1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
8 * This is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License version 2.1, as published by the Free Software
11 * Foundation. See file COPYING.
16 * This is the top level monitor. It runs on each machine in the Monitor
17 * Cluster. The election of a leader for the paxos algorithm only happens
18 * once per machine via the elector. There is a separate paxos instance (state)
19 * kept for each of the system components: Object Store Device (OSD) Monitor,
20 * Placement Group (PG) Monitor, Metadata Server (MDS) Monitor, and Client Monitor.
23 #ifndef CEPH_MONITOR_H
24 #define CEPH_MONITOR_H
30 #include "include/types.h"
31 #include "include/health.h"
32 #include "msg/Messenger.h"
34 #include "common/Timer.h"
36 #include "health_check.h"
41 #include "MonCommand.h"
44 #include "common/config_obs.h"
45 #include "common/LogClient.h"
46 #include "auth/AuthClient.h"
47 #include "auth/AuthServer.h"
48 #include "auth/cephx/CephxKeyServer.h"
49 #include "auth/AuthMethodList.h"
50 #include "auth/KeyRing.h"
51 #include "include/common_fwd.h"
52 #include "messages/MMonCommand.h"
53 #include "mon/MonitorDBStore.h"
54 #include "mgr/MgrClient.h"
56 #include "mon/MonOpRequest.h"
57 #include "common/WorkQueue.h"
59 using namespace TOPNSPC::common
;
61 #define CEPH_MON_PROTOCOL 13 /* cluster internal */
65 l_cluster_first
= 555000,
67 l_cluster_num_mon_quorum
,
73 l_cluster_osd_bytes_used
,
74 l_cluster_osd_bytes_avail
,
77 l_cluster_num_pg_active_clean
,
78 l_cluster_num_pg_active
,
79 l_cluster_num_pg_peering
,
81 l_cluster_num_object_degraded
,
82 l_cluster_num_object_misplaced
,
83 l_cluster_num_object_unfound
,
104 class AdminSocketHook
;
106 #define COMPAT_SET_LOC "feature_set"
108 class Monitor
: public Dispatcher
,
111 public md_config_obs_t
{
114 const char **orig_argv
= nullptr;
119 Messenger
*messenger
;
120 ConnectionRef con_self
;
121 ceph::mutex lock
= ceph::make_mutex("Monitor::lock");
124 ThreadPool cpu_tp
; ///< threadpool for CPU intensive work
126 ceph::mutex auth_lock
= ceph::make_mutex("Monitor::auth_lock");
128 /// true if we have ever joined a quorum. if false, we are either a
129 /// new cluster, a newly joining monitor, or a just-upgraded
131 bool has_ever_joined
;
133 PerfCounters
*logger
, *cluster_logger
;
134 bool cluster_logger_registered
;
136 void register_cluster_logger();
137 void unregister_cluster_logger();
142 set
<entity_addrvec_t
> extra_probe_peers
;
144 LogClient log_client
;
146 LogChannelRef audit_clog
;
148 KeyServer key_server
;
150 AuthMethodList auth_cluster_required
;
151 AuthMethodList auth_service_required
;
155 vector
<MonCommand
> leader_mon_commands
; // quorum leader's commands
156 vector
<MonCommand
> local_mon_commands
; // commands i support
157 bufferlist local_mon_commands_bl
; // encoded version of above
159 vector
<MonCommand
> prenautilus_local_mon_commands
;
160 bufferlist prenautilus_local_mon_commands_bl
;
162 Messenger
*mgr_messenger
;
163 MgrClient mgr_client
;
164 uint64_t mgr_proxy_bytes
= 0; // in-flight proxied mgr command message bytes
165 std::string gss_ktfile_client
{};
170 // -- local storage --
172 MonitorDBStore
*store
;
173 static const string MONITOR_NAME
;
174 static const string MONITOR_STORE_PREFIX
;
176 // -- monitor state --
187 int state
