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.
17 #ifndef CEPH_MDCACHE_H
18 #define CEPH_MDCACHE_H
20 #include <boost/utility/string_view.hpp>
22 #include "include/types.h"
23 #include "include/filepath.h"
24 #include "include/elist.h"
26 #include "osdc/Filer.h"
30 #include "include/Context.h"
31 #include "events/EMetaBlob.h"
32 #include "RecoveryQueue.h"
33 #include "StrayManager.h"
34 #include "MDSContext.h"
38 #include "messages/MClientRequest.h"
39 #include "messages/MMDSSlaveRequest.h"
52 class MMDSCacheRejoin
;
61 struct MMDSFindInoReply
;
63 struct MMDSOpenInoReply
;
67 class MMDSSlaveRequest
;
70 class MMDSFragmentNotify
;
76 // How many inodes currently in stray dentries
78 // How many stray dentries are currently delayed for purge due to refs
79 l_mdc_num_strays_delayed
,
80 // How many stray dentries are currently being enqueued for purge
81 l_mdc_num_strays_enqueuing
,
83 // How many dentries have ever been added to stray dir
85 // How many dentries have been passed on to PurgeQueue
86 l_mdc_strays_enqueued
,
87 // How many strays have been reintegrated?
88 l_mdc_strays_reintegrated
,
89 // How many strays have been migrated?
90 l_mdc_strays_migrated
,
92 // How many inode sizes currently being recovered
93 l_mdc_num_recovering_processing
,
94 // How many inodes currently waiting to have size recovered
95 l_mdc_num_recovering_enqueued
,
96 // How many inodes waiting with elevated priority for recovery
97 l_mdc_num_recovering_prioritized
,
98 // How many inodes ever started size recovery
99 l_mdc_recovery_started
,
100 // How many inodes ever completed size recovery
101 l_mdc_recovery_completed
,
103 l_mdss_ireq_enqueue_scrub
,
104 l_mdss_ireq_exportdir
,
106 l_mdss_ireq_fragmentdir
,
107 l_mdss_ireq_fragstats
,
108 l_mdss_ireq_inodestats
,
114 // flags for predirty_journal_parents()
115 static const int PREDIRTY_PRIMARY
= 1; // primary dn, adjust nested accounting
116 static const int PREDIRTY_DIR
= 2; // update parent dir mtime/size
117 static const int PREDIRTY_SHALLOW
= 4; // only go to immediate parent (for easier rollback)
125 LRU lru
; // dentry lru for expiring items from cache
126 LRU bottom_lru
; // dentries that should be trimmed ASAP
128 ceph::unordered_map
<inodeno_t
,CInode
*> inode_map
; // map of head inodes by ino
129 map
<vinodeno_t
, CInode
*> snap_inode_map
; // map of snap inodes by ino
130 CInode
*root
; // root inode
131 CInode
*myin
; // .ceph/mds%d dir
134 void set_readonly() { readonly
= true; }
136 CInode
*strays
[NUM_STRAY
]; // my stray dir
139 CInode
*get_stray() {
140 return strays
[stray_index
];
143 set
<CInode
*> base_inodes
;
145 std::unique_ptr
<PerfCounters
> logger
;
149 bool exceeded_size_limit
;
152 static uint64_t cache_limit_inodes(void) {
153 return g_conf
->get_val
<int64_t>("mds_cache_size");
155 static uint64_t cache_limit_memory(void) {
156 return g_conf
->get_val
<uint64_t>("mds_cache_memory_limit");
158 static double cache_reservation(void) {
159 return g_conf
->get_val
<double>("mds_cache_reservation");
161 static double cache_mid(void) {
162 return g_conf
->get_val
<double>("mds_cache_mid");
164 static double cache_health_threshold(void) {
165 return g_conf
->get_val
<double>("mds_health_cache_threshold");
167 double cache_toofull_ratio(void) const {
168 uint64_t inode_limit
= cache_limit_inodes();
169 double inode_reserve
= inode_limit
*(1.0-cache_reservation());
170 double memory_reserve
= cache_limit_memory()*(1.0-cache_reservation());
171 return fmax(0.0, fmax((cache_size()-memory_reserve
)/memory_reserve
, inode_limit
== 0 ? 0.0 : (CInode::count()-inode_reserve
)/inode_reserve
));
173 bool cache_toofull(void) const {
174 return cache_toofull_ratio() > 0.0;
176 uint64_t cache_size(void) const {
177 return mempool::get_pool(mempool::mds_co::id
).allocated_bytes();
179 bool cache_overfull(void) const {
180 uint64_t inode_limit
= cache_limit_inodes();
181 return (inode_limit
> 0 && CInode::count() > inode_limit
*cache_health_threshold()) || (cache_size() > cache_limit_memory()*cache_health_threshold());
184 void advance_stray() {
185 stray_index
= (stray_index
+1)%NUM_STRAY
;
188 void activate_stray_manager();
191 * Call this when you know that a CDentry is ready to be passed
192 * on to StrayManager (i.e. this is a stray you've just created)
194 void notify_stray(CDentry
*dn
) {
195 assert(dn
->get_dir()->get_inode()->is_stray());
196 stray_manager
.eval_stray(dn
);
199 void maybe_eval_stray(CInode
*in
, bool delay
=false);
200 void clear_dirty_bits_for_stray(CInode
* diri
);
202 bool is_readonly() { return readonly
; }
203 void force_readonly();
207 int num_shadow_inodes
;
209 int num_inodes_with_caps
;
211 unsigned max_dir_commit_size
;
213 static file_layout_t
gen_default_file_layout(const MDSMap
&mdsmap
);
214 static file_layout_t
gen_default_log_layout(const MDSMap
&mdsmap
);
216 file_layout_t default_file_layout
;
217 file_layout_t default_log_layout
;
219 void register_perfcounters();
221 // -- client leases --
223 static const int client_lease_pools
= 3;
224 float client_lease_durations
[client_lease_pools
];
226 xlist
<ClientLease
*> client_leases
[client_lease_pools
];
228 void touch_client_lease(ClientLease
*r
, int pool
, utime_t ttl
) {
229 client_leases
[pool
].push_back(&r
->item_lease
);
233 void notify_stray_removed()
235 stray_manager
.notify_stray_removed();
238 void notify_stray_created()
240 stray_manager
.notify_stray_created();
243 void eval_remote(CDentry
*dn
)
245 stray_manager
.eval_remote(dn
);
249 uint64_t last_cap_id
;
254 struct discover_info_t
{
266 tid(0), mds(-1), snap(CEPH_NOSNAP
), basei(NULL
),
267 want_base_dir(false), want_xlocked(false) {}
270 basei
->put(MDSCacheObject::PIN_DISCOVERBASE
);
272 void pin_base(CInode
*b
) {
274 basei
->get(MDSCacheObject::PIN_DISCOVERBASE
);
278 map
<ceph_tid_t
, discover_info_t
> discovers
;
279 ceph_tid_t discover_last_tid
;
281 void _send_discover(discover_info_t
& dis
);
282 discover_info_t
& _create_discover(mds_rank_t mds
) {
283 ceph_tid_t t
= ++discover_last_tid
;
284 discover_info_t
& d
= discovers
[t
];
291 map
<int, map
<inodeno_t
, list
<MDSInternalContextBase
*> > > waiting_for_base_ino
;
293 void discover_base_ino(inodeno_t want_ino
, MDSInternalContextBase
*onfinish
, mds_rank_t from
=MDS_RANK_NONE
);
294 void discover_dir_frag(CInode
*base
, frag_t approx_fg
, MDSInternalContextBase
*onfinish
,
295 mds_rank_t from
=MDS_RANK_NONE
);
296 void discover_path(CInode
*base
, snapid_t snap
, filepath want_path
, MDSInternalContextBase
*onfinish
,
297 bool want_xlocked
=false, mds_rank_t from
=MDS_RANK_NONE
);
298 void discover_path(CDir
*base
, snapid_t snap
, filepath want_path
, MDSInternalContextBase
*onfinish
,
299 bool want_xlocked
=false);
300 void kick_discovers(mds_rank_t who
); // after a failure.
