[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 <boost/utility/string_view.hpp>

#include <common/DecayCounter.h>

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

class OSDMap;
class PerfCounters;
class LogEvent;
class EMetaBlob;
class EUpdate;
class MMDSSlaveRequest;
struct SnapInfo;
class MClientRequest;
class MClientReply;
class MDLog;

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
  MDSInternalContext *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 = time::min();
  time reconnect_last_seen = time::min();
  set<client_t> client_reconnect_gather;  // clients i need a reconnect msg from.

  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(Message *m);

  void handle_osd_map();

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

  void handle_client_session(class MClientSession *m);
  time last_recalled() const {
    return last_recall_state;
  }
  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,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);
  void reconnect_clients(MDSInternalContext *reconnect_done_);
  void handle_client_reconnect(class MClientReconnect *m);
  //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 RecallFlags {
    NONE = 0,
    STEADY = (1<<0),
    ENFORCE_MAX = (1<<1),
  };
  std::pair<bool, uint64_t> recall_client_state(MDSGatherBuilder* gather, enum RecallFlags=RecallFlags::NONE);
  void force_clients_readonly();

  // -- requests --
  void handle_client_request(MClientRequest *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, const char *evt);
  void dispatch_client_request(MDRequestRef& mdr);
  void perf_gather_op_latency(const MClientRequest* 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, MClientReply *reply, CInode *in, CDentry *dn,
		      snapid_t snapid,
		      int num_dentries_wanted,
		      MDRequestRef& mdr);

  void encode_empty_dirstat(bufferlist& bl);
  void encode_infinite_lease(bufferlist& bl);
  void encode_null_lease(bufferlist& bl);

  void handle_slave_request(MMDSSlaveRequest *m);
  void handle_slave_request_reply(MMDSSlaveRequest *m);
  void dispatch_slave_request(MDRequestRef& mdr);
  void handle_slave_auth_pin(MDRequestRef& mdr);
  void handle_slave_auth_pin_ack(MDRequestRef& mdr, MMDSSlaveRequest *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, boost::string_view dname);
  CDir *traverse_to_auth_dir(MDRequestRef& mdr, vector<CDentry*> &trace, filepath refpath);
  CDentry *prepare_null_dentry(MDRequestRef& mdr, CDir *dir, boost::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, set<SimpleLock*>& rdlocks, bool want_auth,
			      bool no_want_auth=false,
			      file_layout_t **layout=NULL,
			      bool no_lookup=false);
  CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, int n,
                                    set<SimpleLock*>& rdlocks,
                                    set<SimpleLock*>& wrlocks,
				    set<SimpleLock*>& xlocks, bool okexist,
				    bool mustexist, bool alwaysxlock,
				    file_layout_t **layout=NULL);

  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_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_layout_vxattr(string name, string value, const OSDMap& osdmap,
			  file_layout_t *layout, bool validate=true);
  int parse_quota_vxattr(string name, string value, quota_info_t *quota);
  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,
			 set<SimpleLock*> rdlocks,
			 set<SimpleLock*> wrlocks,
			 set<SimpleLock*> xlocks);
  void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
			    file_layout_t *dir_layout,
			    set<SimpleLock*> rdlocks,
			    set<SimpleLock*> wrlocks,
			    set<SimpleLock*> xlocks);
  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);

  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);
  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, MMDSSlaveRequest *m);
  void do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
  void _link_rollback_finish(MutationRef& mut, MDRequestRef& mdr);

  // 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, MMDSSlaveRequest *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, MMDSSlaveRequest *m);
  void handle_slave_rename_notify_ack(MDRequestRef& mdr, MMDSSlaveRequest *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, bool finish_mdr);

  void evict_cap_revoke_non_responders();
  void handle_conf_change(const struct md_config_t *,
                          const std::set <std::string> &changed);

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

  DecayCounter recall_throttle;
  time last_recall_state;
};

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