[ceph.git] / ceph / src / mds / Locker.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_LOCKER_H
#define CEPH_MDS_LOCKER_H

#include <boost/utility/string_view.hpp>

#include "include/types.h"

#include <map>
#include <list>
#include <set>
using std::map;
using std::list;
using std::set;

class MDSRank;
class Session;
class CDentry;
struct SnapRealm;

class Message;

class MLock;

class Capability;

class SimpleLock;
class ScatterLock;
class LocalLock;

#include "CInode.h"
#include "SimpleLock.h"
#include "Mutation.h"

class Locker {
private:
  MDSRank *mds;
  MDCache *mdcache;
 
 public:
  Locker(MDSRank *m, MDCache *c) : mds(m), mdcache(c) {}  

  SimpleLock *get_lock(int lock_type, MDSCacheObjectInfo &info);
  
  void dispatch(Message *m);
  void handle_lock(MLock *m);

  void tick();

  void nudge_log(SimpleLock *lock);

protected:
  void send_lock_message(SimpleLock *lock, int msg);
  void send_lock_message(SimpleLock *lock, int msg, const bufferlist &data);

  // -- locks --
  void _drop_rdlocks(MutationImpl *mut, set<CInode*> *pneed_issue);
  void _drop_non_rdlocks(MutationImpl *mut, set<CInode*> *pneed_issue);
public:
  void include_snap_rdlocks(set<SimpleLock*>& rdlocks, CInode *in);
  void include_snap_rdlocks_wlayout(set<SimpleLock*>& rdlocks, CInode *in,
                                    file_layout_t **layout);

  bool acquire_locks(MDRequestRef& mdr,
		     set<SimpleLock*> &rdlocks,
		     set<SimpleLock*> &wrlocks,
		     set<SimpleLock*> &xlocks,
		     map<SimpleLock*,mds_rank_t> *remote_wrlocks=NULL,
		     CInode *auth_pin_freeze=NULL,
		     bool auth_pin_nonblock=false);

  void notify_freeze_waiter(MDSCacheObject *o);
  void cancel_locking(MutationImpl *mut, set<CInode*> *pneed_issue);
  void drop_locks(MutationImpl *mut, set<CInode*> *pneed_issue=0);
  void set_xlocks_done(MutationImpl *mut, bool skip_dentry=false);
  void drop_non_rdlocks(MutationImpl *mut, set<CInode*> *pneed_issue=0);
  void drop_rdlocks_for_early_reply(MutationImpl *mut);
  void drop_locks_for_fragment_unfreeze(MutationImpl *mut);

  void eval_gather(SimpleLock *lock, bool first=false, bool *need_issue=0, list<MDSInternalContextBase*> *pfinishers=0);
  void eval(SimpleLock *lock, bool *need_issue);
  void eval_any(SimpleLock *lock, bool *need_issue, list<MDSInternalContextBase*> *pfinishers=0, bool first=false) {
    if (!lock->is_stable())
      eval_gather(lock, first, need_issue, pfinishers);
    else if (lock->get_parent()->is_auth())
      eval(lock, need_issue);
  }

  void eval_scatter_gathers(CInode *in);

  void eval_cap_gather(CInode *in, set<CInode*> *issue_set=0);

  bool eval(CInode *in, int mask, bool caps_imported=false);
  void try_eval(MDSCacheObject *p, int mask);
  void try_eval(SimpleLock *lock, bool *pneed_issue);

  bool _rdlock_kick(SimpleLock *lock, bool as_anon);
  bool rdlock_try(SimpleLock *lock, client_t client, MDSInternalContextBase *c);
  bool rdlock_start(SimpleLock *lock, MDRequestRef& mut, bool as_anon=false);
  void rdlock_finish(SimpleLock *lock, MutationImpl *mut, bool *pneed_issue);
  bool can_rdlock_set(set<SimpleLock*>& locks);
  bool rdlock_try_set(set<SimpleLock*>& locks);
  void rdlock_take_set(set<SimpleLock*>& locks, MutationRef& mut);