305 /* subtree keys and each tree's non-recursive nested subtrees (the "bounds") */
306 map
<CDir
*,set
<CDir
*> > subtrees
;
307 map
<CInode
*,list
<pair
<CDir
*,CDir
*> > > projected_subtree_renames
; // renamed ino -> target dir
309 // adjust subtree auth specification
311 // imports/exports/nested_exports
312 // join/split subtrees as appropriate
314 bool is_subtrees() { return !subtrees
.empty(); }
315 void list_subtrees(list
<CDir
*>& ls
);
316 void adjust_subtree_auth(CDir
*root
, mds_authority_t auth
);
317 void adjust_subtree_auth(CDir
*root
, mds_rank_t a
, mds_rank_t b
=CDIR_AUTH_UNKNOWN
) {
318 adjust_subtree_auth(root
, mds_authority_t(a
,b
));
320 void adjust_bounded_subtree_auth(CDir
*dir
, set
<CDir
*>& bounds
, mds_authority_t auth
);
321 void adjust_bounded_subtree_auth(CDir
*dir
, set
<CDir
*>& bounds
, mds_rank_t a
) {
322 adjust_bounded_subtree_auth(dir
, bounds
, mds_authority_t(a
, CDIR_AUTH_UNKNOWN
));
324 void adjust_bounded_subtree_auth(CDir
*dir
, vector
<dirfrag_t
>& bounds
, mds_authority_t auth
);
325 void adjust_bounded_subtree_auth(CDir
*dir
, vector
<dirfrag_t
>& bounds
, mds_rank_t a
) {
326 adjust_bounded_subtree_auth(dir
, bounds
, mds_authority_t(a
, CDIR_AUTH_UNKNOWN
));
328 void map_dirfrag_set(list
<dirfrag_t
>& dfs
, set
<CDir
*>& result
);
329 void try_subtree_merge(CDir
*root
);
330 void try_subtree_merge_at(CDir
*root
, set
<CInode
*> *to_eval
);
331 void subtree_merge_writebehind_finish(CInode
*in
, MutationRef
& mut
);
332 void eval_subtree_root(CInode
*diri
);
333 CDir
*get_subtree_root(CDir
*dir
);
334 CDir
*get_projected_subtree_root(CDir
*dir
);
335 bool is_leaf_subtree(CDir
*dir
) {
336 assert(subtrees
.count(dir
));
337 return subtrees
[dir
].empty();
339 void remove_subtree(CDir
*dir
);
340 bool is_subtree(CDir
*root
) {
341 return subtrees
.count(root
);
343 void get_subtree_bounds(CDir
*root
, set
<CDir
*>& bounds
);
344 void get_wouldbe_subtree_bounds(CDir
*root
, set
<CDir
*>& bounds
);
345 void verify_subtree_bounds(CDir
*root
, const set
<CDir
*>& bounds
);
346 void verify_subtree_bounds(CDir
*root
, const list
<dirfrag_t
>& bounds
);
348 void project_subtree_rename(CInode
*diri
, CDir
*olddir
, CDir
*newdir
);
349 void adjust_subtree_after_rename(CInode
*diri
, CDir
*olddir
, bool pop
);
351 void get_auth_subtrees(set
<CDir
*>& s
);
352 void get_fullauth_subtrees(set
<CDir
*>& s
);
355 int num_subtrees_fullauth();
356 int num_subtrees_fullnonauth();
360 // delayed cache expire
361 map
<CDir
*, map
<mds_rank_t
, MCacheExpire
*> > delayed_expire
; // subtree root -> expire msg
365 ceph::unordered_map
<metareqid_t
, MDRequestRef
> active_requests
;
368 int get_num_client_requests();
370 MDRequestRef
request_start(MClientRequest
*req
);
371 MDRequestRef
request_start_slave(metareqid_t rid
, __u32 attempt
, Message
*m
);
372 MDRequestRef
request_start_internal(int op
);
373 bool have_request(metareqid_t rid
) {
374 return active_requests
.count(rid
);
376 MDRequestRef
request_get(metareqid_t rid
);
377 void request_pin_ref(MDRequestRef
& r
, CInode
*ref
, vector
<CDentry
*>& trace
);
378 void request_finish(MDRequestRef
& mdr
);
379 void request_forward(MDRequestRef
& mdr
, mds_rank_t mds
, int port
=0);
380 void dispatch_request(MDRequestRef
& mdr
);
381 void request_drop_foreign_locks(MDRequestRef
& mdr
);
382 void request_drop_non_rdlocks(MDRequestRef
& r
);
383 void request_drop_locks(MDRequestRef
& r
);
384 void request_cleanup(MDRequestRef
& r
);
386 void request_kill(MDRequestRef
& r
); // called when session closes
388 // journal/snap helpers
389 CInode
*pick_inode_snap(CInode
*in
, snapid_t follows
);
390 CInode
*cow_inode(CInode
*in
, snapid_t last
);
391 void journal_cow_dentry(MutationImpl
*mut
, EMetaBlob
*metablob
, CDentry
*dn
,
392 snapid_t follows
=CEPH_NOSNAP
,
393 CInode
**pcow_inode
=0, CDentry::linkage_t
*dnl
=0);
394 void journal_cow_inode(MutationRef
& mut
, EMetaBlob
*metablob
, CInode
*in
, snapid_t follows
=CEPH_NOSNAP
,
395 CInode
**pcow_inode
=0);
396 void journal_dirty_inode(MutationImpl
*mut
, EMetaBlob
*metablob
, CInode
*in
, snapid_t follows
=CEPH_NOSNAP
);
398 void project_rstat_inode_to_frag(CInode
*cur
, CDir
*parent
, snapid_t first
,
399 int linkunlink
, SnapRealm
*prealm
);
400 void _project_rstat_inode_to_frag(CInode::mempool_inode
& inode
, snapid_t ofirst
, snapid_t last
,
401 CDir
*parent
, int linkunlink
, bool update_inode
);
402 void project_rstat_frag_to_inode(nest_info_t
& rstat
, nest_info_t
& accounted_rstat
,
403 snapid_t ofirst
, snapid_t last
,
404 CInode
*pin
, bool cow_head
);
405 void broadcast_quota_to_client(CInode
*in
);
406 void predirty_journal_parents(MutationRef mut
, EMetaBlob
*blob
,
407 CInode
*in
, CDir
*parent
,
408 int flags
, int linkunlink
=0,
409 snapid_t follows
=CEPH_NOSNAP
);
412 void add_uncommitted_master(metareqid_t reqid
, LogSegment
*ls
, set
<mds_rank_t
> &slaves
, bool safe
=false) {
413 uncommitted_masters
[reqid
].ls
= ls
;
414 uncommitted_masters
[reqid
].slaves
