduce some debug level

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

os/bluestore: put fsck structures in mempools

This lets us track them individually in debug mode.

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

mempool: do not skip first 2 mempools

This broke when we sorted the pool definitions in the header.

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

os/bluestore: flush osr even on failure in _deferred_replay

This is one less thing for fsck to worry about.

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

os/bluestore: trim cache during fsck

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

os/bluestore: be noisy about bitmap init (it's slow!)

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

Merge pull request #14316 from liewegas/wip-bluestore-fsck-unused

os/bluestore: fsck: verify blob.unused field

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

os/bluestore: use typedef for blob unused (uint16_t)

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

os/bluestore: fsck: verify blob.unused field

Two checks:

- verify that no logical extents reference the portion of the
  blob marked unused, and
- verify that the csum (if present) is zero for any unused
  region.

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

Merge pull request #14157 from liewegas/wip-19379

os/bluestore: fix deferred writes vs collection split race

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

Merge pull request #14258 from liewegas/wip-bluestore-tool

ceph-bluestore-tool: rename from bluefs-tool; improve usage

Merge pull request #14294 from tchaikov/wip-check-last-osd-report-if-luminous

mon/OSDMonitor: check last_osd_report only when the whole cluster is lu…

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

Merge pull request #14303 from liewegas/wip-more-mgr-commands

mgr: move 'osd perf' and 'osd blocked-by' to mgr

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

Merge pull request #14010 from zhangsw/fix-rgw-fetchremote-bug

rgw: version id doesn't work in fetch_remote_obj

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

Merge pull request #13310 from ceph/wip-ceph-disk-fix

ceph-disk: Add fix subcommand

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

rgw_lc: drop a bunch of unused headers

Dropping headers that are already included as a part of rgw_common

Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>

rgw: add support for multipart upload expiration.

Fixes: http://tracker.ceph.com/issues/19088
Signed-off-by: Zhang Shaowen <zhangshaowen@cmss.chinamobile.com>

add RGW ldap auth documentation

Signed-off-by: Harald Klein <hari@vt100.at>

Merge pull request #14314 from rzarzynski/wip-rgw-bulkupload-freebsd

rgw: switch to std::array in RGWBulkUploadOp due to C++11 and FreeBSD.
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #14252 from cbodley/wip-19409

radosgw-admin: new 'global quota' commands update period config
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #14109 from cbodley/wip-rgw-stat-manifest

rgw: stat requests skip compression, manifest handling, etc
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

os/bluestore/KernelDevice: assume rotational if we can't get devname

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

Merge pull request #14281 from tchaikov/wip-19429

qa/tasks/workunit.py: use "overrides" as the default settings of workunit

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

Merge pull request #14297 from Adirl/ibport

msg/async/rdma: Make port number an attribute of the Connection not o…

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

Merge pull request #13766 from dachary/wip-18962-ceph-disk-zap

ceph-disk: Write 10M to all partitions before zapping

Reviewed-by: Sébastien Han <seb@redhat.com>

Merge pull request #13998 from wangzhengyong/aio_complete

rgw: fix memory leak in delete_obj_aio

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

Merge pull request #12883 from mslovy/wip-rbd-import

rbd: import real thin-provision image

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

Merge pull request #14327 from cbodley/wip-12863

rgw: switch from "timegm()" to "internal_timegm()" for better portability

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

Merge pull request #14190 from idryomov/wip-respect-epoch-barrier

osdc/Objecter: respect epoch barrier in _op_submit()

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

Merge pull request #14220 from jcsp/wip-19407

mon: update mgrmap when active goes offline

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

mgr/Mgr: log something at end of handle_mgr_digest

Saw a run that showed the notify_all pg_summary as the last
message for this thread; hopefully we can see if it is
blocked in notify_all() or not.

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

mgr/Mgr: drop lock while fetching config-keys

This avoids a deadlock if some other message arrives before
the command reply and needs to take the mgr lock in dispatch.

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

Merge pull request #12437 from idealguo/remove-duplicate-flush

rgw: remove duplicate flush formatter

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

Merge pull request #13934 from chuang-he/fix_copy_obj_data

rgw: fix configurable write obj window size

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

Merge pull request #11963 from zhangsw/fix-rgw-lifecycle-bug

rgw: LCWorker's worktime is not the same as config rgw_lifecycle_work_time.

