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.
23 #include <boost/utility/string_view.hpp>
25 #include "common/config.h"
26 #include "include/counter.h"
27 #include "include/elist.h"
28 #include "include/types.h"
29 #include "include/lru.h"
30 #include "include/compact_set.h"
32 #include "MDSCacheObject.h"
36 #include "SimpleLock.h"
37 #include "ScatterLock.h"
38 #include "LocalLock.h"
39 #include "Capability.h"
40 #include "SnapRealm.h"
43 #define dout_context g_ceph_context
55 struct ObjectOperation
;
59 ostream
& operator<<(ostream
& out
, const CInode
& in
);
61 struct cinode_lock_info_t
{
66 extern cinode_lock_info_t cinode_lock_info
[];
67 extern int num_cinode_locks
;
71 * Base class for CInode, containing the backing store data and
72 * serialization methods. This exists so that we can read and
73 * handle CInodes from the backing store without hitting all
74 * the business logic in CInode proper.
76 class InodeStoreBase
{
78 typedef inode_t
<mempool::mds_co::pool_allocator
> mempool_inode
;
79 typedef old_inode_t
<mempool::mds_co::pool_allocator
> mempool_old_inode
;
80 typedef mempool::mds_co::compact_map
<snapid_t
, mempool_old_inode
> mempool_old_inode_map
;
81 typedef xattr_map
<mempool::mds_co::pool_allocator
> mempool_xattr_map
; // FIXME bufferptr not in mempool
83 mempool_inode inode
; // the inode itself
84 mempool::mds_co::string symlink
; // symlink dest, if symlink
85 mempool_xattr_map xattrs
;
86 fragtree_t dirfragtree
; // dir frag tree, if any. always consistent with our dirfrag map.
87 mempool_old_inode_map old_inodes
; // key = last, value.first = first
88 snapid_t oldest_snap
= CEPH_NOSNAP
;
89 damage_flags_t damage_flags
= 0;
94 bool is_file() const { return inode
.is_file(); }
95 bool is_symlink() const { return inode
.is_symlink(); }
96 bool is_dir() const { return inode
.is_dir(); }
97 static object_t
get_object_name(inodeno_t ino
, frag_t fg
, const char *suffix
);
99 /* Full serialization for use in ".inode" root inode objects */
100 void encode(bufferlist
&bl
, uint64_t features
, const bufferlist
*snap_blob
=NULL
) const;
101 void decode(bufferlist::iterator
&bl
, bufferlist
& snap_blob
);
103 /* Serialization without ENCODE_START/FINISH blocks for use embedded in dentry */
104 void encode_bare(bufferlist
&bl
, uint64_t features
, const bufferlist
*snap_blob
=NULL
) const;
105 void decode_bare(bufferlist::iterator
&bl
, bufferlist
&snap_blob
, __u8 struct_v
=5);
107 /* For test/debug output */
108 void dump(Formatter
*f
) const;
110 /* For use by offline tools */
111 __u32
hash_dentry_name(boost::string_view dn
);
112 frag_t
pick_dirfrag(boost::string_view dn
);
115 class InodeStore
: public InodeStoreBase
{
117 // FIXME bufferlist not part of mempool
118 bufferlist snap_blob
; // Encoded copy of SnapRealm, because we can't
119 // rehydrate it without full MDCache
120 void encode(bufferlist
&bl
, uint64_t features
) const {
121 InodeStoreBase::encode(bl
, features
, &snap_blob
);
123 void decode(bufferlist::iterator
&bl
) {
124 InodeStoreBase::decode(bl
, snap_blob
);
126 void encode_bare(bufferlist
&bl
, uint64_t features
) const {
127 InodeStoreBase::encode_bare(bl
, features
, &snap_blob
);
129 void decode_bare(bufferlist::iterator
&bl
) {
130 InodeStoreBase::decode_bare(bl
, snap_blob
);
133 static void generate_test_instances(std::list
<InodeStore
*>& ls
);
135 WRITE_CLASS_ENCODER_FEATURES(InodeStore
)
137 // cached inode wrapper
138 class CInode
: public MDSCacheObject
, public InodeStoreBase
, public Counter
<CInode
> {
140 MEMPOOL_CLASS_HELPERS();
142 static const int PIN_DIRFRAG
= -1;
143 static const int PIN_CAPS
= 2; // client caps
144 static const int PIN_IMPORTING
= -4; // importing
145 static const int PIN_OPENINGDIR
= 7;
146 static const int PIN_REMOTEPARENT
= 8;
147 static const int PIN_BATCHOPENJOURNAL
= 9;
148 static const int PIN_SCATTERED
= 10;
149 static const int PIN_STICKYDIRS
= 11;
150 //static const int PIN_PURGING = -12;
151 static const int PIN_FREEZING
= 13;
152 static const int PIN_FROZEN
= 14;
153 static const int PIN_IMPORTINGCAPS
= -15;
154 static const int PIN_PASTSNAPPARENT
= -16;
155 static const int PIN_OPENINGSNAPPARENTS
= 17;
156 static const int PIN_TRUNCATING