  void wrlock_force(SimpleLock *lock, MutationRef& mut);
  bool wrlock_start(SimpleLock *lock, MDRequestRef& mut, bool nowait=false);
  void wrlock_finish(SimpleLock *lock, MutationImpl *mut, bool *pneed_issue);

  void remote_wrlock_start(SimpleLock *lock, mds_rank_t target, MDRequestRef& mut);
  void remote_wrlock_finish(SimpleLock *lock, mds_rank_t target, MutationImpl *mut);

  bool xlock_start(SimpleLock *lock, MDRequestRef& mut);
  void _finish_xlock(SimpleLock *lock, client_t xlocker, bool *pneed_issue);
  void xlock_finish(SimpleLock *lock, MutationImpl *mut, bool *pneed_issue);

  void xlock_export(SimpleLock *lock, MutationImpl *mut);
  void xlock_import(SimpleLock *lock);


  // simple
public:
  void try_simple_eval(SimpleLock *lock);
  bool simple_rdlock_try(SimpleLock *lock, MDSInternalContextBase *con);
protected:
  void simple_eval(SimpleLock *lock, bool *need_issue);
  void handle_simple_lock(SimpleLock *lock, MLock *m);

public:
  bool simple_sync(SimpleLock *lock, bool *need_issue=0);
protected:
  void simple_lock(SimpleLock *lock, bool *need_issue=0);
  void simple_excl(SimpleLock *lock, bool *need_issue=0);
  void simple_xlock(SimpleLock *lock);


  // scatter
public:
  void scatter_eval(ScatterLock *lock, bool *need_issue);        // public for MDCache::adjust_subtree_auth()

  void scatter_tick();
  void scatter_nudge(ScatterLock *lock, MDSInternalContextBase *c, bool forcelockchange=false);

protected:
  void handle_scatter_lock(ScatterLock *lock, MLock *m);
  bool scatter_scatter_fastpath(ScatterLock *lock);
  void scatter_scatter(ScatterLock *lock, bool nowait=false);
  void scatter_tempsync(ScatterLock *lock, bool *need_issue=0);

  void scatter_writebehind(ScatterLock *lock);

  void scatter_writebehind_finish(ScatterLock *lock, MutationRef& mut);

  xlist<ScatterLock*> updated_scatterlocks;
public:
  void mark_updated_scatterlock(ScatterLock *lock);


  void handle_reqrdlock(SimpleLock *lock, MLock *m);


  // caps

  // when to defer processing client cap release or writeback due to being
  // frozen.  the condition must be consistent across handle_client_caps and
  // process_request_cap_release to preserve ordering.
  bool should_defer_client_cap_frozen(CInode *in);

  void process_request_cap_release(MDRequestRef& mdr, client_t client, const ceph_mds_request_release& r,
				   boost::string_view dname);

  void kick_cap_releases(MDRequestRef& mdr);
  void kick_issue_caps(CInode *in, client_t client, ceph_seq_t seq);

  void remove_client_cap(CInode *in, Capability *cap);

  void get_late_revoking_clients(std::list<client_t> *result, double timeout) const;

private:
  bool any_late_revoking_caps(xlist<Capability*> const &revoking, double timeout) const;

 protected:
  bool _need_flush_mdlog(CInode *in, int wanted_caps);
  void adjust_cap_wanted(Capability *cap, int wanted, int issue_seq);
  void handle_client_caps(class MClientCaps *m);
  void _update_cap_fields(CInode *in, int dirty, MClientCaps *m, CInode::mempool_inode *pi);
  void _do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t follows, client_t client, MClientCaps *m, MClientCaps *ack);
  void _do_null_snapflush(CInode *head_in, client_t client, snapid_t last=CEPH_NOSNAP);
  bool _do_cap_update(CInode *in, Capability *cap, int dirty, snapid_t follows, MClientCaps *m,
		      MClientCaps *ack=0, bool *need_flush=NULL);
  void handle_client_cap_release(class MClientCapRelease *m);
  void _do_cap_release(client_t client, inodeno_t ino, uint64_t cap_id, ceph_seq_t mseq, ceph_seq_t seq);
  void caps_tick();

