qa/tasks: add thrash test for persistent write log cache

add thrash test for persistent write log cache. run rbd bench
on persistent write log cache, thrashes rbd bench, test the
recovery function of persistent write log cache.

Signed-off-by: Yin Congmin <congmin.yin@intel.com>

Merge pull request #46514 from adk3798/fqdn-autotune

mgr/cephadm: use host shortname for osd memory autotuning

Reviewed-by: Guillaume Abrioux <gabrioux@redhat.com>
Reviewed-by: Redouane Kachach <rkachach@redhat.com>

Merge pull request #46517 from guits/cephadm-upgrade-status-is_paused

cephadm: add 'is_paused' field in orch status output

Reviewed-by: Adam King <adking@redhat.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge pull request #46539 from adk3798/master-main-cleanup2

mgr/cephadm: some master -> main cleanup

Reviewed-by: David Galloway <dgallowa@redhat.com>
Reviewed-by: Redouane Kachach <rkachach@redhat.com>

Merge pull request #44234 from lxbsz/wip-53504

client: always return ESTALE directly in handle_reply

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Milind Changire <mchangir@redhat.com>

Merge pull request #44247 from neesingh-rh/fix-52982

client: Inode::hold_caps_until is time from monotonic clock now.

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

Merge pull request #46210 from vshankar/wip-54701

mds: do not assert early on when issuing client leases

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #44335 from SMIL-Infra/export-unlinked-dir

mds: fix crash when exporting unlinked dir

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Xiubo Li <xiubli@redhat.com>

Merge pull request #46078 from kotreshhr/fuse-directory-dacs-issue

client/fuse: Fix directory DACs overriding for root

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Ramana Raja <rraja@redhat.com>

Merge pull request #46267 from tshacked/fix_24894

client: allow overwrites to file with size greater than the max_file_size

Reviewed-by: Venky Shankar <vshankar@redhat.com>

Merge pull request #46371 from aliakseimakarau/fs_multifs_fix

mds/fsmap: proper update of filesystems

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Laura Flores <lflores@redhat.com>

Merge pull request #46398 from lxbsz/feature-cleanup

mds, client: remove useless feature required code

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Jos Collin <jcollin@redhat.com>
Reviewed-by: Dhairya Parmar <dparmar@redhat.com>

Merge pull request #46405 from mgfritch/ceph-backport-fw-slash

script/ceph-backport.sh: consume excess forward slash chars

Reviewed-by: Adam King <adking@redhat.com>
Reviewed-by: Kefu Chai <tchaikov@gmail.com>

Merge pull request #46263 from rhcs-dashboard/rbd-mirroring-snapshot

mgr/dashboard: rbd mirroring snapshot support

Reviewed-by: Aashish Sharma <aasharma@redhat.com>
Reviewed-by: Avan Thakkar <athakkar@redhat.com>
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: Deepika Upadhyay <dupadhya@redhat.com>
Reviewed-by: Nizamudeen A <nia@redhat.com>

Merge pull request #46261 from dang/wip-dang-zipper-req-id

RGW - Zipper - Provide default get_new_req_id()

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

Merge pull request #46403 from soumyakoduri/wip-skoduri-syncpolicy-crash

rgw: Avoid dereferencing nullptr while configuring bucket sync policy

Reviewed-by: Casey Bodley <cbodley@redhat.com>
Reviewed-by: Shilpa Jagannath <smanjara@redhat.com>
Reviewed-by: Matt Benjamin <mbenjami@redhat.com>

Merge pull request #46479 from rhcs-dashboard/rbd-mirroring-list-primary

mgr/dashboard: add rbd primary info

Reviewed-by: Sarthak Gupta <sarthak.dev.0702@gmail.com>
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>

Merge pull request #46473 from liu-chunmei/crimson-no-rollback

crimson/osd: don't do rollback when object not exists

Merge pull request #46045 from idryomov/wip-qa-rwl-on-tmpfs

qa/suites/rbd: place cache file on tmpfs for xfstests

Reviewed-by: Deepika Upadhyay <dupadhya@redhat.com>

Merge pull request #46440 from ronen-fr/wip-rf-scrub-trimisbad

osd/scrub: do not start scrubbing if the PG is snap-trimming

Reviewed-by: Samuel Just <sjust@redhat.com>
Reviewed-by: Laura Flores <lflores@redhat.com>

cephadm: add 'is_paused' field in orch status output

it can be convenient for automation tool if `orch status` output reports
whether the upgrade is currently paused.