= 18;
157 static const int PIN_STRAY
= 19; // we pin our stray inode while active
158 static const int PIN_NEEDSNAPFLUSH
= 20;
159 static const int PIN_DIRTYRSTAT
= 21;
160 static const int PIN_EXPORTINGCAPS
= 22;
161 static const int PIN_DIRTYPARENT
= 23;
162 static const int PIN_DIRWAITER
= 24;
163 static const int PIN_SCRUBQUEUE
= 25;
165 const char *pin_name(int p
) const override
{
167 case PIN_DIRFRAG
: return "dirfrag";
168 case PIN_CAPS
: return "caps";
169 case PIN_IMPORTING
: return "importing";
170 case PIN_OPENINGDIR
: return "openingdir";
171 case PIN_REMOTEPARENT
: return "remoteparent";
172 case PIN_BATCHOPENJOURNAL
: return "batchopenjournal";
173 case PIN_SCATTERED
: return "scattered";
174 case PIN_STICKYDIRS
: return "stickydirs";
175 //case PIN_PURGING: return "purging";
176 case PIN_FREEZING
: return "freezing";
177 case PIN_FROZEN
: return "frozen";
178 case PIN_IMPORTINGCAPS
: return "importingcaps";
179 case PIN_EXPORTINGCAPS
: return "exportingcaps";
180 case PIN_PASTSNAPPARENT
: return "pastsnapparent";
181 case PIN_OPENINGSNAPPARENTS
: return "openingsnapparents";
182 case PIN_TRUNCATING
: return "truncating";
183 case PIN_STRAY
: return "stray";
184 case PIN_NEEDSNAPFLUSH
: return "needsnapflush";
185 case PIN_DIRTYRSTAT
: return "dirtyrstat";
186 case PIN_DIRTYPARENT
: return "dirtyparent";
187 case PIN_DIRWAITER
: return "dirwaiter";
188 case PIN_SCRUBQUEUE
: return "scrubqueue";
189 default: return generic_pin_name(p
);
194 static const int STATE_EXPORTING
= (1<<2); // on nonauth bystander.
195 static const int STATE_OPENINGDIR
= (1<<5);
196 static const int STATE_FREEZING
= (1<<7);
197 static const int STATE_FROZEN
= (1<<8);
198 static const int STATE_AMBIGUOUSAUTH
= (1<<9);
199 static const int STATE_EXPORTINGCAPS
= (1<<10);
200 static const int STATE_NEEDSRECOVER
= (1<<11);
201 static const int STATE_RECOVERING
= (1<<12);
202 static const int STATE_PURGING
= (1<<13);
203 static const int STATE_DIRTYPARENT
= (1<<14);
204 static const int STATE_DIRTYRSTAT
= (1<<15);
205 static const int STATE_STRAYPINNED
= (1<<16);
206 static const int STATE_FROZENAUTHPIN
= (1<<17);
207 static const int STATE_DIRTYPOOL
= (1<<18);
208 static const int STATE_REPAIRSTATS
= (1<<19);
209 static const int STATE_MISSINGOBJS
= (1<<20);
210 static const int STATE_EVALSTALECAPS
= (1<<21);
211 static const int STATE_QUEUEDEXPORTPIN
= (1<<22);
212 // orphan inode needs notification of releasing reference
213 static const int STATE_ORPHAN
= STATE_NOTIFYREF
;
215 static const int MASK_STATE_EXPORTED
=
216 (STATE_DIRTY
|STATE_NEEDSRECOVER
|STATE_DIRTYPARENT
|STATE_DIRTYPOOL
);
217 static const int MASK_STATE_EXPORT_KEPT
=
218 (STATE_FROZEN
|STATE_AMBIGUOUSAUTH
|STATE_EXPORTINGCAPS
|STATE_QUEUEDEXPORTPIN
);
221 static const uint64_t WAIT_DIR
= (1<<0);
222 static const uint64_t WAIT_FROZEN
= (1<<1);
223 static const uint64_t WAIT_TRUNC
= (1<<2);
224 static const uint64_t WAIT_FLOCK
= (1<<3);
226 static const uint64_t WAIT_ANY_MASK
= (uint64_t)(-1);
229 static const unsigned EXPORT_NONCE
= 1; // nonce given to replicas created by export
231 ostream
& print_db_line_prefix(ostream
& out
) override
;
236 SnapRealm
*snaprealm
= nullptr;
237 SnapRealm
*containing_realm
= nullptr;
238 snapid_t first
, last
;
239 mempool::mds_co::compact_set
<snapid_t
> dirty_old_rstats
;
241 class scrub_stamp_info_t
{
243 /// version we started our latest scrub (whether in-progress or finished)
244 version_t scrub_start_version
= 0;
245 /// time we started our latest scrub (whether in-progress or finished)
246 utime_t scrub_start_stamp
;
247 /// version we started our most recent finished scrub
248 version_t last_scrub_version
= 0;
249 /// time we started our most recent finished scrub
250 utime_t last_scrub_stamp
;
251 scrub_stamp_info_t() {}
253 scrub_start_version
= last_scrub_version
= 0;
254 scrub_start_stamp
= last_scrub_stamp
= utime_t();
258 class scrub_info_t
: public scrub_stamp_info_t
{
260 CDentry
*scrub_parent
= nullptr;
261 MDSInternalContextBase
*on_finish
= nullptr;
263 bool last_scrub_dirty
= false; /// are our stamps dirty with respect to disk state?
264 bool scrub_in_progress
= false; /// are we currently scrubbing?
