ceph/src/mds/CInode.h

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
   7  *
   8  * This is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License version 2.1, as published by the Free Software
  11  * Foundation.  See file COPYING.
  12  *
  13  */
  14
  15
  16
  17 #ifndef CEPH_CINODE_H
  18 #define CEPH_CINODE_H
  19
  20 #include "common/config.h"
  21 #include "include/counter.h"
  22 #include "include/elist.h"
  23 #include "include/types.h"
  24 #include "include/lru.h"
  25 #include "include/compact_set.h"
  26
  27 #include "MDSCacheObject.h"
  28 #include "flock.h"
  29
  30 #include "CDentry.h"
  31 #include "SimpleLock.h"
  32 #include "ScatterLock.h"
  33 #include "LocalLock.h"
  34 #include "Capability.h"
  35 #include "SnapRealm.h"
  36 #include "Mutation.h"
  37
  38 #include <list>
  39 #include <set>
  40 #include <map>
  41
  42 #define dout_context g_ceph_context
  43
  44 class Context;
  45 class CDentry;
  46 class CDir;
  47 class Message;
  48 class CInode;
  49 class MDCache;
  50 class LogSegment;
  51 struct SnapRealm;
  52 class Session;
  53 class MClientCaps;
  54 struct ObjectOperation;
  55 class EMetaBlob;
  56
  57
  58 ostream& operator<<(ostream& out, const CInode& in);
  59
  60 struct cinode_lock_info_t {
  61   int lock;
  62   int wr_caps;
  63 };
  64
  65 extern cinode_lock_info_t cinode_lock_info[];
  66 extern int num_cinode_locks;
  67
  68
  69 /**
  70  * Base class for CInode, containing the backing store data and
  71  * serialization methods.  This exists so that we can read and
  72  * handle CInodes from the backing store without hitting all
  73  * the business logic in CInode proper.
  74  */
  75 class InodeStoreBase {
  76 public:
  77   inode_t                    inode;        // the inode itself
  78   std::string                symlink;      // symlink dest, if symlink
  79   std::map<std::string, bufferptr> xattrs;
  80   fragtree_t                 dirfragtree;  // dir frag tree, if any.  always consistent with our dirfrag map.
  81   compact_map<snapid_t, old_inode_t> old_inodes;   // key = last, value.first = first
  82   snapid_t                  oldest_snap;
  83   damage_flags_t            damage_flags;
  84
  85   InodeStoreBase() : oldest_snap(CEPH_NOSNAP), damage_flags(0) { }
  86
  87   /* Helpers */
  88   bool is_file() const    { return inode.is_file(); }
  89   bool is_symlink() const { return inode.is_symlink(); }
  90   bool is_dir() const     { return inode.is_dir(); }
  91   static object_t get_object_name(inodeno_t ino, frag_t fg, const char *suffix);
  92
  93   /* Full serialization for use in ".inode" root inode objects */
  94   void encode(bufferlist &bl, uint64_t features, const bufferlist *snap_blob=NULL) const;
  95   void decode(bufferlist::iterator &bl, bufferlist& snap_blob);
  96
  97   /* Serialization without ENCODE_START/FINISH blocks for use embedded in dentry */
  98   void encode_bare(bufferlist &bl, uint64_t features, const bufferlist *snap_blob=NULL) const;
  99   void decode_bare(bufferlist::iterator &bl, bufferlist &snap_blob, __u8 struct_v=5);
 100
 101   /* For test/debug output */
 102   void dump(Formatter *f) const;
 103
 104   /* For use by offline tools */
 105   __u32 hash_dentry_name(const std::string &dn);
 106   frag_t pick_dirfrag(const std::string &dn);
 107 };
 108
 109 class InodeStore : public InodeStoreBase {
 110 public:
 111   bufferlist snap_blob;  // Encoded copy of SnapRealm, because we can't
 112                          // rehydrate it without full MDCache
 113   void encode(bufferlist &bl, uint64_t features) const {
 114     InodeStoreBase::encode(bl, features, &snap_blob);
 115   }
 116   void decode(bufferlist::iterator &bl) {
 117     InodeStoreBase::decode(bl, snap_blob);
 118   }
 119   void encode_bare(bufferlist &bl, uint64_t features) const {
 120     InodeStoreBase::encode_bare(bl, features, &snap_blob);
 121   }
 122   void decode_bare(bufferlist::iterator &bl) {
 123     InodeStoreBase::decode_bare(bl, snap_blob);
 124   }
 125
 126   static void generate_test_instances(std::list<InodeStore*>& ls);
 127 };
 128 WRITE_CLASS_ENCODER_FEATURES(InodeStore)
 129
 130 // cached inode wrapper
 131 class CInode : public MDSCacheObject, public InodeStoreBase, public Counter<CInode> {
 132  public:
 133   // -- pins --
 134   static const int PIN_DIRFRAG =         -1;
 135   static const int PIN_CAPS =             2;  // client caps
 136   static const int PIN_IMPORTING =       -4;  // importing
 137   static const int PIN_OPENINGDIR =       7;
 138   static const int PIN_REMOTEPARENT =     8;
 139   static const int PIN_BATCHOPENJOURNAL = 9;
 140   static const int PIN_SCATTERED =        10;
 141   static const int PIN_STICKYDIRS =       11;
 142   //static const int PIN_PURGING =         -12;
 143   static const int PIN_FREEZING =         13;
 144   static const int PIN_FROZEN =           14;
 145   static const int PIN_IMPORTINGCAPS =   -15;
 146   static const int PIN_PASTSNAPPARENT =  -16;
 147   static const int PIN_OPENINGSNAPPARENTS = 17;
 148   static const int PIN_TRUNCATING =       18;
 149   static const int PIN_STRAY =            19;  // we pin our stray inode while active
 150   static const int PIN_NEEDSNAPFLUSH =    20;
 151   static const int PIN_DIRTYRSTAT =       21;
 152   static const int PIN_EXPORTINGCAPS =    22;
 153   static const int PIN_DIRTYPARENT =      23;
 154   static const int PIN_DIRWAITER =        24;
 155   static const int PIN_SCRUBQUEUE =       25;
 156
 157   const char *pin_name(int p) const override {
 158     switch (p) {
 159     case PIN_DIRFRAG: return "dirfrag";
 160     case PIN_CAPS: return "caps";
 161     case PIN_IMPORTING: return "importing";
 162     case PIN_OPENINGDIR: return "openingdir";
 163     case PIN_REMOTEPARENT: return "remoteparent";
 164     case PIN_BATCHOPENJOURNAL: return "batchopenjournal";
 165     case PIN_SCATTERED: return "scattered";
 166     case PIN_STICKYDIRS: return "stickydirs";
 167       //case PIN_PURGING: return "purging";
 168     case PIN_FREEZING: return "freezing";
 169     case PIN_FROZEN: return "frozen";
 170     case PIN_IMPORTINGCAPS: return "importingcaps";
 171     case PIN_EXPORTINGCAPS: return "exportingcaps";
 172     case PIN_PASTSNAPPARENT: return "pastsnapparent";
 173     case PIN_OPENINGSNAPPARENTS: return "openingsnapparents";
 174     case PIN_TRUNCATING: return "truncating";
 175     case PIN_STRAY: return "stray";
 176     case PIN_NEEDSNAPFLUSH: return "needsnapflush";
 177     case PIN_DIRTYRSTAT: return "dirtyrstat";
 178     case PIN_DIRTYPARENT: return "dirtyparent";
 179     case PIN_DIRWAITER: return "dirwaiter";
 180     case PIN_SCRUBQUEUE: return "scrubqueue";
 181     default: return generic_pin_name(p);
 182     }
 183   }
 184
 185   // -- state --
 186   static const int STATE_EXPORTING =   (1<<2);   // on nonauth bystander.
