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 <string_view>
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"
33 #include "MDSContext.h"
37 #include "SimpleLock.h"
38 #include "ScatterLock.h"
39 #include "LocalLock.h"
40 #include "Capability.h"
41 #include "SnapRealm.h"
44 #include "messages/MClientCaps.h"
46 #define dout_context g_ceph_context
56 struct ObjectOperation
;
60 ostream
& operator<<(ostream
& out
, const CInode
& in
);
62 struct cinode_lock_info_t
{
67 extern cinode_lock_info_t cinode_lock_info
[];
68 extern int num_cinode_locks
;
72 * Base class for CInode, containing the backing store data and
73 * serialization methods. This exists so that we can read and
74 * handle CInodes from the backing store without hitting all
75 * the business logic in CInode proper.
77 class InodeStoreBase
{
79 typedef inode_t
<mempool::mds_co::pool_allocator
> mempool_inode
;
80 typedef old_inode_t
<mempool::mds_co::pool_allocator
> mempool_old_inode
;
81 typedef mempool::mds_co::compact_map
<snapid_t
, mempool_old_inode
> mempool_old_inode_map
;
82 typedef xattr_map
<mempool::mds_co::pool_allocator
> mempool_xattr_map
; // FIXME bufferptr not in mempool
84 mempool_inode inode
; // the inode itself
85 mempool::mds_co::string symlink
; // symlink dest, if symlink
86 mempool_xattr_map xattrs
;
87 fragtree_t dirfragtree
; // dir frag tree, if any. always consistent with our dirfrag map.
88 mempool_old_inode_map old_inodes
; // key = last, value.first = first
89 snapid_t oldest_snap
= CEPH_NOSNAP
;
90 damage_flags_t damage_flags
= 0;
95 bool is_file() const { return inode
.is_file(); }
96 bool is_symlink() const { return inode
.is_symlink(); }
97 bool is_dir() const { return inode
.is_dir(); }
98 static object_t
get_object_name(inodeno_t ino
, frag_t fg
, std::string_view suffix
);
100 /* Full serialization for use in ".inode" root inode objects */
101 void encode(bufferlist
&bl
, uint64_t features
, const bufferlist
*snap_blob
=NULL
) const;
102 void decode(bufferlist::const_iterator
&bl
, bufferlist
& snap_blob
);
104 /* Serialization without ENCODE_START/FINISH blocks for use embedded in dentry */
105 void encode_bare(bufferlist
&bl
, uint64_t features
, const bufferlist
*snap_blob
=NULL
) const;
106 void decode_bare(bufferlist::const_iterator
&bl
, bufferlist
&snap_blob
, __u8 struct_v
=5);
108 /* For test/debug output */
109 void dump(Formatter
*f
) const;
111 /* For use by offline tools */
112 __u32
hash_dentry_name(std::string_view dn
);
113 frag_t
pick_dirfrag(std::string_view dn
);
116 class InodeStore
: public InodeStoreBase
{
118 // FIXME bufferlist not part of mempool
119 bufferlist snap_blob
; // Encoded copy of SnapRealm, because we can't
120 // rehydrate it without full MDCache
121 void encode(bufferlist
&bl
, uint64_t features
) const {
122 InodeStoreBase::encode(bl
, features
, &snap_blob
);
124 void decode(bufferlist::const_iterator
&bl
) {
125 InodeStoreBase::decode(bl
, snap_blob
);
127 void encode_bare(bufferlist
&bl
, uint64_t features
) const {
128 InodeStoreBase::encode_bare(bl
, features
, &snap_blob
);
130 void decode_bare(bufferlist::const_iterator
&bl
) {
131 InodeStoreBase::decode_bare(bl
, snap_blob
);
134 static void generate_test_instances(std::list
<InodeStore
*>& ls
);
136 WRITE_CLASS_ENCODER_FEATURES(InodeStore
)
138 // just for ceph-dencoder
139 class InodeStoreBare
: public InodeStore
{
141 void encode(bufferlist
&bl
, uint64_t features
) const {
142 InodeStore::encode_bare(bl
, features
);
144 void decode(bufferlist::const_iterator
&bl
) {
145 InodeStore::decode_bare(bl
);
147 static void generate_test_instances(std::list
<InodeStoreBare
*>& ls
);
149 WRITE_CLASS_ENCODER_FEATURES(InodeStoreBare
)
151 // cached inode wrapper
152 class CInode
: public MDSCacheObject
, public InodeStoreBase
, public Counter
<CInode
> {
154 MEMPOOL_CLASS_HELPERS();
156 static const int PIN_DIRFRAG
= -1;
157 static const int PIN_CAPS
= 2; // client caps
158 static const int PIN_IMPORTING
= -4; // importing
159 static const int PIN_OPENINGDIR
= 7;
160 static const int PIN_REMOTEPARENT
= 8;
161 static const int PIN_BATCHOPENJOURNAL
= 9;
162 static const int PIN_SCATTERED
= 10;
163 static const int PIN_STICKYDIRS
= 11;
164 //static const int PIN_PURGING = -12;
165 static const int PIN_FREEZING
= 13;
166 static const int PIN_FROZEN
= 14;
167 static const int PIN_IMPORTINGCAPS
= -15;
168 static const int PIN_PASTSNAPPARENT
= -16;
169 static const int PIN_OPENINGSNAPPARENTS
= 17;
170 static const int PIN_TRUNCATING
= 18;
171 static const int PIN_STRAY
= 19; // we pin our stray inode while active