265 bool children_scrubbed
= false;
267 /// my own (temporary) stamps and versions for each dirfrag we have
268 std::map
<frag_t
, scrub_stamp_info_t
> dirfrag_stamps
; // XXX not part of mempool
270 ScrubHeaderRef header
;
275 const scrub_info_t
*scrub_info() const{
281 ScrubHeaderRef
get_scrub_header() {
282 if (scrub_infop
== nullptr) {
285 return scrub_infop
->header
;
289 bool scrub_is_in_progress() const {
290 return (scrub_infop
&& scrub_infop
->scrub_in_progress
);
293 * Start scrubbing on this inode. That could be very short if it's
294 * a file, or take a long time if we're recursively scrubbing a directory.
295 * @pre It is not currently scrubbing
296 * @post it has set up internal scrubbing state
297 * @param scrub_version What version are we scrubbing at (usually, parent
298 * directory's get_projected_version())
300 void scrub_initialize(CDentry
*scrub_parent
,
301 ScrubHeaderRef
& header
,
302 MDSInternalContextBase
*f
);
304 * Get the next dirfrag to scrub. Gives you a frag_t in output param which
305 * you must convert to a CDir (and possibly load off disk).
306 * @param dir A pointer to frag_t, will be filled in with the next dirfrag to
307 * scrub if there is one.
308 * @returns 0 on success, you should scrub the passed-out frag_t right now;
309 * ENOENT: There are no remaining dirfrags to scrub
310 * <0 There was some other error (It will return -ENOTDIR if not a directory)
312 int scrub_dirfrag_next(frag_t
* out_dirfrag
);
314 * Get the currently scrubbing dirfrags. When returned, the
315 * passed-in list will be filled in with all frag_ts which have
316 * been returned from scrub_dirfrag_next but not sent back
317 * via scrub_dirfrag_finished.
319 void scrub_dirfrags_scrubbing(list
<frag_t
> *out_dirfrags
);
321 * Report to the CInode that a dirfrag it owns has been scrubbed. Call
322 * this for every frag_t returned from scrub_dirfrag_next().
323 * @param dirfrag The frag_t that was scrubbed
325 void scrub_dirfrag_finished(frag_t dirfrag
);
327 * Call this once the scrub has been completed, whether it's a full
328 * recursive scrub on a directory or simply the data on a file (or
329 * anything in between).
330 * @param c An out param which is filled in with a Context* that must
333 void scrub_finished(MDSInternalContextBase
**c
);
335 * Report to the CInode that alldirfrags it owns have been scrubbed.
337 void scrub_children_finished() {
338 scrub_infop
->children_scrubbed
= true;
340 void scrub_set_finisher(MDSInternalContextBase
*c
) {
341 assert(!scrub_infop
->on_finish
);
342 scrub_infop
->on_finish
= c
;
347 * Create a scrub_info_t struct for the scrub_infop poitner.
349 void scrub_info_create() const;
351 * Delete the scrub_info_t struct if it's not got any useful data
353 void scrub_maybe_delete_info();
356 bool is_multiversion() const {
357 return snaprealm
|| // other snaprealms will link to me
358 inode
.is_dir() || // links to me in other snaps
359 inode
.nlink
> 1 || // there are remote links, possibly snapped, that will need to find me
360 !old_inodes
.empty(); // once multiversion, always multiversion. until old_inodes gets cleaned out.
362 snapid_t
get_oldest_snap();
364 uint64_t last_journaled
= 0; // log offset for the last time i was journaled
365 //loff_t last_open_journaled; // log offset for the last journaled EOpen
366 utime_t last_dirstat_prop
;
369 // list item node for when we have unpropagated rstat data
370 elist
<CInode
*>::item dirty_rstat_item
;
372 bool is_dirty_rstat() {
373 return state_test(STATE_DIRTYRSTAT
);
375 void mark_dirty_rstat();
376 void clear_dirty_rstat();
378 //bool hack_accessed = false;
379 //utime_t hack_load_stamp;
382 * Projection methods, used to store inode changes until they have been journaled,
383 * at which point they are popped.
385 * project_inode as needed. If you're changing xattrs or sr_t, then pass true
386 * as needed then change the xattrs/snapnode member as needed. (Dirty
387 * exception: project_past_snaprealm_parent allows you to project the
388 * snapnode after doing project_inode (i.e. you don't need to pass
391 * Then, journal. Once journaling is done, pop_and_dirty_projected_inode.
392 * This function will take care of the inode itself, the xattrs, and the snaprealm.