= STATE_INIT
;
190 static const char *get_state_name(int s
) {
192 case STATE_PROBING
: return "probing";
193 case STATE_SYNCHRONIZING
: return "synchronizing";
194 case STATE_ELECTING
: return "electing";
195 case STATE_LEADER
: return "leader";
196 case STATE_PEON
: return "peon";
197 case STATE_SHUTDOWN
: return "shutdown";
198 default: return "???";
201 const char *get_state_name() const {
202 return get_state_name(state
);
205 bool is_init() const { return state
== STATE_INIT
; }
206 bool is_shutdown() const { return state
== STATE_SHUTDOWN
; }
207 bool is_probing() const { return state
== STATE_PROBING
; }
208 bool is_synchronizing() const { return state
== STATE_SYNCHRONIZING
; }
209 bool is_electing() const { return state
== STATE_ELECTING
; }
210 bool is_leader() const { return state
== STATE_LEADER
; }
211 bool is_peon() const { return state
== STATE_PEON
; }
213 const utime_t
&get_leader_since() const;
215 void prepare_new_fingerprint(MonitorDBStore::TransactionRef t
);
217 std::vector
<DaemonHealthMetric
> get_health_metrics();
223 friend class Elector
;
225 /// features we require of peers (based on on-disk compatset)
226 uint64_t required_features
;
228 int leader
; // current leader (to best of knowledge)
229 set
<int> quorum
; // current active set of monitors (if !starting)
230 mono_clock::time_point quorum_since
; // when quorum formed
231 utime_t leader_since
; // when this monitor became the leader, if it is the leader
232 utime_t exited_quorum
; // time detected as not in quorum; 0 if in
234 // map of counts of connected clients, by type and features, for
236 map
<int,FeatureMap
> quorum_feature_map
;
239 * Intersection of quorum member's connection feature bits.
241 uint64_t quorum_con_features
;
243 * Intersection of quorum members mon-specific feature bits
245 mon_feature_t quorum_mon_features
;
247 ceph_release_t quorum_min_mon_release
{ceph_release_t::unknown
};
249 set
<string
> outside_quorum
;
252 * @defgroup Monitor_h_scrub
255 version_t scrub_version
; ///< paxos version we are scrubbing
256 map
<int,ScrubResult
> scrub_result
; ///< results so far
259 * trigger a cross-mon scrub
261 * Verify all mons are storing identical content
265 void handle_scrub(MonOpRequestRef op
);
266 bool _scrub(ScrubResult
*r
,
267 pair
<string
,string
> *start
,
269 void scrub_check_results();
270 void scrub_timeout();
273 void scrub_update_interval(int secs
);
275 Context
*scrub_event
; ///< periodic event to trigger scrub (leader)
276 Context
*scrub_timeout_event
; ///< scrub round timeout (leader)
277 void scrub_event_start();
278 void scrub_event_cancel();
279 void scrub_reset_timeout();
280 void scrub_cancel_timeout();
283 pair
<string
,string
> last_key
; ///< last scrubbed key
286 ScrubState() : finished(false) { }
287 virtual ~ScrubState() { }
289 std::shared_ptr
<ScrubState
> scrub_state
; ///< keeps track of current scrub
292 * @defgroup Monitor_h_sync Synchronization
296 * @} // provider state
298 struct SyncProvider
{
299 entity_addrvec_t addrs
;
300 uint64_t cookie
; ///< unique cookie for this sync attempt
301 utime_t timeout
; ///< when we give up and expire this attempt
302 version_t last_committed
; ///< last paxos version on peer
303 pair
<string
,string
> last_key
; ///< last key sent to (or on) peer
304 bool full
; ///< full scan?
