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.
15 #ifndef CEPH_MDS_SERVER_H
16 #define CEPH_MDS_SERVER_H
26 class MMDSSlaveRequest
;
34 l_mdss_handle_client_request
,
35 l_mdss_handle_slave_request
,
36 l_mdss_handle_client_session
,
37 l_mdss_dispatch_client_request
,
38 l_mdss_req_lookuphash
,
40 l_mdss_req_lookupparent
,
41 l_mdss_req_lookupname
,
43 l_mdss_req_lookupsnap
,
47 l_mdss_req_setdirlayout
,
51 l_mdss_req_setfilelock
,
52 l_mdss_req_getfilelock
,
65 l_mdss_req_renamesnap
,
66 l_mdss_dispatch_slave_request
,
77 // OSDMap full status, used to generate ENOSPC on some operations
80 // State for while in reconnect
81 MDSInternalContext
*reconnect_done
;
82 int failed_reconnects
;
84 friend class MDSContinuation
;
85 friend class ServerContext
;
86 friend class ServerLogContext
;
89 bool terminating_sessions
;
91 explicit Server(MDSRank
*m
);
93 g_ceph_context
->get_perfcounters_collection()->remove(logger
);
95 delete reconnect_done
;
101 void dispatch(Message
*m
);
103 void handle_osd_map();
105 // -- sessions and recovery --
106 utime_t reconnect_start
;
107 set
<client_t
> client_reconnect_gather
; // clients i need a reconnect msg from.
108 bool waiting_for_reconnect(client_t c
) const;
109 void dump_reconnect_status(Formatter
*f
) const;
111 Session
*get_session(Message
*m
);
112 void handle_client_session(class MClientSession
*m
);
113 void _session_logged(Session
*session
, uint64_t state_seq
,
114 bool open
, version_t pv
, interval_set
<inodeno_t
>& inos
,version_t piv
);
115 version_t
prepare_force_open_sessions(map
<client_t
,entity_inst_t
> &cm
,
116 map
<client_t
,uint64_t>& sseqmap
);
117 void finish_force_open_sessions(map
<client_t
,entity_inst_t
> &cm
,
118 map
<client_t
,uint64_t>& sseqmap
,
119 bool dec_import
=true);
120 void flush_client_sessions(set
<client_t
>& client_set
, MDSGatherBuilder
& gather
);
121 void finish_flush_session(Session
*session
, version_t seq
);
122 void terminate_sessions();
123 void find_idle_sessions();
124 void kill_session(Session
*session
, Context
*on_safe
);
125 void journal_close_session(Session
*session
, int state
, Context
*on_safe
);
126 void reconnect_clients(MDSInternalContext
*reconnect_done_
);
127 void handle_client_reconnect(class MClientReconnect
*m
);
128 //void process_reconnect_cap(CInode *in, int from, ceph_mds_cap_reconnect& capinfo);
129 void reconnect_gather_finish();
130 void reconnect_tick();
131 void recover_filelocks(CInode
*in
, bufferlist locks
, int64_t client
);
133 void recall_client_state(float ratio
);
134 void force_clients_readonly();
137 void handle_client_request(MClientRequest
*m
);
139 void journal_and_reply(MDRequestRef
& mdr
, CInode
*tracei
, CDentry
*tracedn
,
140 LogEvent
*le
, MDSLogContextBase
*fin
);
141 void submit_mdlog_entry(LogEvent
*le
, MDSLogContextBase
*fin
,
142 MDRequestRef
& mdr
, const char *evt
);
143 void dispatch_client_request(MDRequestRef
& mdr
);
144 void early_reply(MDRequestRef
& mdr
, CInode
*tracei
, CDentry
*tracedn
);
145 void respond_to_request(MDRequestRef
& mdr
, int r
= 0);
146 void set_trace_dist(Session
*session
, MClientReply
*reply
, CInode
*in
, CDentry
*dn
,
148 int num_dentries_wanted
,
151 void encode_empty_dirstat(bufferlist
& bl
);
152 void encode_infinite_lease(bufferlist
& bl
);
153 void encode_null_lease(bufferlist
& bl
);
155 void handle_slave_request(MMDSSlaveRequest