172 static const int PIN_NEEDSNAPFLUSH
= 20;
173 static const int PIN_DIRTYRSTAT
= 21;
174 static const int PIN_EXPORTINGCAPS
= 22;
175 static const int PIN_DIRTYPARENT
= 23;
176 static const int PIN_DIRWAITER
= 24;
177 static const int PIN_SCRUBQUEUE
= 25;
179 std::string_view
pin_name(int p
) const override
{
181 case PIN_DIRFRAG
: return "dirfrag";
182 case PIN_CAPS
: return "caps";
183 case PIN_IMPORTING
: return "importing";
184 case PIN_OPENINGDIR
: return "openingdir";
185 case PIN_REMOTEPARENT
: return "remoteparent";
186 case PIN_BATCHOPENJOURNAL
: return "batchopenjournal";
187 case PIN_SCATTERED
: return "scattered";
188 case PIN_STICKYDIRS
: return "stickydirs";
189 //case PIN_PURGING: return "purging";
190 case PIN_FREEZING
: return "freezing";
191 case PIN_FROZEN
: return "frozen";
192 case PIN_IMPORTINGCAPS
: return "importingcaps";
193 case PIN_EXPORTINGCAPS
: return "exportingcaps";
194 case PIN_PASTSNAPPARENT
: return "pastsnapparent";
195 case PIN_OPENINGSNAPPARENTS
: return "openingsnapparents";
196 case PIN_TRUNCATING
: return "truncating";
197 case PIN_STRAY
: return "stray";
198 case PIN_NEEDSNAPFLUSH
: return "needsnapflush";
199 case PIN_DIRTYRSTAT
: return "dirtyrstat";
200 case PIN_DIRTYPARENT
: return "dirtyparent";
201 case PIN_DIRWAITER
: return "dirwaiter";
202 case PIN_SCRUBQUEUE
: return "scrubqueue";
203 default: return generic_pin_name(p
);
208 static const int DUMP_INODE_STORE_BASE
= (1 << 0);
209 static const int DUMP_MDS_CACHE_OBJECT
= (1 << 1);
210 static const int DUMP_LOCKS
= (1 << 2);
211 static const int DUMP_STATE
= (1 << 3);
212 static const int DUMP_CAPS
= (1 << 4);
213 static const int DUMP_PATH
= (1 << 5);
214 static const int DUMP_DIRFRAGS
= (1 << 6);
215 static const int DUMP_ALL
= (-1);
216 static const int DUMP_DEFAULT
= DUMP_ALL
& (~DUMP_PATH
) & (~DUMP_DIRFRAGS
);
219 static const int STATE_EXPORTING
= (1<<0); // on nonauth bystander.
220 static const int STATE_OPENINGDIR
= (1<<1);
221 static const int STATE_FREEZING
= (1<<2);
222 static const int STATE_FROZEN
= (1<<3);
223 static const int STATE_AMBIGUOUSAUTH
= (1<<4);
224 static const int STATE_EXPORTINGCAPS
= (1<<5);
225 static const int STATE_NEEDSRECOVER
= (1<<6);
226 static const int STATE_RECOVERING
= (1<<7);
227 static const int STATE_PURGING
= (1<<8);
228 static const int STATE_DIRTYPARENT
= (1<<9);
229 static const int STATE_DIRTYRSTAT
= (1<<10);
230 static const int STATE_STRAYPINNED
= (1<<11);
231 static const int STATE_FROZENAUTHPIN
= (1<<12);
232 static const int STATE_DIRTYPOOL
= (1<<13);
233 static const int STATE_REPAIRSTATS
= (1<<14);
234 static const int STATE_MISSINGOBJS
= (1<<15);
235 static const int STATE_EVALSTALECAPS
= (1<<16);
236 static const int STATE_QUEUEDEXPORTPIN
= (1<<17);
237 static const int STATE_TRACKEDBYOFT
= (1<<18); // tracked by open file table
238 // orphan inode needs notification of releasing reference
239 static const int STATE_ORPHAN
= STATE_NOTIFYREF
;
241 static const int MASK_STATE_EXPORTED
=
242 (STATE_DIRTY
|STATE_NEEDSRECOVER
|STATE_DIRTYPARENT
|STATE_DIRTYPOOL
);
243 static const int MASK_STATE_EXPORT_KEPT
=
244 (STATE_FROZEN
|STATE_AMBIGUOUSAUTH
|STATE_EXPORTINGCAPS
|
245 STATE_QUEUEDEXPORTPIN
|STATE_TRACKEDBYOFT
);
248 static const uint64_t WAIT_DIR
= (1<<0);
249 static const uint64_t WAIT_FROZEN
= (1<<1);
250 static const uint64_t WAIT_TRUNC
= (1<<2);
251 static const uint64_t WAIT_FLOCK
= (1<<3);
253 static const uint64_t WAIT_ANY_MASK
= (uint64_t)(-1);
256 static const unsigned EXPORT_NONCE
= 1; // nonce given to replicas created by export
258 ostream
& print_db_line_prefix(ostream
& out
) override
;
263 SnapRealm
*snaprealm
= nullptr;
264 SnapRealm
*containing_realm
= nullptr;
265 snapid_t first
, last
;
266 mempool::mds_co::compact_set
<snapid_t
> dirty_old_rstats
;
268 class scrub_stamp_info_t
{
270 /// version we started our latest scrub (whether in-progress or finished)
271 version_t scrub_start_version
= 0;
272 /// time we started our latest scrub (whether in-progress or finished)
273 utime_t scrub_start_stamp
;
274 /// version we started our most recent finished scrub
275 version_t last_scrub_version
= 0;
276 /// time we started our most recent finished scrub
277 utime_t last_scrub_stamp
;
278 scrub_stamp_info_t() {}
280 scrub_start_version
= last_scrub_version
= 0;
281 scrub_start_stamp
= last_scrub_stamp
= utime_t();
285 class scrub_info_t
: public scrub_stamp_info_t
{
287 CDentry
*scrub_parent
= nullptr;
288 MDSContext
*on_finish
= nullptr;
290 bool last_scrub_dirty
= false; /// are our stamps dirty with respect to disk state?