395 class projected_inode
{
398 std::unique_ptr
<mempool_xattr_map
> xattrs
;
399 std::unique_ptr
<sr_t
> snapnode
;
401 projected_inode() = delete;
402 projected_inode(const mempool_inode
&in
) : inode(in
) {}
406 mempool::mds_co::list
<projected_inode
> projected_nodes
; // projected values (only defined while dirty)
407 size_t num_projected_xattrs
= 0;
408 size_t num_projected_srnodes
= 0;
410 sr_t
&project_snaprealm(projected_inode
&pi
);
412 CInode::projected_inode
&project_inode(bool xattr
= false, bool snap
= false);
413 void pop_and_dirty_projected_inode(LogSegment
*ls
);
415 projected_inode
*get_projected_node() {
416 if (projected_nodes
.empty())
419 return &projected_nodes
.back();
422 version_t
get_projected_version() const {
423 if (projected_nodes
.empty())
424 return inode
.version
;
426 return projected_nodes
.back().inode
.version
;
428 bool is_projected() const {
429 return !projected_nodes
.empty();
432 const mempool_inode
*get_projected_inode() const {
433 if (projected_nodes
.empty())
436 return &projected_nodes
.back().inode
;
438 mempool_inode
*get_projected_inode() {
439 if (projected_nodes
.empty())
442 return &projected_nodes
.back().inode
;
444 mempool_inode
*get_previous_projected_inode() {
445 assert(!projected_nodes
.empty());
446 auto it
= projected_nodes
.rbegin();
448 if (it
!= projected_nodes
.rend())
454 mempool_xattr_map
*get_projected_xattrs() {
455 if (num_projected_xattrs
> 0) {
456 for (auto it
= projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
458 return it
->xattrs
.get();
462 mempool_xattr_map
*get_previous_projected_xattrs() {
463 if (num_projected_xattrs
> 0) {
464 for (auto it
= ++projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
466 return it
->xattrs
.get();
471 const sr_t
*get_projected_srnode() const {
472 if (num_projected_srnodes
> 0) {
473 for (auto it
= projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
475 return it
->snapnode
.get();
478 return &snaprealm
->srnode
;
482 sr_t
*get_projected_srnode() {
483 if (num_projected_srnodes
> 0) {
484 for (auto it
= projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
486 return it
->snapnode
.get();
489 return &snaprealm
->srnode
;
493 void project_past_snaprealm_parent(SnapRealm
*newparent
);
496 void pop_projected_snaprealm(sr_t
*next_snaprealm
);
499 mempool_old_inode
& cow_old_inode(snapid_t follows
, bool cow_head
);
500 void split_old_inode(snapid_t snap
);
501 mempool_old_inode
*pick_old_inode(snapid_t last
);
502 void pre_cow_old_inode();
503 void purge_stale_snap_data(const std::set
<snapid_t
>& snaps
);
505 // -- cache infrastructure --
507 mempool::mds_co::compact_map
<frag_t
,CDir
*> dirfrags
; // cached dir fragments under this Inode
509 //for the purpose of quickly determining whether there's a subtree root or exporting dir
510 int num_subtree_roots
= 0;
511 int num_exporting_dirs
= 0;
513 int stickydir_ref
= 0;
514 scrub_info_t
*scrub_infop
= nullptr;
517 bool has_dirfrags() { return !dirfrags
.empty(); }
518 CDir
* get_dirfrag(frag_t fg
) {
519 if (dirfrags
.count(fg
)) {
520 //assert(g_conf->debug_mds < 2 || dirfragtree.is_leaf(fg)); // performance hack FIXME
525 bool get_dirfrags_under(frag_t fg
, std::list
<CDir
*>& ls
);
526 CDir
* get_approx_dirfrag(frag_t fg
);
528 template<typename Container
>
529 void get_dirfrags(Container
& ls
) const {
531 for (const auto &p
: dirfrags
)
532 ls
.push_back(p
.second
);
534 template<typename Container
>
535 void get_nested_dirfrags(Container
& ls
) const {
536 // dirfrags in same subtree
537 for (const auto &p
: dirfrags
) {
538 typename
Container::value_type dir
= p
.second
;
539 if (!dir
->is_subtree_root())
543 template<typename Container
>
544 void get_subtree_dirfrags(Container
& ls
) {
545 // dirfrags that are roots of new subtrees
546 for (const auto &p
: dirfrags
) {
547 typename
Container::value_type dir
= p
.second
;
548 if (dir
->is_subtree_root())
553 CDir
*get_or_open_dirfrag(MDCache
*mdcache
, frag_t fg
);
554 CDir
*add_dirfrag(CDir
*dir
);
555 void close_dirfrag(frag_t fg
);
556 void close_dirfrags();
557 bool has_subtree_root_dirfrag(int auth
=-1);
558 bool has_subtree_or_exporting_dirfrag();
560 void force_dirfrags();
561 void verify_dirfrags();
563 void get_stickydirs();
564 void put_stickydirs();
567 // parent dentries in cache
568 CDentry
*parent
= nullptr; // primary link
569 mempool::mds_co::compact_set
<CDentry
*> remote_parents
; // if hard linked
571 mempool::mds_co::list
<CDentry
*> projected_parent
; // for in-progress rename, (un)link, etc.
