[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 "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

  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 --
  utime_t  reconnect_start;
  set<client_t> client_reconnect_gather;  // clients i need a reconnect msg from.
  bool waiting_for_reconnect(client_t c) const;
  void dump_reconnect_status(Formatter *f) const;

  void handle_client_session(class MClientSession *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,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);

  void recall_client_state(void);
  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);
};

#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
7c673cae FG	20	#include "MDSRank.h"
	21	#include "Mutation.h"
	22
	23	class OSDMap;
	24	class PerfCounters;
	25	class LogEvent;
	26	class EMetaBlob;
	27	class EUpdate;
	28	class MMDSSlaveRequest;
	29	struct SnapInfo;
	30	class MClientRequest;
	31	class MClientReply;
	32	class MDLog;
	33
	34	enum {
	35	l_mdss_first = 1000,
d2e6a577 FG	36	l_mdss_dispatch_client_request,
d2e6a577 FG	37	l_mdss_dispatch_slave_request,
7c673cae	38	l_mdss_handle_client_request,
7c673cae	39	l_mdss_handle_client_session,
d2e6a577	40	l_mdss_handle_slave_request,
91327a77 AA	41	l_mdss_req_create_latency,
	42	l_mdss_req_getattr_latency,
	43	l_mdss_req_getfilelock_latency,
	44	l_mdss_req_link_latency,
	45	l_mdss_req_lookup_latency,
	46	l_mdss_req_lookuphash_latency,
	47	l_mdss_req_lookupino_latency,
	48	l_mdss_req_lookupname_latency,
	49	l_mdss_req_lookupparent_latency,
	50	l_mdss_req_lookupsnap_latency,
	51	l_mdss_req_lssnap_latency,
	52	l_mdss_req_mkdir_latency,
	53	l_mdss_req_mknod_latency,
	54	l_mdss_req_mksnap_latency,
	55	l_mdss_req_open_latency,
	56	l_mdss_req_readdir_latency,
	57	l_mdss_req_rename_latency,
	58	l_mdss_req_renamesnap_latency,
	59	l_mdss_req_rmdir_latency,
	60	l_mdss_req_rmsnap_latency,
	61	l_mdss_req_rmxattr_latency,
	62	l_mdss_req_setattr_latency,
	63	l_mdss_req_setdirlayout_latency,
	64	l_mdss_req_setfilelock_latency,
	65	l_mdss_req_setlayout_latency,
	66	l_mdss_req_setxattr_latency,
	67	l_mdss_req_symlink_latency,
	68	l_mdss_req_unlink_latency,
	69	l_mdss_cap_revoke_eviction,
7c673cae FG	70	l_mdss_last,
	71	};
	72
	73	class Server {
91327a77 AA	74	public:
	75	using clock = ceph::coarse_mono_clock;
	76	using time = ceph::coarse_mono_time;
	77
7c673cae FG	78	private:
	79	MDSRank *mds;
	80	MDCache *mdcache;
	81	MDLog *mdlog;
	82	PerfCounters *logger;
	83
	84	// OSDMap full status, used to generate ENOSPC on some operations
	85	bool is_full;
	86
	87	// State for while in reconnect
	88	MDSInternalContext *reconnect_done;
	89	int failed_reconnects;
31f18b77 FG	90	bool reconnect_evicting; // true if I am waiting for evictions to complete
31f18b77 FG	91	// before proceeding to reconnect_gather_finish
7c673cae	92
91327a77 AA	93	double cap_revoke_eviction_timeout = 0;
91327a77 AA	94
7c673cae FG	95	friend class MDSContinuation;
	96	friend class ServerContext;
	97	friend class ServerLogContext;
	98
	99	public:
	100	bool terminating_sessions;
	101
	102	explicit Server(MDSRank *m);
	103	~Server() {
	104	g_ceph_context->get_perfcounters_collection()->remove(logger);
	105	delete logger;
	106	delete reconnect_done;
	107	}
	108
	109	void create_logger();
	110
	111	// message handler
	112	void dispatch(Message *m);
	113
	114	void handle_osd_map();
	115
	116	// -- sessions and recovery --
	117	utime_t reconnect_start;
	118	set<client_t> client_reconnect_gather; // clients i need a reconnect msg from.