 187   static const int STATE_OPENINGDIR =  (1<<5);
 188   static const int STATE_FREEZING =    (1<<7);
 189   static const int STATE_FROZEN =      (1<<8);
 190   static const int STATE_AMBIGUOUSAUTH = (1<<9);
 191   static const int STATE_EXPORTINGCAPS = (1<<10);
 192   static const int STATE_NEEDSRECOVER = (1<<11);
 193   static const int STATE_RECOVERING =   (1<<12);
 194   static const int STATE_PURGING =     (1<<13);
 195   static const int STATE_DIRTYPARENT =  (1<<14);
 196   static const int STATE_DIRTYRSTAT =  (1<<15);
 197   static const int STATE_STRAYPINNED = (1<<16);
 198   static const int STATE_FROZENAUTHPIN = (1<<17);
 199   static const int STATE_DIRTYPOOL =   (1<<18);
 200   static const int STATE_REPAIRSTATS = (1<<19);
 201   static const int STATE_MISSINGOBJS = (1<<20);
 202   static const int STATE_EVALSTALECAPS = (1<<21);
 203   // orphan inode needs notification of releasing reference
 204   static const int STATE_ORPHAN =       STATE_NOTIFYREF;
 205
 206   static const int MASK_STATE_EXPORTED =
 207     (STATE_DIRTY|STATE_NEEDSRECOVER|STATE_DIRTYPARENT|STATE_DIRTYPOOL);
 208   static const int MASK_STATE_EXPORT_KEPT =
 209     (STATE_FROZEN|STATE_AMBIGUOUSAUTH|STATE_EXPORTINGCAPS);
 210
 211   // -- waiters --
 212   static const uint64_t WAIT_DIR         = (1<<0);
 213   static const uint64_t WAIT_FROZEN      = (1<<1);
 214   static const uint64_t WAIT_TRUNC       = (1<<2);
 215   static const uint64_t WAIT_FLOCK       = (1<<3);
 216
 217   static const uint64_t WAIT_ANY_MASK   = (uint64_t)(-1);
 218
 219   // misc
 220   static const unsigned EXPORT_NONCE = 1; // nonce given to replicas created by export
 221
 222   ostream& print_db_line_prefix(ostream& out) override;
 223
 224  public:
 225   MDCache *mdcache;
 226
 227   SnapRealm        *snaprealm;
 228   SnapRealm        *containing_realm;
 229   snapid_t          first, last;
 230   compact_set<snapid_t> dirty_old_rstats;
 231
 232   class scrub_stamp_info_t {
 233   public:
 234     /// version we started our latest scrub (whether in-progress or finished)
 235     version_t scrub_start_version;
 236     /// time we started our latest scrub (whether in-progress or finished)
 237     utime_t scrub_start_stamp;
 238     /// version we started our most recent finished scrub
 239     version_t last_scrub_version;
 240     /// time we started our most recent finished scrub
 241     utime_t last_scrub_stamp;
 242     scrub_stamp_info_t() : scrub_start_version(0), last_scrub_version(0) {}
 243     void reset() {
 244       scrub_start_version = 0;
 245       scrub_start_stamp = utime_t();
 246     }
 247   };
 248
 249   class scrub_info_t : public scrub_stamp_info_t {
 250   public:
 251     CDentry *scrub_parent;
 252     MDSInternalContextBase *on_finish;
 253
 254     bool last_scrub_dirty; /// are our stamps dirty with respect to disk state?
 255     bool scrub_in_progress; /// are we currently scrubbing?
 256     bool children_scrubbed;
 257
 258     /// my own (temporary) stamps and versions for each dirfrag we have
 259     std::map<frag_t, scrub_stamp_info_t> dirfrag_stamps;
 260
 261     ScrubHeaderRefConst header;
 262
 263     scrub_info_t() : scrub_stamp_info_t(),
 264         scrub_parent(NULL), on_finish(NULL),
 265         last_scrub_dirty(false), scrub_in_progress(false),
 266         children_scrubbed(false) {}
 267   };
 268
 269   const scrub_info_t *scrub_info() const{
 270     if (!scrub_infop)
 271       scrub_info_create();
 272     return scrub_infop;
 273   }
 274
 275   bool scrub_is_in_progress() const {
 276     return (scrub_infop && scrub_infop->scrub_in_progress);
 277   }
 278   /**
 279    * Start scrubbing on this inode. That could be very short if it's
 280    * a file, or take a long time if we're recursively scrubbing a directory.