291 bool scrub_in_progress
= false; /// are we currently scrubbing?
292 bool children_scrubbed
= false;
294 /// my own (temporary) stamps and versions for each dirfrag we have
295 std::map
<frag_t
, scrub_stamp_info_t
> dirfrag_stamps
; // XXX not part of mempool
297 ScrubHeaderRef header
;
302 const scrub_info_t
*scrub_info() const{
308 ScrubHeaderRef
get_scrub_header() {
309 if (scrub_infop
== nullptr) {
312 return scrub_infop
->header
;
316 bool scrub_is_in_progress() const {
317 return (scrub_infop
&& scrub_infop
->scrub_in_progress
);
320 * Start scrubbing on this inode. That could be very short if it's
321 * a file, or take a long time if we're recursively scrubbing a directory.
322 * @pre It is not currently scrubbing
323 * @post it has set up internal scrubbing state
324 * @param scrub_version What version are we scrubbing at (usually, parent
325 * directory's get_projected_version())
327 void scrub_initialize(CDentry
*scrub_parent
,
328 ScrubHeaderRef
& header
,
331 * Get the next dirfrag to scrub. Gives you a frag_t in output param which
332 * you must convert to a CDir (and possibly load off disk).
333 * @param dir A pointer to frag_t, will be filled in with the next dirfrag to
334 * scrub if there is one.
335 * @returns 0 on success, you should scrub the passed-out frag_t right now;
336 * ENOENT: There are no remaining dirfrags to scrub
337 * <0 There was some other error (It will return -ENOTDIR if not a directory)
339 int scrub_dirfrag_next(frag_t
* out_dirfrag
);
341 * Get the currently scrubbing dirfrags. When returned, the
342 * passed-in list will be filled in with all frag_ts which have
343 * been returned from scrub_dirfrag_next but not sent back
344 * via scrub_dirfrag_finished.
346 void scrub_dirfrags_scrubbing(frag_vec_t
*out_dirfrags
);
348 * Report to the CInode that a dirfrag it owns has been scrubbed. Call
349 * this for every frag_t returned from scrub_dirfrag_next().
350 * @param dirfrag The frag_t that was scrubbed
352 void scrub_dirfrag_finished(frag_t dirfrag
);
354 * Call this once the scrub has been completed, whether it's a full
355 * recursive scrub on a directory or simply the data on a file (or
356 * anything in between).
357 * @param c An out param which is filled in with a Context* that must
360 void scrub_finished(MDSContext
**c
);
362 void scrub_aborted(MDSContext
**c
);
365 * Report to the CInode that alldirfrags it owns have been scrubbed.
367 void scrub_children_finished() {
368 scrub_infop
->children_scrubbed
= true;
370 void scrub_set_finisher(MDSContext
*c
) {
371 ceph_assert(!scrub_infop
->on_finish
);
372 scrub_infop
->on_finish
= c
;
377 * Create a scrub_info_t struct for the scrub_infop pointer.
379 void scrub_info_create() const;
381 * Delete the scrub_info_t struct if it's not got any useful data
383 void scrub_maybe_delete_info();
386 bool is_multiversion() const {
387 return snaprealm
|| // other snaprealms will link to me
388 inode
.is_dir() || // links to me in other snaps
389 inode
.nlink
> 1 || // there are remote links, possibly snapped, that will need to find me
390 !old_inodes
.empty(); // once multiversion, always multiversion. until old_inodes gets cleaned out.
392 snapid_t
get_oldest_snap();
394 uint64_t last_journaled
= 0; // log offset for the last time i was journaled
395 //loff_t last_open_journaled; // log offset for the last journaled EOpen
396 utime_t last_dirstat_prop
;
399 // list item node for when we have unpropagated rstat data
400 elist
<CInode
*>::item dirty_rstat_item
;
402 bool is_dirty_rstat() {
403 return state_test(STATE_DIRTYRSTAT
);
405 void mark_dirty_rstat();
406 void clear_dirty_rstat();
408 //bool hack_accessed = false;
409 //utime_t hack_load_stamp;
412 * Projection methods, used to store inode changes until they have been journaled,
413 * at which point they are popped.
415 * project_inode as needed. If you're changing xattrs or sr_t, then pass true
416 * as needed then change the xattrs/snapnode member as needed. (Dirty
417 * exception: project_past_snaprealm_parent allows you to project the
418 * snapnode after doing project_inode (i.e. you don't need to pass
421 * Then, journal. Once journaling is done, pop_and_dirty_projected_inode.
422 * This function will take care of the inode itself, the xattrs, and the snaprealm.