305 MonitorDBStore::Synchronizer synchronizer
; ///< iterator
307 SyncProvider() : cookie(0), last_committed(0), full(false) {}
309 void reset_timeout(CephContext
*cct
, int grace
) {
310 timeout
= ceph_clock_now();
315 map
<uint64_t, SyncProvider
> sync_providers
; ///< cookie -> SyncProvider for those syncing from us
316 uint64_t sync_provider_count
; ///< counter for issued cookies to keep them unique
319 * @} // requester state
321 entity_addrvec_t sync_provider
; ///< who we are syncing from
322 uint64_t sync_cookie
; ///< 0 if we are starting, non-zero otherwise
323 bool sync_full
; ///< true if we are a full sync, false for recent catch-up
324 version_t sync_start_version
; ///< last_committed at sync start
325 Context
*sync_timeout_event
; ///< timeout event
328 * floor for sync source
330 * When we sync we forget about our old last_committed value which
331 * can be dangerous. For example, if we have a cluster of:
337 * If something forces us to sync (say, corruption, or manual
338 * intervention, or bug), we forget last_committed, and might abort.
339 * If mon.a happens to be down when we come back, we will see:
344 * and sync from mon.b, at which point a+b will both have lc 80 and
345 * come online with a majority holding out of date commits.
347 * Avoid this by preserving our old last_committed value prior to
348 * sync and never going backwards.
350 version_t sync_last_committed_floor
;
353 * Obtain the synchronization target prefixes in set form.
355 * We consider a target prefix all those that are relevant when
356 * synchronizing two stores. That is, all those that hold paxos service's
357 * versions, as well as paxos versions, or any control keys such as the
358 * first or last committed version.
360 * Given the current design, this function should return the name of all and
361 * any available paxos service, plus the paxos name.
363 * @returns a set of strings referring to the prefixes being synchronized
365 set
<string
> get_sync_targets_names();
368 * Reset the monitor's sync-related data structures for syncing *from* a peer
370 void sync_reset_requester();
373 * Reset sync state related to allowing others to sync from us
375 void sync_reset_provider();
378 * Caled when a sync attempt times out (requester-side)
383 * Get the latest monmap for backup purposes during sync
385 void sync_obtain_latest_monmap(bufferlist
&bl
);
390 * Start pulling committed state from another monitor.
392 * @param entity where to pull committed state from
393 * @param full whether to do a full sync or just catch up on recent paxos
395 void sync_start(entity_addrvec_t
&addrs
, bool full
);
399 * force a sync on next mon restart
401 void sync_force(Formatter
*f
);
405 * store critical state for safekeeping during sync
407 * We store a few things on the side that we don't want to get clobbered by sync. This
408 * includes the latest monmap and a lower bound on last_committed.
410 void sync_stash_critical_state(MonitorDBStore::TransactionRef tx
);
413 * reset the sync timeout
415 * This is used on the client to restart if things aren't progressing
417 void sync_reset_timeout();
420 * trim stale sync provider state
422 * If someone is syncing from us and hasn't talked to us recently, expire their state.
424 void sync_trim_providers();
429 * Finish up a sync after we've gotten all of the chunks.
431 * @param last_committed final last_committed value from provider
433 void sync_finish(version_t last_committed
);
436 * request the next chunk from the provider
438 void sync_get_next_chunk();
441 * handle sync message
443 * @param m Sync message with operation type MMonSync::OP_START_CHUNKS
445 void handle_sync(MonOpRequestRef op
);
447 void _sync_reply_no_cookie(MonOpRequestRef op
);
449 void handle_sync_get_cookie(MonOpRequestRef op
);
450 void handle_sync_get_chunk(MonOpRequestRef op
);
451 void handle_sync_finish(MonOpRequestRef op
);
453 void handle_sync_cookie(MonOpRequestRef op
);
454 void handle_sync_forward(MonOpRequestRef op
);
455 void handle_sync_chunk(MonOpRequestRef op
);
456 void handle_sync_no_cookie(MonOpRequestRef op
);
459 * @} // Synchronization
462 list
<Context
*> waitfor_quorum
;
463 list
<Context
*> maybe_wait_for_quorum
;
466 * @defgroup Monitor_h_TimeCheck Monitor Clock Drift Early Warning System
469 * We use time checks to keep track of any clock drifting going on in the
470 * cluster. This is accomplished by periodically ping each monitor in the
471 * quorum and register its response time on a map, assessing how much its
472 * clock has drifted. We also take this opportunity to assess the latency
475 * This mechanism works as follows:
477 * - Leader sends out a 'PING' message to each other monitor in the quorum.
