[ceph.git] / ceph / src / mds / Server.h

// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
// vim: ts=8 sw=2 smarttab
/*
 * Ceph - scalable distributed file system
 *
 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
 *
 * This is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License version 2.1, as published by the Free Software 
 * Foundation.  See file COPYING.
 * 
 */

#ifndef CEPH_MDS_SERVER_H
#define CEPH_MDS_SERVER_H

#include <string_view>

#include <common/DecayCounter.h>

#include "messages/MClientReconnect.h"
#include "messages/MClientReply.h"
#include "messages/MClientRequest.h"
#include "messages/MClientSession.h"
#include "messages/MClientSnap.h"
#include "messages/MClientReclaim.h"
#include "messages/MClientReclaimReply.h"
#include "messages/MLock.h"

#include "MDSRank.h"
#include "Mutation.h"
#include "MDSContext.h"

class OSDMap;
class PerfCounters;
class LogEvent;
class EMetaBlob;
class EUpdate;
class MDLog;
struct SnapInfo;

enum {
  l_mdss_first = 1000,
  l_mdss_dispatch_client_request,
  l_mdss_dispatch_slave_request,
  l_mdss_handle_client_request,
  l_mdss_handle_client_session,
  l_mdss_handle_slave_request,
  l_mdss_req_create_latency,
  l_mdss_req_getattr_latency,
  l_mdss_req_getfilelock_latency,
  l_mdss_req_link_latency,
  l_mdss_req_lookup_latency,
  l_mdss_req_lookuphash_latency,
  l_mdss_req_lookupino_latency,
  l_mdss_req_lookupname_latency,
  l_mdss_req_lookupparent_latency,
  l_mdss_req_lookupsnap_latency,
  l_mdss_req_lssnap_latency,
  l_mdss_req_mkdir_latency,
  l_mdss_req_mknod_latency,
  l_mdss_req_mksnap_latency,
  l_mdss_req_open_latency,
  l_mdss_req_readdir_latency,
  l_mdss_req_rename_latency,
  l_mdss_req_renamesnap_latency,
  l_mdss_req_rmdir_latency,
  l_mdss_req_rmsnap_latency,
  l_mdss_req_rmxattr_latency,
  l_mdss_req_setattr_latency,
  l_mdss_req_setdirlayout_latency,
  l_mdss_req_setfilelock_latency,
  l_mdss_req_setlayout_latency,
  l_mdss_req_setxattr_latency,
  l_mdss_req_symlink_latency,
  l_mdss_req_unlink_latency,
  l_mdss_cap_revoke_eviction,
  l_mdss_last,
};

class Server {
public:
  using clock = ceph::coarse_mono_clock;
  using time = ceph::coarse_mono_time;

private:
  MDSRank *mds;
  MDCache *mdcache;
  MDLog *mdlog;
  PerfCounters *logger;

  // OSDMap full status, used to generate ENOSPC on some operations
  bool is_full;

  // State for while in reconnect
  MDSContext *reconnect_done;
  int failed_reconnects;
  bool reconnect_evicting;  // true if I am waiting for evictions to complete
                            // before proceeding to reconnect_gather_finish
  time reconnect_start = clock::zero();
  time reconnect_last_seen = clock::zero();
  set<client_t> client_reconnect_gather;  // clients i need a reconnect msg from.

  feature_bitset_t supported_features;
  feature_bitset_t required_client_features;

  bool replay_unsafe_with_closed_session = false;
  double cap_revoke_eviction_timeout = 0;

  friend class MDSContinuation;
  friend class ServerContext;
  friend class ServerLogContext;

public:
  bool terminating_sessions;

  explicit Server(MDSRank *m);
  ~Server() {
    g_ceph_context->get_perfcounters_collection()->remove(logger);
    delete logger;
    delete reconnect_done;
  }

  void create_logger();

  // message handler
  void dispatch(const Message::const_ref &m);

  void handle_osd_map();

  // -- sessions and recovery --
  bool waiting_for_reconnect(client_t c) const;
  void dump_reconnect_status(Formatter *f) const;

  time last_recalled() const {
    return last_recall_state;
  }

  void handle_client_session(const MClientSession::const_ref &m);
  void _session_logged(Session *session, uint64_t state_seq, 
		       bool open, version_t pv, interval_set<inodeno_t>& inos,version_t piv);
  version_t prepare_force_open_sessions(map<client_t,entity_inst_t> &cm,
					map<client_t,client_metadata_t>& cmm,
					map<client_t,pair<Session*,uint64_t> >& smap);
  void finish_force_open_sessions(const map<client_t,pair<Session*,uint64_t> >& smap,
				  bool dec_import=true);
  void flush_client_sessions(set<client_t>& client_set, MDSGatherBuilder& gather);
  void finish_flush_session(Session *session, version_t seq);
  void terminate_sessions();
  void find_idle_sessions();
  void kill_session(Session *session, Context *on_safe);
  size_t apply_blacklist(const std::set<entity_addr_t> &blacklist);
  void journal_close_session(Session *session, int state, Context *on_safe);