573 mds_authority_t inode_auth
= CDIR_AUTH_DEFAULT
;
575 // -- distributed state --
578 using cap_map
= mempool::mds_co::map
<client_t
, Capability
*>;
579 cap_map client_caps
; // client -> caps
580 mempool::mds_co::compact_map
<int32_t, int32_t> mds_caps_wanted
; // [auth] mds -> caps wanted
581 int replica_caps_wanted
= 0; // [replica] what i've requested from auth
584 mempool::mds_co::compact_map
<int, mempool::mds_co::set
<client_t
> > client_snap_caps
; // [auth] [snap] dirty metadata we still need from the head
585 mempool::mds_co::compact_map
<snapid_t
, mempool::mds_co::set
<client_t
> > client_need_snapflush
;
587 void add_need_snapflush(CInode
*snapin
, snapid_t snapid
, client_t client
);
588 void remove_need_snapflush(CInode
*snapin
, snapid_t snapid
, client_t client
);
589 bool split_need_snapflush(CInode
*cowin
, CInode
*in
);
593 ceph_lock_state_t
*fcntl_locks
= nullptr;
594 ceph_lock_state_t
*flock_locks
= nullptr;
596 ceph_lock_state_t
*get_fcntl_lock_state() {
598 fcntl_locks
= new ceph_lock_state_t(g_ceph_context
, CEPH_LOCK_FCNTL
);
601 void clear_fcntl_lock_state() {
605 ceph_lock_state_t
*get_flock_lock_state() {
607 flock_locks
= new ceph_lock_state_t(g_ceph_context
, CEPH_LOCK_FLOCK
);
610 void clear_flock_lock_state() {
614 void clear_file_locks() {
615 clear_fcntl_lock_state();
616 clear_flock_lock_state();
618 void _encode_file_locks(bufferlist
& bl
) const {
619 bool has_fcntl_locks
= fcntl_locks
&& !fcntl_locks
->empty();
620 ::encode(has_fcntl_locks
, bl
);
622 ::encode(*fcntl_locks
, bl
);
623 bool has_flock_locks
= flock_locks
&& !flock_locks
->empty();
624 ::encode(has_flock_locks
, bl
);
626 ::encode(*flock_locks
, bl
);
628 void _decode_file_locks(bufferlist::iterator
& p
) {
629 bool has_fcntl_locks
;
630 ::decode(has_fcntl_locks
, p
);
632 ::decode(*get_fcntl_lock_state(), p
);
634 clear_fcntl_lock_state();
635 bool has_flock_locks
;
636 ::decode(has_flock_locks
, p
);
638 ::decode(*get_flock_lock_state(), p
);
640 clear_flock_lock_state();
643 // LogSegment lists i (may) belong to
645 elist
<CInode
*>::item item_dirty
;
646 elist
<CInode
*>::item item_caps
;
647 elist
<CInode
*>::item item_open_file
;
648 elist
<CInode
*>::item item_dirty_parent
;
649 elist
<CInode
*>::item item_dirty_dirfrag_dir
;
650 elist
<CInode
*>::item item_dirty_dirfrag_nest
;
651 elist
<CInode
*>::item item_dirty_dirfrag_dirfragtree
;
652 elist
<CInode
*>::item item_scrub
;
654 // also update RecoveryQueue::RecoveryQueue() if you change this
655 elist
<CInode
*>::item
& item_recover_queue
= item_dirty_dirfrag_dir
;
656 elist
<CInode
*>::item
& item_recover_queue_front
= item_dirty_dirfrag_nest
;
659 int auth_pin_freeze_allowance
= 0;
661 inode_load_vec_t pop
;
662 elist
<CInode
*>::item item_pop_lru
;
667 friend class Migrator
;
668 friend class MDCache
;
669 friend class StrayManager
;
671 friend class CInodeExport
;
673 // ---------------------------
675 CInode(MDCache
*c
, bool auth
=true, snapid_t f
=2, snapid_t l
=CEPH_NOSNAP
) :
680 item_open_file(this),
681 item_dirty_parent(this),
682 item_dirty_dirfrag_dir(this),
683 item_dirty_dirfrag_nest(this),
684 item_dirty_dirfrag_dirfragtree(this),
685 pop(ceph_clock_now()),
686 versionlock(this, &versionlock_type
),
687 authlock(this, &authlock_type
),
688 linklock(this, &linklock_type
),
689 dirfragtreelock(this, &dirfragtreelock_type
),
690 filelock(this, &filelock_type
),
691 xattrlock(this, &xattrlock_type
),
692 snaplock(this, &snaplock_type
),
693 nestlock(this, &nestlock_type
),
694 flocklock(this, &flocklock_type
),
695 policylock(this, &policylock_type
)
697 if (auth
) state_set(STATE_AUTH
);
703 assert(num_projected_xattrs
== 0);
704 assert(num_projected_srnodes
== 0);
705 assert(num_subtree_roots
== 0);
706 assert(num_exporting_dirs
== 0);
711 bool is_root() const { return inode
.ino
== MDS_INO_ROOT
; }
712 bool is_stray() const { return MDS_INO_IS_STRAY(inode
.ino
); }
713 mds_rank_t
get_stray_owner() const {
714 return (mds_rank_t
)MDS_INO_STRAY_OWNER(inode
.ino
);
716 bool is_mdsdir() const { return MDS_INO_IS_MDSDIR(inode
.ino
); }
717 bool is_base() const { return is_root() || is_mdsdir(); }
718 bool is_system() const { return inode
.ino
< MDS_INO_SYSTEM_BASE
; }
719 bool is_normal() const { return !(is_base() || is_system() || is_stray()); }
721 bool is_head() const { return last
== CEPH_NOSNAP
; }
723 // note: this overloads MDSCacheObject
724 bool is_ambiguous_auth() const {
725 return state_test(STATE_AMBIGUOUSAUTH
) ||
726 MDSCacheObject::is_ambiguous_auth();
728 void set_ambiguous_auth() {
729 state_set(STATE_AMBIGUOUSAUTH
);
731 void clear_ambiguous_auth(std::list
<MDSInternalContextBase
*>& finished
);
732 void clear_ambiguous_auth();
734 inodeno_t
ino() const { return inode
.ino
; }
735 vinodeno_t
vino() const { return vinodeno_t(inode
.ino
, last
); }
736 int d_type() const { return IFTODT(inode
.mode
); }
738 mempool_inode
& get_inode() { return inode
; }
739 CDentry
* get_parent_dn() { return parent
; }
740 const CDentry
* get_parent_dn() const { return parent
; }
741 const CDentry
* get_projected_parent_dn() const { return !projected_parent
.empty() ? projected_parent
.back() : parent
; }
742 CDentry
* get_projected_parent_dn() { return !projected_parent
.empty() ? projected_parent
.back() : parent
; }
743 CDir
*get_parent_dir();
744 const CDir
*get_projected_parent_dir() const;
745 CDir
*get_projected_parent_dir();
746 CInode
*get_parent_inode();
748 bool is_lt(const MDSCacheObject
*r
) const override
{
749 const CInode
*o
= static_cast<const CInode
*>(r
);
750 return ino() < o
->ino() ||
751 (ino() == o
->ino() && last
< o
->last
);
755 bool is_projected_ancestor_of(CInode
*other
);
757 void make_path_string(std::string
& s
, bool projected
=false, const CDentry
*use_parent
=NULL
) const;
758 void make_path(filepath
& s
, bool projected
=false) const;
759 void name_stray_dentry(std::string
& dname
);
762 version_t
get_version() const { return inode
.version
; }
764 version_t
pre_dirty();
765 void _mark_dirty(LogSegment
*ls
);
766 void mark_dirty(version_t projected_dirv
, LogSegment
*ls
);
769 void store(MDSInternalContextBase
*fin
);
770 void _stored(int r
, version_t cv
, Context
*fin
);
772 * Flush a CInode to disk. This includes the backtrace, the parent
773 * directory's link, and the Inode object itself (if a base directory).