= slaves
;
415 uncommitted_masters
[reqid
].safe
= safe
;
417 void wait_for_uncommitted_master(metareqid_t reqid
, MDSInternalContextBase
*c
) {
418 uncommitted_masters
[reqid
].waiters
.push_back(c
);
420 bool have_uncommitted_master(metareqid_t reqid
, mds_rank_t from
) {
421 auto p
= uncommitted_masters
.find(reqid
);
422 return p
!= uncommitted_masters
.end() && p
->second
.slaves
.count(from
) > 0;
424 void log_master_commit(metareqid_t reqid
);
425 void logged_master_update(metareqid_t reqid
);
426 void _logged_master_commit(metareqid_t reqid
);
427 void committed_master_slave(metareqid_t r
, mds_rank_t from
);
428 void finish_committed_masters();
430 void _logged_slave_commit(mds_rank_t from
, metareqid_t reqid
);
434 set
<mds_rank_t
> recovery_set
;
437 void set_recovery_set(set
<mds_rank_t
>& s
);
438 void handle_mds_failure(mds_rank_t who
);
439 void handle_mds_recovery(mds_rank_t who
);
443 // from EImportStart w/o EImportFinish during journal replay
444 map
<dirfrag_t
, vector
<dirfrag_t
> > my_ambiguous_imports
;
446 map
<mds_rank_t
, map
<dirfrag_t
, vector
<dirfrag_t
> > > other_ambiguous_imports
;
448 map
<mds_rank_t
, map
<metareqid_t
, MDSlaveUpdate
*> > uncommitted_slave_updates
; // slave: for replay.
449 map
<CInode
*, int> uncommitted_slave_rename_olddir
; // slave: preserve the non-auth dir until seeing commit.
450 map
<CInode
*, int> uncommitted_slave_unlink
; // slave: preserve the unlinked inode until seeing commit.
452 // track master requests whose slaves haven't acknowledged commit
454 set
<mds_rank_t
> slaves
;
456 list
<MDSInternalContextBase
*> waiters
;
460 umaster() : ls(NULL
), safe(false), committing(false), recovering(false) {}
462 map
<metareqid_t
, umaster
> uncommitted_masters
; // master: req -> slave set
464 set
<metareqid_t
> pending_masters
;
465 map
<int, set
<metareqid_t
> > ambiguous_slave_updates
;
467 friend class ESlaveUpdate
;
468 friend class ECommitted
;
470 bool resolves_pending
;
471 set
<mds_rank_t
> resolve_gather
; // nodes i need resolves from
472 set
<mds_rank_t
> resolve_ack_gather
; // nodes i need a resolve_ack from
473 map
<metareqid_t
, mds_rank_t
> need_resolve_rollback
; // rollbacks i'm writing to the journal
474 map
<mds_rank_t
, MMDSResolve
*> delayed_resolve
;
476 void handle_resolve(MMDSResolve
*m
);
477 void handle_resolve_ack(MMDSResolveAck
*m
);
478 void process_delayed_resolve();
479 void discard_delayed_resolve(mds_rank_t who
);
480 void maybe_resolve_finish();
481 void disambiguate_my_imports();
482 void disambiguate_other_imports();
483 void trim_unlinked_inodes();
484 void add_uncommitted_slave_update(metareqid_t reqid
, mds_rank_t master
, MDSlaveUpdate
*);
485 void finish_uncommitted_slave_update(metareqid_t reqid
, mds_rank_t master
);
486 MDSlaveUpdate
* get_uncommitted_slave_update(metareqid_t reqid
, mds_rank_t master
);
488 void recalc_auth_bits(bool replay
);
489 void remove_inode_recursive(CInode
*in
);
491 bool is_ambiguous_slave_update(metareqid_t reqid
, mds_rank_t master
) {
492 auto p
= ambiguous_slave_updates
.find(master
);
493 return p
!= ambiguous_slave_updates
.end() && p
->second
.count(reqid
);
495 void add_ambiguous_slave_update(metareqid_t reqid
, mds_rank_t master
) {
496 ambiguous_slave_updates
[master
].insert(reqid
);
498 void remove_ambiguous_slave_update(metareqid_t reqid
, mds_rank_t master
) {
499 auto p
= ambiguous_slave_updates
.find(master
);
500 auto q
= p
->second
.find(reqid
);
501 assert(q
!= p
->second
.end());
503 if (p
->second
.empty())
504 ambiguous_slave_updates
.erase(p
);
507 void add_rollback(metareqid_t reqid
, mds_rank_t master
) {
508 need_resolve_rollback
[reqid
] = master
;
510 void finish_rollback(metareqid_t reqid
);
513 void add_ambiguous_import(dirfrag_t base
, const vector
<dirfrag_t
>& bounds
);
514 void add_ambiguous_import(CDir
*base
, const set
<CDir
*>& bounds
);
515 bool have_ambiguous_import(dirfrag_t base
) {
516 return my_ambiguous_imports
.count(base
);
518 void get_ambiguous_import_bounds(dirfrag_t base
, vector
<dirfrag_t
>& bounds
) {
519 assert(my_ambiguous_imports
.count(base
));
520 bounds
= my_ambiguous_imports
[base
];
522 void cancel_ambiguous_import(CDir
*);
523 void finish_ambiguous_import(dirfrag_t dirino
);
524 void resolve_start(MDSInternalContext
*resolve_done_
);
525 void send_resolves();
526 void send_slave_resolves();
527 void send_subtree_resolves();
528 void maybe_send_pending_resolves() {
529 if (resolves_pending
)
530 send_subtree_resolves();
533 void _move_subtree_map_bound(dirfrag_t df
, dirfrag_t oldparent
, dirfrag_t newparent
,
534 map
<dirfrag_t
,vector
<dirfrag_t
> >& subtrees
);
535 ESubtreeMap
*create_subtree_map();
538 void clean_open_file_lists();
542 bool rejoins_pending
;
543 set
<mds_rank_t
> rejoin_gather
; // nodes from whom i need a rejoin
544 set
<mds_rank_t
> rejoin_sent
; // nodes i sent a rejoin to
545 set
<mds_rank_t
> rejoin_ack_sent
; // nodes i sent a rejoin to
546 set
<mds_rank_t
> rejoin_ack_gather