425 class projected_inode
{
427 static sr_t
* const UNDEF_SRNODE
;
430 std::unique_ptr
<mempool_xattr_map
> xattrs
;
431 sr_t
*snapnode
= UNDEF_SRNODE
;
433 projected_inode() = delete;
434 explicit projected_inode(const mempool_inode
&in
) : inode(in
) {}
438 mempool::mds_co::list
<projected_inode
> projected_nodes
; // projected values (only defined while dirty)
439 size_t num_projected_xattrs
= 0;
440 size_t num_projected_srnodes
= 0;
443 CInode::projected_inode
&project_inode(bool xattr
= false, bool snap
= false);
444 void pop_and_dirty_projected_inode(LogSegment
*ls
);
446 projected_inode
*get_projected_node() {
447 if (projected_nodes
.empty())
450 return &projected_nodes
.back();
453 version_t
get_projected_version() const {
454 if (projected_nodes
.empty())
455 return inode
.version
;
457 return projected_nodes
.back().inode
.version
;
459 bool is_projected() const {
460 return !projected_nodes
.empty();
463 const mempool_inode
*get_projected_inode() const {
464 if (projected_nodes
.empty())
467 return &projected_nodes
.back().inode
;
469 mempool_inode
*get_projected_inode() {
470 if (projected_nodes
.empty())
473 return &projected_nodes
.back().inode
;
475 mempool_inode
*get_previous_projected_inode() {
476 ceph_assert(!projected_nodes
.empty());
477 auto it
= projected_nodes
.rbegin();
479 if (it
!= projected_nodes
.rend())
485 mempool_xattr_map
*get_projected_xattrs() {
486 if (num_projected_xattrs
> 0) {
487 for (auto it
= projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
489 return it
->xattrs
.get();
493 mempool_xattr_map
*get_previous_projected_xattrs() {
494 if (num_projected_xattrs
> 0) {
495 for (auto it
= ++projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
497 return it
->xattrs
.get();
502 sr_t
*prepare_new_srnode(snapid_t snapid
);
503 void project_snaprealm(sr_t
*new_srnode
);
504 sr_t
*project_snaprealm(snapid_t snapid
=0) {
505 sr_t
* new_srnode
= prepare_new_srnode(snapid
);
506 project_snaprealm(new_srnode
);
509 const sr_t
*get_projected_srnode() const {
510 if (num_projected_srnodes
> 0) {
511 for (auto it
= projected_nodes
.rbegin(); it
!= projected_nodes
.rend(); ++it
)
512 if (it
->snapnode
!= projected_inode::UNDEF_SRNODE
)
516 return &snaprealm
->srnode
;
521 void mark_snaprealm_global(sr_t
*new_srnode
);
522 void clear_snaprealm_global(sr_t
*new_srnode
);
523 bool is_projected_snaprealm_global() const;
525 void record_snaprealm_past_parent(sr_t
*new_snap
, SnapRealm
*newparent
);
526 void record_snaprealm_parent_dentry(sr_t
*new_snap
, SnapRealm
*newparent
,
527 CDentry
*dn
, bool primary_dn
);
528 void project_snaprealm_past_parent(SnapRealm
*newparent
);
529 void early_pop_projected_snaprealm();
532 void pop_projected_snaprealm(sr_t
*next_snaprealm
, bool early
);
535 mempool_old_inode
& cow_old_inode(snapid_t follows
, bool cow_head
);
536 void split_old_inode(snapid_t snap
);
537 mempool_old_inode
*pick_old_inode(snapid_t last
);
538 void pre_cow_old_inode();
539 bool has_snap_data(snapid_t s
);
540 void purge_stale_snap_data(const std::set
<snapid_t
>& snaps
);
542 // -- cache infrastructure --
544 mempool::mds_co::compact_map
<frag_t
,CDir
*> dirfrags
; // cached dir fragments under this Inode
546 //for the purpose of quickly determining whether there's a subtree root or exporting dir
547 int num_subtree_roots
= 0;
548 int num_exporting_dirs
= 0;
550 int stickydir_ref
= 0;
551 scrub_info_t
*scrub_infop
= nullptr;
554 bool has_dirfrags() { return !dirfrags
.empty(); }
555 CDir
* get_dirfrag(frag_t fg
) {
556 auto pi
= dirfrags
.find(fg
);
557 if (pi
!= dirfrags
.end()) {
558 //assert(g_conf()->debug_mds < 2 || dirfragtree.is_leaf(fg)); // performance hack FIXME
563 bool get_dirfrags_under(frag_t fg
, std::list
<CDir
*>& ls
);
564 CDir
* get_approx_dirfrag(frag_t fg
);
566 template<typename Container
>
567 void get_dirfrags(Container
& ls
) const {
569 if constexpr (std::is_same_v
<Container
, std::vector
<CDir
*>>)
570 ls
.reserve(ls
.size() + dirfrags
.size());
571 for (const auto &p
: dirfrags
)
572 ls
.push_back(p
.second
);
574 template<typename Container
>
575 void get_nested_dirfrags(Container
& ls
) const {
576 // dirfrags in same subtree
577 if constexpr (std::is_same_v
<Container
, std::vector
<CDir
*>>)
578 ls
.reserve(ls
.size() + dirfrags
.size() - num_subtree_roots
);
579 for (const auto &p
: dirfrags
) {
580 typename
Container::value_type dir
= p
.second
;
581 if (!dir
->is_subtree_root())
585 template<typename Container
>
586 void get_subtree_dirfrags(Container
& ls
) {
587 // dirfrags that are roots of new subtrees
588 if constexpr (std::is_same_v
<Container
, std::vector
<CDir
*>>)
589 ls
.reserve(ls
.size() + num_subtree_roots
);
590 for (const auto &p
: dirfrags
) {
591 typename
Container::value_type dir
= p
.second
;
592 if (dir
->is_subtree_root())
597 CDir
*get_or_open_dirfrag(MDCache
*mdcache
, frag_t fg
);
598 CDir
*add_dirfrag(CDir
*dir
);
599 void close_dirfrag(frag_t fg
);
600 void close_dirfrags();
601 bool has_subtree_root_dirfrag(int auth
=-1);
602 bool has_subtree_or_exporting_dirfrag();
604 void force_dirfrags();
605 void verify_dirfrags();
607 void get_stickydirs();
608 void put_stickydirs();
611 // parent dentries in cache
612 CDentry
*parent
= nullptr; // primary link
613 mempool::mds_co::compact_set
<CDentry
*> remote_parents
; // if hard linked
615 mempool::mds_co::list
<CDentry
*> projected_parent
; // for in-progress rename, (un)link, etc.