478 * The message is timestamped with the leader's current time. The leader's
479 * current time is recorded in a map, associated with each peon's
481 * - The peon replies to the leader with a timestamped 'PONG' message.
482 * - The leader calculates a delta between the peon's timestamp and its
483 * current time and stashes it.
484 * - The leader also calculates the time it took to receive the 'PONG'
485 * since the 'PING' was sent, and stashes an approximate latency estimate.
486 * - Once all the quorum members have pong'ed, the leader will share the
487 * clock skew and latency maps with all the monitors in the quorum.
489 map
<int, utime_t
> timecheck_waiting
;
490 map
<int, double> timecheck_skews
;
491 map
<int, double> timecheck_latencies
;
492 // odd value means we are mid-round; even value means the round has
494 version_t timecheck_round
;
495 unsigned int timecheck_acks
;
496 utime_t timecheck_round_start
;
497 friend class HealthMonitor
;
498 /* When we hit a skew we will start a new round based off of
499 * 'mon_timecheck_skew_interval'. Each new round will be backed off
500 * until we hit 'mon_timecheck_interval' -- which is the typical
501 * interval when not in the presence of a skew.
503 * This variable tracks the number of rounds with skews since last clean
504 * so that we can report to the user and properly adjust the backoff.
506 uint64_t timecheck_rounds_since_clean
;
510 Context
*timecheck_event
;
512 void timecheck_start();
513 void timecheck_finish();
514 void timecheck_start_round();
515 void timecheck_finish_round(bool success
= true);
516 void timecheck_cancel_round();
517 void timecheck_cleanup();
518 void timecheck_reset_event();
519 void timecheck_check_skews();
520 void timecheck_report();
522 health_status_t
timecheck_status(ostringstream
&ss
,
523 const double skew_bound
,
524 const double latency
);
525 void handle_timecheck_leader(MonOpRequestRef op
);
526 void handle_timecheck_peon(MonOpRequestRef op
);
527 void handle_timecheck(MonOpRequestRef op
);
530 * Returns 'true' if this is considered to be a skew; 'false' otherwise.
532 bool timecheck_has_skew(const double skew_bound
, double *abs
) const {
533 double abs_skew
= std::fabs(skew_bound
);
536 return (abs_skew
> g_conf()->mon_clock_drift_allowed
);
543 * Handle ping messages from others.
545 void handle_ping(MonOpRequestRef op
);
547 Context
*probe_timeout_event
= nullptr; // for probing
549 void reset_probe_timeout();
550 void cancel_probe_timeout();
551 void probe_timeout(int r
);
553 void _apply_compatset_features(CompatSet
&new_features
);
557 int get_leader() const { return leader
; }
558 string
get_leader_name() {
559 return quorum
.empty() ? string() : monmap
->get_name(*quorum
.begin());
561 const set
<int>& get_quorum() const { return quorum
; }
562 list
<string
> get_quorum_names() {
564 for (set
<int>::iterator p
= quorum
.begin(); p
!= quorum
.end(); ++p
)
565 q
.push_back(monmap
->get_name(*p
));
568 uint64_t get_quorum_con_features() const {
569 return quorum_con_features
;
571 mon_feature_t
get_quorum_mon_features() const {
572 return quorum_mon_features
;
574 uint64_t get_required_features() const {
575 return required_features
;
577 mon_feature_t
get_required_mon_features() const {
578 return monmap
->get_required_features();
580 void apply_quorum_to_compatset_features();
581 void apply_monmap_to_compatset_features();
582 void calc_quorum_requirements();
584 void get_combined_feature_map(FeatureMap
*fm
);
587 void _reset(); ///< called from bootstrap, start_, or join_election
588 void wait_for_paxos_write();
589 void _finish_svc_election(); ///< called by {win,lose}_election
593 void join_election();
594 void start_election();
595 void win_standalone_election();
596 // end election (called by Elector)
597 void win_election(epoch_t epoch
, const set
<int>& q
,
599 const mon_feature_t
& mon_features
,
600 ceph_release_t min_mon_release
,
601 const map
<int,Metadata
>& metadata
);
602 void lose_election(epoch_t epoch
, set
<int>& q
, int l
,
604 const mon_feature_t
& mon_features
,
605 ceph_release_t min_mon_release
);
606 // end election (called by Elector)
607 void finish_election();
609 void update_logger();
612 * Vector holding the Services serviced by this Monitor.
