qa: wrap quiesce verification to dump debugging on error

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

qa: update quiesce tests for control via locallock

- Verify multirank quiesce incl. caps issued.
- Unset splitauth experimental

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

qa: set archive path in vstart_runner

For storing misc. test artifacts.

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

qa: refactor CephFSMount.kill_background to optionally kill all background jobs

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

qa: use kwarg for rank parameter

Otherwise it gets included in the *args list. This is necessary after commit
`qa: simplify calls to (rank|mds)_(tell|asok)`.

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

qa: simplify calls to (rank|mds)_(tell|asok)

Instead of requiring the caller to put the arguments in a list, allow passing
as regular arguments.

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

Revert "pybind/mgr/volumes: block quiesce for critical .meta file"

This reverts commit 16e50abbd42a4b4f740816ba5f60bc259a6a96bf.

This flag is no longer necessary as the volumes plugin issues quiesce calls
against the data (i.e. root) directory of the subvolume rather than the
subvolume directory (with its associated .meta file).

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

mds: remove is_root indication on quiesce_inode op

This is no longer necessary with the change to a LocalLock quiescelock.

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

mds: prevent new lock cache cons when invalidating an existing one

The previous scheme invalidated a lock cache and then immediately removed it
from its Capability list. The lock cache would eventually be deleted but a new
one could be constructed shortly after. The main reason for this is that simply
invalidating the lock cache does not drive a state change in the local locks
preventing new writers. This is mostly important for acquiring the quiescelock.

This commit also corrects a bug where a MDLockCache would be created for a
given opcode type (like create) when the capability does not have the issued
cap (CEPH_CAP_DIR_CREATE). The bug would not cause any negative side-effects
but would hold locks unnecessarily when only MDS ops (and not the client
executing ops asynchronously) are acquiring the locks.

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

mds: use XLOCK_WAIT For local lock xlockers

This avoids waking up all waiters when only WAIT_XLOCK waiters should wake.

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

mds: prevent new wrlocks on LocalLock if there exists any xlock waiter

Otherwise, an xlock waiter can become starved as a LocalLock supports multiple
writers.

Strictly speaking, a new lock state would be appropriate for this but we cheat
frequently with the LocalLock -- there is only one state. All transition checks
are already manually performed by the Locker.

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

mds: block import discover when parent directory inode is quiesced

This is to prevent two racing ranks quiescing some root from exporting a tree
under a completely quiesced directory (inode). The state of that imported tree
may take time to quiesce and cause the root to be QUIESCED before all inodes
under it are actually quiesced.

If a dirfrag to be imported is discovered before the parent is quiesced, then
the quiesce traversal will issue a quiesce_inode op normally for parent which
will attempt to authpin the parent. That will block if the export is still
in-progress (causing quiesce to wait for the export to finish or abort).

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

mds: avoid issuing exclusive caps to clients lacking w caps

Signed-off-by: Leonid Usov <leonid.usov@ibm.com>

mds: print lock cache during invalidation

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

mds: use inodeno_t to track quiesce requests

If a CInode is removed from cache before the quiesce_inode request can process
it (and pin it in cache), a new CInode may be created with the same address.
That pointer still exists in MutationImpl::quiesce_ops and would prevent
issuing a quiesce_inode op for the new inode.

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

mds: dispatch quiesce_inode ops after dir traversal

Mostly to avoid the possibility of the CDir::items changing during iteration
when calling sub-ops.

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

mds: remove quiescelock handling for SimpleLock type

This is no longer necessary with the conversion to LocalLock.

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

mds: quiescelock as local lock + cap masking

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Signed-off-by: Leonid Usov <leonid.usov@ibm.com>

qa: run quiesce unit tests in fs:functional

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Fixes: https://tracker.ceph.com/issues/63664

qa: add quiesce protocol unit tests

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Fixes: https://tracker.ceph.com/issues/63664

qa: detect partial migrations during large config of dist epin

This method would wrongly "succeed" when looking for setup of distributed
ephemerally pinned directory fragments. If the migrator splits a subtree during
the course of migration (to reduce the migration size) then the operation may
not actually be complete.