617 mds_authority_t inode_auth
= CDIR_AUTH_DEFAULT
;
619 // -- distributed state --
622 using mempool_cap_map
= mempool::mds_co::map
<client_t
, Capability
>;
623 mempool_cap_map client_caps
; // client -> caps
624 mempool::mds_co::compact_map
<int32_t, int32_t> mds_caps_wanted
; // [auth] mds -> caps wanted
625 int replica_caps_wanted
= 0; // [replica] what i've requested from auth
626 int num_caps_wanted
= 0;
629 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
630 mempool::mds_co::compact_map
<snapid_t
, mempool::mds_co::set
<client_t
> > client_need_snapflush
;
632 void add_need_snapflush(CInode
*snapin
, snapid_t snapid
, client_t client
);
633 void remove_need_snapflush(CInode
*snapin
, snapid_t snapid
, client_t client
);
634 bool split_need_snapflush(CInode
*cowin
, CInode
*in
);
638 ceph_lock_state_t
*fcntl_locks
= nullptr;
639 ceph_lock_state_t
*flock_locks
= nullptr;
641 ceph_lock_state_t
*get_fcntl_lock_state() {
643 fcntl_locks
= new ceph_lock_state_t(g_ceph_context
, CEPH_LOCK_FCNTL
);
646 void clear_fcntl_lock_state() {
650 ceph_lock_state_t
*get_flock_lock_state() {
652 flock_locks
= new ceph_lock_state_t(g_ceph_context
, CEPH_LOCK_FLOCK
);
655 void clear_flock_lock_state() {
659 void clear_file_locks() {
660 clear_fcntl_lock_state();
661 clear_flock_lock_state();
663 void _encode_file_locks(bufferlist
& bl
) const {
665 bool has_fcntl_locks
= fcntl_locks
&& !fcntl_locks
->empty();
666 encode(has_fcntl_locks
, bl
);
668 encode(*fcntl_locks
, bl
);
669 bool has_flock_locks
= flock_locks
&& !flock_locks
->empty();
670 encode(has_flock_locks
, bl
);
672 encode(*flock_locks
, bl
);
674 void _decode_file_locks(bufferlist::const_iterator
& p
) {
676 bool has_fcntl_locks
;
677 decode(has_fcntl_locks
, p
);
679 decode(*get_fcntl_lock_state(), p
);
681 clear_fcntl_lock_state();
682 bool has_flock_locks
;
683 decode(has_flock_locks
, p
);
685 decode(*get_flock_lock_state(), p
);
687 clear_flock_lock_state();
690 // LogSegment lists i (may) belong to
692 elist
<CInode
*>::item item_dirty
;
693 elist
<CInode
*>::item item_caps
;
694 elist
<CInode
*>::item item_open_file
;
695 elist
<CInode
*>::item item_dirty_parent
;
696 elist
<CInode
*>::item item_dirty_dirfrag_dir
;
697 elist
<CInode
*>::item item_dirty_dirfrag_nest
;
698 elist
<CInode
*>::item item_dirty_dirfrag_dirfragtree
;
699 elist
<CInode
*>::item item_scrub
;
701 // also update RecoveryQueue::RecoveryQueue() if you change this
702 elist
<CInode
*>::item
& item_recover_queue
= item_dirty_dirfrag_dir
;
703 elist
<CInode
*>::item
& item_recover_queue_front
= item_dirty_dirfrag_nest
;
706 int auth_pin_freeze_allowance
= 0;
708 inode_load_vec_t pop
;
709 elist
<CInode
*>::item item_pop_lru
;
714 friend class Migrator
;
715 friend class MDCache
;
716 friend class StrayManager
;
718 friend class CInodeExport
;
720 // ---------------------------
722 CInode(MDCache
*c
, bool auth
=true, snapid_t f
=2, snapid_t l
=CEPH_NOSNAP
);
727 ceph_assert(num_projected_xattrs
== 0);
728 ceph_assert(num_projected_srnodes
== 0);
729 ceph_assert(num_caps_wanted
== 0);
730 ceph_assert(num_subtree_roots
== 0);
731 ceph_assert(num_exporting_dirs
== 0);
736 bool is_root() const { return inode
.ino
== MDS_INO_ROOT
; }
737 bool is_stray() const { return MDS_INO_IS_STRAY(inode
.ino
); }
738 mds_rank_t
get_stray_owner() const {
739 return (mds_rank_t
)MDS_INO_STRAY_OWNER(inode
.ino
);
741 bool is_mdsdir() const { return MDS_INO_IS_MDSDIR(inode
.ino
); }
742 bool is_base() const { return MDS_INO_IS_BASE(inode
.ino
); }
743 bool is_system() const { return inode
.ino
< MDS_INO_SYSTEM_BASE
; }
744 bool is_normal() const { return !(is_base() || is_system() || is_stray()); }
746 bool is_head() const { return last
== CEPH_NOSNAP
; }
748 // note: this overloads MDSCacheObject
749 bool is_ambiguous_auth() const {
750 return state_test(STATE_AMBIGUOUSAUTH
) ||
751 MDSCacheObject::is_ambiguous_auth();
753 void set_ambiguous_auth() {
754 state_set(STATE_AMBIGUOUSAUTH
);
756 void clear_ambiguous_auth(MDSContext::vec
& finished
);
757 void clear_ambiguous_auth();
759 inodeno_t
ino() const { return inode
.ino
; }
760 vinodeno_t
vino() const { return vinodeno_t(inode
.ino
, last
); }
761 int d_type() const { return IFTODT(inode
.mode
); }
763 mempool_inode
& get_inode() { return inode
; }
764 const mempool_inode
& get_inode() const { return inode
; }
765 CDentry
* get_parent_dn() { return parent
; }
766 const CDentry
* get_parent_dn() const { return parent
; }
767 CDentry
* get_projected_parent_dn() { return !projected_parent
.empty() ? projected_parent
.back() : parent
; }
768 const CDentry
* get_projected_parent_dn() const { return !projected_parent
.empty() ? projected_parent