; // nodes from whom i need a rejoin ack
547 map
<mds_rank_t
,map
<inodeno_t
,map
<client_t
,Capability::Import
> > > rejoin_imported_caps
;
548 map
<inodeno_t
,pair
<mds_rank_t
,map
<client_t
,Capability::Export
> > > rejoin_slave_exports
;
549 map
<client_t
,entity_inst_t
> rejoin_client_map
;
551 map
<inodeno_t
,map
<client_t
,cap_reconnect_t
> > cap_exports
; // ino -> client -> capex
552 map
<inodeno_t
,mds_rank_t
> cap_export_targets
; // ino -> auth mds
554 map
<inodeno_t
,map
<client_t
,map
<mds_rank_t
,cap_reconnect_t
> > > cap_imports
; // ino -> client -> frommds -> capex
555 set
<inodeno_t
> cap_imports_missing
;
556 map
<inodeno_t
, list
<MDSInternalContextBase
*> > cap_reconnect_waiters
;
557 int cap_imports_num_opening
;
559 set
<CInode
*> rejoin_undef_inodes
;
560 set
<CInode
*> rejoin_potential_updated_scatterlocks
;
561 set
<CDir
*> rejoin_undef_dirfrags
;
562 map
<mds_rank_t
, set
<CInode
*> > rejoin_unlinked_inodes
;
564 vector
<CInode
*> rejoin_recover_q
, rejoin_check_q
;
565 list
<SimpleLock
*> rejoin_eval_locks
;
566 list
<MDSInternalContextBase
*> rejoin_waiters
;
568 void rejoin_walk(CDir
*dir
, MMDSCacheRejoin
*rejoin
);
569 void handle_cache_rejoin(MMDSCacheRejoin
*m
);
570 void handle_cache_rejoin_weak(MMDSCacheRejoin
*m
);
571 CInode
* rejoin_invent_inode(inodeno_t ino
, snapid_t last
);
572 CDir
* rejoin_invent_dirfrag(dirfrag_t df
);
573 void handle_cache_rejoin_strong(MMDSCacheRejoin
*m
);
574 void rejoin_scour_survivor_replicas(mds_rank_t from
, MMDSCacheRejoin
*ack
,
575 set
<vinodeno_t
>& acked_inodes
,
576 set
<SimpleLock
*>& gather_locks
);
577 void handle_cache_rejoin_ack(MMDSCacheRejoin
*m
);
578 void rejoin_send_acks();
579 void rejoin_trim_undef_inodes();
580 void maybe_send_pending_rejoins() {
582 rejoin_send_rejoins();
584 std::unique_ptr
<MDSInternalContext
> rejoin_done
;
585 std::unique_ptr
<MDSInternalContext
> resolve_done
;
587 void rejoin_start(MDSInternalContext
*rejoin_done_
);
588 void rejoin_gather_finish();
589 void rejoin_send_rejoins();
590 void rejoin_export_caps(inodeno_t ino
, client_t client
, const cap_reconnect_t
& icr
,
592 cap_exports
[ino
][client
] = icr
;
593 cap_export_targets
[ino
] = target
;
595 void rejoin_recovered_caps(inodeno_t ino
, client_t client
, const cap_reconnect_t
& icr
,
596 mds_rank_t frommds
=MDS_RANK_NONE
) {
597 cap_imports
[ino
][client
][frommds
] = icr
;
599 const cap_reconnect_t
*get_replay_cap_reconnect(inodeno_t ino
, client_t client
) {
600 if (cap_imports
.count(ino
) &&
601 cap_imports
[ino
].count(client
) &&
602 cap_imports
[ino
][client
].count(MDS_RANK_NONE
)) {
603 return &cap_imports
[ino
][client
][MDS_RANK_NONE
];
607 void remove_replay_cap_reconnect(inodeno_t ino
, client_t client
) {
608 assert(cap_imports
[ino
].size() == 1);
609 assert(cap_imports
[ino
][client
].size() == 1);
610 cap_imports
.erase(ino
);
612 void wait_replay_cap_reconnect(inodeno_t ino
, MDSInternalContextBase
*c
) {
613 cap_reconnect_waiters
[ino
].push_back(c
);
616 // [reconnect/rejoin caps]
617 struct reconnected_cap_info_t
{
619 snapid_t snap_follows
;
621 reconnected_cap_info_t() :
622 realm_ino(0), snap_follows(0), dirty_caps(0) {}
624 map
<inodeno_t
,map
<client_t
, reconnected_cap_info_t
> > reconnected_caps
; // inode -> client -> snap_follows,realmino
625 map
<inodeno_t
,map
<client_t
, snapid_t
> > reconnected_snaprealms
; // realmino -> client -> realmseq
627 void add_reconnected_cap(client_t client
, inodeno_t ino
, const cap_reconnect_t
& icr
) {
628 reconnected_cap_info_t
&info
= reconnected_caps
[ino
][client
];
629 info
.realm_ino
= inodeno_t(icr
.capinfo
.snaprealm
);
630 info
.snap_follows
= icr
.snap_follows
;
632 void set_reconnected_dirty_caps(client_t client
, inodeno_t ino
, int dirty
) {
633 reconnected_cap_info_t
&info
= reconnected_caps
[ino
][client
];
634 info
.dirty_caps
|= dirty
;
636 void add_reconnected_snaprealm(client_t client
, inodeno_t ino
, snapid_t seq
) {
637 reconnected_snaprealms
[ino
][client
] = seq
;
640 friend class C_MDC_RejoinOpenInoFinish
;
641 friend class C_MDC_RejoinSessionsOpened
;
642 void rejoin_open_ino_finish(inodeno_t ino
, int ret
);
643 void rejoin_open_sessions_finish(map
<client_t
,entity_inst_t
> client_map
,
644 map
<client_t
,uint64_t>& sseqmap
);
645 bool process_imported_caps();
646 void choose_lock_states_and_reconnect_caps();
647 void prepare_realm_split(SnapRealm
*realm
, client_t client
, inodeno_t ino
,
648 map
<client_t
,MClientSnap
*>& splits
);
649 void do_realm_invalidate_and_update_notify(CInode
*in
, int snapop
, bool nosend
=false);
650 void send_snaps(map
<client_t
,MClientSnap
*>& splits
);
651 Capability
* rejoin_import_cap(CInode
*in
, client_t client
, const cap_reconnect_t
& icr
, mds_rank_t frommds
);
652 void finish_snaprealm_reconnect(client_t client
, SnapRealm
*realm
, snapid_t seq
);
653 void try_reconnect_cap(CInode
*in
, Session
*session
);
654 void export_remaining_imported_caps();
656 // cap imports. delayed snap parent opens.