  set<client_t> client_reclaim_gather;
  size_t get_num_pending_reclaim() const { return client_reclaim_gather.size(); }
  Session *find_session_by_uuid(std::string_view uuid);
  void reclaim_session(Session *session, const MClientReclaim::const_ref &m);
  void finish_reclaim_session(Session *session, const MClientReclaimReply::ref &reply=nullptr);
  void handle_client_reclaim(const MClientReclaim::const_ref &m);

  void reconnect_clients(MDSContext *reconnect_done_);
  void handle_client_reconnect(const MClientReconnect::const_ref &m);
  void infer_supported_features(Session *session, client_metadata_t& client_metadata);
  void update_required_client_features();

  //void process_reconnect_cap(CInode *in, int from, ceph_mds_cap_reconnect& capinfo);
  void reconnect_gather_finish();
  void reconnect_tick();
  void recover_filelocks(CInode *in, bufferlist locks, int64_t client);

  enum class RecallFlags : uint64_t {
    NONE = 0,
    STEADY = (1<<0),
    ENFORCE_MAX = (1<<1),
    TRIM = (1<<2),
    ENFORCE_LIVENESS = (1<<3),
  };
  std::pair<bool, uint64_t> recall_client_state(MDSGatherBuilder* gather, RecallFlags=RecallFlags::NONE);
  void force_clients_readonly();

  // -- requests --
  void handle_client_request(const MClientRequest::const_ref &m);

  void journal_and_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn,
			 LogEvent *le, MDSLogContextBase *fin);
  void submit_mdlog_entry(LogEvent *le, MDSLogContextBase *fin,
                          MDRequestRef& mdr, std::string_view event);
  void dispatch_client_request(MDRequestRef& mdr);
  void perf_gather_op_latency(const MClientRequest::const_ref &req, utime_t lat);
  void early_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn);
  void respond_to_request(MDRequestRef& mdr, int r = 0);
  void set_trace_dist(Session *session, const MClientReply::ref &reply, CInode *in, CDentry *dn,
		      snapid_t snapid,
		      int num_dentries_wanted,
		      MDRequestRef& mdr);


  void handle_slave_request(const MMDSSlaveRequest::const_ref &m);
  void handle_slave_request_reply(const MMDSSlaveRequest::const_ref &m);
  void dispatch_slave_request(MDRequestRef& mdr);
  void handle_slave_auth_pin(MDRequestRef& mdr);
  void handle_slave_auth_pin_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);

  // some helpers
  bool check_fragment_space(MDRequestRef& mdr, CDir *in);
  bool check_access(MDRequestRef& mdr, CInode *in, unsigned mask);
  bool _check_access(Session *session, CInode *in, unsigned mask, int caller_uid, int caller_gid, int setattr_uid, int setattr_gid);
  CDir *validate_dentry_dir(MDRequestRef& mdr, CInode *diri, std::string_view dname);
  CDir *traverse_to_auth_dir(MDRequestRef& mdr, vector<CDentry*> &trace, filepath refpath);
  CDentry *prepare_null_dentry(MDRequestRef& mdr, CDir *dir, std::string_view dname, bool okexist=false);
  CDentry *prepare_stray_dentry(MDRequestRef& mdr, CInode *in);
  CInode* prepare_new_inode(MDRequestRef& mdr, CDir *dir, inodeno_t useino, unsigned mode,
			    file_layout_t *layout=NULL);
  void journal_allocated_inos(MDRequestRef& mdr, EMetaBlob *blob);
  void apply_allocated_inos(MDRequestRef& mdr, Session *session);

  CInode* rdlock_path_pin_ref(MDRequestRef& mdr, int n, MutationImpl::LockOpVec& lov,
			      bool want_auth, bool no_want_auth=false,
			      file_layout_t **layout=nullptr,
			      bool no_lookup=false);
  CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, int n,
				    MutationImpl::LockOpVec& lov,
				    bool okexist, bool mustexist, bool alwaysxlock,
				    file_layout_t **layout=nullptr);

  CDir* try_open_auth_dirfrag(CInode *diri, frag_t fg, MDRequestRef& mdr);