 281    * @pre It is not currently scrubbing
 282    * @post it has set up internal scrubbing state
 283    * @param scrub_version What version are we scrubbing at (usually, parent
 284    * directory's get_projected_version())
 285    */
 286   void scrub_initialize(CDentry *scrub_parent,
 287                         const ScrubHeaderRefConst& header,
 288                         MDSInternalContextBase *f);
 289   /**
 290    * Get the next dirfrag to scrub. Gives you a frag_t in output param which
 291    * you must convert to a CDir (and possibly load off disk).
 292    * @param dir A pointer to frag_t, will be filled in with the next dirfrag to
 293    * scrub if there is one.
 294    * @returns 0 on success, you should scrub the passed-out frag_t right now;
 295    * ENOENT: There are no remaining dirfrags to scrub
 296    * <0 There was some other error (It will return -ENOTDIR if not a directory)
 297    */
 298   int scrub_dirfrag_next(frag_t* out_dirfrag);
 299   /**
 300    * Get the currently scrubbing dirfrags. When returned, the
 301    * passed-in list will be filled in with all frag_ts which have
 302    * been returned from scrub_dirfrag_next but not sent back
 303    * via scrub_dirfrag_finished.
 304    */
 305   void scrub_dirfrags_scrubbing(list<frag_t> *out_dirfrags);
 306   /**
 307    * Report to the CInode that a dirfrag it owns has been scrubbed. Call
 308    * this for every frag_t returned from scrub_dirfrag_next().
 309    * @param dirfrag The frag_t that was scrubbed
 310    */
 311   void scrub_dirfrag_finished(frag_t dirfrag);
 312   /**
 313    * Call this once the scrub has been completed, whether it's a full
 314    * recursive scrub on a directory or simply the data on a file (or
 315    * anything in between).
 316    * @param c An out param which is filled in with a Context* that must
 317    * be complete()ed.
 318    */
 319   void scrub_finished(MDSInternalContextBase **c);
 320   /**
 321    * Report to the CInode that alldirfrags it owns have been scrubbed.
 322    */
 323   void scrub_children_finished() {
 324     scrub_infop->children_scrubbed = true;
 325   }
 326   void scrub_set_finisher(MDSInternalContextBase *c) {
 327     assert(!scrub_infop->on_finish);
 328     scrub_infop->on_finish = c;
 329   }
 330
 331 private:
 332   /**
 333    * Create a scrub_info_t struct for the scrub_infop poitner.
 334    */
 335   void scrub_info_create() const;
 336   /**
 337    * Delete the scrub_info_t struct if it's not got any useful data
 338    */
 339   void scrub_maybe_delete_info();
 340 public:
 341
 342   bool is_multiversion() const {
 343     return snaprealm ||  // other snaprealms will link to me
 344       inode.is_dir() ||  // links to me in other snaps
 345       inode.nlink > 1 || // there are remote links, possibly snapped, that will need to find me
 346       !old_inodes.empty(); // once multiversion, always multiversion.  until old_inodes gets cleaned out.
 347   }
 348   snapid_t get_oldest_snap();
 349
 350   uint64_t last_journaled;       // log offset for the last time i was journaled
 351   //loff_t last_open_journaled;  // log offset for the last journaled EOpen
 352   utime_t last_dirstat_prop;
 353
 354
 355   // list item node for when we have unpropagated rstat data
 356   elist<CInode*>::item dirty_rstat_item;
 357
 358   bool is_dirty_rstat() {
 359     return state_test(STATE_DIRTYRSTAT);
 360   }
 361   void mark_dirty_rstat();
 362   void clear_dirty_rstat();
 363
 364   //bool hack_accessed;
 365   //utime_t hack_load_stamp;
 366
 367   /**
 368    * Projection methods, used to store inode changes until they have been journaled,
 369    * at which point they are popped.
 370    * Usage:
 371    * project_inode as needed. If you're also projecting xattrs, pass
 372    * in an xattr map (by pointer), then edit the map.
 373    * If you're also projecting the snaprealm, call project_snaprealm after
 374    * calling project_inode, and modify the snaprealm as necessary.
 375    *
 376    * Then, journal. Once journaling is done, pop_and_dirty_projected_inode.
 377    * This function will take care of the inode itself, the xattrs, and the snaprealm.