Fixes: https://tracker.ceph.com/issues/55843
Signed-off-by: Guillaume Abrioux <gabrioux@redhat.com>

Merge pull request #46538 from adamemerson/wip-remove-unbuilt-rgw

rgw: Remove rgw_json_enc.cc which is not built

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

Merge pull request #46396 from xxhdx1985126/wip-backref-buffer-optimization

crimson/os/seastore/backref: don't merge in-cache backrefs that has already been released

Reviewed-by: Samuel Just <sjust@redhat.com>
Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>

mgr/cephadm: some master -> main cleanup

Signed-off-by: Adam King <adking@redhat.com>

Merge pull request #46537 from cbodley/wip-55901

rgw: fix un/signed comparison warnings in rgw_admin.cc

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

Merge pull request #46536 from cbodley/wip-55898

rgw: fix un/signed comparison warnings in rgw_sync.cc

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

rgw: Remove rgw_json_enc.cc which is not built

This was removed from the build in a refactor but not removed from the tree.

Signed-off-by: Adam C. Emerson <aemerson@redhat.com>

rgw: fix un/signed comparison warnings in rgw_admin.cc

Fixes: https://tracker.ceph.com/issues/55901
Signed-off-by: Casey Bodley <cbodley@redhat.com>

mgr/cephadm: use hostname from crush map for osd memory autotuning

Fixes: https://tracker.ceph.com/issues/55841
Signed-off-by: Adam King <adking@redhat.com>

Merge pull request #46376 from adk3798/cache-size

mgr/cephadm: store device info separately from rest of host cache

Reviewed-by: Redouane Kachach <rkachach@redhat.com>

Merge pull request #46502 from adk3798/default-image-master-main

cephadm: master -> main updates

Reviewed-by: Michael Fritch <mfritch@suse.com>
Reviewed-by: Redouane Kachach <rkachach@redhat.com>

RGW - Zipper - Provide default get_new_req_id()

Now that RadorStore's get_new_req_id() just returns a random number, use
that as the default implementation for SAL.

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

rgw: fix un/signed comparison warnings in rgw_sync.cc

Fixes: https://tracker.ceph.com/issues/55898
Signed-off-by: Casey Bodley <cbodley@redhat.com>

mgr/dashboard: snapshot mirroring from dashboard

Enable snapshot mirroring from the Pools -> Image

Also show the mirror-snapshot in the image where snapshot is enabled

When parsing images if an image has the snapshot mode enabled, it will
try to  run commands that don't work with that mode. The solution was
not running those for now and appending the mode in the get call.

Fixes: https://tracker.ceph.com/issues/55648
Signed-off-by: Pere Diaz Bou <pdiazbou@redhat.com>
Signed-off-by: Nizamudeen A <nia@redhat.com>
Signed-off-by: Aashish Sharma <aasharma@redhat.com>
Signed-off-by: Avan Thakkar <athakkar@redhat.com>

Merge pull request #46452 from ceph/fix-codeowners_api-master

codeowners: remove api team

Reviewed-by: Sarthak Gupta <sarthak.dev.0702@gmail.com>
Reviewed-by: nSedrickm <NOT@FOUND>
Reviewed-by: Nizamudeen A <nia@redhat.com>

mds: do not assert early on when issuing client leases

Move the relevant ceph_assert()'s when the MDS is issuing leases
to clients since clients may not be capable of reading projected
linkages.

Fixes: http://tracker.ceph.com/issues/54701
Signed-off-by: Venky Shankar <vshankar@redhat.com>

Merge PR #46516 into main

* refs/pull/46516/head:
doc/dev/developer_guide/testing_integration_tests: document how to test custom kernels

Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Anthony D Atri <anthony.datri@gmail.com>
Reviewed-by: Venky Shankar <vshankar@redhat.com>

mds, client: remove useless feature required code

This code was added but never ever used.