657 // realm inode -> client -> cap inodes needing to split to this realm
658 map
<CInode
*,set
<CInode
*> > missing_snap_parents
;
659 map
<client_t
,set
<CInode
*> > delayed_imported_caps
;
661 void do_cap_import(Session
*session
, CInode
*in
, Capability
*cap
,
662 uint64_t p_cap_id
, ceph_seq_t p_seq
, ceph_seq_t p_mseq
,
663 int peer
, int p_flags
);
664 void do_delayed_cap_imports();
665 void rebuild_need_snapflush(CInode
*head_in
, SnapRealm
*realm
, client_t client
,
666 snapid_t snap_follows
);
667 void check_realm_past_parents(SnapRealm
*realm
, bool reconnect
);
668 void open_snap_parents();
670 bool open_undef_inodes_dirfrags();
671 void opened_undef_inode(CInode
*in
);
672 void opened_undef_dirfrag(CDir
*dir
) {
673 rejoin_undef_dirfrags
.erase(dir
);
676 void reissue_all_caps();
680 friend class Migrator
;
681 friend class MDBalancer
;
683 // StrayManager needs to be able to remove_inode() from us
684 // when it is done purging
685 friend class StrayManager
;
687 // File size recovery
689 RecoveryQueue recovery_queue
;
690 void identify_files_to_recover();
692 void start_files_to_recover();
693 void do_file_recover();
694 void queue_file_recover(CInode
*in
);
695 void _queued_file_recover_cow(CInode
*in
, MutationRef
& mut
);
698 std::unique_ptr
<Migrator
> migrator
;
701 explicit MDCache(MDSRank
*m
, PurgeQueue
&purge_queue_
);
708 CInode
*get_root() { return root
; }
709 CInode
*get_myin() { return myin
; }
711 size_t get_cache_size() { return lru
.lru_get_size(); }
714 bool trim(uint64_t count
=0);
716 void trim_lru(uint64_t count
, map
<mds_rank_t
, MCacheExpire
*>& expiremap
);
717 bool trim_dentry(CDentry
*dn
, map
<mds_rank_t
, MCacheExpire
*>& expiremap
);
718 void trim_dirfrag(CDir
*dir
, CDir
*con
,
719 map
<mds_rank_t
, MCacheExpire
*>& expiremap
);
720 bool trim_inode(CDentry
*dn
, CInode
*in
, CDir
*con
,
721 map
<mds_rank_t
,class MCacheExpire
*>& expiremap
);
722 void send_expire_messages(map
<mds_rank_t
, MCacheExpire
*>& expiremap
);
723 void trim_non_auth(); // trim out trimmable non-auth items
725 bool trim_non_auth_subtree(CDir
*directory
);
726 void standby_trim_segment(LogSegment
*ls
);
727 void try_trim_non_auth_subtree(CDir
*dir
);
728 bool can_trim_non_auth_dirfrag(CDir
*dir
) {
729 return my_ambiguous_imports
.count((dir
)->dirfrag()) == 0 &&
730 uncommitted_slave_rename_olddir
.count(dir
->inode
) == 0;
734 * For all unreferenced inodes, dirs, dentries below an inode, compose
735 * expiry messages. This is used when giving up all replicas of entities
736 * for an MDS peer in the 'stopping' state, such that the peer can
737 * empty its cache and finish shutting down.
739 * We have to make sure we're only expiring un-referenced items to
740 * avoid interfering with ongoing stray-movement (we can't distinguish
741 * between the "moving my strays" and "waiting for my cache to empty"
742 * phases within 'stopping')
744 * @return false if we completed cleanly, true if caller should stop
745 * expiring because we hit something with refs.