 378    */
 379
 380   struct projected_inode_t {
 381     inode_t *inode;
 382     std::map<std::string,bufferptr> *xattrs;
 383     sr_t *snapnode;
 384
 385     projected_inode_t()
 386       : inode(NULL), xattrs(NULL), snapnode(NULL) {}
 387     projected_inode_t(inode_t *in, sr_t *sn)
 388       : inode(in), xattrs(NULL), snapnode(sn) {}
 389     projected_inode_t(inode_t *in, std::map<std::string, bufferptr> *xp = NULL, sr_t *sn = NULL)
 390       : inode(in), xattrs(xp), snapnode(sn) {}
 391   };
 392   std::list<projected_inode_t*> projected_nodes;   // projected values (only defined while dirty)
 393   int num_projected_xattrs;
 394   int num_projected_srnodes;
 395
 396   inode_t *project_inode(std::map<std::string,bufferptr> *px=0);
 397   void pop_and_dirty_projected_inode(LogSegment *ls);
 398
 399   projected_inode_t *get_projected_node() {
 400     if (projected_nodes.empty())
 401       return NULL;
 402     else
 403       return projected_nodes.back();
 404   }
 405
 406   version_t get_projected_version() const {
 407     if (projected_nodes.empty())
 408       return inode.version;
 409     else
 410       return projected_nodes.back()->inode->version;
 411   }
 412   bool is_projected() const {
 413     return !projected_nodes.empty();
 414   }
 415
 416   const inode_t *get_projected_inode() const {
 417     if (projected_nodes.empty())
 418       return &inode;
 419     else
 420       return projected_nodes.back()->inode;
 421   }
 422   inode_t *get_projected_inode() {
 423     if (projected_nodes.empty())
 424       return &inode;
 425     else
 426       return projected_nodes.back()->inode;
 427   }
 428   inode_t *get_previous_projected_inode() {
 429     assert(!projected_nodes.empty());
 430     std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
 431     ++p;
 432     if (p != projected_nodes.rend())
 433       return (*p)->inode;
 434     else
 435       return &inode;
 436   }
 437
 438   std::map<std::string,bufferptr> *get_projected_xattrs() {
 439     if (num_projected_xattrs > 0) {
 440       for (std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
 441            p != projected_nodes.rend();
 442            ++p)
 443         if ((*p)->xattrs)
 444           return (*p)->xattrs;
 445     }
 446     return &xattrs;
 447   }
 448   std::map<std::string,bufferptr> *get_previous_projected_xattrs() {
 449       std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
 450     for (++p;  // skip the most recent projected value
 451          p != projected_nodes.rend();
 452          ++p)
 453       if ((*p)->xattrs)
 454         return (*p)->xattrs;
 455     return &xattrs;
 456   }
 457
 458   sr_t *project_snaprealm(snapid_t snapid=0);
 459   const sr_t *get_projected_srnode() const {
 460     if (num_projected_srnodes > 0) {
 461       for (std::list<projected_inode_t*>::const_reverse_iterator p = projected_nodes.rbegin();
 462           p != projected_nodes.rend();
 463           ++p)
 464         if ((*p)->snapnode)
 465           return (*p)->snapnode;
 466     }
 467     if (snaprealm)
 468       return &snaprealm->srnode;
 469     else
 470       return NULL;
 471   }
 472   sr_t *get_projected_srnode() {
 473     if (num_projected_srnodes > 0) {
 474       for (std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
 475            p != projected_nodes.rend();
 476            ++p)
 477         if ((*p)->snapnode)
 478           return (*p)->snapnode;
 479     }
 480     if (snaprealm)
 481       return &snaprealm->srnode;
 482     else
 483       return NULL;
 484   }
 485   void project_past_snaprealm_parent(SnapRealm *newparent);
 486
 487 private:
 488   void pop_projected_snaprealm(sr_t *next_snaprealm);
 489
 490 public:
 491   old_inode_t& cow_old_inode(snapid_t follows, bool cow_head);
 492   void split_old_inode(snapid_t snap);
 493   old_inode_t *pick_old_inode(snapid_t last);
 494   void pre_cow_old_inode();
 495   void purge_stale_snap_data(const std::set<snapid_t>& snaps);
 496
 497   // -- cache infrastructure --
 498 private:
 499   compact_map<frag_t,CDir*> dirfrags; // cached dir fragments under this Inode
 500   int stickydir_ref;
 501   scrub_info_t *scrub_infop;
 502
 503 public:
 504   bool has_dirfrags() { return !dirfrags.empty(); }
 505   CDir* get_dirfrag(frag_t fg) {
 506     if (dirfrags.count(fg)) {
 507       //assert(g_conf->debug_mds < 2 || dirfragtree.is_leaf(fg)); // performance hack FIXME
 508       return dirfrags[fg];
 509     } else
 510       return NULL;
 511   }
 512   bool get_dirfrags_under(frag_t fg, std::list<CDir*>& ls);
 513   CDir* get_approx_dirfrag(frag_t fg);
 514   void get_dirfrags(std::list<CDir*>& ls);
 515   void get_nested_dirfrags(std::list<CDir*>& ls);
 516   void get_subtree_dirfrags(std::list<CDir*>& ls);
 517   CDir *get_or_open_dirfrag(MDCache *mdcache, frag_t fg);
 518   CDir *add_dirfrag(CDir *dir);
 519   void close_dirfrag(frag_t fg);
 520   void close_dirfrags();
 521   bool has_subtree_root_dirfrag(int auth=-1);
 522   bool has_subtree_or_exporting_dirfrag();
 523
 524   void force_dirfrags();
 525   void verify_dirfrags();
 526
 527   void get_stickydirs();
 528   void put_stickydirs();
 529
 530  protected:
 531   // parent dentries in cache
 532   CDentry         *parent;             // primary link
 533   compact_set<CDentry*>    remote_parents;     // if hard linked
 534
 535   std::list<CDentry*>   projected_parent;   // for in-progress rename, (un)link, etc.