.back() : parent
; }
769 const CDentry
* get_oldest_parent_dn() const {
772 return !projected_parent
.empty() ? projected_parent
.front(): NULL
;
774 CDir
*get_parent_dir();
775 const CDir
*get_projected_parent_dir() const;
776 CDir
*get_projected_parent_dir();
777 CInode
*get_parent_inode();
779 bool is_lt(const MDSCacheObject
*r
) const override
{
780 const CInode
*o
= static_cast<const CInode
*>(r
);
781 return ino() < o
->ino() ||
782 (ino() == o
->ino() && last
< o
->last
);
786 bool is_ancestor_of(const CInode
*other
) const;
787 bool is_projected_ancestor_of(const CInode
*other
) const;
789 void make_path_string(std::string
& s
, bool projected
=false, const CDentry
*use_parent
=NULL
) const;
790 void make_path(filepath
& s
, bool projected
=false) const;
791 void name_stray_dentry(std::string
& dname
);
794 version_t
get_version() const { return inode
.version
; }
796 version_t
pre_dirty();
797 void _mark_dirty(LogSegment
*ls
);
798 void mark_dirty(version_t projected_dirv
, LogSegment
*ls
);
801 void store(MDSContext
*fin
);
802 void _stored(int r
, version_t cv
, Context
*fin
);
804 * Flush a CInode to disk. This includes the backtrace, the parent
805 * directory's link, and the Inode object itself (if a base directory).
806 * @pre is_auth() on both the inode and its containing directory
807 * @pre can_auth_pin()
808 * @param fin The Context to call when the flush is completed.
810 void flush(MDSContext
*fin
);
811 void fetch(MDSContext
*fin
);
812 void _fetched(bufferlist
& bl
, bufferlist
& bl2
, Context
*fin
);
815 void build_backtrace(int64_t pool
, inode_backtrace_t
& bt
);
816 void store_backtrace(MDSContext
*fin
, int op_prio
=-1);
817 void _stored_backtrace(int r
, version_t v
, Context
*fin
);
818 void fetch_backtrace(Context
*fin
, bufferlist
*backtrace
);
821 * Return the pool ID where we currently write backtraces for
822 * this inode (in addition to inode.old_pools)
824 * @returns a pool ID >=0
826 int64_t get_backtrace_pool() const;
828 void mark_dirty_parent(LogSegment
*ls
, bool dirty_pool
=false);
829 void clear_dirty_parent();
830 void verify_diri_backtrace(bufferlist
&bl
, int err
);
831 bool is_dirty_parent() { return state_test(STATE_DIRTYPARENT
); }
832 bool is_dirty_pool() { return state_test(STATE_DIRTYPOOL
); }
834 void encode_snap_blob(bufferlist
&bl
);
835 void decode_snap_blob(const bufferlist
&bl
);
836 void encode_store(bufferlist
& bl
, uint64_t features
);
837 void decode_store(bufferlist::const_iterator
& bl
);
839 void encode_replica(mds_rank_t rep
, bufferlist
& bl
, uint64_t features
, bool need_recover
) {
840 ceph_assert(is_auth());
842 __u32 nonce
= add_replica(rep
);
846 _encode_base(bl
, features
);
847 _encode_locks_state_for_replica(bl
, need_recover
);
849 void decode_replica(bufferlist::const_iterator
& p
, bool is_new
) {
853 replica_nonce
= nonce
;
856 _decode_locks_state(p
, is_new
);
861 mempool::mds_co::compact_map
<frag_t
, MDSContext::vec
> waiting_on_dir
;
863 void add_dir_waiter(frag_t fg
, MDSContext
*c
);
864 void take_dir_waiting(frag_t fg
, MDSContext::vec
& ls
);
865 bool is_waiting_for_dir(frag_t fg
) {
866 return waiting_on_dir
.count(fg
);
868 void add_waiter(uint64_t tag
, MDSContext
*c
) override
;
869 void take_waiting(uint64_t tag
, MDSContext::vec
& ls
) override
;
871 // -- encode/decode helpers --
872 void _encode_base(bufferlist
& bl
, uint64_t features
);
873 void _decode_base(bufferlist::const_iterator
& p
);
874 void _encode_locks_full(bufferlist
& bl
);
875 void _decode_locks_full(bufferlist::const_iterator
& p
);
876 void _encode_locks_state_for_replica(bufferlist
& bl
, bool need_recover
);
877 void _encode_locks_state_for_rejoin(bufferlist
& bl
, int rep
);
878 void _decode_locks_state(bufferlist::const_iterator
& p
, bool is_new
);
879 void _decode_locks_rejoin(bufferlist::const_iterator
& p
, MDSContext::vec
& waiters
,
880 std::list
<SimpleLock
*>& eval_locks
, bool survivor
);
882 // -- import/export --
883 void encode_export(bufferlist
& bl
);
884 void finish_export();
885 void abort_export() {
886 put(PIN_TEMPEXPORTING
);
887 ceph_assert(state_test(STATE_EXPORTINGCAPS
));
888 state_clear(STATE_EXPORTINGCAPS
);
889 put(PIN_EXPORTINGCAPS
);
891 void decode_import(bufferlist::const_iterator
& p
, LogSegment
*ls
);
894 // for giving to clients
895 int encode_inodestat(bufferlist
& bl
, Session
*session
, SnapRealm
*realm
,
896 snapid_t snapid
=CEPH_NOSNAP
, unsigned max_bytes
=0,
897 int getattr_wants
=0);
898 void encode_cap_message(const MClientCaps::ref
&m
, Capability
*cap
);
903 static LockType versionlock_type
;
904 static LockType authlock_type
;
905 static LockType linklock_type
;