774 * @pre is_auth() on both the inode and its containing directory
775 * @pre can_auth_pin()
776 * @param fin The Context to call when the flush is completed.
778 void flush(MDSInternalContextBase
*fin
);
779 void fetch(MDSInternalContextBase
*fin
);
780 void _fetched(bufferlist
& bl
, bufferlist
& bl2
, Context
*fin
);
783 void build_backtrace(int64_t pool
, inode_backtrace_t
& bt
);
784 void store_backtrace(MDSInternalContextBase
*fin
, int op_prio
=-1);
785 void _stored_backtrace(int r
, version_t v
, Context
*fin
);
786 void fetch_backtrace(Context
*fin
, bufferlist
*backtrace
);
789 * Return the pool ID where we currently write backtraces for
790 * this inode (in addition to inode.old_pools)
792 * @returns a pool ID >=0
794 int64_t get_backtrace_pool() const;
796 void mark_dirty_parent(LogSegment
*ls
, bool dirty_pool
=false);
797 void clear_dirty_parent();
798 void verify_diri_backtrace(bufferlist
&bl
, int err
);
799 bool is_dirty_parent() { return state_test(STATE_DIRTYPARENT
); }
800 bool is_dirty_pool() { return state_test(STATE_DIRTYPOOL
); }
802 void encode_snap_blob(bufferlist
&bl
);
803 void decode_snap_blob(bufferlist
&bl
);
804 void encode_store(bufferlist
& bl
, uint64_t features
);
805 void decode_store(bufferlist::iterator
& bl
);
807 void encode_replica(mds_rank_t rep
, bufferlist
& bl
, uint64_t features
, bool need_recover
) {
811 if (!is_replicated())
812 replicate_relax_locks();
814 __u32 nonce
= add_replica(rep
);
817 _encode_base(bl
, features
);
818 _encode_locks_state_for_replica(bl
, need_recover
);
820 void decode_replica(bufferlist::iterator
& p
, bool is_new
) {
823 replica_nonce
= nonce
;
826 _decode_locks_state(p
, is_new
);
831 mempool::mds_co::compact_map
<frag_t
, std::list
<MDSInternalContextBase
*> > waiting_on_dir
;
833 void add_dir_waiter(frag_t fg
, MDSInternalContextBase
*c
);
834 void take_dir_waiting(frag_t fg
, std::list
<MDSInternalContextBase
*>& ls
);
835 bool is_waiting_for_dir(frag_t fg
) {
836 return waiting_on_dir
.count(fg
);
838 void add_waiter(uint64_t tag
, MDSInternalContextBase
*c
) override
;
839 void take_waiting(uint64_t tag
, std::list
<MDSInternalContextBase
*>& ls
) override
;
841 // -- encode/decode helpers --
842 void _encode_base(bufferlist
& bl
, uint64_t features
);
843 void _decode_base(bufferlist::iterator
& p
);
844 void _encode_locks_full(bufferlist
& bl
);
845 void _decode_locks_full(bufferlist::iterator
& p
);
846 void _encode_locks_state_for_replica(bufferlist
& bl
, bool need_recover
);
847 void _encode_locks_state_for_rejoin(bufferlist
& bl
, int rep
);
848 void _decode_locks_state(bufferlist::iterator
& p
, bool is_new
);
849 void _decode_locks_rejoin(bufferlist::iterator
& p
, std::list
<MDSInternalContextBase
*>& waiters
,
850 std::list
<SimpleLock
*>& eval_locks
, bool survivor
);
852 // -- import/export --
853 void encode_export(bufferlist
& bl
);
854 void finish_export(utime_t now
);
855 void abort_export() {
856 put(PIN_TEMPEXPORTING
);
857 assert(state_test(STATE_EXPORTINGCAPS
));
858 state_clear(STATE_EXPORTINGCAPS
);
859 put(PIN_EXPORTINGCAPS
);
861 void decode_import(bufferlist::iterator
& p
, LogSegment
*ls
);
864 // for giving to clients
865 int encode_inodestat(bufferlist
& bl
, Session
*session
, SnapRealm
*realm
,
866 snapid_t snapid
=CEPH_NOSNAP
, unsigned max_bytes
=0,
867 int getattr_wants
=0);
868 void encode_cap_message(MClientCaps
*m
, Capability
*cap
);
873 static LockType versionlock_type
;
874 static LockType authlock_type
;
875 static LockType linklock_type
;
876 static LockType dirfragtreelock_type
;
877 static LockType filelock_type
;
878 static LockType xattrlock_type
;
879 static LockType snaplock_type
;
880 static LockType nestlock_type
;
881 static LockType flocklock_type
;
882 static LockType policylock_type
;
884 // FIXME not part of mempool
885 LocalLock versionlock
;
888 ScatterLock dirfragtreelock
;
889 ScatterLock filelock
;
890 SimpleLock xattrlock
;
892 ScatterLock nestlock
;
893 SimpleLock flocklock
;
894 SimpleLock policylock
;
896 SimpleLock
* get_lock(int type
) override
{
898 case CEPH_LOCK_IFILE
: return &filelock
;
899 case CEPH_LOCK_IAUTH
: return &authlock
;
900 case CEPH_LOCK_ILINK
: return &linklock
;
901 case CEPH_LOCK_IDFT
: return &dirfragtreelock
;