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

mgr/Mgr: fix map waits in init

If we go active multiple times we may already have an fsmap
subscription registered via MonClient, which means that we
won't receive a net new FSMap during the init phase.  This is
trivial to reproduce; just 'ceph mgr fail x' and when it
activates the second time init() will get stuck waiting.

The core issue is that monc subs are stateful, but we forget
the various maps with the previous Mgr incarnation.  Sipmle
workaround is to reset the mon connection as we init.

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

mgr/Mgr: use cond (not context) for FSMap wait

This is simpler.

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

Switch from "timegm()" to "internal_timegm()"

Changed timegm() to internal_timegm() for better portability

Fixes: http://tracker.ceph.com/issues/12863
Signed-off-by: Rishabh Kumar <kris.kr296@gmail.com>

Merge pull request #14042 from pritha-srivastava/wip-rgw-fix-accounting

rgw: Added code to correctly account for bytes sent/ received during a 'PUT' operation.

Reviewed-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>

Merge pull request #14125 from zhangsw/fix-rgw-cache-normalname-bug

rgw: add pool namespace to cache's key so that system obj can have unique key

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

Merge pull request #14072 from trociny/wip-notify

librbd: Notifier::notify API improvement

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

mon/MgrMonitor: encode MgrMap with mon quroum features

Fixes: http://tracker.ceph.com/issues/19491
Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #14160 from Liuchang0812/wip-19363-fix-lc-two-response

rgw/lifecycle: do not send lifecycle rules when GetLifeCycle failed

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

rgw: switch to std::array in RGWBulkUploadOp due to C++11 and FreeBSD.

Before this patch RGWBulkUploadOp::terminal_errors was declared as
std::initializer_list<int>. Unfortunately, a constexpr constructor
for it is available since C++14, not C++11. This was causing build
failures on FreeBSD.

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

os/bluestore: do not extend blobs with unused bitmap

If we resize the blob we need to adjust the resolution of the
unused bitmap, and that is only possible for some bit patterns.
For now just ignore blobs with unused blocks.

Add an assert in add_tail() so that we don't forget that
add_tail is (probably) where we'd would do that adjustment.

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

os/bluestore: some whitespace

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

os/bluestore: mark_used for wctx->writes

The refactored blob reuse code needs to mark blobs used when
processing wctx->writes.

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

Bug fix for http://tracker.ceph.com/issues/18576
Refer to http://lttng.org/docs/v2.9/ on dlclose side effects. Workaround is to use
RTLD_NODELETE flag when calling dlopen.

Signed-off-by: Anjaneya Chagam <anjaneya.chagam@intel.com>

os/bluestore: prefix fsck errors with 'fsck error:'

Greppable!

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

msg/async/rdma: Make port number an attribute of the Connection not of the Device

Since multiple connections on different ports could exist, shouldn't use
device->active_port, instead use conn->ibport.
Or in other words, now Device object doesn't have an active_port,
instead every port specific action (create_qp, get_lid, get_gid etc.)
need to specify the port number. The information about the port number
is known to the connection (RDMAConnectedSocket*) who is the caller of
those actions.

Issue: 995322
Change-Id: I482cb87c04ba99845dc44f6dd0547835fe814ebf
Signed-off-by: Amir Vadai <amir@vadai.me>

Merge pull request #14310 from tchaikov/wip-yield-for-ppc

common/simple_spin: use __ppc_yield() on all powerpc archs

Reviewed-by: Brad Hubbard <bhubbard@redhat.com>
Tested-by: Andrew Solomon <asolomon@us.ibm.com>

Merge pull request #14289 from tchaikov/wip-conditionalize-crc32c

cmake,crc32c: conditionalize crc32c on different archs

Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Willem Jan Withagen <wjw@digiware.nl>
Tested-by: Andrew Solomon <asolomon@us.ibm.com>

Merge pull request #14275 from ceph/evelu-cephdisk-race

ceph-disk: Adding retry loop in get_partition_dev()

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

Merge pull request #14185 from DHB-liuhong/objects

mon/PGMap: make si units more readable in PGMap summary

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

Merge pull request #14313 from wjin/clean_mgr

mgr: rm nonused main function

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

ceph-disk: Adding retry loop in get_partition_dev()

There is very rare cases where get_partition_dev() is called before the actual partition is available in /sys/block/<device>.