  // requests on existing inodes.
  void handle_client_getattr(MDRequestRef& mdr, bool is_lookup);
  void handle_client_lookup_ino(MDRequestRef& mdr,
				bool want_parent, bool want_dentry);
  void _lookup_snap_ino(MDRequestRef& mdr);
  void _lookup_ino_2(MDRequestRef& mdr, int r);
  void handle_client_readdir(MDRequestRef& mdr);
  void handle_client_file_setlock(MDRequestRef& mdr);
  void handle_client_file_readlock(MDRequestRef& mdr);

  void handle_client_setattr(MDRequestRef& mdr);
  void handle_client_setlayout(MDRequestRef& mdr);
  void handle_client_setdirlayout(MDRequestRef& mdr);

  int parse_quota_vxattr(string name, string value, quota_info_t *quota);
  void create_quota_realm(CInode *in);
  int parse_layout_vxattr(string name, string value, const OSDMap& osdmap,
			  file_layout_t *layout, bool validate=true);
  int check_layout_vxattr(MDRequestRef& mdr,
                          string name,
                          string value,
                          file_layout_t *layout);
  void handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
			 file_layout_t *dir_layout,
			 MutationImpl::LockOpVec& lov);
  void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
			    file_layout_t *dir_layout,
			    MutationImpl::LockOpVec& lov);
  void handle_client_setxattr(MDRequestRef& mdr);
  void handle_client_removexattr(MDRequestRef& mdr);

  void handle_client_fsync(MDRequestRef& mdr);

  // open
  void handle_client_open(MDRequestRef& mdr);
  void handle_client_openc(MDRequestRef& mdr);  // O_CREAT variant.
  void do_open_truncate(MDRequestRef& mdr, int cmode);  // O_TRUNC variant.

  // namespace changes
  void handle_client_mknod(MDRequestRef& mdr);
  void handle_client_mkdir(MDRequestRef& mdr);
  void handle_client_symlink(MDRequestRef& mdr);

  // link
  void handle_client_link(MDRequestRef& mdr);
  void _link_local(MDRequestRef& mdr, CDentry *dn, CInode *targeti);
  void _link_local_finish(MDRequestRef& mdr, CDentry *dn, CInode *targeti,
			  version_t, version_t, bool);

  void _link_remote(MDRequestRef& mdr, bool inc, CDentry *dn, CInode *targeti);
  void _link_remote_finish(MDRequestRef& mdr, bool inc, CDentry *dn, CInode *targeti,
			   version_t);

  void handle_slave_link_prep(MDRequestRef& mdr);
  void _logged_slave_link(MDRequestRef& mdr, CInode *targeti, bool adjust_realm);
  void _commit_slave_link(MDRequestRef& mdr, int r, CInode *targeti);
  void _committed_slave(MDRequestRef& mdr);  // use for rename, too
  void handle_slave_link_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
  void do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
  void _link_rollback_finish(MutationRef& mut, MDRequestRef& mdr,
			     map<client_t,MClientSnap::ref>& split);

  // unlink
  void handle_client_unlink(MDRequestRef& mdr);
  bool _dir_is_nonempty_unlocked(MDRequestRef& mdr, CInode *rmdiri);
  bool _dir_is_nonempty(MDRequestRef& mdr, CInode *rmdiri);
  void _unlink_local(MDRequestRef& mdr, CDentry *dn, CDentry *straydn);
  void _unlink_local_finish(MDRequestRef& mdr,
			    CDentry *dn, CDentry *straydn,
			    version_t);
  bool _rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, vector<CDentry*>& trace, CDentry *straydn);
  void handle_slave_rmdir_prep(MDRequestRef& mdr);
  void _logged_slave_rmdir(MDRequestRef& mdr, CDentry *srcdn, CDentry *straydn);
  void _commit_slave_rmdir(MDRequestRef& mdr, int r, CDentry *straydn);
  void handle_slave_rmdir_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);
  void do_rmdir_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
  void _rmdir_rollback_finish(MDRequestRef& mdr, metareqid_t reqid, CDentry *dn, CDentry *straydn);

  // rename
  void handle_client_rename(MDRequestRef& mdr);
  void _rename_finish(MDRequestRef& mdr,
		      CDentry *srcdn, CDentry *destdn, CDentry *straydn);

  void handle_client_lssnap(MDRequestRef& mdr);
  void handle_client_mksnap(MDRequestRef& mdr);
  void _mksnap_finish(MDRequestRef& mdr, CInode *diri, SnapInfo &info);
  void handle_client_rmsnap(MDRequestRef& mdr);
  void _rmsnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
  void handle_client_renamesnap(MDRequestRef& mdr);
  void _renamesnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);