 536
 537   mds_authority_t inode_auth;
 538
 539   // -- distributed state --
 540 protected:
 541   // file capabilities
 542   std::map<client_t, Capability*> client_caps;         // client -> caps
 543   compact_map<int32_t, int32_t>      mds_caps_wanted;     // [auth] mds -> caps wanted
 544   int                   replica_caps_wanted; // [replica] what i've requested from auth
 545
 546 public:
 547   compact_map<int, std::set<client_t> > client_snap_caps;     // [auth] [snap] dirty metadata we still need from the head
 548   compact_map<snapid_t, std::set<client_t> > client_need_snapflush;
 549
 550   void add_need_snapflush(CInode *snapin, snapid_t snapid, client_t client);
 551   void remove_need_snapflush(CInode *snapin, snapid_t snapid, client_t client);
 552   bool split_need_snapflush(CInode *cowin, CInode *in);
 553
 554 protected:
 555
 556   ceph_lock_state_t *fcntl_locks;
 557   ceph_lock_state_t *flock_locks;
 558
 559   ceph_lock_state_t *get_fcntl_lock_state() {
 560     if (!fcntl_locks)
 561       fcntl_locks = new ceph_lock_state_t(g_ceph_context, CEPH_LOCK_FCNTL);
 562     return fcntl_locks;
 563   }
 564   void clear_fcntl_lock_state() {
 565     delete fcntl_locks;
 566     fcntl_locks = NULL;
 567   }
 568   ceph_lock_state_t *get_flock_lock_state() {
 569     if (!flock_locks)
 570       flock_locks = new ceph_lock_state_t(g_ceph_context, CEPH_LOCK_FLOCK);
 571     return flock_locks;
 572   }
 573   void clear_flock_lock_state() {
 574     delete flock_locks;
 575     flock_locks = NULL;
 576   }
 577   void clear_file_locks() {
 578     clear_fcntl_lock_state();
 579     clear_flock_lock_state();
 580   }
 581   void _encode_file_locks(bufferlist& bl) const {
 582     bool has_fcntl_locks = fcntl_locks && !fcntl_locks->empty();
 583     ::encode(has_fcntl_locks, bl);
 584     if (has_fcntl_locks)
 585       ::encode(*fcntl_locks, bl);
 586     bool has_flock_locks = flock_locks && !flock_locks->empty();
 587     ::encode(has_flock_locks, bl);
 588     if (has_flock_locks)
 589       ::encode(*flock_locks, bl);
 590   }
 591   void _decode_file_locks(bufferlist::iterator& p) {
 592     bool has_fcntl_locks;
 593     ::decode(has_fcntl_locks, p);
 594     if (has_fcntl_locks)
 595       ::decode(*get_fcntl_lock_state(), p);
 596     else
 597       clear_fcntl_lock_state();
 598     bool has_flock_locks;
 599     ::decode(has_flock_locks, p);
 600     if (has_flock_locks)
 601       ::decode(*get_flock_lock_state(), p);
 602     else
 603       clear_flock_lock_state();
 604   }
 605
 606   // LogSegment lists i (may) belong to
 607 public:
 608   elist<CInode*>::item item_dirty;
 609   elist<CInode*>::item item_caps;
 610   elist<CInode*>::item item_open_file;
 611   elist<CInode*>::item item_dirty_parent;
 612   elist<CInode*>::item item_dirty_dirfrag_dir;
 613   elist<CInode*>::item item_dirty_dirfrag_nest;
 614   elist<CInode*>::item item_dirty_dirfrag_dirfragtree;
 615   elist<CInode*>::item item_scrub;
 616
 617 public:
 618   int auth_pin_freeze_allowance;
 619
 620   inode_load_vec_t pop;
 621
 622   // friends
 623   friend class Server;
 624   friend class Locker;
 625   friend class Migrator;
 626   friend class MDCache;
 627   friend class StrayManager;
 628   friend class CDir;
 629   friend class CInodeExport;
 630   friend class C_CInode_ExportPin;
 631
 632   // ---------------------------
 633   CInode(MDCache *c, bool auth=true, snapid_t f=2, snapid_t l=CEPH_NOSNAP) :
 634     mdcache(c),
 635     snaprealm(0), containing_realm(0),
 636     first(f), last(l),
 637     last_journaled(0), //last_open_journaled(0),
 638     //hack_accessed(true),
 639     num_projected_xattrs(0),
 640     num_projected_srnodes(0),
 641     stickydir_ref(0),
 642     scrub_infop(NULL),
 643     parent(0),
 644     inode_auth(CDIR_AUTH_DEFAULT),
 645     replica_caps_wanted(0),
 646     fcntl_locks(0), flock_locks(0),
 647     item_dirty(this), item_caps(this), item_open_file(this), item_dirty_parent(this),
 648     item_dirty_dirfrag_dir(this),
 649     item_dirty_dirfrag_nest(this),
 650     item_dirty_dirfrag_dirfragtree(this),
 651     auth_pin_freeze_allowance(0),
 652     pop(ceph_clock_now()),
 653     versionlock(this, &versionlock_type),
 654     authlock(this, &authlock_type),
 655     linklock(this, &linklock_type),
 656     dirfragtreelock(this, &dirfragtreelock_type),
 657     filelock(this, &filelock_type),
 658     xattrlock(this, &xattrlock_type),
 659     snaplock(this, &snaplock_type),
 660     nestlock(this, &nestlock_type),
 661     flocklock(this, &flocklock_type),
 662     policylock(this, &policylock_type),
 663     loner_cap(-1), want_loner_cap(-1)
 664   {
 665     state = 0;
 666     if (auth) state_set(STATE_AUTH);
 667   }
 668   ~CInode() override {
 669     close_dirfrags();
 670     close_snaprealm();
 671     clear_file_locks();
 672     assert(num_projected_xattrs == 0);
 673     assert(num_projected_srnodes == 0);
 674   }
 675
 676
 677   // -- accessors --
 678   bool is_root() const { return inode.ino == MDS_INO_ROOT; }
 679   bool is_stray() const { return MDS_INO_IS_STRAY(inode.ino); }
 680   mds_rank_t get_stray_owner() const {
 681     return (mds_rank_t)MDS_INO_STRAY_OWNER(inode.ino);
 682   }
 683   bool is_mdsdir() const { return MDS_INO_IS_MDSDIR(inode.ino); }
 684   bool is_base() const { return is_root() || is_mdsdir(); }
 685   bool is_system() const { return inode.ino < MDS_INO_SYSTEM_BASE; }
 686   bool is_normal() const { return !(is_base() || is_system() || is_stray()); }
 687
 688   bool is_head() const { return last == CEPH_NOSNAP; }
 689
 690   // note: this overloads MDSCacheObject
 691   bool is_ambiguous_auth() const {
 692     return state_test(STATE_AMBIGUOUSAUTH) ||
 693       MDSCacheObject::is_ambiguous_auth();
 694   }
 695   void set_ambiguous_auth() {
 696     state_set(STATE_AMBIGUOUSAUTH);
 697   }
 698   void clear_ambiguous_auth(std::list<MDSInternalContextBase*>& finished);
 699   void clear_ambiguous_auth();
 700
 701   inodeno_t ino() const { return inode.ino; }
 702   vinodeno_t vino() const { return vinodeno_t(inode.ino, last); }
 703   int d_type() const { return IFTODT(inode.mode); }
 704
 705   inode_t& get_inode() { return inode; }
 706   CDentry* get_parent_dn() { return parent; }
 707   const CDentry* get_parent_dn() const { return parent; }
 708   const CDentry* get_projected_parent_dn() const { return !projected_parent.empty() ? projected_parent.back() : parent; }
 709   CDentry* get_projected_parent_dn() { return !projected_parent.empty() ? projected_parent.back() : parent; }
 710   CDir *get_parent_dir();
 711   const CDir *get_projected_parent_dir() const;
 712   CDir *get_projected_parent_dir();
 713   CInode *get_parent_inode();
 714
 715   bool is_lt(const MDSCacheObject *r) const override {
 716     const CInode *o = static_cast<const CInode*>(r);
 717     return ino() < o->ino() ||
 718       (ino() == o->ino() && last < o->last);
 719   }
 720
 721   // -- misc --
 722   bool is_projected_ancestor_of(CInode *other);
 723
 724   void make_path_string(std::string& s, bool projected=false, const CDentry *use_parent=NULL) const;
 725   void make_path(filepath& s, bool projected=false) const;
 726   void name_stray_dentry(std::string& dname);
 727
 728   // -- dirtyness --
 729   version_t get_version() const { return inode.version; }
 730
 731   version_t pre_dirty();
 732   void _mark_dirty(LogSegment *ls);
 733   void mark_dirty(version_t projected_dirv, LogSegment *ls);
 734   void mark_clean();
 735
 736   void store(MDSInternalContextBase *fin);
 737   void _stored(int r, version_t cv, Context *fin);
 738   /**
 739    * Flush a CInode to disk. This includes the backtrace, the parent
 740    * directory's link, and the Inode object itself (if a base directory).