906 static LockType dirfragtreelock_type
;
907 static LockType filelock_type
;
908 static LockType xattrlock_type
;
909 static LockType snaplock_type
;
910 static LockType nestlock_type
;
911 static LockType flocklock_type
;
912 static LockType policylock_type
;
914 // FIXME not part of mempool
915 LocalLock versionlock
;
918 ScatterLock dirfragtreelock
;
919 ScatterLock filelock
;
920 SimpleLock xattrlock
;
922 ScatterLock nestlock
;
923 SimpleLock flocklock
;
924 SimpleLock policylock
;
926 SimpleLock
* get_lock(int type
) override
{
928 case CEPH_LOCK_IFILE
: return &filelock
;
929 case CEPH_LOCK_IAUTH
: return &authlock
;
930 case CEPH_LOCK_ILINK
: return &linklock
;
931 case CEPH_LOCK_IDFT
: return &dirfragtreelock
;
932 case CEPH_LOCK_IXATTR
: return &xattrlock
;
933 case CEPH_LOCK_ISNAP
: return &snaplock
;
934 case CEPH_LOCK_INEST
: return &nestlock
;
935 case CEPH_LOCK_IFLOCK
: return &flocklock
;
936 case CEPH_LOCK_IPOLICY
: return &policylock
;
941 void set_object_info(MDSCacheObjectInfo
&info
) override
;
942 void encode_lock_state(int type
, bufferlist
& bl
) override
;
943 void decode_lock_state(int type
, const bufferlist
& bl
) override
;
945 void _finish_frag_update(CDir
*dir
, MutationRef
& mut
);
947 void clear_dirty_scattered(int type
) override
;
948 bool is_dirty_scattered();
949 void clear_scatter_dirty(); // on rejoin ack
951 void start_scatter(ScatterLock
*lock
);
952 void finish_scatter_update(ScatterLock
*lock
, CDir
*dir
,
953 version_t inode_version
, version_t dir_accounted_version
);
954 void finish_scatter_gather_update(int type
);
955 void finish_scatter_gather_update_accounted(int type
, MutationRef
& mut
, EMetaBlob
*metablob
);
958 void open_snaprealm(bool no_split
=false);
959 void close_snaprealm(bool no_join
=false);
960 SnapRealm
*find_snaprealm() const;
961 void encode_snap(bufferlist
& bl
);
962 void decode_snap(bufferlist::const_iterator
& p
);
966 client_t loner_cap
= -1, want_loner_cap
= -1;
968 client_t
get_loner() const { return loner_cap
; }
969 client_t
get_wanted_loner() const { return want_loner_cap
; }
971 // this is the loner state our locks should aim for
972 client_t
get_target_loner() const {
973 if (loner_cap
== want_loner_cap
)
979 client_t
calc_ideal_loner();
980 void set_loner_cap(client_t l
);
981 bool choose_ideal_loner();
982 bool try_set_loner();
983 bool try_drop_loner();
985 // choose new lock state during recovery, based on issued caps
986 void choose_lock_state(SimpleLock
*lock
, int allissued
);
987 void choose_lock_states(int dirty_caps
);
989 int count_nonstale_caps() {
991 for (const auto &p
: client_caps
) {
992 if (!p
.second
.is_stale())
997 bool multiple_nonstale_caps() {
999 for (const auto &p
: client_caps
) {
1000 if (!p
.second
.is_stale()) {
1009 bool is_any_caps() { return !client_caps
.empty(); }
1010 bool is_any_nonstale_caps() { return count_nonstale_caps(); }
1012 const mempool::mds_co::compact_map
<int32_t,int32_t>& get_mds_caps_wanted() const { return mds_caps_wanted
; }
1013 void set_mds_caps_wanted(mempool::mds_co::compact_map
<int32_t,int32_t>& m
);
1014 void set_mds_caps_wanted(mds_rank_t mds
, int32_t wanted
);
1016 const mempool_cap_map
& get_client_caps() const { return client_caps
; }
1017 Capability
*get_client_cap(client_t client
) {
1018 auto client_caps_entry
= client_caps
.find(client
);
1019 if (client_caps_entry
!= client_caps
.end())
1020 return &client_caps_entry
->second
;
1023 int get_client_cap_pending(client_t client
) const {
1024 auto client_caps_entry
= client_caps
.find(client
);
1025 if (client_caps_entry
!= client_caps
.end()) {
1026 return client_caps_entry
->second
.pending();
1032 int get_num_caps_wanted() const { return num_caps_wanted
; }
1033 void adjust_num_caps_wanted(int d
);
1035 Capability
*add_client_cap(client_t client
, Session
*session
, SnapRealm
*conrealm
=0);
1036 void remove_client_cap(client_t client
);
1037 void move_to_realm(SnapRealm
*realm
);
1039 Capability
*reconnect_cap(client_t client
, const cap_reconnect_t
& icr
, Session
*session
);
1040 void clear_client_caps_after_export();
1041 void export_client_caps(std::map
<client_t
,Capability::Export
>& cl
);
1044 int get_caps_liked() const;
1045 int get_caps_allowed_ever() const;
1046 int get_caps_allowed_by_type(int type
) const;
1047 int get_caps_careful() const;
1048 int get_xlocker_mask(client_t client
) const;
1049 int get_caps_allowed_for_client(Session
*s
, Capability
*cap
, mempool_inode
*file_i
) const;
1051 // caps issued, wanted
1052 int get_caps_issued(int *ploner
= 0, int *pother
= 0, int *pxlocker
= 0,
1053 int shift
= 0, int mask
= -1);
1054 bool is_any_caps_wanted() const;
1055 int get_caps_wanted(int *ploner