Signed-off-by: Xiubo Li <xiubli@redhat.com>

Merge pull request #46165 from rishabh-d-dave/qa-omit-sudo

qa/cephfs: set omit_sudo False when sudo is set to True

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Dhairya Parmar <dparmar@redhat.com>
Reviewed-by: Nikhilkumar Shelke <nshelke@redhat.com>

Merge pull request #46168 from rishabh-d-dave/fix-caps-helper

qa/cephfs: fix minor bug in caps_helper.py's run_mon_cap_tests()

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Ramana Raja <rraja@redhat.com>

Merge pull request #40434 from rishabh-d-dave/fs-refactor-method-in-mount

qa/cephfs: modify get_key_from_keyfile() in mount.py

Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
Reviewed-by: Jos Collin <jcollin@redhat.com>

mgr/dashboard: add rbd primary info

Signed-off-by: Pere Diaz Bou <pdiazbou@redhat.com>

Merge pull request #46522 from tchaikov/wip-crimson-logging

crimson/osd: reset logger before exit

Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com>

Merge pull request #46416 from tchaikov/wip-debian-dh-python3

debian: python3 related cleanups

Reviewed-by: Redouane Kachach <rkachach@redhat.com>
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>

crimson/osd: reset logger before exit

* extract the code to set logging fstream into a dedicated function
* do not reset logging until the end of the seastar application.

before this change, `reset_logger` is created in the
`if (auto log_file = local_conf()->log_file; !log_file.empty())` branch,
so its life cycle ends when the `if` block ends. in other words,
the cerr fstream is used for logging after the `if` block ends.
this is not the expected behavior.

after this changge, `reset_logger` is created out of the `if` block.
so we won't reset the logger back to `cerr` until the lambda passed to
`seastar::async()` exits.

Signed-off-by: Kefu Chai <tchaikov@gmail.com>

crimson/os/seastore/backref: tolerate removing non-existing backrefs

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/cache: don't index already removed backref entries in Cache::backref_buffer

This is needed by extent splitting, and can avoid inserting/removing
unnecessary backrefs

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

Merge pull request #46483 from yaarith/rook-telemetry

mgr/telemetry: add Rook data

Reviewed-by: Laura Flores <lflores@redhat.com>

Merge pull request #46417 from xxhdx1985126/wip-gc-parallel-live_extent_retrieval

crimson/os/seastore/segment_cleaner: parallel live extents retrieval

Reviewed-by: Samuel Just <sjust@redhat.com>

doc/dev/developer_guide/testing_integration_tests: document how to test custom kernels

Fixes: https://tracker.ceph.com/issues/55530
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

qa/suites/rbd: place cache file on tmpfs for xfstests

The RWL mode needs DAX and is dog slow otherwise -- qemu_xfstests.yaml
job always hits the 6 hour max_job_time limit.

As our tmpfs instance is limited and qemu_xfstests.yaml opens three
images at the same time, reduce the "big cache" size to 5G.  This facet
was added to iron out 32-bit head/tail pointer issues and 5G still does
the job there.

Going through the loop device is needed because tmpfs doesn't support
O_DIRECT.

Fixes: https://tracker.ceph.com/issues/55400
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

Merge pull request #46480 from cfsnyder/wip-cfsnyder-device-classes-in-service-spec

python-common: allow crush device class to be set from osd service spec

Reviewed-by: Adam King <adking@redhat.com>

test/crimson/seastore: add test case for parallel extent retrieval

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

crimson/os/seastore/cache: make access to Transaction::read_set atomic

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

Merge pull request #46505 from rhcs-dashboard/fix-backports_main-main

script/ceph-backport.sh: deal with main branch

Reviewed-by: Laura Flores <lflores@redhat.com>
Reviewed-by: nmshelke <NOT@FOUND>
Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #46186 from rhcs-dashboard/add-daemon-logs

mgr/dashboard: Add daemon logs tab to Logs component

Reviewed-by: Avan Thakkar <athakkar@redhat.com>
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: Nizamudeen A <nia@redhat.com>
Reviewed-by: sunilangadi2 <NOT@FOUND>

Merge pull request #46283 from MrFreezeex/mixin-config

ceph-mixin: fix linting issue and add cluster template support

Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: Pere Diaz Bou <pdiazbou@redhat.com>