 741    * @pre is_auth() on both the inode and its containing directory
 742    * @pre can_auth_pin()
 743    * @param fin The Context to call when the flush is completed.
 744    */
 745   void flush(MDSInternalContextBase *fin);
 746   void fetch(MDSInternalContextBase *fin);
 747   void _fetched(bufferlist& bl, bufferlist& bl2, Context *fin);
 748
 749
 750   void build_backtrace(int64_t pool, inode_backtrace_t& bt);
 751   void store_backtrace(MDSInternalContextBase *fin, int op_prio=-1);
 752   void _stored_backtrace(int r, version_t v, Context *fin);
 753   void fetch_backtrace(Context *fin, bufferlist *backtrace);
 754 protected:
 755   /**
 756    * Return the pool ID where we currently write backtraces for
 757    * this inode (in addition to inode.old_pools)
 758    *
 759    * @returns a pool ID >=0
 760    */
 761   int64_t get_backtrace_pool() const;
 762 public:
 763   void _mark_dirty_parent(LogSegment *ls, bool dirty_pool=false);
 764   void clear_dirty_parent();
 765   void verify_diri_backtrace(bufferlist &bl, int err);
 766   bool is_dirty_parent() { return state_test(STATE_DIRTYPARENT); }
 767   bool is_dirty_pool() { return state_test(STATE_DIRTYPOOL); }
 768
 769   void encode_snap_blob(bufferlist &bl);
 770   void decode_snap_blob(bufferlist &bl);
 771   void encode_store(bufferlist& bl, uint64_t features);
 772   void decode_store(bufferlist::iterator& bl);
 773
 774   void encode_replica(mds_rank_t rep, bufferlist& bl, uint64_t features) {
 775     assert(is_auth());
 776
 777     // relax locks?
 778     if (!is_replicated())
 779       replicate_relax_locks();
 780
 781     __u32 nonce = add_replica(rep);
 782     ::encode(nonce, bl);
 783
 784     _encode_base(bl, features);
 785     _encode_locks_state_for_replica(bl);
 786   }
 787   void decode_replica(bufferlist::iterator& p, bool is_new) {
 788     __u32 nonce;
 789     ::decode(nonce, p);
 790     replica_nonce = nonce;
 791
 792     _decode_base(p);
 793     _decode_locks_state(p, is_new);
 794   }
 795
 796   // -- waiting --
 797 protected:
 798   compact_map<frag_t, std::list<MDSInternalContextBase*> > waiting_on_dir;
 799 public:
 800   void add_dir_waiter(frag_t fg, MDSInternalContextBase *c);
 801   void take_dir_waiting(frag_t fg, std::list<MDSInternalContextBase*>& ls);
 802   bool is_waiting_for_dir(frag_t fg) {
 803     return waiting_on_dir.count(fg);
 804   }
 805   void add_waiter(uint64_t tag, MDSInternalContextBase *c) override;
 806   void take_waiting(uint64_t tag, std::list<MDSInternalContextBase*>& ls) override;
 807
 808   // -- encode/decode helpers --
 809   void _encode_base(bufferlist& bl, uint64_t features);
 810   void _decode_base(bufferlist::iterator& p);
 811   void _encode_locks_full(bufferlist& bl);
 812   void _decode_locks_full(bufferlist::iterator& p);
 813   void _encode_locks_state_for_replica(bufferlist& bl);
 814   void _encode_locks_state_for_rejoin(bufferlist& bl, int rep);
 815   void _decode_locks_state(bufferlist::iterator& p, bool is_new);
 816   void _decode_locks_rejoin(bufferlist::iterator& p, std::list<MDSInternalContextBase*>& waiters,
 817                             std::list<SimpleLock*>& eval_locks);
 818
 819   // -- import/export --
 820   void encode_export(bufferlist& bl);
 821   void finish_export(utime_t now);
 822   void abort_export() {
 823     put(PIN_TEMPEXPORTING);
 824     assert(state_test(STATE_EXPORTINGCAPS));
 825     state_clear(STATE_EXPORTINGCAPS);
 826     put(PIN_EXPORTINGCAPS);
 827   }
 828   void decode_import(bufferlist::iterator& p, LogSegment *ls);
 829
 830
 831   // for giving to clients
 832   int encode_inodestat(bufferlist& bl, Session *session, SnapRealm *realm,
 833                        snapid_t snapid=CEPH_NOSNAP, unsigned max_bytes=0,
 834                        int getattr_wants=0);
 835   void encode_cap_message(MClientCaps *m, Capability *cap);
 836
 837
 838   // -- locks --
 839 public:
 840   static LockType versionlock_type;
 841   static LockType authlock_type;
 842   static LockType linklock_type;
 843   static LockType dirfragtreelock_type;
 844   static LockType filelock_type;
 845   static LockType xattrlock_type;
 846   static LockType snaplock_type;
 847   static LockType nestlock_type;
 848   static LockType flocklock_type;
 849   static LockType policylock_type;
 850
 851   LocalLock  versionlock;
 852   SimpleLock authlock;
 853   SimpleLock linklock;
 854   ScatterLock dirfragtreelock;
 855   ScatterLock filelock;
 856   SimpleLock xattrlock;
 857   SimpleLock snaplock;
 858   ScatterLock nestlock;
 859   SimpleLock flocklock;
 860   SimpleLock policylock;
 861
 862   SimpleLock* get_lock(int type) override {
 863     switch (type) {
 864     case CEPH_LOCK_IFILE: return &filelock;
 865     case CEPH_LOCK_IAUTH: return &authlock;
 866     case CEPH_LOCK_ILINK: return &linklock;
 867     case CEPH_LOCK_IDFT: return &dirfragtreelock;
 868     case CEPH_LOCK_IXATTR: return &xattrlock;
 869     case CEPH_LOCK_ISNAP: return &snaplock;
 870     case CEPH_LOCK_INEST: return &nestlock;
 871     case CEPH_LOCK_IFLOCK: return &flocklock;
 872     case CEPH_LOCK_IPOLICY: return &policylock;