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

qa: use stdin-killer to timeout run_shell_payload

- simplify argument processing / forwarding
- use stdin-killer to kill all sub-processes of the shell
- do not needlessly use run_shell to execute the command as it adds a timeout
  to the stdout/stderr processing
- provide a stdin (PIPE) by default otherwise teuthology's code closes stdin
  and triggers stdin-killer to timeout the shell.
- use a 15 minute timeout by default

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

qa: simplify run_shell argument processing

No functional changes.

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

doc: add dev docs for quiesce protocol

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

pybind/mgr/volumes: block quiesce for critical .meta file

This file must remain accessible during quiesce to allow snapshots to be taken
on a subvolume.

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

mds: add vxattr to block quiesce on an inode

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

mds: convert encoded ephemeral dist pin to flags

To use space efficiently, convert this field to a 8 bit flags field that is
backward compatible.

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

mds: add counter to throttle quiesce

So a storm of quiesce operations do not affect normal MDS operations.

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

mds: add quiesce set feature flag

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Fixes: https://tracker.ceph.com/issues/63664

mds: skip non-head inodes for quiesce

These are accessible only via snapshots.

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

mds: add quiesce op

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Fixes: https://tracker.ceph.com/issues/63664

mds: print all SimpleLock flags in debug output

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

mds: pretty print mutation when dumping lock

Now looks like:

    ... (iquiesce xlock x=1 by request(mds.0:2785 nref=6)) ...

instead of:

    ... (iquiesce xlock x=1 by 0xdeadbeef)) ...

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

mds: add new inode quiescelock

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

mds: use 128 bits for waiters on MDSCacheObject

Adding a new inode lock will overflow inode wait bits into the MDSCacheObject
wait bits. Make space for the quiescelock.

This includes a minor refactor to no longer attempt scoping the set of masks we
test in MDSCacheObject::waiting when calling MDSCacheObject::is_waiter_for.
This optimization wasn't worth the overhead and would be awkard to keep as
std::bitset cannot be used as a key for a std::multimap (easily).  Instead, we
use the sequence number as a key which helps us to avoid allocating another map
whenever we call MDSCacheObject::take_waiting.

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

mds: provide mechanism to authpin while freezing

When a subtree is freezing, it's no longer possible to acquire new authpins.
This is a problem when a compound request like quiescing a subtree is trying to
acquire authpins for each sub-op. This creates a situation where some quiesce
sub-ops complete with authpins (thereby preventing the tree from becoming
"frozen") and new sub-ops cannot acquire authpins (because the tree is
"freezing"). To circumvent this, allow some authpin requests to proceed if
FLAG_BYPASSFREEZING is set.

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

mds: add command to get specific op

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

mds: finish request before completing internal req

Importantly, this marks the request dead so it cannot be killed.

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

mds: complete internal op if killed

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

mds: avoid killing dead requests

This is important when you have:

* sub-op A finishes and resumes parent
* parent kills all sub-ops (including sub-op A)

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

mds: add command to kill request

Fixes: https://tracker.ceph.com/issues/63374
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

mds: add path argument to `ops` and `dump tree` to stream result to local file

This file can be collected during testing to avoid sending a large JSON result
over the `ceph tell`/asok interface.

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

mds: print internal_request filepaths if present

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

mds: add more information to debug message

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

mds: remove redundant parenthesis

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

mds: implement Mutation::dump method

So we can dump it from asok commands.

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

mds: make LockType fields const

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

mds: annotate mdr with try_rdlock_snap_layout failure

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

mds: refactor if into switch

To make it easier to read and match other constructs in this code.

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

mds: call Locker method using this

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

mds: simplify assert

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

mds: dump locks passed to Locker::acquire_locks

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

mds: add LockOp::print method for debugging

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

mds: use new insert template via print

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

mds: add request result to mutation for analysis by tests

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

mds: add comment on locking order rules

From a past commit.