Merge pull request #46454 from idryomov/wip-rbd-unlink-newest-snap-at-capacity

librbd: unlink newest mirror snapshot when at capacity, bump capacity

Reviewed-by: Arthur Outhenin-Chalandre <arthur.outhenin-chalandre@cern.ch>
Reviewed-by: Mykola Golub <mgolub@suse.com>

Merge pull request #46434 from idryomov/wip-rbd-preserve-non-primary-snap

rbd-mirror: don't prune non-primary snapshot when restarting delta sync

Reviewed-by: Arthur Outhenin-Chalandre <arthur.outhenin-chalandre@cern.ch>
Reviewed-by: Mykola Golub <mgolub@suse.com>

qa/cephfs: modify get_key_from_keyfile() in mount.py

CephFSMount.get_key_from_keyfile() should raise an exception instead of
returning None if key is not found in keyring file.

Fixes: https://tracker.ceph.com/issues/50010
Signed-off-by: Rishabh Dave <ridave@redhat.com>

Merge pull request #46411 from pcuzner/add-serial-numbers

cephadm: Add server serial info to gather-facts

Reviewed-by: Adam King <adking@redhat.com>

Merge pull request #46444 from rkachach/fix_issue_55800

mgr/cephadm: check if a service exists before trying to restart it

Reviewed-by: Adam King <adking@redhat.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>

Merge pull request #46445 from rkachach/fix_issue_55801

mgr/cephadm: capture exception when not able to list upgrade tags

Reviewed-by: Adam King <adking@redhat.com>

Merge pull request #46481 from guits/cephadm-custom-names-osd-adoption

cephadm: fix osd adoption with custom cluster name

Reviewed-by: Adam King <adking@redhat.com>

qa/suites/rbd: refactor persistent-writeback-cache suite

Rename to pwl-cache, introduce home subdirectory and 4-cache-path.yaml.
No functional changes.

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

Merge pull request #39002 from ceph/wip-rgw-multisite-reshard

rgw multisite: bucket reshard work in progress

Reviewed-by: Matt Benjamin <mbenjami@redhat.com>

Merge pull request #45470 from ceph/wip-setx

run-backend-api-tests.sh: set -x for Jenkins job debugging

Reviewed-by: Ernesto Puerta <epuertat@redhat.com>

Merge pull request #44684 from zenomri/wip-omri-tracing-compiled

tracer: set tracing compiled in by default

script/ceph-backport.sh: deal with main branch

Signed-off-by: Ernesto Puerta <epuertat@redhat.com>

Merge pull request #43216 from k0ste/fix_47537

doc/mgr: Document wildcard to expose Prometheus metrics for all RBD pools and namespaces

python-common: allow crush device class to be set from osd service spec

Adds crush_device_class parameter to DriveGroupSpec so that device class
can be set via service specs with cephadm.

Fixes: https://tracker.ceph.com/issues/55813
Signed-off-by: Cory Snyder <csnyder@iland.com>

cephadm: master -> main updates

As part of the work to change the name of the
"master" branch to "main" in Ceph repos a few things
should change in the cephadm binary as well

Signed-off-by: Adam King <adking@redhat.com>

Merge pull request #46019 from yushu20171007/fix_issue_55422

common: notify all when max backlog reached in OutputDataSocket

Reviewed-by: Matt Benjamin <mbenjami@redhat.com>

Merge pull request #46501 from rhcs-dashboard/fix-55826-master

qa: fix teuthology master branch ref

Reviewed-by: amathuria <NOT@FOUND>
Reviewed-by: Avan Thakkar <athakkar@redhat.com>

qa: fix teuthology master branch ref

Fixes: https://tracker.ceph.com/issues/55826
Signed-off-by: Ernesto Puerta <epuertat@redhat.com>

doc/mgr: Document wildcard to expose Prometheus metrics for all RBD pools and namespaces

Fixes: https://tracker.ceph.com/issues/47537
Signed-off-by: Konstantin Shalygin <k0ste@k0ste.ru>

Merge pull request #46272 from sshambar/bug-55664

cephadm: preserve cephadm user during RPM upgrade

Reviewed-by: Kefu Chai <tchaikov@gmail.com>
Reviewed-by: Michael Fritch <mfritch@suse.com>

crimson/osd: don't do rollback when object not exists

otherwise the follwoing operation's obc change will be discarded.