614 vector
<std::unique_ptr
<PaxosService
>> paxos_service
;
616 class MDSMonitor
*mdsmon() {
617 return (class MDSMonitor
*)paxos_service
[PAXOS_MDSMAP
].get();
620 class MonmapMonitor
*monmon() {
621 return (class MonmapMonitor
*)paxos_service
[PAXOS_MONMAP
].get();
624 class OSDMonitor
*osdmon() {
625 return (class OSDMonitor
*)paxos_service
[PAXOS_OSDMAP
].get();
628 class AuthMonitor
*authmon() {
629 return (class AuthMonitor
*)paxos_service
[PAXOS_AUTH
].get();
632 class LogMonitor
*logmon() {
633 return (class LogMonitor
*) paxos_service
[PAXOS_LOG
].get();
636 class MgrMonitor
*mgrmon() {
637 return (class MgrMonitor
*) paxos_service
[PAXOS_MGR
].get();
640 class MgrStatMonitor
*mgrstatmon() {
641 return (class MgrStatMonitor
*) paxos_service
[PAXOS_MGRSTAT
].get();
644 class HealthMonitor
*healthmon() {
645 return (class HealthMonitor
*) paxos_service
[PAXOS_HEALTH
].get();
648 class ConfigMonitor
*configmon() {
649 return (class ConfigMonitor
*) paxos_service
[PAXOS_CONFIG
].get();
653 friend class OSDMonitor
;
654 friend class MDSMonitor
;
655 friend class MonmapMonitor
;
656 friend class LogMonitor
;
657 friend class ConfigKeyService
;
659 QuorumService
*config_key_service
;
662 MonSessionMap session_map
;
663 ceph::mutex session_map_lock
= ceph::make_mutex("Monitor::session_map_lock");
664 AdminSocketHook
*admin_hook
;
666 template<typename Func
, typename
...Args
>
667 void with_session_map(Func
&& func
) {
668 std::lock_guard
l(session_map_lock
);
669 std::forward
<Func
>(func
)(session_map
);
671 void send_latest_monmap(Connection
*con
);
674 void handle_get_version(MonOpRequestRef op
);
675 void handle_subscribe(MonOpRequestRef op
);
676 void handle_mon_get_map(MonOpRequestRef op
);
678 static void _generate_command_map(cmdmap_t
& cmdmap
,
679 map
<string
,string
> ¶m_str_map
);
680 static const MonCommand
*_get_moncommand(
681 const string
&cmd_prefix
,
682 const vector
<MonCommand
>& cmds
);
683 bool _allowed_command(MonSession
*s
, const string
& module
,
684 const string
& prefix
,
685 const cmdmap_t
& cmdmap
,
686 const map
<string
,string
>& param_str_map
,
687 const MonCommand
*this_cmd
);
688 void get_mon_status(Formatter
*f
);
689 void _quorum_status(Formatter
*f
, ostream
& ss
);
690 bool _add_bootstrap_peer_hint(std::string_view cmd
, const cmdmap_t
& cmdmap
,
692 void handle_tell_command(MonOpRequestRef op
);
693 void handle_command(MonOpRequestRef op
);
694 void handle_route(MonOpRequestRef op
);
696 void handle_mon_metadata(MonOpRequestRef op
);
697 int get_mon_metadata(int mon
, Formatter
*f
, ostream
& err
);
698 int print_nodes(Formatter
*f
, ostream
& err
);
700 // Accumulate metadata across calls to update_mon_metadata
701 map
<int, Metadata
> mon_metadata
;
702 map
<int, Metadata
> pending_metadata
;
707 struct health_cache_t
{
708 health_status_t overall
;
712 // health_status_t doesn't really have a NONE value and we're not
713 // okay with setting something else (say, HEALTH_ERR). so just
717 } health_status_cache
;
719 Context
*health_tick_event
= nullptr;
720 Context
*health_interval_event
= nullptr;
722 void health_tick_start();
723 void health_tick_stop();
724 ceph::real_clock::time_point
