ceph/src/client/Client.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 #ifndef CEPH_CLIENT_H
  17 #define CEPH_CLIENT_H
  18
  19 #include "include/types.h"
  20
  21 // stl
  22 #include <string>
  23 #include <memory>
  24 #include <set>
  25 #include <map>
  26 #include <fstream>
  27 using std::set;
  28 using std::map;
  29 using std::fstream;
  30
  31 #include "include/unordered_set.h"
  32 #include "include/unordered_map.h"
  33 #include "include/filepath.h"
  34 #include "include/interval_set.h"
  35 #include "include/lru.h"
  36 #include "mds/mdstypes.h"
  37 #include "msg/Dispatcher.h"
  38 #include "msg/Messenger.h"
  39
  40 #include "common/Mutex.h"
  41 #include "common/Timer.h"
  42 #include "common/Finisher.h"
  43 #include "common/compiler_extensions.h"
  44 #include "common/cmdparse.h"
  45 #include "common/CommandTable.h"
  46
  47 #include "osdc/ObjectCacher.h"
  48
  49 #include "InodeRef.h"
  50 #include "UserPerm.h"
  51 #include "include/cephfs/ceph_statx.h"
  52
  53 class FSMap;
  54 class FSMapUser;
  55 class MonClient;
  56
  57 class CephContext;
  58 class MClientReply;
  59 class MClientRequest;
  60 class MClientSession;
  61 class MClientRequest;
  62 class MClientRequestForward;
  63 struct MClientLease;
  64 class MClientCaps;
  65 class MClientCapRelease;
  66
  67 struct DirStat;
  68 struct LeaseStat;
  69 struct InodeStat;
  70
  71 class Filer;
  72 class Objecter;
  73 class WritebackHandler;
  74
  75 class PerfCounters;
  76 class MDSMap;
  77 class Message;
  78
  79 enum {
  80   l_c_first = 20000,
  81   l_c_reply,
  82   l_c_lat,
  83   l_c_wrlat,
  84   l_c_last,
  85 };
  86
  87
  88 class MDSCommandOp : public CommandOp
  89 {
  90   public:
  91   mds_gid_t     mds_gid;
  92
  93   MDSCommandOp(ceph_tid_t t) : CommandOp(t) {}
  94 };
  95
  96 /* error code for ceph_fuse */
  97 #define CEPH_FUSE_NO_MDS_UP    -(1<<2) /* no mds up deteced in ceph_fuse */
  98
  99 // ============================================
 100 // types for my local metadata cache
 101 /* basic structure:
 102
 103  - Dentries live in an LRU loop.  they get expired based on last access.
 104       see include/lru.h.  items can be bumped to "mid" or "top" of list, etc.
 105  - Inode has ref count for each Fh, Dir, or Dentry that points to it.
 106  - when Inode ref goes to 0, it's expired.
 107  - when Dir is empty, it's removed (and it's Inode ref--)
 108
 109 */
 110
 111 /* getdir result */
 112 struct DirEntry {
 113   string d_name;
 114   struct stat st;
 115   int stmask;
 116   explicit DirEntry(const string &s) : d_name(s), stmask(0) {}
 117   DirEntry(const string &n, struct stat& s, int stm) : d_name(n), st(s), stmask(stm) {}
 118 };
 119
 120 struct Cap;
 121 class Dir;
 122 class Dentry;
 123 struct SnapRealm;
 124 struct Fh;
 125 struct CapSnap;
 126
 127 struct MetaSession;
 128 struct MetaRequest;
 129 class ceph_lock_state_t;
 130
 131
 132 typedef void (*client_ino_callback_t)(void *handle, vinodeno_t ino, int64_t off, int64_t len);
 133
 134 typedef void (*client_dentry_callback_t)(void *handle, vinodeno_t dirino,
 135                                          vinodeno_t ino, string& name);
 136 typedef int (*client_remount_callback_t)(void *handle);
 137
 138 typedef int (*client_getgroups_callback_t)(void *handle, gid_t **sgids);
 139 typedef void(*client_switch_interrupt_callback_t)(void *handle, void *data);
 140 typedef mode_t (*client_umask_callback_t)(void *handle);
 141
 142 struct client_callback_args {
 143   void *handle;
 144   client_ino_callback_t ino_cb;
 145   client_dentry_callback_t dentry_cb;
 146   client_switch_interrupt_callback_t switch_intr_cb;
 147   client_remount_callback_t remount_cb;
 148   client_getgroups_callback_t getgroups_cb;
 149   client_umask_callback_t umask_cb;
 150 };
 151
 152 // ========================================================