*m
);
156 void handle_slave_request_reply(MMDSSlaveRequest
*m
);
157 void dispatch_slave_request(MDRequestRef
& mdr
);
158 void handle_slave_auth_pin(MDRequestRef
& mdr
);
159 void handle_slave_auth_pin_ack(MDRequestRef
& mdr
, MMDSSlaveRequest
*ack
);
162 bool check_fragment_space(MDRequestRef
& mdr
, CDir
*in
);
163 bool check_access(MDRequestRef
& mdr
, CInode
*in
, unsigned mask
);
164 bool _check_access(Session
*session
, CInode
*in
, unsigned mask
, int caller_uid
, int caller_gid
, int setattr_uid
, int setattr_gid
);
165 CDir
*validate_dentry_dir(MDRequestRef
& mdr
, CInode
*diri
, const string
& dname
);
166 CDir
*traverse_to_auth_dir(MDRequestRef
& mdr
, vector
<CDentry
*> &trace
, filepath refpath
);
167 CDentry
*prepare_null_dentry(MDRequestRef
& mdr
, CDir
*dir
, const string
& dname
, bool okexist
=false);
168 CDentry
*prepare_stray_dentry(MDRequestRef
& mdr
, CInode
*in
);
169 CInode
* prepare_new_inode(MDRequestRef
& mdr
, CDir
*dir
, inodeno_t useino
, unsigned mode
,
170 file_layout_t
*layout
=NULL
);
171 void journal_allocated_inos(MDRequestRef
& mdr
, EMetaBlob
*blob
);
172 void apply_allocated_inos(MDRequestRef
& mdr
, Session
*session
);
174 CInode
* rdlock_path_pin_ref(MDRequestRef
& mdr
, int n
, set
<SimpleLock
*>& rdlocks
, bool want_auth
,
175 bool no_want_auth
=false,
176 file_layout_t
**layout
=NULL
,
177 bool no_lookup
=false);
178 CDentry
* rdlock_path_xlock_dentry(MDRequestRef
& mdr
, int n
,
179 set
<SimpleLock
*>& rdlocks
,
180 set
<SimpleLock
*>& wrlocks
,
181 set
<SimpleLock
*>& xlocks
, bool okexist
,
182 bool mustexist
, bool alwaysxlock
,
183 file_layout_t
**layout
=NULL
);
185 CDir
* try_open_auth_dirfrag(CInode
*diri
, frag_t fg
, MDRequestRef
& mdr
);
188 // requests on existing inodes.
189 void handle_client_getattr(MDRequestRef
& mdr
, bool is_lookup
);
190 void handle_client_lookup_ino(MDRequestRef
& mdr
,
191 bool want_parent
, bool want_dentry
);
192 void _lookup_ino_2(MDRequestRef
& mdr
, int r
);
193 void handle_client_readdir(MDRequestRef
& mdr
);
194 void handle_client_file_setlock(MDRequestRef
& mdr
);
195 void handle_client_file_readlock(MDRequestRef
& mdr
);
197 void handle_client_setattr(MDRequestRef
& mdr
);
198 void handle_client_setlayout(MDRequestRef
& mdr
);
199 void handle_client_setdirlayout(MDRequestRef
& mdr
);
201 int parse_layout_vxattr(string name
, string value
, const OSDMap
& osdmap
,
202 file_layout_t
*layout
, bool validate
=true);
203 int parse_quota_vxattr(string name
, string value
, quota_info_t
*quota
);
204 int check_layout_vxattr(MDRequestRef
& mdr
,
207 file_layout_t
*layout
);
208 void handle_set_vxattr(MDRequestRef
& mdr
, CInode
*cur
,
209 file_layout_t
*dir_layout
,
210 set
<SimpleLock
*> rdlocks
,
211 set
<SimpleLock
*> wrlocks
,
212 set
<SimpleLock
*> xlocks
);
213 void handle_remove_vxattr(MDRequestRef
& mdr
, CInode
*cur
,
214 file_layout_t
*dir_layout
,
215 set
<SimpleLock
*> rdlocks
,
216 set
<SimpleLock
*> wrlocks
,
217 set
<SimpleLock
*> xlocks
);
218 void handle_client_setxattr(MDRequestRef
& mdr
);
219 void handle_client_removexattr(MDRequestRef
& mdr
);
221 void handle_client_fsync(MDRequestRef
& mdr
);
224 void handle_client_open(MDRequestRef
& mdr
);
225 void handle_client_openc(MDRequestRef
& mdr
); // O_CREAT variant.
226 void do_open_truncate(MDRequestRef
& mdr
, int cmode
); // O_TRUNC variant.