health_interval_calc_next_update();
725 void health_interval_start();
726 void health_interval_stop();
727 void health_events_cleanup();
729 void health_to_clog_update_conf(const std::set
<std::string
> &changed
);
731 void do_health_to_clog_interval();
732 void do_health_to_clog(bool force
= false);
735 const health_check_map_t
& updated
,
736 const health_check_map_t
& previous
,
737 MonitorDBStore::TransactionRef t
);
741 class HealthCheckLogStatus
{
743 health_status_t severity
;
744 std::string last_message
;
745 utime_t updated_at
= 0;
746 HealthCheckLogStatus(health_status_t severity_
,
747 const std::string
&last_message_
,
749 : severity(severity_
),
750 last_message(last_message_
),
751 updated_at(updated_at_
)
754 std::map
<std::string
, HealthCheckLogStatus
> health_check_log_times
;
758 void get_cluster_status(stringstream
&ss
, Formatter
*f
);
760 void reply_command(MonOpRequestRef op
, int rc
, const string
&rs
, version_t version
);
761 void reply_command(MonOpRequestRef op
, int rc
, const string
&rs
, bufferlist
& rdata
, version_t version
);
763 void reply_tell_command(MonOpRequestRef op
, int rc
, const string
&rs
);
767 void handle_probe(MonOpRequestRef op
);
769 * Handle a Probe Operation, replying with our name, quorum and known versions.
771 * We use the MMonProbe message class for anything and everything related with
772 * Monitor probing. One of the operations relates directly with the probing
773 * itself, in which we receive a probe request and to which we reply with
774 * our name, our quorum and the known versions for each Paxos service. Thus the
775 * redundant function name. This reply will obviously be sent to the one
776 * probing/requesting these infos.
778 * @todo Add @pre and @post
780 * @param m A Probe message, with an operation of type Probe.
782 void handle_probe_probe(MonOpRequestRef op
);
783 void handle_probe_reply(MonOpRequestRef op
);
786 struct RoutedRequest
{
788 bufferlist request_bl
;
791 uint64_t con_features
;
794 RoutedRequest() : tid(0), session(NULL
), con_features(0) {}
800 uint64_t routed_request_tid
;
801 map
<uint64_t, RoutedRequest
*> routed_requests
;
803 void forward_request_leader(MonOpRequestRef op
);
804 void handle_forward(MonOpRequestRef op
);
805 void send_reply(MonOpRequestRef op
, Message
*reply
);
806 void no_reply(MonOpRequestRef op
);
807 void resend_routed_requests();
808 void remove_session(MonSession
*s
);
809 void remove_all_sessions();
810 void waitlist_or_zap_client(MonOpRequestRef op
);
812 void send_mon_message(Message
*m
, int rank
);
815 struct C_Command
: public C_MonOp
{
821 C_Command(Monitor
*_mm
, MonOpRequestRef _op
, int r
, string s
, version_t v
) :
822 C_MonOp(_op
), mon(_mm
), rc(r
), rs(s
), version(v
){}
823 C_Command(Monitor
*_mm
, MonOpRequestRef _op
, int r
, string s
, bufferlist rd
, version_t v
) :
824 C_MonOp(_op
), mon(_mm
), rc(r
), rs(s
), rdata(rd
), version(v
){}
826 void _finish(int r
) override
{
827 auto m
= op
->get_req
<MMonCommand
>();
830 if (!op
->get_req()->get_connection()) {
831 ss
<< "connection dropped for command ";
833 MonSession
*s
= op
->get_session();
835 // if client drops we may not have a session to draw information from.