  // helpers
  bool _rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, set<mds_rank_t> &witnesse,
			       vector<CDentry*>& srctrace, vector<CDentry*>& dsttrace, CDentry *straydn);
  version_t _rename_prepare_import(MDRequestRef& mdr, CDentry *srcdn, bufferlist *client_map_bl);
  bool _need_force_journal(CInode *diri, bool empty);
  void _rename_prepare(MDRequestRef& mdr,
		       EMetaBlob *metablob, bufferlist *client_map_bl,
		       CDentry *srcdn, CDentry *destdn, CDentry *straydn);
  /* set not_journaling=true if you're going to discard the results --
   * this bypasses the asserts to make sure we're journaling the right
   * things on the right nodes */
  void _rename_apply(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn);

  // slaving
  void handle_slave_rename_prep(MDRequestRef& mdr);
  void handle_slave_rename_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
  void handle_slave_rename_notify_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
  void _slave_rename_sessions_flushed(MDRequestRef& mdr);
  void _logged_slave_rename(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
  void _commit_slave_rename(MDRequestRef& mdr, int r, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
  void do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr, bool finish_mdr=false);
  void _rename_rollback_finish(MutationRef& mut, MDRequestRef& mdr, CDentry *srcdn, version_t srcdnpv,
			       CDentry *destdn, CDentry *staydn, map<client_t,MClientSnap::ref> splits[2],
			       bool finish_mdr);

  void evict_cap_revoke_non_responders();
  void handle_conf_change(const std::set<std::string>& changed);

private:
  void reply_client_request(MDRequestRef& mdr, const MClientReply::ref &reply);
  void flush_session(Session *session, MDSGatherBuilder *gather);

  DecayCounter recall_throttle;
  time last_recall_state;
};

static inline constexpr auto operator|(Server::RecallFlags a, Server::RecallFlags b) {
  using T = std::underlying_type<Server::RecallFlags>::type;
  return static_cast<Server::RecallFlags>(static_cast<T>(a) | static_cast<T>(b));
}
static inline constexpr auto operator&(Server::RecallFlags a, Server::RecallFlags b) {
  using T = std::underlying_type<Server::RecallFlags>::type;
  return static_cast<Server::RecallFlags>(static_cast<T>(a) & static_cast<T>(b));
}
static inline std::ostream& operator<<(std::ostream& os, const Server::RecallFlags& f) {
  using T = std::underlying_type<Server::RecallFlags>::type;
  return os << "0x" << std::hex << static_cast<T>(f) << std::dec;
}
static inline constexpr bool operator!(const Server::RecallFlags& f) {
  using T = std::underlying_type<Server::RecallFlags>::type;
  return static_cast<T>(f) == static_cast<T>(0);
}