747 bool expire_recursive(
749 std::map
<mds_rank_t
, MCacheExpire
*>& expiremap
);
751 void trim_client_leases();
752 void check_memory_usage();
754 utime_t last_recall_state
;
758 set
<inodeno_t
> shutdown_exported_strays
;
760 void shutdown_start();
761 void shutdown_check();
762 bool shutdown_pass();
763 bool shutdown_export_strays();
764 bool shutdown(); // clear cache (ie at shutodwn)
766 bool did_shutdown_log_cap
;
769 bool have_inode(vinodeno_t vino
) {
770 if (vino
.snapid
== CEPH_NOSNAP
)
771 return inode_map
.count(vino
.ino
) ? true : false;
773 return snap_inode_map
.count(vino
) ? true : false;
775 bool have_inode(inodeno_t ino
, snapid_t snap
=CEPH_NOSNAP
) {
776 return have_inode(vinodeno_t(ino
, snap
));
778 CInode
* get_inode(vinodeno_t vino
) {
779 if (vino
.snapid
== CEPH_NOSNAP
) {
780 auto p
= inode_map
.find(vino
.ino
);
781 if (p
!= inode_map
.end())
784 auto p
= snap_inode_map
.find(vino
);
785 if (p
!= snap_inode_map
.end())
790 CInode
* get_inode(inodeno_t ino
, snapid_t s
=CEPH_NOSNAP
) {
791 return get_inode(vinodeno_t(ino
, s
));
794 CDir
* get_dirfrag(dirfrag_t df
) {
795 CInode
*in
= get_inode(df
.ino
);
798 return in
->get_dirfrag(df
.frag
);
800 CDir
* get_dirfrag(inodeno_t ino
, boost::string_view dn
) {
801 CInode
*in
= get_inode(ino
);
804 frag_t fg
= in
->pick_dirfrag(dn
);
805 return in
->get_dirfrag(fg
);
807 CDir
* get_force_dirfrag(dirfrag_t df
, bool replay
) {
808 CInode
*diri
= get_inode(df
.ino
);
811 CDir
*dir
= force_dir_fragment(diri
, df
.frag
, replay
);
813 dir
= diri
->get_dirfrag(df
.frag
);
817 MDSCacheObject
*get_object(MDSCacheObjectInfo
&info
);
822 void add_inode(CInode
*in
);
824 void remove_inode(CInode
*in
);
826 void touch_inode(CInode
*in
) {
827 if (in
->get_parent_dn())
828 touch_dentry(in
->get_projected_parent_dn());
831 void touch_dentry(CDentry
*dn
) {
832 if (dn
->state_test(CDentry::STATE_BOTTOMLRU
)) {
833 bottom_lru
.lru_midtouch(dn
);
838 lru
.lru_midtouch(dn
);
841 void touch_dentry_bottom(CDentry
*dn
) {
842 if (dn
->state_test(CDentry::STATE_BOTTOMLRU
))
844 lru
.lru_bottouch(dn
);
848 void inode_remove_replica(CInode
*in
, mds_rank_t rep
, bool rejoin
,
849 set
<SimpleLock
*>& gather_locks
);
850 void dentry_remove_replica(CDentry
*dn
, mds_rank_t rep
, set
<SimpleLock
*>& gather_locks
);
852 void rename_file(CDentry
*srcdn
, CDentry
*destdn
);
856 void truncate_inode(CInode
*in
, LogSegment
*ls
);
857 void _truncate_inode(CInode
*in
, LogSegment
*ls
);
858 void truncate_inode_finish(CInode
*in
, LogSegment
*ls
);
859 void truncate_inode_logged(CInode
*in
, MutationRef
& mut
);
861 void add_recovered_truncate(CInode
*in
, LogSegment
*ls
);
862 void remove_recovered_truncate(CInode
*in
, LogSegment
*ls
);
863 void start_recovered_truncates();
867 CDir
*get_auth_container(CDir
*in
);
868 CDir
*get_export_container(CDir
*dir
);
869 void find_nested_exports(CDir
*dir
, set
<CDir
*>& s
);
870 void find_nested_exports_under(CDir
*import
, CDir
*dir
, set
<CDir
*>& s
);
874 bool opening_root
, open
;
875 list
<MDSInternalContextBase
*> waiting_for_open
;
879 void create_unlinked_system_inode(CInode
*in
, inodeno_t ino
,
881 CInode
*create_system_inode(inodeno_t ino
, int mode
);
882 CInode
*create_root_inode();
884 void create_empty_hierarchy(MDSGather
*gather
);
885 void create_mydir_hierarchy(MDSGather
*gather
);
887 bool is_open() { return open
; }
888 void wait_for_open(MDSInternalContextBase
*c
) {
889 waiting_for_open
.push_back(c
);
892 void open_root_inode(MDSInternalContextBase
*c
);
894 void open_mydir_inode(MDSInternalContextBase
*c
);
895 void populate_mydir();
897 void _create_system_file(CDir
*dir
, const char *name
, CInode
*in
, MDSInternalContextBase
*fin
);
898 void _create_system_file_finish(MutationRef
& mut
, CDentry
*dn
,
899 version_t dpv
, MDSInternalContextBase
*fin
);
901 void open_foreign_mdsdir(inodeno_t ino
, MDSInternalContextBase
*c
);
902 CDir
*get_stray_dir(CInode
*in
);
903 CDentry
*get_or_create_stray_dentry(CInode
*in
);
905 MDSInternalContextBase
*_get_waiter(MDRequestRef
& mdr
, Message
*req
, MDSInternalContextBase
*fin
);
908 * Find the given dentry (and whether it exists or not), its ancestors,
909 * and get them all into memory and usable on this MDS. This function
910 * makes a best-effort attempt to load everything; if it needs to
911 * go away and do something then it will put the request on a waitlist.
912 * It prefers the mdr, then the req, then the fin. (At least one of these
915 * At least one of the params mdr, req, and fin must be non-null.
917 * @param mdr The MDRequest associated with the path. Can be null.
918 * @param req The Message associated with the path. Can be null.
919 * @param fin The Context associated with the path. Can be null.
920 * @param path The path to traverse to.
921 * @param pdnvec Data return parameter -- on success, contains a
922 * vector of dentries. On failure, is either empty or contains the
923 * full trace of traversable dentries.
924 * @param pin Data return parameter -- if successful, points to the inode
925 * associated with filepath. If unsuccessful, is null.
926 * @param onfail Specifies different lookup failure behaviors. If set to
927 * MDS_TRAVERSE_DISCOVERXLOCK, path_traverse will succeed on null
928 * dentries (instead of returning -ENOENT). If set to
929 * MDS_TRAVERSE_FORWARD, it will forward the request to the auth
930 * MDS if that becomes appropriate (ie, if it doesn't know the contents
931 * of a directory). If set to MDS_TRAVERSE_DISCOVER, it
932 * will attempt to look up the path from a different MDS (and bring them
933 * into its cache as replicas).
935 * @returns 0 on success, 1 on "not done yet", 2 on "forwarding", -errno otherwise.
936 * If it returns 1, the requester associated with this call has been placed
937 * on the appropriate waitlist, and it should unwind itself and back out.
938 * If it returns 2 the request has been forwarded, and again the requester
939 * should unwind itself and back out.