 873     }
 874     return 0;
 875   }
 876
 877   void set_object_info(MDSCacheObjectInfo &info) override;
 878   void encode_lock_state(int type, bufferlist& bl) override;
 879   void decode_lock_state(int type, bufferlist& bl) override;
 880
 881   void _finish_frag_update(CDir *dir, MutationRef& mut);
 882
 883   void clear_dirty_scattered(int type) override;
 884   bool is_dirty_scattered();
 885   void clear_scatter_dirty();  // on rejoin ack
 886
 887   void start_scatter(ScatterLock *lock);
 888   void finish_scatter_update(ScatterLock *lock, CDir *dir,
 889                              version_t inode_version, version_t dir_accounted_version);
 890   void finish_scatter_gather_update(int type);
 891   void finish_scatter_gather_update_accounted(int type, MutationRef& mut, EMetaBlob *metablob);
 892
 893   // -- snap --
 894   void open_snaprealm(bool no_split=false);
 895   void close_snaprealm(bool no_join=false);
 896   SnapRealm *find_snaprealm() const;
 897   void encode_snap(bufferlist& bl);
 898   void decode_snap(bufferlist::iterator& p);
 899
 900   // -- caps -- (new)
 901   // client caps
 902   client_t loner_cap, want_loner_cap;
 903
 904   client_t get_loner() const { return loner_cap; }
 905   client_t get_wanted_loner() const { return want_loner_cap; }
 906
 907   // this is the loner state our locks should aim for
 908   client_t get_target_loner() const {
 909     if (loner_cap == want_loner_cap)
 910       return loner_cap;
 911     else
 912       return -1;
 913   }
 914
 915   client_t calc_ideal_loner();
 916   client_t choose_ideal_loner();
 917   bool try_set_loner();
 918   void set_loner_cap(client_t l);
 919   bool try_drop_loner();
 920
 921   // choose new lock state during recovery, based on issued caps
 922   void choose_lock_state(SimpleLock *lock, int allissued);
 923   void choose_lock_states(int dirty_caps);
 924
 925   int count_nonstale_caps() {
 926     int n = 0;
 927     for (std::map<client_t,Capability*>::iterator it = client_caps.begin();
 928          it != client_caps.end();
 929          ++it)
 930       if (!it->second->is_stale())
 931         n++;
 932     return n;
 933   }
 934   bool multiple_nonstale_caps() {
 935     int n = 0;
 936     for (std::map<client_t,Capability*>::iterator it = client_caps.begin();
 937          it != client_caps.end();
 938          ++it)
 939       if (!it->second->is_stale()) {
 940         if (n)
 941           return true;
 942         n++;
 943       }
 944     return false;
 945   }
 946
 947   bool is_any_caps() { return !client_caps.empty(); }
 948   bool is_any_nonstale_caps() { return count_nonstale_caps(); }
 949
 950   const compact_map<int32_t,int32_t>& get_mds_caps_wanted() const { return mds_caps_wanted; }
 951   compact_map<int32_t,int32_t>& get_mds_caps_wanted() { return mds_caps_wanted; }
 952
 953   const std::map<client_t,Capability*>& get_client_caps() const { return client_caps; }
 954   Capability *get_client_cap(client_t client) {
 955     auto client_caps_entry = client_caps.find(client);
 956     if (client_caps_entry != client_caps.end())
 957       return client_caps_entry->second;
 958     return 0;
 959   }
 960   int get_client_cap_pending(client_t client) const {
 961     auto client_caps_entry = client_caps.find(client);
 962     if (client_caps_entry != client_caps.end()) {
 963       return client_caps_entry->second->pending();
 964     } else {
 965       return 0;
 966     }
 967   }
 968
 969   Capability *add_client_cap(client_t client, Session *session, SnapRealm *conrealm=0);
 970   void remove_client_cap(client_t client);
 971   void move_to_realm(SnapRealm *realm);
 972
 973   Capability *reconnect_cap(client_t client, const cap_reconnect_t& icr, Session *session);
 974   void clear_client_caps_after_export();
 975   void export_client_caps(std::map<client_t,Capability::Export>& cl);
 976
 977   // caps allowed
 978   int get_caps_liked() const;
 979   int get_caps_allowed_ever() const;
 980   int get_caps_allowed_by_type(int type) const;
 981   int get_caps_careful() const;
 982   int get_xlocker_mask(client_t client) const;
 983   int get_caps_allowed_for_client(Session *s, inode_t *file_i) const;
 984
 985   // caps issued, wanted
 986   int get_caps_issued(int *ploner = 0, int *pother = 0, int *pxlocker = 0,
 987                       int shift = 0, int mask = -1);
 988   bool is_any_caps_wanted() const;
 989   int get_caps_wanted(int *ploner = 0, int *pother = 0, int shift = 0, int mask = -1) const;
 990   bool issued_caps_need_gather(SimpleLock *lock);
 991   void replicate_relax_locks();
 992
 993   // -- authority --
 994   mds_authority_t authority() const override;
 995
 996   // -- auth pins --
 997   void adjust_nested_auth_pins(int a, void *by);
 998   bool can_auth_pin() const override;
 999   void auth_pin(void *by) override;
1000   void auth_unpin(void *by) override;
1001
1002   // -- freeze --
1003   bool is_freezing_inode() const { return state_test(STATE_FREEZING); }
1004   bool is_frozen_inode() const { return state_test(STATE_FROZEN); }
1005   bool is_frozen_auth_pin() const { return state_test(STATE_FROZENAUTHPIN); }