Signed-off-by: chunmei-liu <chunmei.liu@intel.com>

script/ceph-backport.sh: consume excess forward slash chars

when attempting to determine the upstream remote
(e.g. `git@github.com:/ceph/ceph`)

Fixes: https://tracker.ceph.com/issues/55775
Signed-off-by: Michael Fritch <mfritch@suse.com>

Merge pull request #46488 from jtlayton/teuth-branch-fix

qa: remove .teuthology_branch file

qa: remove .teuthology_branch file

This was originally added to help support the py2 -> py3 conversion.
That's long since complete so we should be able to just remove this file
now.

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

Merge pull request #46487 from jtlayton/teuth-branch-fix

qa: fix .teuthology_branch file in qa/

test/rgw/multisite: enable zonegroup resharding feature

qa/tasks/rgw_multisite.py uses 'zonegroup set' to create zonegroups from
their json format. this doesn't enable any of the supported zonegroup
features by default, so this adds the 'enabled_features' field to the
json representations

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

qa: fix .teuthology_branch file in qa/

According to teuthology-suite:

  -t <branch>, --teuthology-branch <branch>
                              The teuthology branch to run against.
                              Default value is determined in the next order.
                              There is TEUTH_BRANCH environment variable set.
                              There is `qa/.teuthology_branch` present in
                              the suite repo and contains non-empty string.
                              There is `teuthology_branch` present in one of
                              the user or system `teuthology.yaml` configuration
                              files respectively, otherwise use `main`.

The .teuthology_branch file in the qa/ dir currently points at "master".
Change it to point to "main".

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

cephadm: fix osd adoption with custom cluster name

When adopting Ceph OSD containers from a Ceph cluster with a custom name, it fails
because the name isn't propagated in unit.run.
The idea here is to change the lvm metadata and enforce 'ceph.cluster_name=ceph'
given that cephadm doesn't support custom names anyway.

Fixes: https://tracker.ceph.com/issues/55654
Signed-off-by: Adam King <adking@redhat.com>
Co-authored-by: Guillaume Abrioux <gabrioux@redhat.com>

Merge pull request #46474 from idryomov/wip-rbd-codeowners

CODEOWNERS: add RBD team

Reviewed-by: Deepika Upadhyay <dupadhya@redhat.com>

CODEOWNERS: add RBD team

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

crimson/os/seastore/segment_cleaner: retrieve different live extents in parallel

Signed-off-by: Xuehan Xu <xxhdx1985126@gmail.com>

mgr/cephadm: capture exception when not able to list upgrade tags
Fixes: https://tracker.ceph.com/issues/55801
Signed-off-by: Redouane Kachach <rkachach@redhat.com>

osd/scrub: do not start scrubbing if the PG is snap-trimming

Both 'snap-trim' and 'snaptrim-wait' PG states now prevent
scrub from starting.

Background:
A PG should not be scrubbed and trimmed concurrently. Unlike
write operations, snap trimming does not verify that a targeted
object is not in the executing scrub's chunk.
The trimmer always checked for active scrubs before starting. The
scrubber - did not. This PR fixes that omission.

Fixes: https://tracker.ceph.com/issues/49525
Signed-off-by: Ronen Friedman <rfriedma@redhat.com>

mgr/telemetry: add Rook data

Add the first Rook data collection to telemetry's basic channel.

We choose to nag with this collection since we wish to know the volume
of Rook deployments in the wild.

The next Rook collections should have consecutive numbers (basic_rook_v02,
basic_rook_v03, ...).

See tracker below for more details.