  // Maintain a global list to quickly find if any caps are late revoking
  xlist<Capability*> revoking_caps;
  // Maintain a per-client list to find clients responsible for late ones quickly
  std::map<client_t, xlist<Capability*> > revoking_caps_by_client;

  // local
public:
  void local_wrlock_grab(LocalLock *lock, MutationRef& mut);
protected:
  bool local_wrlock_start(LocalLock *lock, MDRequestRef& mut);
  void local_wrlock_finish(LocalLock *lock, MutationImpl *mut);
  bool local_xlock_start(LocalLock *lock, MDRequestRef& mut);
  void local_xlock_finish(LocalLock *lock, MutationImpl *mut);


  // file
public:
  void file_eval(ScatterLock *lock, bool *need_issue);
protected:
  void handle_file_lock(ScatterLock *lock, MLock *m);
  void scatter_mix(ScatterLock *lock, bool *need_issue=0);
  void file_excl(ScatterLock *lock, bool *need_issue=0);
  void file_xsyn(SimpleLock *lock, bool *need_issue=0);

public:
  void file_recover(ScatterLock *lock);

private:
  xlist<ScatterLock*> updated_filelocks;
public:
  void mark_updated_Filelock(ScatterLock *lock);

  // -- file i/o --
public:
  version_t issue_file_data_version(CInode *in);
  Capability* issue_new_caps(CInode *in, int mode, Session *session, SnapRealm *conrealm, bool is_replay);
  bool issue_caps(CInode *in, Capability *only_cap=0);
  void issue_caps_set(set<CInode*>& inset);
  void issue_truncate(CInode *in);
  void revoke_stale_caps(Session *session);
  void resume_stale_caps(Session *session);
  void remove_stale_leases(Session *session);

public:
  void request_inode_file_caps(CInode *in);
protected:
  void handle_inode_file_caps(class MInodeFileCaps *m);

  void file_update_finish(CInode *in, MutationRef& mut, bool share_max, bool issue_client_cap,
			  client_t client, MClientCaps *ack);
private:
  uint64_t calc_new_max_size(CInode::mempool_inode *pi, uint64_t size);
public:
  void calc_new_client_ranges(CInode *in, uint64_t size, bool update,
			      CInode::mempool_inode::client_range_map* new_ranges,
			      bool *max_increased);
  bool check_inode_max_size(CInode *in, bool force_wrlock=false,
                            uint64_t newmax=0, uint64_t newsize=0,
			    utime_t mtime=utime_t());
  void share_inode_max_size(CInode *in, Capability *only_cap=0);

private:
  friend class C_MDL_CheckMaxSize;
  friend class C_MDL_RequestInodeFileCaps;
  friend class C_Locker_FileUpdate_finish;
  friend class C_Locker_RetryCapRelease;
  friend class C_Locker_Eval;
  friend class C_Locker_ScatterWB;
  friend class LockerContext;
  friend class LockerLogContext;

  
  // -- client leases --
public:
  void handle_client_lease(struct MClientLease *m);

  void issue_client_lease(CDentry *dn, client_t client, bufferlist &bl, utime_t now, Session *session);
  void revoke_client_leases(SimpleLock *lock);
};