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

mds: allow specifying rdlock position

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

mds: remove dead method

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

common: provide a template for object dumps

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

common: support long running ops without slow warnings

A quiesce op in the MDS may run for an indeterminate amount of time. Warnings
should not be generated.

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

common: simplify loop

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

common: add JSONFormatterFile class

To stream JSON to a file for collection (for testing).  The idea here that
trying to send a very large JSON result over `ceph tell` is best to be avoided.

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

common: use more efficient vector for stack

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

include: use larger int for large gathers

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

Merge pull request #56266 from cbodley/wip-qa-rgw-lua-pool-application

qa/rgw: ignore POOL_APP_NOT_ENABLED in rgw/lua subsuite

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

Merge pull request #56220 from linuxbox2/wip-nfs-rename

rgw_file: fix mv/rename cases broken by zipper integration

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

Merge pull request #56225 from linuxbox2/wip-cors-fix

rgw: Add missing empty checks to the split string in is_string_in_set().

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

Merge pull request #56044 from pr0ton11/rgw-disable-signature-url

RGW: allow user disabling presigned urls in rgw configuration

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

Merge pull request #55795 from kchheda3/wip-rename-notif-events

rgw/notification: Make the LifeCycle & Replication events AWS Compatible and Support s3:Replication:Delete event

Reviewed-by: Yuval Lifshitz <ylifshit@redhat.com>
Reviewed-by: Matt Benjamin <mbenjamin@redhat.com>

Merge pull request #56333 from molpako/patch-3

doc/cephfs: fix architecture link to correct relative path

Reviewed-by: Zac Dover <zac.dover@proton.me>

Merge pull request #55595 from VallariAg/wip-nvmeof-test-v3

qa/suite/rbd/nvmeof: Deploy multiple gateways and namespaces

Reviewed-by: Barak Davidov <barakda@il.ibm.com>
Reviewed-by: Aviv Caro <Aviv.Caro@ibm.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>

doc/cephfs: fix architecture link to correct relative path

Signed-off-by: molpako <35787116+molpako@users.noreply.github.com>

Merge pull request #55864 from rosinL/wip-speedup-check-generated

test/encoding/check-generated: parallel testing by type

Merge pull request #56203 from athanatos/sjust/wip-protoc-dependencies

crimson: add crimson-osd rpm and deb runtime dependencies for protobuf inherited from seastar

Merge pull request #56070 from phlogistonjohn/jjm-cephadm-sudo-cmd-auditing

cephadm: sudo cmd auditing

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

Merge pull request #56178 from adk3798/cephadm-pub-net-config-check-out-of-date

mgr/cephadm: refresh public_network for config checks before checking

Reviewed-by: John Mulligan <jmulligan@redhat.com>

Merge pull request #56175 from asm0deuz/bootstrap_no_container

cephadm: bootstrap fails if no container engine installed

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

Merge pull request #56147 from adk3798/stopped-daemon-shell

cephadm: fix `cephadm shell --name <daemon-name>` for stopped/failed daemon

Reviewed-by: John Mulligan <jmulligan@redhat.com>

Merge pull request #56143 from insatomcat/timemaster

cephadm.py: add timemaster to timesync services list

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

Merge pull request #56075 from phlogistonjohn/jjm-enable-black-cephadm

cephadm: enforce black formatting for cephadmlib

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

Merge pull request #55883 from thegreenbear/ingress-haproxy-bind-ipv4v6

Ingress service: make HAProxy to listen on IPv4 and IPv6 …

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

Merge pull request #55874 from adk3798/only-bind-to-network

mgr/cephadm: allow grafana and prometheus to only bind to specific network

Reviewed-by: John Mulligan <jmulligan@redhat.com>

Merge pull request #55645 from ronen-fr/wip-rf-nextobj

osd: improved get_objects_by_prefixes() ergonomics

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

qa/suite/rbd/nvmeof: Deploy multiple gateways and namespaces

1. Deploy 2 gateways on different nodes, then check for multi-path.
    To add another gateway, only "roles" need to be changed in job yaml.
2. Create "n" nvmeof namespaces, configured by 'namespaces_count'
3. Rename qa/suites/rbd/nvmeof/cluster/fixed-3.yaml to fixed-4.yaml
    which contains 2 gateways and 2 initiators.