	119	bool waiting_for_reconnect(client_t c) const;
	120	void dump_reconnect_status(Formatter *f) const;
	121
7c673cae FG	122	void handle_client_session(class MClientSession *m);
	123	void _session_logged(Session *session, uint64_t state_seq,
	124	bool open, version_t pv, interval_set<inodeno_t>& inos,version_t piv);
	125	version_t prepare_force_open_sessions(map<client_t,entity_inst_t> &cm,
28e407b8 AA	126	map<client_t,pair<Session*,uint64_t> >& smap);
28e407b8 AA	127	void finish_force_open_sessions(const map<client_t,pair<Session*,uint64_t> >& smap,
7c673cae FG	128	bool dec_import=true);
	129	void flush_client_sessions(set<client_t>& client_set, MDSGatherBuilder& gather);
	130	void finish_flush_session(Session *session, version_t seq);
	131	void terminate_sessions();
	132	void find_idle_sessions();
	133	void kill_session(Session session, Context on_safe);
31f18b77	134	size_t apply_blacklist(const std::set<entity_addr_t> &blacklist);
7c673cae FG	135	void journal_close_session(Session session, int state, Context on_safe);
	136	void reconnect_clients(MDSInternalContext *reconnect_done_);
	137	void handle_client_reconnect(class MClientReconnect *m);
	138	//void process_reconnect_cap(CInode *in, int from, ceph_mds_cap_reconnect& capinfo);
	139	void reconnect_gather_finish();
	140	void reconnect_tick();
	141	void recover_filelocks(CInode *in, bufferlist locks, int64_t client);
	142
181888fb	143	void recall_client_state(void);
7c673cae FG	144	void force_clients_readonly();
	145
	146	// -- requests --
	147	void handle_client_request(MClientRequest *m);
	148
	149	void journal_and_reply(MDRequestRef& mdr, CInode tracei, CDentry tracedn,
	150	LogEvent le, MDSLogContextBase fin);
	151	void submit_mdlog_entry(LogEvent le, MDSLogContextBase fin,
	152	MDRequestRef& mdr, const char *evt);
	153	void dispatch_client_request(MDRequestRef& mdr);
91327a77	154	void perf_gather_op_latency(const MClientRequest* req, utime_t lat);
7c673cae FG	155	void early_reply(MDRequestRef& mdr, CInode tracei, CDentry tracedn);
	156	void respond_to_request(MDRequestRef& mdr, int r = 0);
	157	void set_trace_dist(Session session, MClientReply reply, CInode in, CDentry dn,
	158	snapid_t snapid,
	159	int num_dentries_wanted,
	160	MDRequestRef& mdr);
	161
	162	void encode_empty_dirstat(bufferlist& bl);
	163	void encode_infinite_lease(bufferlist& bl);
	164	void encode_null_lease(bufferlist& bl);
	165
	166	void handle_slave_request(MMDSSlaveRequest *m);
	167	void handle_slave_request_reply(MMDSSlaveRequest *m);
	168	void dispatch_slave_request(MDRequestRef& mdr);
	169	void handle_slave_auth_pin(MDRequestRef& mdr);
	170	void handle_slave_auth_pin_ack(MDRequestRef& mdr, MMDSSlaveRequest *ack);
	171
	172	// some helpers
	173	bool check_fragment_space(MDRequestRef& mdr, CDir *in);
	174	bool check_access(MDRequestRef& mdr, CInode *in, unsigned mask);
	175	bool _check_access(Session session, CInode in, unsigned mask, int caller_uid, int caller_gid, int setattr_uid, int setattr_gid);
94b18763	176	CDir validate_dentry_dir(MDRequestRef& mdr, CInode diri, boost::string_view dname);
7c673cae	177	CDir traverse_to_auth_dir(MDRequestRef& mdr, vector<CDentry> &trace, filepath refpath);
94b18763	178	CDentry prepare_null_dentry(MDRequestRef& mdr, CDir dir, boost::string_view dname, bool okexist=false);
7c673cae FG	179	CDentry prepare_stray_dentry(MDRequestRef& mdr, CInode in);
	180	CInode* prepare_new_inode(MDRequestRef& mdr, CDir *dir, inodeno_t useino, unsigned mode,
	181	file_layout_t *layout=NULL);
	182	void journal_allocated_inos(MDRequestRef& mdr, EMetaBlob *blob);
	183	void apply_allocated_inos(MDRequestRef& mdr, Session *session);
	184
	185	CInode* rdlock_path_pin_ref(MDRequestRef& mdr, int n, set<SimpleLock*>& rdlocks, bool want_auth,
	186	bool no_want_auth=false,
	187	file_layout_t **layout=NULL,
	188	bool no_lookup=false);
	189	CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, int n,
	190	set<SimpleLock*>& rdlocks,
	191	set<SimpleLock*>& wrlocks,
	192	set<SimpleLock*>& xlocks, bool okexist,
	193	bool mustexist, bool alwaysxlock,
	194	file_layout_t **layout=NULL);
	195
	196	CDir* try_open_auth_dirfrag(CInode *diri, frag_t fg, MDRequestRef& mdr);
	197
	198
	199	// requests on existing inodes.