#endif
Commit	Line	Data
7c673cae FG	1	// -- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t --
	2	// vim: ts=8 sw=2 smarttab
	3	/*
	4	* Ceph - scalable distributed file system
	5	*
	6	* Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
	7	*
	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.
	12	*
	13	*/
	14
	15	#ifndef CEPH_MDS_SERVER_H
	16	#define CEPH_MDS_SERVER_H
	17
11fdf7f2	18	#include <string_view>
94b18763	19
a8e16298 TL	20	#include <common/DecayCounter.h>
a8e16298 TL	21
11fdf7f2 TL	22	#include "messages/MClientReconnect.h"
	23	#include "messages/MClientReply.h"
	24	#include "messages/MClientRequest.h"
	25	#include "messages/MClientSession.h"
	26	#include "messages/MClientSnap.h"
	27	#include "messages/MClientReclaim.h"
	28	#include "messages/MClientReclaimReply.h"
	29	#include "messages/MLock.h"
	30
7c673cae FG	31	#include "MDSRank.h"
7c673cae FG	32	#include "Mutation.h"
11fdf7f2	33	#include "MDSContext.h"
7c673cae FG	34
	35	class OSDMap;
	36	class PerfCounters;
	37	class LogEvent;
	38	class EMetaBlob;
	39	class EUpdate;
7c673cae	40	class MDLog;
11fdf7f2	41	struct SnapInfo;
7c673cae FG	42
	43	enum {
	44	l_mdss_first = 1000,
d2e6a577 FG	45	l_mdss_dispatch_client_request,
d2e6a577 FG	46	l_mdss_dispatch_slave_request,
7c673cae	47	l_mdss_handle_client_request,
7c673cae	48	l_mdss_handle_client_session,
d2e6a577	49	l_mdss_handle_slave_request,
91327a77 AA	50	l_mdss_req_create_latency,
	51	l_mdss_req_getattr_latency,
	52	l_mdss_req_getfilelock_latency,
	53	l_mdss_req_link_latency,
	54	l_mdss_req_lookup_latency,
	55	l_mdss_req_lookuphash_latency,
	56	l_mdss_req_lookupino_latency,
	57	l_mdss_req_lookupname_latency,
	58	l_mdss_req_lookupparent_latency,
	59	l_mdss_req_lookupsnap_latency,
	60	l_mdss_req_lssnap_latency,
	61	l_mdss_req_mkdir_latency,
	62	l_mdss_req_mknod_latency,
	63	l_mdss_req_mksnap_latency,
	64	l_mdss_req_open_latency,
	65	l_mdss_req_readdir_latency,
	66	l_mdss_req_rename_latency,
	67	l_mdss_req_renamesnap_latency,
	68	l_mdss_req_rmdir_latency,
	69	l_mdss_req_rmsnap_latency,
	70	l_mdss_req_rmxattr_latency,
	71	l_mdss_req_setattr_latency,
	72	l_mdss_req_setdirlayout_latency,
	73	l_mdss_req_setfilelock_latency,
	74	l_mdss_req_setlayout_latency,
	75	l_mdss_req_setxattr_latency,
	76	l_mdss_req_symlink_latency,
	77	l_mdss_req_unlink_latency,
	78	l_mdss_cap_revoke_eviction,
7c673cae FG	79	l_mdss_last,
	80	};
	81
	82	class Server {
91327a77 AA	83	public:
	84	using clock = ceph::coarse_mono_clock;
	85	using time = ceph::coarse_mono_time;
	86
7c673cae FG	87	private:
	88	MDSRank *mds;
	89	MDCache *mdcache;
	90	MDLog *mdlog;
	91	PerfCounters *logger;
	92
	93	// OSDMap full status, used to generate ENOSPC on some operations
	94	bool is_full;
	95
	96	// State for while in reconnect
11fdf7f2	97	MDSContext *reconnect_done;
7c673cae	98	int failed_reconnects;
31f18b77 FG	99	bool reconnect_evicting; // true if I am waiting for evictions to complete
31f18b77 FG	100	// before proceeding to reconnect_gather_finish
11fdf7f2 TL	101	time reconnect_start = clock::zero();
11fdf7f2 TL	102	time reconnect_last_seen = clock::zero();
f64942e4	103	set<client_t> client_reconnect_gather; // clients i need a reconnect msg from.
7c673cae	104
11fdf7f2 TL	105	feature_bitset_t supported_features;
	106	feature_bitset_t required_client_features;
	107
92f5a8d4	108	bool replay_unsafe_with_closed_session = false;
91327a77 AA	109	double cap_revoke_eviction_timeout = 0;
91327a77 AA	110
7c673cae FG	111	friend class MDSContinuation;
	112	friend class ServerContext;
	113	friend class ServerLogContext;
	114
	115	public:
	116	bool terminating_sessions;
	117
	118	explicit Server(MDSRank *m);
	119	~Server() {
	120	g_ceph_context->get_perfcounters_collection()->remove(logger);
	121	delete logger;
	122	delete reconnect_done;
	123	}
	124
	125	void create_logger();
	126
	127	// message handler
11fdf7f2	128	void dispatch(const Message::const_ref &m);
7c673cae FG	129
	130	void handle_osd_map();
	131
	132	// -- sessions and recovery --
7c673cae FG	133	bool waiting_for_reconnect(client_t c) const;
	134	void dump_reconnect_status(Formatter *f) const;
	135
a8e16298 TL	136	time last_recalled() const {
	137	return last_recall_state;
	138	}
11fdf7f2 TL	139
11fdf7f2 TL	140	void handle_client_session(const MClientSession::const_ref &m);