837 ss
<< "from='" << s
->name
<< " " << s
->addrs
<< "' "
838 << "entity='" << s
->entity_name
<< "' ";
840 ss
<< "session dropped for command ";
846 cmdmap_from_json(m
->cmd
, &cmdmap
, ds
);
847 cmd_getval(cmdmap
, "prefix", prefix
);
848 if (prefix
!= "config set" && prefix
!= "config-key set")
849 ss
<< "cmd='" << m
->cmd
<< "': finished";
851 mon
->audit_clog
->info() << ss
.str();
852 mon
->reply_command(op
, rc
, rs
, rdata
, version
);
854 else if (r
== -ECANCELED
)
856 else if (r
== -EAGAIN
)
857 mon
->dispatch_op(op
);
859 ceph_abort_msg("bad C_Command return value");
864 class C_RetryMessage
: public C_MonOp
{
867 C_RetryMessage(Monitor
*m
, MonOpRequestRef op
) :
868 C_MonOp(op
), mon(m
) { }
870 void _finish(int r
) override
{
871 if (r
== -EAGAIN
|| r
>= 0)
872 mon
->dispatch_op(op
);
873 else if (r
== -ECANCELED
)
876 ceph_abort_msg("bad C_RetryMessage return value");
880 //ms_dispatch handles a lot of logic and we want to reuse it
881 //on forwarded messages, so we create a non-locking version for this class
882 void _ms_dispatch(Message
*m
);
883 bool ms_dispatch(Message
*m
) override
{
884 std::lock_guard l
{lock
};
888 void dispatch_op(MonOpRequestRef op
);
889 //mon_caps is used for un-connected messages from monitors
891 bool get_authorizer(int dest_type
, AuthAuthorizer
**authorizer
);
892 public: // for AuthMonitor msgr1:
893 int ms_handle_authentication(Connection
*con
) override
;
895 void ms_handle_accept(Connection
*con
) override
;
896 bool ms_handle_reset(Connection
*con
) override
;
897 void ms_handle_remote_reset(Connection
*con
) override
{}
898 bool ms_handle_refused(Connection
*con
) override
;
901 int get_auth_request(
903 AuthConnectionMeta
*auth_meta
,
905 vector
<uint32_t> *preferred_modes
,
906 bufferlist
*out
) override
;
907 int handle_auth_reply_more(
909 AuthConnectionMeta
*auth_meta
,
910 const bufferlist
& bl
,
911 bufferlist
*reply
) override
;
912 int handle_auth_done(
914 AuthConnectionMeta
*auth_meta
,
917 const bufferlist
& bl
,
918 CryptoKey
*session_key
,
919 std::string
*connection_secret
) override
;
920 int handle_auth_bad_method(
922 AuthConnectionMeta
*auth_meta
,
923 uint32_t old_auth_method
,
925 const std::vector
<uint32_t>& allowed_methods
,
926 const std::vector
<uint32_t>& allowed_modes
) override
;
929 int handle_auth_request(
931 AuthConnectionMeta
*auth_meta
,
933 uint32_t auth_method
,
934 const bufferlist
& bl
,
935 bufferlist
*reply
) override
;
938 int write_default_keyring(bufferlist
& bl
);
939 void extract_save_mon_key(KeyRing
& keyring
);
941 void collect_metadata(Metadata
*m
);
942 void update_mon_metadata(int from
, Metadata
&& m
);
944 void count_metadata(const string
& field
, Formatter
*f
);
945 void count_metadata(const string
& field
, map
<string
,int> *out
);
948 static CompatSet
get_initial_supported_features();
949 static CompatSet
get_supported_features();
950 static CompatSet
get_legacy_features();
951 /// read the ondisk features into the CompatSet pointed to by read_features
952 static void read_features_off_disk(MonitorDBStore
*store
, CompatSet
*read_features
);
953 void read_features();
954 void write_features(MonitorDBStore::TransactionRef t
);
956 OpTracker op_tracker
;
959 Monitor(CephContext
*cct_
, string nm
, MonitorDBStore
*s
,
960 Messenger