941 int path_traverse(MDRequestRef
& mdr
, Message
*req
, MDSInternalContextBase
*fin
, const filepath
& path
,
942 vector
<CDentry
*> *pdnvec
, CInode
**pin
, int onfail
);
944 CInode
*cache_traverse(const filepath
& path
);
946 void open_remote_dirfrag(CInode
*diri
, frag_t fg
, MDSInternalContextBase
*fin
);
947 CInode
*get_dentry_inode(CDentry
*dn
, MDRequestRef
& mdr
, bool projected
=false);
949 bool parallel_fetch(map
<inodeno_t
,filepath
>& pathmap
, set
<inodeno_t
>& missing
);
950 bool parallel_fetch_traverse_dir(inodeno_t ino
, filepath
& path
,
951 set
<CDir
*>& fetch_queue
, set
<inodeno_t
>& missing
,
952 C_GatherBuilder
&gather_bld
);
954 void open_remote_dentry(CDentry
*dn
, bool projected
, MDSInternalContextBase
*fin
,
955 bool want_xlocked
=false);
956 void _open_remote_dentry_finish(CDentry
*dn
, inodeno_t ino
, MDSInternalContextBase
*fin
,
957 bool want_xlocked
, int r
);
959 void make_trace(vector
<CDentry
*>& trace
, CInode
*in
);
962 struct open_ino_info_t
{
963 vector
<inode_backpointer_t
> ancestors
;
964 set
<mds_rank_t
> checked
;
966 mds_rank_t auth_hint
;
968 bool fetch_backtrace
;
975 list
<MDSInternalContextBase
*> waiters
;
976 open_ino_info_t() : checking(MDS_RANK_NONE
), auth_hint(MDS_RANK_NONE
),
977 check_peers(true), fetch_backtrace(true), discover(false),
978 want_replica(false), want_xlocked(false), tid(0), pool(-1),
981 ceph_tid_t open_ino_last_tid
;
982 map
<inodeno_t
,open_ino_info_t
> opening_inodes
;
984 void _open_ino_backtrace_fetched(inodeno_t ino
, bufferlist
& bl
, int err
);
985 void _open_ino_parent_opened(inodeno_t ino
, int ret
);
986 void _open_ino_traverse_dir(inodeno_t ino
, open_ino_info_t
& info
, int err
);
987 void _open_ino_fetch_dir(inodeno_t ino
, MMDSOpenIno
*m
, CDir
*dir
, bool parent
);
988 int open_ino_traverse_dir(inodeno_t ino
, MMDSOpenIno
*m
,
989 vector
<inode_backpointer_t
>& ancestors
,
990 bool discover
, bool want_xlocked
, mds_rank_t
*hint
);
991 void open_ino_finish(inodeno_t ino
, open_ino_info_t
& info
, int err
);
992 void do_open_ino(inodeno_t ino
, open_ino_info_t
& info
, int err
);
993 void do_open_ino_peer(inodeno_t ino
, open_ino_info_t
& info
);
994 void handle_open_ino(MMDSOpenIno
*m
, int err
=0);
995 void handle_open_ino_reply(MMDSOpenInoReply
*m
);
996 friend class C_IO_MDC_OpenInoBacktraceFetched
;
997 friend struct C_MDC_OpenInoTraverseDir
;
998 friend struct C_MDC_OpenInoParentOpened
;
1001 void kick_open_ino_peers(mds_rank_t who
);
1002 void open_ino(inodeno_t ino
, int64_t pool
, MDSInternalContextBase
*fin
,
1003 bool want_replica
=true, bool want_xlocked
=false);
1005 // -- find_ino_peer --
1006 struct find_ino_peer_info_t
{
1009 MDSInternalContextBase
*fin
;
1011 mds_rank_t checking
;
1012 set
<mds_rank_t
> checked
;
1014 find_ino_peer_info_t() : tid(0), fin(NULL
), hint(MDS_RANK_NONE
), checking(MDS_RANK_NONE
) {}
1017 map
<ceph_tid_t
, find_ino_peer_info_t
> find_ino_peer
;
1018 ceph_tid_t find_ino_peer_last_tid
;
1020 void find_ino_peers(inodeno_t ino
, MDSInternalContextBase
*c
, mds_rank_t hint
=MDS_RANK_NONE
);
1021 void _do_find_ino_peer(find_ino_peer_info_t
& fip
);
1022 void handle_find_ino(MMDSFindIno
*m
);
1023 void handle_find_ino_reply(MMDSFindInoReply
*m
);
1024 void kick_find_ino_peers(mds_rank_t who
);
1028 void snaprealm_create(MDRequestRef
& mdr
, CInode
*in
);
1029 void _snaprealm_create_finish(MDRequestRef
& mdr
, MutationRef
& mut
, CInode
*in
);
1033 void fetch_backtrace(inodeno_t ino
, int64_t pool
, bufferlist
& bl
, Context
*fin
);
1034 uint64_t get_num_strays() const { return stray_manager
.get_num_strays(); }
1037 void scan_stray_dir(dirfrag_t next
=dirfrag_t());
1038 StrayManager stray_manager
;
1039 friend struct C_MDC_RetryScanStray
;
1040 friend class C_IO_MDC_FetchedBacktrace
;
1044 void dispatch(Message
*m
);
1048 void handle_discover(MDiscover
*dis
);
1049 void handle_discover_reply(MDiscoverReply
*m
);
1050 friend class C_MDC_Join
;
1053 void replicate_dir(CDir
*dir
, mds_rank_t to
, bufferlist
& bl
);
1054 void replicate_dentry(CDentry
*dn
, mds_rank_t to
, bufferlist
& bl
);
1055 void replicate_inode(CInode
*in
, mds_rank_t to
, bufferlist
& bl
,
1058 CDir
* add_replica_dir(bufferlist::iterator
& p
, CInode
*diri
, mds_rank_t from
, list
<MDSInternalContextBase
*>& finished
);
1059 CDentry
*add_replica_dentry(bufferlist::iterator
& p
, CDir
*dir
, list
<MDSInternalContextBase
*>& finished
);
1060 CInode
*add_replica_inode(bufferlist::iterator
& p
, CDentry
*dn
, list
<MDSInternalContextBase
*>& finished
);
1062 void replicate_stray(CDentry
*straydn
, mds_rank_t who
, bufferlist
& bl
);
1063 CDentry
*add_replica_stray(bufferlist
&bl
, mds_rank_t from
);
1067 void send_dentry_link(CDentry
*dn
, MDRequestRef
& mdr
);
1068 void send_dentry_unlink(CDentry
*dn
, CDentry
*straydn
, MDRequestRef
& mdr
);
1070 void handle_dentry_link(MDentryLink
*m
);
1071 void handle_dentry_unlink(MDentryUnlink
*m
);
1074 // -- fragmenting --
1080 list
<MDSInternalContextBase
*> waiters
;
1081 list
<frag_t
> old_frags
;
1082 bufferlist rollback
;