7c673cae FG	141	void _session_logged(Session *session, uint64_t state_seq,
	142	bool open, version_t pv, interval_set<inodeno_t>& inos,version_t piv);
	143	version_t prepare_force_open_sessions(map<client_t,entity_inst_t> &cm,
11fdf7f2	144	map<client_t,client_metadata_t>& cmm,
28e407b8 AA	145	map<client_t,pair<Session*,uint64_t> >& smap);
28e407b8 AA	146	void finish_force_open_sessions(const map<client_t,pair<Session*,uint64_t> >& smap,
7c673cae FG	147	bool dec_import=true);
	148	void flush_client_sessions(set<client_t>& client_set, MDSGatherBuilder& gather);
	149	void finish_flush_session(Session *session, version_t seq);
	150	void terminate_sessions();
	151	void find_idle_sessions();
	152	void kill_session(Session session, Context on_safe);
31f18b77	153	size_t apply_blacklist(const std::set<entity_addr_t> &blacklist);
7c673cae	154	void journal_close_session(Session session, int state, Context on_safe);
11fdf7f2 TL	155
	156	set<client_t> client_reclaim_gather;
	157	size_t get_num_pending_reclaim() const { return client_reclaim_gather.size(); }
	158	Session *find_session_by_uuid(std::string_view uuid);
	159	void reclaim_session(Session *session, const MClientReclaim::const_ref &m);
	160	void finish_reclaim_session(Session *session, const MClientReclaimReply::ref &reply=nullptr);
	161	void handle_client_reclaim(const MClientReclaim::const_ref &m);
	162
	163	void reconnect_clients(MDSContext *reconnect_done_);
	164	void handle_client_reconnect(const MClientReconnect::const_ref &m);
	165	void infer_supported_features(Session *session, client_metadata_t& client_metadata);
	166	void update_required_client_features();
	167
7c673cae FG	168	//void process_reconnect_cap(CInode *in, int from, ceph_mds_cap_reconnect& capinfo);
	169	void reconnect_gather_finish();
	170	void reconnect_tick();
	171	void recover_filelocks(CInode *in, bufferlist locks, int64_t client);
	172
92f5a8d4	173	enum class RecallFlags : uint64_t {
a8e16298 TL	174	NONE = 0,
	175	STEADY = (1<<0),
	176	ENFORCE_MAX = (1<<1),
92f5a8d4 TL	177	TRIM = (1<<2),
92f5a8d4 TL	178	ENFORCE_LIVENESS = (1<<3),
a8e16298	179	};
92f5a8d4	180	std::pair<bool, uint64_t> recall_client_state(MDSGatherBuilder* gather, RecallFlags=RecallFlags::NONE);
7c673cae FG	181	void force_clients_readonly();
	182
	183	// -- requests --
11fdf7f2	184	void handle_client_request(const MClientRequest::const_ref &m);
7c673cae FG	185
	186	void journal_and_reply(MDRequestRef& mdr, CInode tracei, CDentry tracedn,
	187	LogEvent le, MDSLogContextBase fin);
	188	void submit_mdlog_entry(LogEvent le, MDSLogContextBase fin,
11fdf7f2	189	MDRequestRef& mdr, std::string_view event);
7c673cae	190	void dispatch_client_request(MDRequestRef& mdr);
11fdf7f2	191	void perf_gather_op_latency(const MClientRequest::const_ref &req, utime_t lat);
7c673cae FG	192	void early_reply(MDRequestRef& mdr, CInode tracei, CDentry tracedn);
7c673cae FG	193	void respond_to_request(MDRequestRef& mdr, int r = 0);
11fdf7f2	194	void set_trace_dist(Session session, const MClientReply::ref &reply, CInode in, CDentry *dn,
7c673cae FG	195	snapid_t snapid,
	196	int num_dentries_wanted,
	197	MDRequestRef& mdr);
	198
7c673cae	199
11fdf7f2 TL	200	void handle_slave_request(const MMDSSlaveRequest::const_ref &m);
11fdf7f2 TL	201	void handle_slave_request_reply(const MMDSSlaveRequest::const_ref &m);
7c673cae FG	202	void dispatch_slave_request(MDRequestRef& mdr);
7c673cae FG	203	void handle_slave_auth_pin(MDRequestRef& mdr);
11fdf7f2	204	void handle_slave_auth_pin_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);
7c673cae FG	205
	206	// some helpers
	207	bool check_fragment_space(MDRequestRef& mdr, CDir *in);
	208	bool check_access(MDRequestRef& mdr, CInode *in, unsigned mask);
	209	bool _check_access(Session session, CInode in, unsigned mask, int caller_uid, int caller_gid, int setattr_uid, int setattr_gid);
11fdf7f2	210	CDir validate_dentry_dir(MDRequestRef& mdr, CInode diri, std::string_view dname);
7c673cae	211	CDir traverse_to_auth_dir(MDRequestRef& mdr, vector<CDentry> &trace, filepath refpath);
11fdf7f2	212	CDentry prepare_null_dentry(MDRequestRef& mdr, CDir dir, std::string_view dname, bool okexist=false);
7c673cae FG	213	CDentry prepare_stray_dentry(MDRequestRef& mdr, CInode in);