	200	void handle_client_getattr(MDRequestRef& mdr, bool is_lookup);
	201	void handle_client_lookup_ino(MDRequestRef& mdr,
	202	bool want_parent, bool want_dentry);
	203	void _lookup_ino_2(MDRequestRef& mdr, int r);
	204	void handle_client_readdir(MDRequestRef& mdr);
	205	void handle_client_file_setlock(MDRequestRef& mdr);
	206	void handle_client_file_readlock(MDRequestRef& mdr);
	207
	208	void handle_client_setattr(MDRequestRef& mdr);
	209	void handle_client_setlayout(MDRequestRef& mdr);
	210	void handle_client_setdirlayout(MDRequestRef& mdr);
	211
	212	int parse_layout_vxattr(string name, string value, const OSDMap& osdmap,
	213	file_layout_t *layout, bool validate=true);
	214	int parse_quota_vxattr(string name, string value, quota_info_t *quota);
	215	int check_layout_vxattr(MDRequestRef& mdr,
	216	string name,
	217	string value,
	218	file_layout_t *layout);
	219	void handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
	220	file_layout_t *dir_layout,
	221	set<SimpleLock*> rdlocks,
	222	set<SimpleLock*> wrlocks,
	223	set<SimpleLock*> xlocks);
	224	void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
	225	file_layout_t *dir_layout,
	226	set<SimpleLock*> rdlocks,
	227	set<SimpleLock*> wrlocks,
	228	set<SimpleLock*> xlocks);
	229	void handle_client_setxattr(MDRequestRef& mdr);
	230	void handle_client_removexattr(MDRequestRef& mdr);
	231
	232	void handle_client_fsync(MDRequestRef& mdr);
	233
	234	// open
	235	void handle_client_open(MDRequestRef& mdr);
	236	void handle_client_openc(MDRequestRef& mdr); // O_CREAT variant.
	237	void do_open_truncate(MDRequestRef& mdr, int cmode); // O_TRUNC variant.