1083 ufragment() : bits(0), committed(false), ls(NULL
) {}
1085 map
<dirfrag_t
, ufragment
> uncommitted_fragments
;
1087 struct fragment_info_t
{
1090 list
<CDir
*> resultfrags
;
1092 // for deadlock detection
1094 utime_t last_cum_auth_pins_change
;
1095 int last_cum_auth_pins
;
1096 int num_remote_waiters
; // number of remote authpin waiters
1097 fragment_info_t() : bits(0), all_frozen(false), last_cum_auth_pins(0), num_remote_waiters(0) {}
1098 bool is_fragmenting() { return !resultfrags
.empty(); }
1100 map
<dirfrag_t
,fragment_info_t
> fragments
;
1102 void adjust_dir_fragments(CInode
*diri
, frag_t basefrag
, int bits
,
1103 list
<CDir
*>& frags
, list
<MDSInternalContextBase
*>& waiters
, bool replay
);
1104 void adjust_dir_fragments(CInode
*diri
,
1105 list
<CDir
*>& srcfrags
,
1106 frag_t basefrag
, int bits
,
1107 list
<CDir
*>& resultfrags
,
1108 list
<MDSInternalContextBase
*>& waiters
,
1110 CDir
*force_dir_fragment(CInode
*diri
, frag_t fg
, bool replay
=true);
1111 void get_force_dirfrag_bound_set(vector
<dirfrag_t
>& dfs
, set
<CDir
*>& bounds
);
1113 bool can_fragment(CInode
*diri
, list
<CDir
*>& dirs
);
1114 void fragment_freeze_dirs(list
<CDir
*>& dirs
);
1115 void fragment_mark_and_complete(MDRequestRef
& mdr
);
1116 void fragment_frozen(MDRequestRef
& mdr
, int r
);
1117 void fragment_unmark_unfreeze_dirs(list
<CDir
*>& dirs
);
1118 void dispatch_fragment_dir(MDRequestRef
& mdr
);
1119 void _fragment_logged(MDRequestRef
& mdr
);
1120 void _fragment_stored(MDRequestRef
& mdr
);
1121 void _fragment_committed(dirfrag_t f
, list
<CDir
*>& resultfrags
);
1122 void _fragment_finish(dirfrag_t f
, list
<CDir
*>& resultfrags
);
1124 friend class EFragment
;
1125 friend class C_MDC_FragmentFrozen
;
1126 friend class C_MDC_FragmentMarking
;
1127 friend class C_MDC_FragmentPrep
;
1128 friend class C_MDC_FragmentStore
;
1129 friend class C_MDC_FragmentCommit
;
1130 friend class C_IO_MDC_FragmentFinish
;
1132 void handle_fragment_notify(MMDSFragmentNotify
*m
);
1134 void add_uncommitted_fragment(dirfrag_t basedirfrag
, int bits
, list
<frag_t
>& old_frag
,
1135 LogSegment
*ls
, bufferlist
*rollback
=NULL
);
1136 void finish_uncommitted_fragment(dirfrag_t basedirfrag
, int op
);
1137 void rollback_uncommitted_fragment(dirfrag_t basedirfrag
, list
<frag_t
>& old_frags
);
1139 void wait_for_uncommitted_fragment(dirfrag_t dirfrag
, MDSInternalContextBase
*c
) {
1140 assert(uncommitted_fragments
.count(dirfrag
));
1141 uncommitted_fragments
[dirfrag
].waiters
.push_back(c
);
1143 void split_dir(CDir
*dir
, int byn
);
1144 void merge_dir(CInode
*diri
, frag_t fg
);
1145 void rollback_uncommitted_fragments();
1147 void find_stale_fragment_freeze();
1148 void fragment_freeze_inc_num_waiters(CDir
*dir
);
1149 bool fragment_are_all_frozen(CDir
*dir
);
1150 int get_num_fragmenting_dirs() { return fragments
.size(); }
1153 //int send_inode_updates(CInode *in);
1154 //void handle_inode_update(MInodeUpdate *m);
1156 int send_dir_updates(CDir
*in
, bool bcast
=false);
1157 void handle_dir_update(MDirUpdate
*m
);
1159 // -- cache expiration --
1160 void handle_cache_expire(MCacheExpire
*m
);
1161 void process_delayed_expire(CDir
*dir
);
1162 void discard_delayed_expire(CDir
*dir
);
1165 int dump_cache(boost::string_view fn
, Formatter
*f
,
1166 boost::string_view dump_root
= "",
1169 int dump_cache() { return dump_cache(NULL
, NULL
); }
1170 int dump_cache(boost::string_view filename
);
1171 int dump_cache(Formatter
*f
);
1172 int dump_cache(boost::string_view dump_root
, int depth
, Formatter
*f
);
1174 int cache_status(Formatter
*f
);
1176 void dump_resolve_status(Formatter
*f
) const;
1177 void dump_rejoin_status(Formatter
*f
) const;
1182 void show_subtrees(int dbl
=10);
1184 CInode
*hack_pick_random_inode() {
1185 assert(!inode_map
.empty());
1186 int n
= rand() % inode_map
.size();
1187 auto p
= inode_map
.begin();
1193 void flush_dentry_work(MDRequestRef
& mdr
);
1195 * Resolve path to a dentry and pass it onto the ScrubStack.
1197 * TODO: return enough information to the original mdr formatter
1198 * and completion that they can subsequeuntly check the progress of
1199 * this scrub (we won't block them on a whole scrub as it can take a very
1202 void enqueue_scrub_work(MDRequestRef
& mdr
);
1203 void repair_inode_stats_work(MDRequestRef
& mdr
);
1204 void repair_dirfrag_stats_work(MDRequestRef
& mdr
);
1205 friend class C_MDC_RepairDirfragStats
;
1207 void flush_dentry(boost::string_view path
, Context
*fin
);
1209 * Create and start an OP_ENQUEUE_SCRUB
1211 void enqueue_scrub(boost::string_view path
, boost::string_view tag
,
1212 bool force
, bool recursive
, bool repair
,
1213 Formatter
*f
, Context
*fin
);
1214 void repair_inode_stats(CInode
*diri
);
1215 void repair_dirfrag_stats(CDir
*dir
);
1218 /* Because exports may fail, this set lets us keep track of inodes that need exporting. */
1219 std::set
<CInode
*> export_pin_queue
;
1222 class C_MDS_RetryRequest
: public MDSInternalContext
{
1226 C_MDS_RetryRequest(MDCache
*c
, MDRequestRef
& r
);
1227 void finish(int r
) override
;