#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_LOCKER_H
	16	#define CEPH_MDS_LOCKER_H
	17
94b18763 FG	18	#include <boost/utility/string_view.hpp>
94b18763 FG	19
7c673cae FG	20	#include "include/types.h"
	21
	22	#include <map>
	23	#include <list>
	24	#include <set>
	25	using std::map;
	26	using std::list;
	27	using std::set;
	28
	29	class MDSRank;
	30	class Session;
7c673cae FG	31	class CDentry;
	32	struct SnapRealm;
	33
	34	class Message;
	35
	36	class MLock;
	37
	38	class Capability;
	39
	40	class SimpleLock;
	41	class ScatterLock;
	42	class LocalLock;
	43
94b18763	44	#include "CInode.h"
7c673cae FG	45	#include "SimpleLock.h"
	46	#include "Mutation.h"
	47
	48	class Locker {
	49	private:
	50	MDSRank *mds;
	51	MDCache *mdcache;
	52
	53	public:
	54	Locker(MDSRank m, MDCache c) : mds(m), mdcache(c) {}
	55
	56	SimpleLock *get_lock(int lock_type, MDSCacheObjectInfo &info);
	57
	58	void dispatch(Message *m);
	59	void handle_lock(MLock *m);
	60
	61	void tick();
	62
	63	void nudge_log(SimpleLock *lock);
	64
	65	protected:
	66	void send_lock_message(SimpleLock *lock, int msg);
	67	void send_lock_message(SimpleLock *lock, int msg, const bufferlist &data);
	68
	69	// -- locks --
	70	void _drop_rdlocks(MutationImpl mut, set<CInode> *pneed_issue);
	71	void _drop_non_rdlocks(MutationImpl mut, set<CInode> *pneed_issue);
	72	public:
	73	void include_snap_rdlocks(set<SimpleLock>& rdlocks, CInode in);
	74	void include_snap_rdlocks_wlayout(set<SimpleLock>& rdlocks, CInode in,
	75	file_layout_t **layout);
	76
	77	bool acquire_locks(MDRequestRef& mdr,
	78	set<SimpleLock*> &rdlocks,
	79	set<SimpleLock*> &wrlocks,
	80	set<SimpleLock*> &xlocks,
	81	map<SimpleLock,mds_rank_t> remote_wrlocks=NULL,
	82	CInode *auth_pin_freeze=NULL,
	83	bool auth_pin_nonblock=false);
	84
224ce89b	85	void notify_freeze_waiter(MDSCacheObject *o);
7c673cae FG	86	void cancel_locking(MutationImpl mut, set<CInode> *pneed_issue);
	87	void drop_locks(MutationImpl mut, set<CInode> *pneed_issue=0);
	88	void set_xlocks_done(MutationImpl *mut, bool skip_dentry=false);
	89	void drop_non_rdlocks(MutationImpl mut, set<CInode> *pneed_issue=0);
b32b8144	90	void drop_rdlocks_for_early_reply(MutationImpl *mut);
a8e16298	91	void drop_locks_for_fragment_unfreeze(MutationImpl *mut);
7c673cae FG	92
	93	void eval_gather(SimpleLock lock, bool first=false, bool need_issue=0, list<MDSInternalContextBase> pfinishers=0);
	94	void eval(SimpleLock lock, bool need_issue);
	95	void eval_any(SimpleLock lock, bool need_issue, list<MDSInternalContextBase> pfinishers=0, bool first=false) {
	96	if (!lock->is_stable())
	97	eval_gather(lock, first, need_issue, pfinishers);
	98	else if (lock->get_parent()->is_auth())
	99	eval(lock, need_issue);
	100	}
	101
	102	void eval_scatter_gathers(CInode *in);
	103
	104	void eval_cap_gather(CInode in, set<CInode> *issue_set=0);
	105
	106	bool eval(CInode *in, int mask, bool caps_imported=false);
	107	void try_eval(MDSCacheObject *p, int mask);
	108	void try_eval(SimpleLock lock, bool pneed_issue);
	109
	110	bool _rdlock_kick(SimpleLock *lock, bool as_anon);
	111	bool rdlock_try(SimpleLock lock, client_t client, MDSInternalContextBase c);
	112	bool rdlock_start(SimpleLock *lock, MDRequestRef& mut, bool as_anon=false);
	113	void rdlock_finish(SimpleLock lock, MutationImpl mut, bool *pneed_issue);
	114	bool can_rdlock_set(set<SimpleLock*>& locks);
	115	bool rdlock_try_set(set<SimpleLock*>& locks);
	116	void rdlock_take_set(set<SimpleLock*>& locks, MutationRef& mut);
	117