	214	CInode* prepare_new_inode(MDRequestRef& mdr, CDir *dir, inodeno_t useino, unsigned mode,
	215	file_layout_t *layout=NULL);
	216	void journal_allocated_inos(MDRequestRef& mdr, EMetaBlob *blob);
	217	void apply_allocated_inos(MDRequestRef& mdr, Session *session);
	218
11fdf7f2 TL	219	CInode* rdlock_path_pin_ref(MDRequestRef& mdr, int n, MutationImpl::LockOpVec& lov,
	220	bool want_auth, bool no_want_auth=false,
	221	file_layout_t **layout=nullptr,
7c673cae FG	222	bool no_lookup=false);
7c673cae FG	223	CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, int n,
11fdf7f2 TL	224	MutationImpl::LockOpVec& lov,
	225	bool okexist, bool mustexist, bool alwaysxlock,
	226	file_layout_t **layout=nullptr);
7c673cae FG	227
	228	CDir* try_open_auth_dirfrag(CInode *diri, frag_t fg, MDRequestRef& mdr);
	229
	230
	231	// requests on existing inodes.
	232	void handle_client_getattr(MDRequestRef& mdr, bool is_lookup);
	233	void handle_client_lookup_ino(MDRequestRef& mdr,
	234	bool want_parent, bool want_dentry);
11fdf7f2	235	void _lookup_snap_ino(MDRequestRef& mdr);
7c673cae FG	236	void _lookup_ino_2(MDRequestRef& mdr, int r);
	237	void handle_client_readdir(MDRequestRef& mdr);
	238	void handle_client_file_setlock(MDRequestRef& mdr);
	239	void handle_client_file_readlock(MDRequestRef& mdr);
	240
	241	void handle_client_setattr(MDRequestRef& mdr);
	242	void handle_client_setlayout(MDRequestRef& mdr);
	243	void handle_client_setdirlayout(MDRequestRef& mdr);
	244
11fdf7f2 TL	245	int parse_quota_vxattr(string name, string value, quota_info_t *quota);
11fdf7f2 TL	246	void create_quota_realm(CInode *in);
7c673cae FG	247	int parse_layout_vxattr(string name, string value, const OSDMap& osdmap,
7c673cae FG	248	file_layout_t *layout, bool validate=true);
7c673cae FG	249	int check_layout_vxattr(MDRequestRef& mdr,
	250	string name,
	251	string value,
	252	file_layout_t *layout);
	253	void handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
	254	file_layout_t *dir_layout,
11fdf7f2	255	MutationImpl::LockOpVec& lov);
7c673cae FG	256	void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
7c673cae FG	257	file_layout_t *dir_layout,
11fdf7f2	258	MutationImpl::LockOpVec& lov);
7c673cae FG	259	void handle_client_setxattr(MDRequestRef& mdr);
	260	void handle_client_removexattr(MDRequestRef& mdr);
	261
	262	void handle_client_fsync(MDRequestRef& mdr);
	263
	264	// open
	265	void handle_client_open(MDRequestRef& mdr);
	266	void handle_client_openc(MDRequestRef& mdr); // O_CREAT variant.
	267	void do_open_truncate(MDRequestRef& mdr, int cmode); // O_TRUNC variant.
	268
	269	// namespace changes
	270	void handle_client_mknod(MDRequestRef& mdr);
	271	void handle_client_mkdir(MDRequestRef& mdr);
	272	void handle_client_symlink(MDRequestRef& mdr);
	273
	274	// link
	275	void handle_client_link(MDRequestRef& mdr);
	276	void _link_local(MDRequestRef& mdr, CDentry dn, CInode targeti);
11fdf7f2 TL	277	void _link_local_finish(MDRequestRef& mdr, CDentry dn, CInode targeti,
11fdf7f2 TL	278	version_t, version_t, bool);
7c673cae FG	279
	280	void _link_remote(MDRequestRef& mdr, bool inc, CDentry dn, CInode targeti);
	281	void _link_remote_finish(MDRequestRef& mdr, bool inc, CDentry dn, CInode targeti,
	282	version_t);
	283
	284	void handle_slave_link_prep(MDRequestRef& mdr);
11fdf7f2	285	void _logged_slave_link(MDRequestRef& mdr, CInode *targeti, bool adjust_realm);
7c673cae FG	286	void _commit_slave_link(MDRequestRef& mdr, int r, CInode *targeti);
7c673cae FG	287	void _committed_slave(MDRequestRef& mdr); // use for rename, too
11fdf7f2	288	void handle_slave_link_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
7c673cae	289	void do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
11fdf7f2 TL	290	void _link_rollback_finish(MutationRef& mut, MDRequestRef& mdr,
11fdf7f2 TL	291	map<client_t,MClientSnap::ref>& split);
7c673cae FG	292
	293	// unlink
	294	void handle_client_unlink(MDRequestRef& mdr);
	295	bool _dir_is_nonempty_unlocked(MDRequestRef& mdr, CInode *rmdiri);
	296	bool _dir_is_nonempty(MDRequestRef& mdr, CInode *rmdiri);
	297	void _unlink_local(MDRequestRef& mdr, CDentry dn, CDentry straydn);
	298	void _unlink_local_finish(MDRequestRef& mdr,
	299	CDentry dn, CDentry straydn,
	300	version_t);