1006   bool is_frozen() const override;
1007   bool is_frozen_dir() const;
1008   bool is_freezing() const override;
1009
1010   /* Freeze the inode. auth_pin_allowance lets the caller account for any
1011    * auth_pins it is itself holding/responsible for. */
1012   bool freeze_inode(int auth_pin_allowance=0);
1013   void unfreeze_inode(std::list<MDSInternalContextBase*>& finished);
1014   void unfreeze_inode();
1015
1016   void freeze_auth_pin();
1017   void unfreeze_auth_pin();
1018
1019   // -- reference counting --
1020   void bad_put(int by) override {
1021     generic_dout(0) << " bad put " << *this << " by " << by << " " << pin_name(by) << " was " << ref
1022 #ifdef MDS_REF_SET
1023                     << " (" << ref_map << ")"
1024 #endif
1025                     << dendl;
1026 #ifdef MDS_REF_SET
1027     assert(ref_map[by] > 0);
1028 #endif
1029     assert(ref > 0);
1030   }
1031   void bad_get(int by) override {
1032     generic_dout(0) << " bad get " << *this << " by " << by << " " << pin_name(by) << " was " << ref
1033 #ifdef MDS_REF_SET
1034                     << " (" << ref_map << ")"
1035 #endif
1036                     << dendl;
1037 #ifdef MDS_REF_SET
1038     assert(ref_map[by] >= 0);
1039 #endif
1040   }
1041   void first_get() override;
1042   void last_put() override;
1043   void _put() override;
1044
1045
1046   // -- hierarchy stuff --
1047 public:
1048   void set_primary_parent(CDentry *p) {
1049     assert(parent == 0);
1050     parent = p;
1051   }
1052   void remove_primary_parent(CDentry *dn) {
1053     assert(dn == parent);
1054     parent = 0;
1055   }
1056   void add_remote_parent(CDentry *p);
1057   void remove_remote_parent(CDentry *p);
1058   int num_remote_parents() {
1059     return remote_parents.size();
1060   }
1061
1062   void push_projected_parent(CDentry *dn) {
1063     projected_parent.push_back(dn);
1064   }
1065   void pop_projected_parent() {
1066     assert(projected_parent.size());
1067     parent = projected_parent.front();
1068     projected_parent.pop_front();
1069   }
1070
1071 private:
1072   void maybe_export_pin();
1073 public:
1074   void set_export_pin(mds_rank_t rank);
1075   mds_rank_t get_export_pin(bool inherit=true) const;
1076   bool is_exportable(mds_rank_t dest) const;
1077
1078   void print(ostream& out) override;
1079   void dump(Formatter *f) const;
1080
1081   /**
1082    * @defgroup Scrubbing and fsck
1083    * @{
1084    */
1085
1086   /**
1087    * Report the results of validation against a particular inode.
1088    * Each member is a pair of bools.
1089    * <member>.first represents if validation was performed against the member.
1090    * <member.second represents if the member passed validation.
1091    * performed_validation is set to true if the validation was actually
1092    * run. It might not be run if, for instance, the inode is marked as dirty.
1093    * passed_validation is set to true if everything that was checked
1094    * passed its validation.
1095    */
1096   struct validated_data {
1097     template<typename T>struct member_status {
1098       bool checked;
1099       bool passed;
1100       int ondisk_read_retval;
1101       T ondisk_value;
1102       T memory_value;
1103       std::stringstream error_str;
1104       member_status() : checked(false), passed(false),
1105           ondisk_read_retval(0) {}
1106     };
1107
1108     bool performed_validation;
1109     bool passed_validation;
1110
1111     struct raw_stats_t {
1112       frag_info_t dirstat;
1113       nest_info_t rstat;
1114     };
1115
1116     member_status<inode_backtrace_t> backtrace;
1117     member_status<inode_t> inode;
1118     member_status<raw_stats_t> raw_stats;
1119
1120     validated_data() : performed_validation(false),
1121         passed_validation(false) {}
1122
1123     void dump(Formatter *f) const;
1124   };
1125
1126   /**
1127    * Validate that the on-disk state of an inode matches what
1128    * we expect from our memory state. Currently this checks that:
1129    * 1) The backtrace associated with the file data exists and is correct
1130    * 2) For directories, the actual inode metadata matches our memory state,
1131    * 3) For directories, the rstats match
1132    *
1133    * @param results A freshly-created validated_data struct, with values set
1134    * as described in the struct documentation.
1135    * @param mdr The request to be responeded upon the completion of the
1136    * validation (or NULL)
1137    * @param fin Context to call back on completion (or NULL)
1138    */
1139   void validate_disk_state(validated_data *results,
1140                            MDSInternalContext *fin);
1141   static void dump_validation_results(const validated_data& results,
1142                                       Formatter *f);
1143 private:
1144   bool _validate_disk_state(class ValidationContinuation *c,
1145                             int rval, int stage);
1146   friend class ValidationContinuation;
1147   /** @} Scrubbing and fsck */
1148 };
1149
1150 ostream& operator<<(ostream& out, const CInode::scrub_stamp_info_t& si);
1151
1152 #undef dout_context
1153 #endif