Fixes: https://tracker.ceph.com/issues/55740
Signed-off-by: Yaarit Hatuka <yaarit@redhat.com>

Merge pull request #46382 from rzarzynski/wip-crimson-op-tracking-3

crimson/osd: add support for historic & slow op tracking

Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #46437 from cyx1231st/wip-seastore-tune-and-fixes

crimson/os/seastore/segment_cleaner: tune and fixes around reclaiming

Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #46193 from ljflores/wip-zero-detection-off-by-default

os/bluestore: turn bluestore zero block detection off by default

rgw: restore check for empty olh name on reshard

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

test/rgw: fix test case for empty-OLH-name cleanup

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

Merge pull request #46367 from 0xavi0/dbstore-default-dbdir-rgw-data

rgw/dbstore: change default value of dbstore_db_dir to /var/lib/ceph/radosgw

Reviewed-by: Soumya Koduri <skoduri@redhat.com>

Merge pull request #46395 from cbodley/wip-backport-create-issue-assigned-to

backport-create-issue: copy 'Assignee' of original issue to backports

Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: Ilya Dryomov <idryomov@redhat.com>
Reviewed-by: Venky Shankar <vshankar@redhat.com>
Reviewed-by: Neha Ojha <nojha@redhat.com>

Merge pull request #46415 from neha-ojha/wip-cw-core

.github/CODEOWNERS: tag core devs on core PRs

Reviewed-by: Laura Flores <lflores@redhat.com>

librbd: unlink newest mirror snapshot when at capacity, bump capacity

CreatePrimaryRequest::unlink_peer() invoked via "rbd mirror image
snapshot" command or via rbd_support mgr module when creating a new
scheduled mirror snapshot at rbd_mirroring_max_mirroring_snapshots
capacity on the primary cluster can race with Replayer::unlink_peer()
invoked by rbd-mirror when finishing syncing an older snapshot on the
secondary cluster.  Consider the following:

   [ primary: primary-snap1, primary-snap2, primary-snap3
     secondary: non-primary-snap1 (complete), non-primary-snap2 (syncing) ]

0. rbd-mirror is syncing snap1..snap2 delta
1. rbd_support creates primary-snap4
2. due to rbd_mirroring_max_mirroring_snapshots == 3, rbd_support picks
   primary-snap3 for unlinking
3. rbd-mirror finishes syncing snap1..snap2 delta and marks
   non-primary-snap2 complete

   [ snap1 (the old base) is no longer needed on either cluster ]

4. rbd-mirror unlinks and removes primary-snap1
5. rbd-mirror removes non-primary-snap1
6. rbd-mirror picks snap2 as the new base
7. rbd-mirror creates non-primary-snap3 and starts syncing snap2..snap3
   delta

   [ primary: primary-snap2, primary-snap3, primary-snap4
     secondary: non-primary-snap2 (complete), non-primary-snap3 (syncing) ]

8. rbd_support unlinks and removes primary-snap3 which is in-use by
   rbd-mirror

If snap trimming on the primary cluster kicks in soon enough, the
secondary image becomes corrupted: rbd-mirror would eventually finish
"syncing" non-primary-snap3 and mark it complete in spite of bogus data
in the HEAD -- the primary cluster OSDs would start returning ENOENT
for snap trimmed objects.  Luckily, rbd-mirror's attempt to pick snap3
as the new base would wedge the replayer with "split-brain detected:
failed to find matching non-primary snapshot in remote image" error.

Before commit a888bff8d00e ("librbd/mirror: tweak which snapshot is
unlinked when at capacity") this could happen pretty much all the time
as it was the second oldest snapshot that was unlinked.  This commit
changed it to be the third oldest snapshot, turning this into a more
narrow but still very much possible to hit race.

Unfortunately this race condition appears to be inherent to the way
snapshot-based mirroring is currently implemented:

a. when mirror snapshots are created on the producer side of the
   snapshot queue, they are already linked
b. mirror snapshots can be concurrently unlinked/removed on both
   sides of the snapshot queue by non-cooperating clients (local
   rbd_mirror_image_create_snapshot() vs remote rbd-mirror)
c. with mirror peer links off the list due to (a), there is no
   existing way for rbd-mirror to persistently mark a snapshot as
   in-use

As a workaround, bump rbd_mirroring_max_mirroring_snapshots to 5 and
always unlink the newest snapshot (i.e. slot 4) instead of the third
oldest snapshot (i.e. slot 2).  Hopefully this gives enough leeway,
as rbd-mirror would need to sync two snapshots (i.e. transition from
syncing 0-1 to 1-2 and then to 2-3) before potentially colliding with
rbd_mirror_image_create_snapshot() on slot 4.

Fixes: https://tracker.ceph.com/issues/55803
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>