	238
	239	// namespace changes
	240	void handle_client_mknod(MDRequestRef& mdr);
	241	void handle_client_mkdir(MDRequestRef& mdr);
	242	void handle_client_symlink(MDRequestRef& mdr);
243
244	// link
245	void handle_client_link(MDRequestRef& mdr);
246	void _link_local(MDRequestRef& mdr, CDentry dn, CInode targeti);
247	void _link_local_finish(MDRequestRef& mdr,
248	CDentry dn, CInode targeti,
249	version_t, version_t);
250
251	void _link_remote(MDRequestRef& mdr, bool inc, CDentry dn, CInode targeti);
252	void _link_remote_finish(MDRequestRef& mdr, bool inc, CDentry dn, CInode targeti,
253	version_t);
254
255	void handle_slave_link_prep(MDRequestRef& mdr);
256	void _logged_slave_link(MDRequestRef& mdr, CInode *targeti);
257	void _commit_slave_link(MDRequestRef& mdr, int r, CInode *targeti);
258	void _committed_slave(MDRequestRef& mdr); // use for rename, too
259	void handle_slave_link_prep_ack(MDRequestRef& mdr, MMDSSlaveRequest *m);
260	void do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
261	void _link_rollback_finish(MutationRef& mut, MDRequestRef& mdr);
262
263	// unlink
264	void handle_client_unlink(MDRequestRef& mdr);
265	bool _dir_is_nonempty_unlocked(MDRequestRef& mdr, CInode *rmdiri);
266	bool _dir_is_nonempty(MDRequestRef& mdr, CInode *rmdiri);
267	void _unlink_local(MDRequestRef& mdr, CDentry dn, CDentry straydn);
268	void _unlink_local_finish(MDRequestRef& mdr,
269	CDentry dn, CDentry straydn,
270	version_t);
271	bool _rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, vector<CDentry>& trace, CDentry straydn);
272	void handle_slave_rmdir_prep(MDRequestRef& mdr);
273	void _logged_slave_rmdir(MDRequestRef& mdr, CDentry srcdn, CDentry straydn);
31f18b77	274	void _commit_slave_rmdir(MDRequestRef& mdr, int r, CDentry *straydn);
7c673cae FG	275	void handle_slave_rmdir_prep_ack(MDRequestRef& mdr, MMDSSlaveRequest *ack);
	276	void do_rmdir_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
	277	void _rmdir_rollback_finish(MDRequestRef& mdr, metareqid_t reqid, CDentry dn, CDentry straydn);
	278
	279	// rename
	280	void handle_client_rename(MDRequestRef& mdr);
	281	void _rename_finish(MDRequestRef& mdr,
	282	CDentry srcdn, CDentry destdn, CDentry *straydn);
	283
	284	void handle_client_lssnap(MDRequestRef& mdr);
	285	void handle_client_mksnap(MDRequestRef& mdr);
	286	void _mksnap_finish(MDRequestRef& mdr, CInode *diri, SnapInfo &info);
	287	void handle_client_rmsnap(MDRequestRef& mdr);
	288	void _rmsnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
	289	void handle_client_renamesnap(MDRequestRef& mdr);
	290	void _renamesnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
	291
	292
	293	// helpers
	294	bool _rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, set<mds_rank_t> &witnesse,
	295	vector<CDentry>& srctrace, vector<CDentry>& dsttrace, CDentry *straydn);
	296	version_t _rename_prepare_import(MDRequestRef& mdr, CDentry srcdn, bufferlist client_map_bl);
	297	bool _need_force_journal(CInode *diri, bool empty);
	298	void _rename_prepare(MDRequestRef& mdr,
	299	EMetaBlob metablob, bufferlist client_map_bl,
	300	CDentry srcdn, CDentry destdn, CDentry *straydn);
	301	/* set not_journaling=true if you're going to discard the results --
	302	* this bypasses the asserts to make sure we're journaling the right
	303	* things on the right nodes */
	304	void _rename_apply(MDRequestRef& mdr, CDentry srcdn, CDentry destdn, CDentry *straydn);
	305
	306	// slaving
	307	void handle_slave_rename_prep(MDRequestRef& mdr);
	308	void handle_slave_rename_prep_ack(MDRequestRef& mdr, MMDSSlaveRequest *m);
	309	void handle_slave_rename_notify_ack(MDRequestRef& mdr, MMDSSlaveRequest *m);
	310	void _slave_rename_sessions_flushed(MDRequestRef& mdr);
	311	void _logged_slave_rename(MDRequestRef& mdr, CDentry srcdn, CDentry destdn, CDentry *straydn);
	312	void _commit_slave_rename(MDRequestRef& mdr, int r, CDentry srcdn, CDentry destdn, CDentry *straydn);
	313	void do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr, bool finish_mdr=false);
	314	void _rename_rollback_finish(MutationRef& mut, MDRequestRef& mdr, CDentry *srcdn, version_t srcdnpv,
	315	CDentry destdn, CDentry staydn, bool finish_mdr);
	316
91327a77 AA	317	void evict_cap_revoke_non_responders();
	318	void handle_conf_change(const struct md_config_t *,
	319	const std::set <std::string> &changed);
	320
7c673cae FG	321	private:
	322	void reply_client_request(MDRequestRef& mdr, MClientReply *reply);
	323	};
	324
	325	#endif