	118	void wrlock_force(SimpleLock *lock, MutationRef& mut);
	119	bool wrlock_start(SimpleLock *lock, MDRequestRef& mut, bool nowait=false);
	120	void wrlock_finish(SimpleLock lock, MutationImpl mut, bool *pneed_issue);
	121
	122	void remote_wrlock_start(SimpleLock *lock, mds_rank_t target, MDRequestRef& mut);
	123	void remote_wrlock_finish(SimpleLock lock, mds_rank_t target, MutationImpl mut);
	124
	125	bool xlock_start(SimpleLock *lock, MDRequestRef& mut);
	126	void _finish_xlock(SimpleLock lock, client_t xlocker, bool pneed_issue);
	127	void xlock_finish(SimpleLock lock, MutationImpl mut, bool *pneed_issue);
	128
	129	void xlock_export(SimpleLock lock, MutationImpl mut);
	130	void xlock_import(SimpleLock *lock);
	131
	132
	133	// simple
	134	public:
	135	void try_simple_eval(SimpleLock *lock);
	136	bool simple_rdlock_try(SimpleLock lock, MDSInternalContextBase con);
	137	protected:
	138	void simple_eval(SimpleLock lock, bool need_issue);
	139	void handle_simple_lock(SimpleLock lock, MLock m);
	140
	141	public:
	142	bool simple_sync(SimpleLock lock, bool need_issue=0);
	143	protected:
	144	void simple_lock(SimpleLock lock, bool need_issue=0);
	145	void simple_excl(SimpleLock lock, bool need_issue=0);
	146	void simple_xlock(SimpleLock *lock);
	147
	148
	149	// scatter
	150	public:
	151	void scatter_eval(ScatterLock lock, bool need_issue); // public for MDCache::adjust_subtree_auth()
	152
	153	void scatter_tick();
	154	void scatter_nudge(ScatterLock lock, MDSInternalContextBase c, bool forcelockchange=false);
	155
156	protected:
157	void handle_scatter_lock(ScatterLock lock, MLock m);
158	bool scatter_scatter_fastpath(ScatterLock *lock);
159	void scatter_scatter(ScatterLock *lock, bool nowait=false);
160	void scatter_tempsync(ScatterLock lock, bool need_issue=0);
161
162	void scatter_writebehind(ScatterLock *lock);
163
164	void scatter_writebehind_finish(ScatterLock *lock, MutationRef& mut);
165
166	xlist<ScatterLock*> updated_scatterlocks;
167	public:
168	void mark_updated_scatterlock(ScatterLock *lock);
169
170
171	void handle_reqrdlock(SimpleLock lock, MLock m);
172
173
174
175	// caps
176
177	// when to defer processing client cap release or writeback due to being
178	// frozen. the condition must be consistent across handle_client_caps and
179	// process_request_cap_release to preserve ordering.
180	bool should_defer_client_cap_frozen(CInode *in);
181
182	void process_request_cap_release(MDRequestRef& mdr, client_t client, const ceph_mds_request_release& r,
94b18763	183	boost::string_view dname);
7c673cae FG	184
	185	void kick_cap_releases(MDRequestRef& mdr);
	186	void kick_issue_caps(CInode *in, client_t client, ceph_seq_t seq);
	187
a8e16298	188	void remove_client_cap(CInode in, Capability cap);
7c673cae	189
91327a77 AA	190	void get_late_revoking_clients(std::list<client_t> *result, double timeout) const;
	191
	192	private:
	193	bool any_late_revoking_caps(xlist<Capability*> const &revoking, double timeout) const;
7c673cae FG	194
	195	protected:
	196	bool _need_flush_mdlog(CInode *in, int wanted_caps);
	197	void adjust_cap_wanted(Capability *cap, int wanted, int issue_seq);
	198	void handle_client_caps(class MClientCaps *m);
94b18763	199	void _update_cap_fields(CInode in, int dirty, MClientCaps m, CInode::mempool_inode *pi);
7c673cae	200	void _do_snap_update(CInode in, snapid_t snap, int dirty, snapid_t follows, client_t client, MClientCaps m, MClientCaps *ack);
c07f9fc5	201	void _do_null_snapflush(CInode *head_in, client_t client, snapid_t last=CEPH_NOSNAP);