It appear that waiting a very short is usually enough to get the partition beein populated.

Analysis:
update_partition() is supposed to be enough to avoid any racing between events sent by parted/sgdisk/partprobe and
the actual creation on the /sys/block/<device>/* entrypoint.
On our CI that race occurs pretty often but trying to reproduce it locally never been possible.

This patch is almost a workaround rather than a fix to the real problem.
It offer retrying after a very short to be make a chance the device to appear.
This approach have been succesful on the CI.

Note his patch is not changing the timing when the device is perfectly created on time and just differ by a 1/5th up to 2 seconds when the bug occurs.

A typical output from the build running on a CI with that code.
command_check_call: Running command: /usr/bin/udevadm settle --timeout=600
get_dm_uuid: get_dm_uuid /dev/sda uuid path is /sys/dev/block/8:0/dm/uuid
get_partition_dev: Try 1/10 : partition 2 for /dev/sda does not in /sys/block/sda
get_partition_dev: Found partition 2 for /dev/sda after 1 tries
        get_dm_uuid: get_dm_uuid /dev/sda uuid path is /sys/dev/block/8:0/dm/uuid
get_dm_uuid: get_dm_uuid /dev/sda2 uuid path is /sys/dev/block/8:2/dm/uuid

fixes: #19428

Signed-off-by: Erwan Velu <erwan@redhat.com>

mgr: rm nonused main function

Signed-off-by: Wei Jin <wjin.cn@gmail.com>

Merge pull request #14174 from shinobu-x/wp-sk-typo-osdmonitor-20170328

mon/OSDMonitor: Clean up: delete extra S signature for plural

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

ceph-disk: Populate mount options when running "list"

Also tidy up by moving duplicated code into a function

Fixes: http://tracker.ceph.com/issues/17331
Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

crc32c,compressor: only compile functions on supported arch

so we can still compile if certain header file (<sys/auxv.h>) used
for detecting cpu capabilities on ppc is missing on x86 host machine.

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

common/simple_spin: use __ppc_yield() on all powerpc archs

__ppc_yield() is declared in sys/platform/ppc.h by glibc, for better
portability we just use the inlined assembly here. the shared resource
hints are supported by PowerPC ISA 2.06 but on older PowerPC cores, they
are no-ops. so it's fine to do this way.

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

test: add test for rbd import,import-diff,copy with sparse-size option

Signed-off-by: Ning Yao <yaoning@unitedstack.com>

librbd: add sparse copy support

Fixes: fix issue #15648
http://tracker.ceph.com/issues/15648

Signed-off-by: yaoning <yaoning@unitedstack.com>

rbd: import/import-diff real thin-provision image

currently, import/import-diff check whether object is zero based on the whole object
therefore, holes inside object will be filled with zero

Signed-off-by: yaoning <yaoning@unitedstack.com>

rbd: add sparse size support in rbd

add --sparse_size in rbd command options
add calc_sparse_extent in rbd utils

Signed-off-by: yaoning <yaoning@unitedstack.com>

client/SyntheticClient.cc: Fix warning in random_walk

Fixes: http://tracker.ceph.com/issues/19445
Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

Merge pull request #12243 from rzarzynski/wip-rgw-bulkupload

rgw: add support for the BulkUpload of Swift API
Reviewed-by: Orit Wasserman <owasserm@redhat.com>

doc/rgw: update section on global quotas

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

radosgw-admin: new 'global quota' commands update period config

Fixes: http://tracker.ceph.com/issues/19409
Signed-off-by: Casey Bodley <cbodley@redhat.com>

rgw: load period config separately when there's no current period

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

rgw: synchronize period config with period updates

read period config from local storage on RGWPeriod::update(), and write
the new config to storage on RGWPeriod::reflect()

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

rgw: allow RGWPeriodConfig to be stored separately

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

rgw: constify some rgw_pool args

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

rgw: remove unused RGWPeriod::update(RGWZoneGroupMap)

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

Merge pull request #14120 from wangzhengyong/uninitialized

rgw: fix uninitialized fields

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