	301	bool _rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, vector<CDentry>& trace, CDentry straydn);
	302	void handle_slave_rmdir_prep(MDRequestRef& mdr);
	303	void _logged_slave_rmdir(MDRequestRef& mdr, CDentry srcdn, CDentry straydn);
31f18b77	304	void _commit_slave_rmdir(MDRequestRef& mdr, int r, CDentry *straydn);
11fdf7f2	305	void handle_slave_rmdir_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);
7c673cae FG	306	void do_rmdir_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
	307	void _rmdir_rollback_finish(MDRequestRef& mdr, metareqid_t reqid, CDentry dn, CDentry straydn);
	308
	309	// rename
	310	void handle_client_rename(MDRequestRef& mdr);
	311	void _rename_finish(MDRequestRef& mdr,
	312	CDentry srcdn, CDentry destdn, CDentry *straydn);
	313
	314	void handle_client_lssnap(MDRequestRef& mdr);
	315	void handle_client_mksnap(MDRequestRef& mdr);
	316	void _mksnap_finish(MDRequestRef& mdr, CInode *diri, SnapInfo &info);
	317	void handle_client_rmsnap(MDRequestRef& mdr);
	318	void _rmsnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
	319	void handle_client_renamesnap(MDRequestRef& mdr);
	320	void _renamesnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
	321
	322
	323	// helpers
	324	bool _rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, set<mds_rank_t> &witnesse,
	325	vector<CDentry>& srctrace, vector<CDentry>& dsttrace, CDentry *straydn);
	326	version_t _rename_prepare_import(MDRequestRef& mdr, CDentry srcdn, bufferlist client_map_bl);
	327	bool _need_force_journal(CInode *diri, bool empty);
	328	void _rename_prepare(MDRequestRef& mdr,
	329	EMetaBlob metablob, bufferlist client_map_bl,
	330	CDentry srcdn, CDentry destdn, CDentry *straydn);
	331	/* set not_journaling=true if you're going to discard the results --
	332	* this bypasses the asserts to make sure we're journaling the right
	333	* things on the right nodes */
	334	void _rename_apply(MDRequestRef& mdr, CDentry srcdn, CDentry destdn, CDentry *straydn);
	335
	336	// slaving
	337	void handle_slave_rename_prep(MDRequestRef& mdr);
11fdf7f2 TL	338	void handle_slave_rename_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
11fdf7f2 TL	339	void handle_slave_rename_notify_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
7c673cae FG	340	void _slave_rename_sessions_flushed(MDRequestRef& mdr);
	341	void _logged_slave_rename(MDRequestRef& mdr, CDentry srcdn, CDentry destdn, CDentry *straydn);
	342	void _commit_slave_rename(MDRequestRef& mdr, int r, CDentry srcdn, CDentry destdn, CDentry *straydn);
	343	void do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr, bool finish_mdr=false);
	344	void _rename_rollback_finish(MutationRef& mut, MDRequestRef& mdr, CDentry *srcdn, version_t srcdnpv,
11fdf7f2 TL	345	CDentry destdn, CDentry staydn, map<client_t,MClientSnap::ref> splits[2],
11fdf7f2 TL	346	bool finish_mdr);
7c673cae	347
91327a77	348	void evict_cap_revoke_non_responders();
92f5a8d4	349	void handle_conf_change(const std::set<std::string>& changed);
91327a77	350
7c673cae	351	private:
11fdf7f2	352	void reply_client_request(MDRequestRef& mdr, const MClientReply::ref &reply);
f64942e4	353	void flush_session(Session session, MDSGatherBuilder gather);
a8e16298 TL	354
	355	DecayCounter recall_throttle;
	356	time last_recall_state;
7c673cae FG	357	};
7c673cae FG	358
92f5a8d4 TL	359	static inline constexpr auto operator\|(Server::RecallFlags a, Server::RecallFlags b) {
	360	using T = std::underlying_type<Server::RecallFlags>::type;
	361	return static_cast<Server::RecallFlags>(static_cast<T>(a) \| static_cast<T>(b));
	362	}
	363	static inline constexpr auto operator&(Server::RecallFlags a, Server::RecallFlags b) {
	364	using T = std::underlying_type<Server::RecallFlags>::type;
	365	return static_cast<Server::RecallFlags>(static_cast<T>(a) & static_cast<T>(b));
	366	}
	367	static inline std::ostream& operator<<(std::ostream& os, const Server::RecallFlags& f) {
	368	using T = std::underlying_type<Server::RecallFlags>::type;
	369	return os << "0x" << std::hex << static_cast<T>(f) << std::dec;
	370	}
	371	static inline constexpr bool operator!(const Server::RecallFlags& f) {
	372	using T = std::underlying_type<Server::RecallFlags>::type;
	373	return static_cast<T>(f) == static_cast<T>(0);
	374	}
	375
7c673cae	376	#endif