*m
, Messenger
*mgr_m
, MonMap
*map
);
963 static int check_features(MonitorDBStore
*store
);
966 const char** get_tracked_conf_keys() const override
;
967 void handle_conf_change(const ConfigProxy
& conf
,
968 const std::set
<std::string
> &changed
) override
;
970 void update_log_clients();
971 int sanitize_options();
975 void refresh_from_paxos(bool *need_bootstrap
);
979 void handle_signal(int sig
);
981 int mkfs(bufferlist
& osdmapbl
);
984 * check cluster_fsid file
986 * @return EEXIST if file exists and doesn't match, 0 on match, or negative error code
991 * write cluster_fsid file
993 * @return 0 on success, or negative error code
996 int write_fsid(MonitorDBStore::TransactionRef t
);
998 int do_admin_command(std::string_view command
, const cmdmap_t
& cmdmap
,
1004 // don't allow copying
1005 Monitor(const Monitor
& rhs
);
1006 Monitor
& operator=(const Monitor
&rhs
);
1009 static void format_command_descriptions(const std::vector
<MonCommand
> &commands
,
1014 const std::vector
<MonCommand
> &get_local_commands(mon_feature_t f
) {
1015 if (f
.contains_all(ceph::features::mon::FEATURE_NAUTILUS
)) {
1016 return local_mon_commands
;
1018 return prenautilus_local_mon_commands
;
1021 const bufferlist
& get_local_commands_bl(mon_feature_t f
) {
1022 if (f
.contains_all(ceph::features::mon::FEATURE_NAUTILUS
)) {
1023 return local_mon_commands_bl
;
1025 return prenautilus_local_mon_commands_bl
;
1028 void set_leader_commands(const std::vector
<MonCommand
>& cmds
) {
1029 leader_mon_commands
= cmds
;
1032 bool is_keyring_required();
1035 #define CEPH_MON_FEATURE_INCOMPAT_BASE CompatSet::Feature (1, "initial feature set (~v.18)")
1036 #define CEPH_MON_FEATURE_INCOMPAT_GV CompatSet::Feature (2, "global version sequencing (v0.52)")
1037 #define CEPH_MON_FEATURE_INCOMPAT_SINGLE_PAXOS CompatSet::Feature (3, "single paxos with k/v store (v0.\?)")
1038 #define CEPH_MON_FEATURE_INCOMPAT_OSD_ERASURE_CODES CompatSet::Feature(4, "support erasure code pools")
1039 #define CEPH_MON_FEATURE_INCOMPAT_OSDMAP_ENC CompatSet::Feature(5, "new-style osdmap encoding")
1040 #define CEPH_MON_FEATURE_INCOMPAT_ERASURE_CODE_PLUGINS_V2 CompatSet::Feature(6, "support isa/lrc erasure code")
1041 #define CEPH_MON_FEATURE_INCOMPAT_ERASURE_CODE_PLUGINS_V3 CompatSet::Feature(7, "support shec erasure code")
1042 #define CEPH_MON_FEATURE_INCOMPAT_KRAKEN CompatSet::Feature(8, "support monmap features")
1043 #define CEPH_MON_FEATURE_INCOMPAT_LUMINOUS CompatSet::Feature(9, "luminous ondisk layout")
1044 #define CEPH_MON_FEATURE_INCOMPAT_MIMIC CompatSet::Feature(10, "mimic ondisk layout")
1045 #define CEPH_MON_FEATURE_INCOMPAT_NAUTILUS CompatSet::Feature(11, "nautilus ondisk layout")
1046 #define CEPH_MON_FEATURE_INCOMPAT_OCTOPUS CompatSet::Feature(12, "octopus ondisk layout")
1047 // make sure you add your feature to Monitor::get_supported_features
1052 * new C_MonContext{...}
1056 * new C_MonContext(...)
1058 * because of gcc bug [1].
1060 * [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85883
1062 template<typename T
>
1063 class C_MonContext
: public LambdaContext
<T
> {
1065 C_MonContext(const Monitor
* m
, T
&& f
) :
1066 LambdaContext
<T
>(std::forward
<T
>(f
)),
1069 void finish(int r
) override
{
1070 if (mon
->is_shutdown())
1072 LambdaContext
<T
>::finish(r
);