Signed-off-by: Vallari Agrawal <val.agl002@gmail.com>

Merge pull request #56109 from anthonyeleven/improve-monitoring

doc/cephadm/services: Improve monitoring.rst

Reviewed-by: Zac Dover <zac.dover@proton.me>

Merge pull request #55877 from adk3798/cephadm-pull-usage

cephadm: improve cephadm pull usage message

Reviewed-by: Guillaume Abrioux <gabrioux@ibm.com>
Reviewed-by: John Mulligan <jmulligan@redhat.com>

Merge pull request #55615 from rhcs-dashboard/grafana-mount-dashboards

mgr/cephadm: mount grafana dashboards from the path

Reviewed-by: Adam King <adking@redhat.com>
Reviewed-by: Pedro Gonzalez Gomez <pegonzal@redhat.com>

Merge pull request #52760 from thotz/add-healthchecker-timeout-haproxy

cephadm: adjust the ingress ha proxy health check interval

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

Merge pull request #56189 from rhcs-dashboard/fix-64913-main

mgr/dashboard: Allow DELETE method api request, X-TOTAL-COUNT header in CORS config in dashboard

Reviewed-by: Ankush Behl <cloudbehl@gmail.com>
Reviewed-by: Nizamudeen A <nia@redhat.com>

Merge pull request #56250 from myoungwon/wip-laod-metadata-during-gc

crimson/os/seastore: cache metadata during trimming to prevent from disk read

Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>
Reviewed-by: Chunmei Liu <chunmei.liu@intel.com>

Merge pull request #56204 from zmc/do-cmake-args

do_cmake.sh: Use value of ARGS at runtime

osd/SnapMapper: add common logs

Signed-off-by: Matan Breizman <mbreizma@redhat.com>
(cherry picked from commit f302be018ebef365d8876d14185c4bdb246f22d8)

osd: improved get_objects_by_prefixes() ergonomics

Improved call signatures for get_next_objects_to_trim() &
get_objects_by_prefixes().

Also: as as get_next_objects_to_trim() has only a single
failure mode, we should not try to handle two distinct failures
in its callers' code.

Signed-off-by: Ronen Friedman <rfriedma@redhat.com>

Merge pull request #55846 from athanatos/sjust/for-review/wip-crimson-coroutine-support

crimson: add coroutine support for errorated and interruptible futures

Reviewed-by: Matan Breizman <mbreizma@redhat.com>
Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>
Reviewed-by: Chunmei Liu <chunmei.liu@intel.com>

Merge pull request #56034 from tchaikov/wip-msgr-dont-panic-at-hot-unplug

msg: do not abort if driver->del_event() returns -ENOENT

Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
Reviewed-by: Kefu Chai <tchaikov@gmail.com>

Merge pull request #55896 from ronen-fr/wip-rf-compactref

osd: avoid calling a virtual function in a ctor

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

Merge pull request #55872 from aclamk/wip-aclamk-bs-fix-unused

os/bluestore: Mute warnings

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

Merge pull request #54492 from Matan-B/wip-matanb-op-thread-timeout

qa/suites/rados/verify/validater/valgrind: increase op thread timeout

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

Merge pull request #56240 from zdover23/wip-doc-2024-03-17-rados-config-common

doc/rados: update common.rst

Reviewed-by: Anthony D'Atri <anthony.datri@gmail.com>

qa/rgw: ignore POOL_APP_NOT_ENABLED in rgw/lua subsuite

> cluster [WRN] Health check failed: 1 pool(s) do not have an application enabled (POOL_APP_NOT_ENABLED)" in cluster log

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