= 0, int *pother
= 0, int shift
= 0, int mask
= -1) const;
1056 bool issued_caps_need_gather(SimpleLock
*lock
);
1059 mds_authority_t
authority() const override
;
1062 bool can_auth_pin(int *err_ret
=nullptr) const override
;
1063 void auth_pin(void *by
) override
;
1064 void auth_unpin(void *by
) override
;
1067 bool is_freezing_inode() const { return state_test(STATE_FREEZING
); }
1068 bool is_frozen_inode() const { return state_test(STATE_FROZEN
); }
1069 bool is_frozen_auth_pin() const { return state_test(STATE_FROZENAUTHPIN
); }
1070 bool is_frozen() const override
;
1071 bool is_frozen_dir() const;
1072 bool is_freezing() const override
;
1074 /* Freeze the inode. auth_pin_allowance lets the caller account for any
1075 * auth_pins it is itself holding/responsible for. */
1076 bool freeze_inode(int auth_pin_allowance
=0);
1077 void unfreeze_inode(MDSContext::vec
& finished
);
1078 void unfreeze_inode();
1080 void freeze_auth_pin();
1081 void unfreeze_auth_pin();
1083 // -- reference counting --
1084 void bad_put(int by
) override
{
1085 generic_dout(0) << " bad put " << *this << " by " << by
<< " " << pin_name(by
) << " was " << ref
1087 << " (" << ref_map
<< ")"
1091 ceph_assert(ref_map
[by
] > 0);
1093 ceph_assert(ref
> 0);
1095 void bad_get(int by
) override
{
1096 generic_dout(0) << " bad get " << *this << " by " << by
<< " " << pin_name(by
) << " was " << ref
1098 << " (" << ref_map
<< ")"
1102 ceph_assert(ref_map
[by
] >= 0);
1105 void first_get() override
;
1106 void last_put() override
;
1107 void _put() override
;
1110 // -- hierarchy stuff --
1112 void set_primary_parent(CDentry
*p
) {
1113 ceph_assert(parent
== 0 ||
1114 g_conf().get_val
<bool>("mds_hack_allow_loading_invalid_metadata"));
1117 void remove_primary_parent(CDentry
*dn
) {
1118 ceph_assert(dn
== parent
);
1121 void add_remote_parent(CDentry
*p
);
1122 void remove_remote_parent(CDentry
*p
);
1123 int num_remote_parents() {
1124 return remote_parents
.size();
1127 void push_projected_parent(CDentry
*dn
) {
1128 projected_parent
.push_back(dn
);
1130 void pop_projected_parent() {
1131 ceph_assert(projected_parent
.size());
1132 parent
= projected_parent
.front();
1133 projected_parent
.pop_front();
1137 void maybe_export_pin(bool update
=false);
1138 void set_export_pin(mds_rank_t rank
);
1139 mds_rank_t
get_export_pin(bool inherit
=true) const;
1140 bool is_exportable(mds_rank_t dest
) const;
1142 void print(ostream
& out
) override
;
1143 void dump(Formatter
*f
, int flags
= DUMP_DEFAULT
) const;
1146 * @defgroup Scrubbing and fsck
1151 * Report the results of validation against a particular inode.
1152 * Each member is a pair of bools.
1153 * <member>.first represents if validation was performed against the member.
1154 * <member.second represents if the member passed validation.
1155 * performed_validation is set to true if the validation was actually
1156 * run. It might not be run if, for instance, the inode is marked as dirty.
1157 * passed_validation is set to true if everything that was checked
1158 * passed its validation.
1160 struct validated_data
{
1161 template<typename T
>struct member_status
{
1162 bool checked
= false;
1163 bool passed
= false;
1164 bool repaired
= false;
1165 int ondisk_read_retval
= 0;
1168 std::stringstream error_str
;
1171 bool performed_validation
= false;
1172 bool passed_validation
= false;
1174 struct raw_stats_t
{
1175 frag_info_t dirstat
;
1179 member_status
<inode_backtrace_t
> backtrace
;
1180 member_status
<mempool_inode
> inode
; // XXX should not be in mempool; wait for pmr
1181 member_status
<raw_stats_t
> raw_stats
;
1185 void dump(Formatter
*f
) const;
1187 bool all_damage_repaired() const;
1191 * Validate that the on-disk state of an inode matches what
1192 * we expect from our memory state. Currently this checks that:
1193 * 1) The backtrace associated with the file data exists and is correct
1194 * 2) For directories, the actual inode metadata matches our memory state,
1195 * 3) For directories, the rstats match
1197 * @param results A freshly-created validated_data struct, with values set
1198 * as described in the struct documentation.
1199 * @param mdr The request to be responeded upon the completion of the
1200 * validation (or NULL)
1201 * @param fin Context to call back on completion (or NULL)
1203 void validate_disk_state(validated_data
*results
,
1205 static void dump_validation_results(const validated_data
& results
,
1208 bool _validate_disk_state(class ValidationContinuation
*c
,
1209 int rval
, int stage
);
1210 friend class ValidationContinuation
;
1211 /** @} Scrubbing and fsck */
1214 ostream
& operator<<(ostream
& out
, const CInode::scrub_stamp_info_t
& si
);