902 case CEPH_LOCK_IXATTR
: return &xattrlock
;
903 case CEPH_LOCK_ISNAP
: return &snaplock
;
904 case CEPH_LOCK_INEST
: return &nestlock
;
905 case CEPH_LOCK_IFLOCK
: return &flocklock
;
906 case CEPH_LOCK_IPOLICY
: return &policylock
;
911 void set_object_info(MDSCacheObjectInfo
&info
) override
;
912 void encode_lock_state(int type
, bufferlist
& bl
) override
;
913 void decode_lock_state(int type
, bufferlist
& bl
) override
;
915 void _finish_frag_update(CDir
*dir
, MutationRef
& mut
);
917 void clear_dirty_scattered(int type
) override
;
918 bool is_dirty_scattered();
919 void clear_scatter_dirty(); // on rejoin ack
921 void start_scatter(ScatterLock
*lock
);
922 void finish_scatter_update(ScatterLock
*lock
, CDir
*dir
,
923 version_t inode_version
, version_t dir_accounted_version
);
924 void finish_scatter_gather_update(int type
);
925 void finish_scatter_gather_update_accounted(int type
, MutationRef
& mut
, EMetaBlob
*metablob
);
928 void open_snaprealm(bool no_split
=false);
929 void close_snaprealm(bool no_join
=false);
930 SnapRealm
*find_snaprealm() const;
931 void encode_snap(bufferlist
& bl
);
932 void decode_snap(bufferlist::iterator
& p
);
936 client_t loner_cap
= -1, want_loner_cap
= -1;
938 client_t
get_loner() const { return loner_cap
; }
939 client_t
get_wanted_loner() const { return want_loner_cap
; }
941 // this is the loner state our locks should aim for
942 client_t
get_target_loner() const {
943 if (loner_cap
== want_loner_cap
)
949 client_t
calc_ideal_loner();
950 void set_loner_cap(client_t l
);
951 bool choose_ideal_loner();
952 bool try_set_loner();
953 bool try_drop_loner();
955 // choose new lock state during recovery, based on issued caps
956 void choose_lock_state(SimpleLock
*lock
, int allissued
);
957 void choose_lock_states(int dirty_caps
);
959 int count_nonstale_caps() {
961 for (const auto &p
: client_caps
) {
962 if (!p
.second
->is_stale())
967 bool multiple_nonstale_caps() {
969 for (const auto &p
: client_caps
) {
970 if (!p
.second
->is_stale()) {
979 bool is_any_caps() { return !client_caps
.empty(); }
980 bool is_any_nonstale_caps() { return count_nonstale_caps(); }
982 const mempool::mds_co::compact_map
<int32_t,int32_t>& get_mds_caps_wanted() const { return mds_caps_wanted
; }
983 mempool::mds_co::compact_map
<int32_t,int32_t>& get_mds_caps_wanted() { return mds_caps_wanted
; }
985 const cap_map
& get_client_caps() const { return client_caps
; }
986 Capability
*get_client_cap(client_t client
) {
987 auto client_caps_entry
= client_caps
.find(client
);
988 if (client_caps_entry
!= client_caps
.end())
989 return client_caps_entry
->second
;
992 int get_client_cap_pending(client_t client
) const {
993 auto client_caps_entry
= client_caps
.find(client
);
994 if (client_caps_entry
!= client_caps
.end()) {
995 return client_caps_entry
->second
->pending();
1001 Capability
*add_client_cap(client_t client
, Session
*session
, SnapRealm
*conrealm
=0);
1002 void remove_client_cap(client_t client
);
1003 void move_to_realm(SnapRealm
*realm
);
1005 Capability
*reconnect_cap(client_t client
, const cap_reconnect_t
& icr
, Session
*session
);
1006 void clear_client_caps_after_export();
1007 void export_client_caps(std::map
<client_t
,Capability::Export
>& cl
);
1010 int get_caps_liked() const;
1011 int get_caps_allowed_ever() const;
1012 int get_caps_allowed_by_type(int type
) const;
1013 int get_caps_careful() const;
1014 int get_xlocker_mask(client_t client
) const;
1015 int get_caps_allowed_for_client(Session
*s
, mempool_inode
*file_i
) const;
1017 // caps issued, wanted
1018 int get_caps_issued(int *ploner
= 0, int *pother
= 0, int *pxlocker
= 0,
1019 int shift
= 0, int mask
= -1);
1020 bool is_any_caps_wanted() const;
1021 int get_caps_wanted(int *ploner
= 0, int *pother
= 0, int shift
= 0, int mask
= -1) const;
1022 bool issued_caps_need_gather(SimpleLock
*lock
);
1023 void replicate_relax_locks();
1026 mds_authority_t
authority() const override
;
1029 void adjust_nested_auth_pins(int a
, void *by
);
1030 bool can_auth_pin(int *err_ret
=nullptr) const override
;
1031 void auth_pin(void *by
) override
;
1032 void auth_unpin(void *by
) override
;
1035 bool is_freezing_inode() const { return state_test(STATE_FREEZING
); }