229 void handle_client_mknod(MDRequestRef
& mdr
);
230 void handle_client_mkdir(MDRequestRef
& mdr
);
231 void handle_client_symlink(MDRequestRef
& mdr
);
234 void handle_client_link(MDRequestRef
& mdr
);
235 void _link_local(MDRequestRef
& mdr
, CDentry
*dn
, CInode
*targeti
);
236 void _link_local_finish(MDRequestRef
& mdr
,
237 CDentry
*dn
, CInode
*targeti
,
238 version_t
, version_t
);
240 void _link_remote(MDRequestRef
& mdr
, bool inc
, CDentry
*dn
, CInode
*targeti
);
241 void _link_remote_finish(MDRequestRef
& mdr
, bool inc
, CDentry
*dn
, CInode
*targeti
,
244 void handle_slave_link_prep(MDRequestRef
& mdr
);
245 void _logged_slave_link(MDRequestRef
& mdr
, CInode
*targeti
);
246 void _commit_slave_link(MDRequestRef
& mdr
, int r
, CInode
*targeti
);
247 void _committed_slave(MDRequestRef
& mdr
); // use for rename, too
248 void handle_slave_link_prep_ack(MDRequestRef
& mdr
, MMDSSlaveRequest
*m
);
249 void do_link_rollback(bufferlist
&rbl
, mds_rank_t master
, MDRequestRef
& mdr
);
250 void _link_rollback_finish(MutationRef
& mut
, MDRequestRef
& mdr
);
253 void handle_client_unlink(MDRequestRef
& mdr
);
254 bool _dir_is_nonempty_unlocked(MDRequestRef
& mdr
, CInode
*rmdiri
);
255 bool _dir_is_nonempty(MDRequestRef
& mdr
, CInode
*rmdiri
);
256 void _unlink_local(MDRequestRef
& mdr
, CDentry
*dn
, CDentry
*straydn
);
257 void _unlink_local_finish(MDRequestRef
& mdr
,
258 CDentry
*dn
, CDentry
*straydn
,
260 bool _rmdir_prepare_witness(MDRequestRef
& mdr
, mds_rank_t who
, vector
<CDentry
*>& trace
, CDentry
*straydn
);
261 void handle_slave_rmdir_prep(MDRequestRef
& mdr
);
262 void _logged_slave_rmdir(MDRequestRef
& mdr
, CDentry
*srcdn
, CDentry
*straydn
);
263 void _commit_slave_rmdir(MDRequestRef
& mdr
, int r
);
264 void handle_slave_rmdir_prep_ack(MDRequestRef
& mdr
, MMDSSlaveRequest
*ack
);
265 void do_rmdir_rollback(bufferlist
&rbl
, mds_rank_t master
, MDRequestRef
& mdr
);
266 void _rmdir_rollback_finish(MDRequestRef
& mdr
, metareqid_t reqid
, CDentry
*dn
, CDentry
*straydn
);
269 void handle_client_rename(MDRequestRef
& mdr
);
270 void _rename_finish(MDRequestRef
& mdr
,
271 CDentry
*srcdn
, CDentry
*destdn
, CDentry
*straydn
);
273 void handle_client_lssnap(MDRequestRef
& mdr
);
274 void handle_client_mksnap(MDRequestRef
& mdr
);
275 void _mksnap_finish(MDRequestRef
& mdr
, CInode
*diri
, SnapInfo
&info
);
276 void handle_client_rmsnap(MDRequestRef
& mdr
);
277 void _rmsnap_finish(MDRequestRef
& mdr
, CInode
*diri
, snapid_t snapid
);
278 void handle_client_renamesnap(MDRequestRef
& mdr
);
279 void _renamesnap_finish(MDRequestRef
& mdr
, CInode
*diri
, snapid_t snapid
);
283 bool _rename_prepare_witness(MDRequestRef
& mdr
, mds_rank_t who
, set
<mds_rank_t
> &witnesse
,
284 vector
<CDentry
*>& srctrace
, vector
<CDentry
*>& dsttrace
, CDentry
*straydn
);
285 version_t
_rename_prepare_import(MDRequestRef
& mdr
, CDentry
*srcdn
, bufferlist
*client_map_bl
);
286 bool _need_force_journal(CInode
*diri
, bool empty
);
287 void _rename_prepare(MDRequestRef
& mdr
,
288 EMetaBlob
*metablob
, bufferlist
*client_map_bl
,
289 CDentry
*srcdn
, CDentry
*destdn
, CDentry
*straydn
);
290 /* set not_journaling=true if you're going to discard the results --
291 * this bypasses the asserts to make sure we're journaling the right
292 * things on the right nodes */
293 void _rename_apply(MDRequestRef
& mdr
, CDentry
*srcdn
, CDentry
*destdn
, CDentry
*straydn
);
296 void handle_slave_rename_prep(MDRequestRef
& mdr
);
297 void handle_slave_rename_prep_ack(MDRequestRef
& mdr
, MMDSSlaveRequest
*m
);
298 void handle_slave_rename_notify_ack(MDRequestRef
& mdr
, MMDSSlaveRequest
*m
);
299 void _slave_rename_sessions_flushed(MDRequestRef
& mdr
);
300 void _logged_slave_rename(MDRequestRef
& mdr
, CDentry
*srcdn
, CDentry
*destdn
, CDentry
*straydn
);
301 void _commit_slave_rename(MDRequestRef
& mdr
, int r
, CDentry
*srcdn
, CDentry
*destdn
, CDentry
*straydn
);
302 void do_rename_rollback(bufferlist
&rbl
, mds_rank_t master
, MDRequestRef
& mdr
, bool finish_mdr
=false);
303 void _rename_rollback_finish(MutationRef
& mut
, MDRequestRef
& mdr
, CDentry
*srcdn
, version_t srcdnpv
,
304 CDentry
*destdn
, CDentry
*staydn
, bool finish_mdr
);
307 void reply_client_request(MDRequestRef
& mdr
, MClientReply
*reply
);