7c673cae FG	202	bool _do_cap_update(CInode in, Capability cap, int dirty, snapid_t follows, MClientCaps *m,
	203	MClientCaps ack=0, bool need_flush=NULL);
	204	void handle_client_cap_release(class MClientCapRelease *m);
	205	void _do_cap_release(client_t client, inodeno_t ino, uint64_t cap_id, ceph_seq_t mseq, ceph_seq_t seq);
	206	void caps_tick();
	207
	208	// Maintain a global list to quickly find if any caps are late revoking
	209	xlist<Capability*> revoking_caps;
	210	// Maintain a per-client list to find clients responsible for late ones quickly
	211	std::map<client_t, xlist<Capability*> > revoking_caps_by_client;
	212
	213	// local
	214	public:
	215	void local_wrlock_grab(LocalLock *lock, MutationRef& mut);
	216	protected:
	217	bool local_wrlock_start(LocalLock *lock, MDRequestRef& mut);
	218	void local_wrlock_finish(LocalLock lock, MutationImpl mut);
	219	bool local_xlock_start(LocalLock *lock, MDRequestRef& mut);
	220	void local_xlock_finish(LocalLock lock, MutationImpl mut);
	221
	222
	223	// file
	224	public:
	225	void file_eval(ScatterLock lock, bool need_issue);
	226	protected:
	227	void handle_file_lock(ScatterLock lock, MLock m);
	228	void scatter_mix(ScatterLock lock, bool need_issue=0);
	229	void file_excl(ScatterLock lock, bool need_issue=0);
	230	void file_xsyn(SimpleLock lock, bool need_issue=0);
	231
	232	public:
	233	void file_recover(ScatterLock *lock);
	234
	235	private:
	236	xlist<ScatterLock*> updated_filelocks;
	237	public:
	238	void mark_updated_Filelock(ScatterLock *lock);
	239
	240	// -- file i/o --
	241	public:
	242	version_t issue_file_data_version(CInode *in);
	243	Capability* issue_new_caps(CInode in, int mode, Session session, SnapRealm *conrealm, bool is_replay);
	244	bool issue_caps(CInode in, Capability only_cap=0);
	245	void issue_caps_set(set<CInode*>& inset);
	246	void issue_truncate(CInode *in);
	247	void revoke_stale_caps(Session *session);
7c673cae FG	248	void resume_stale_caps(Session *session);
	249	void remove_stale_leases(Session *session);
	250
	251	public:
	252	void request_inode_file_caps(CInode *in);
	253	protected:
	254	void handle_inode_file_caps(class MInodeFileCaps *m);
	255
	256	void file_update_finish(CInode *in, MutationRef& mut, bool share_max, bool issue_client_cap,
	257	client_t client, MClientCaps *ack);
31f18b77	258	private:
94b18763	259	uint64_t calc_new_max_size(CInode::mempool_inode *pi, uint64_t size);
7c673cae	260	public:
a8e16298	261	void calc_new_client_ranges(CInode *in, uint64_t size, bool update,
94b18763	262	CInode::mempool_inode::client_range_map* new_ranges,
7c673cae FG	263	bool *max_increased);
	264	bool check_inode_max_size(CInode *in, bool force_wrlock=false,
	265	uint64_t newmax=0, uint64_t newsize=0,
	266	utime_t mtime=utime_t());
	267	void share_inode_max_size(CInode in, Capability only_cap=0);
	268
	269	private:
	270	friend class C_MDL_CheckMaxSize;
	271	friend class C_MDL_RequestInodeFileCaps;
	272	friend class C_Locker_FileUpdate_finish;
	273	friend class C_Locker_RetryCapRelease;
	274	friend class C_Locker_Eval;
	275	friend class C_Locker_ScatterWB;
	276	friend class LockerContext;
	277	friend class LockerLogContext;
	278
	279
	280	// -- client leases --
	281	public:
	282	void handle_client_lease(struct MClientLease *m);
	283
	284	void issue_client_lease(CDentry dn, client_t client, bufferlist &bl, utime_t now, Session session);
	285	void revoke_client_leases(SimpleLock *lock);
	286	};
	287
	288
	289	#endif