1036 bool is_frozen_inode() const { return state_test(STATE_FROZEN
); }
1037 bool is_frozen_auth_pin() const { return state_test(STATE_FROZENAUTHPIN
); }
1038 bool is_frozen() const override
;
1039 bool is_frozen_dir() const;
1040 bool is_freezing() const override
;
1042 /* Freeze the inode. auth_pin_allowance lets the caller account for any
1043 * auth_pins it is itself holding/responsible for. */
1044 bool freeze_inode(int auth_pin_allowance
=0);
1045 void unfreeze_inode(std::list
<MDSInternalContextBase
*>& finished
);
1046 void unfreeze_inode();
1048 void freeze_auth_pin();
1049 void unfreeze_auth_pin();
1051 // -- reference counting --
1052 void bad_put(int by
) override
{
1053 generic_dout(0) << " bad put " << *this << " by " << by
<< " " << pin_name(by
) << " was " << ref
1055 << " (" << ref_map
<< ")"
1059 assert(ref_map
[by
] > 0);
1063 void bad_get(int by
) override
{
1064 generic_dout(0) << " bad get " << *this << " by " << by
<< " " << pin_name(by
) << " was " << ref
1066 << " (" << ref_map
<< ")"
1070 assert(ref_map
[by
] >= 0);
1073 void first_get() override
;
1074 void last_put() override
;
1075 void _put() override
;
1078 // -- hierarchy stuff --
1080 void set_primary_parent(CDentry
*p
) {
1081 assert(parent
== 0 ||
1082 g_conf
->get_val
<bool>("mds_hack_allow_loading_invalid_metadata"));
1085 void remove_primary_parent(CDentry
*dn
) {
1086 assert(dn
== parent
);
1089 void add_remote_parent(CDentry
*p
);
1090 void remove_remote_parent(CDentry
*p
);
1091 int num_remote_parents() {
1092 return remote_parents
.size();
1095 void push_projected_parent(CDentry
*dn
) {
1096 projected_parent
.push_back(dn
);
1098 void pop_projected_parent() {
1099 assert(projected_parent
.size());
1100 parent
= projected_parent
.front();
1101 projected_parent
.pop_front();
1105 void maybe_export_pin(bool update
=false);
1106 void set_export_pin(mds_rank_t rank
);
1107 mds_rank_t
get_export_pin(bool inherit
=true) const;
1108 bool is_exportable(mds_rank_t dest
) const;
1110 void print(ostream
& out
) override
;
1111 void dump(Formatter
*f
) const;
1114 * @defgroup Scrubbing and fsck
1119 * Report the results of validation against a particular inode.
1120 * Each member is a pair of bools.
1121 * <member>.first represents if validation was performed against the member.
1122 * <member.second represents if the member passed validation.
1123 * performed_validation is set to true if the validation was actually
1124 * run. It might not be run if, for instance, the inode is marked as dirty.
1125 * passed_validation is set to true if everything that was checked
1126 * passed its validation.
1128 struct validated_data
{
1129 template<typename T
>struct member_status
{
1130 bool checked
= false;
1131 bool passed
= false;
1132 bool repaired
= false;
1133 int ondisk_read_retval
= 0;
1136 std::stringstream error_str
;
1139 bool performed_validation
= false;
1140 bool passed_validation
= false;
1142 struct raw_stats_t
{
1143 frag_info_t dirstat
;
1147 member_status
<inode_backtrace_t
> backtrace
;
1148 member_status
<mempool_inode
> inode
; // XXX should not be in mempool; wait for pmr
1149 member_status
<raw_stats_t
> raw_stats
;
1153 void dump(Formatter
*f
) const;
1155 bool all_damage_repaired() const;
1159 * Validate that the on-disk state of an inode matches what
1160 * we expect from our memory state. Currently this checks that:
1161 * 1) The backtrace associated with the file data exists and is correct
1162 * 2) For directories, the actual inode metadata matches our memory state,
1163 * 3) For directories, the rstats match
1165 * @param results A freshly-created validated_data struct, with values set
1166 * as described in the struct documentation.
1167 * @param mdr The request to be responeded upon the completion of the
1168 * validation (or NULL)
1169 * @param fin Context to call back on completion (or NULL)
1171 void validate_disk_state(validated_data
*results
,
1172 MDSInternalContext
*fin
);
1173 static void dump_validation_results(const validated_data
& results
,
1176 bool _validate_disk_state(class ValidationContinuation
*c
,
1177 int rval
, int stage
);
1178 friend class ValidationContinuation
;
1179 /** @} Scrubbing and fsck */
1182 ostream
& operator<<(ostream
& out
, const CInode::scrub_stamp_info_t
& si
);