 153 // client interface
 154
 155 struct dir_result_t {
 156   static const int SHIFT = 28;
 157   static const int64_t MASK = (1 << SHIFT) - 1;
 158   static const int64_t HASH = 0xFFULL << (SHIFT + 24); // impossible frag bits
 159   static const loff_t END = 1ULL << (SHIFT + 32);
 160
 161   static uint64_t make_fpos(unsigned h, unsigned l, bool hash) {
 162     uint64_t v =  ((uint64_t)h<< SHIFT) | (uint64_t)l;
 163     if (hash)
 164       v |= HASH;
 165     else
 166       assert((v & HASH) != HASH);
 167     return v;
 168   }
 169   static unsigned fpos_high(uint64_t p) {
 170     unsigned v = (p & (END-1)) >> SHIFT;
 171     if ((p & HASH) == HASH)
 172       return ceph_frag_value(v);
 173     return v;
 174   }
 175   static unsigned fpos_low(uint64_t p) {
 176     return p & MASK;
 177   }
 178   static int fpos_cmp(uint64_t l, uint64_t r) {
 179     int c = ceph_frag_compare(fpos_high(l), fpos_high(r));
 180     if (c)
 181       return c;
 182     if (fpos_low(l) == fpos_low(r))
 183       return 0;
 184     return fpos_low(l) < fpos_low(r) ? -1 : 1;
 185   }
 186
 187   InodeRef inode;
 188   int64_t offset;        // hash order:
 189                          //   (0xff << 52) | ((24 bits hash) << 28) |
 190                          //   (the nth entry has hash collision);
 191                          // frag+name order;
 192                          //   ((frag value) << 28) | (the nth entry in frag);
 193
 194   unsigned next_offset;  // offset of next chunk (last_name's + 1)
 195   string last_name;      // last entry in previous chunk
 196
 197   uint64_t release_count;
 198   uint64_t ordered_count;
 199   unsigned cache_index;
 200   int start_shared_gen;  // dir shared_gen at start of readdir
 201   UserPerm perms;
 202
 203   frag_t buffer_frag;
 204
 205   struct dentry {
 206     int64_t offset;
 207     string name;
 208     InodeRef inode;
 209     dentry(int64_t o) : offset(o) {}
 210     dentry(int64_t o, const string& n, const InodeRef& in) :
 211       offset(o), name(n), inode(in) {}
 212   };
 213   struct dentry_off_lt {
 214     bool operator()(const dentry& d, int64_t off) const {
 215       return dir_result_t::fpos_cmp(d.offset, off) < 0;
 216     }
 217   };
 218   vector<dentry> buffer;
 219
 220   explicit dir_result_t(Inode *in, const UserPerm& perms);
 221
 222   unsigned offset_high() { return fpos_high(offset); }
 223   unsigned offset_low() { return fpos_low(offset); }
 224
 225   void set_end() { offset |= END; }
 226   bool at_end() { return (offset & END); }
 227
 228   void set_hash_order() { offset |= HASH; }
 229   bool hash_order() { return (offset & HASH) == HASH; }
 230
 231   bool is_cached() {
 232     if (buffer.empty())
 233       return false;
 234     if (hash_order()) {
 235       return buffer_frag.contains(offset_high());
 236     } else {
 237       return buffer_frag == frag_t(offset_high());
 238     }
 239   }
 240
 241   void reset() {
 242     last_name.clear();
 243     next_offset = 2;
 244     offset = 0;
 245     ordered_count = 0;
 246     cache_index = 0;
 247     buffer.clear();
 248   }
 249 };
 250
 251 class Client : public Dispatcher, public md_config_obs_t {
 252  public:
 253   using Dispatcher::cct;
 254
 255   std::unique_ptr<PerfCounters> logger;
 256
 257   class CommandHook : public AdminSocketHook {
 258     Client *m_client;
 259   public:
 260     explicit CommandHook(Client *client);
 261     bool call(std::string command, cmdmap_t &cmdmap, std::string format,
 262               bufferlist& out) override;
 263   };
 264   CommandHook m_command_hook;
 265
 266   // cluster descriptors
 267   std::unique_ptr<MDSMap> mdsmap;
 268
 269   SafeTimer timer;
 270
 271   void *callback_handle;
 272   client_switch_interrupt_callback_t switch_interrupt_cb;
 273   client_remount_callback_t remount_cb;
 274   client_ino_callback_t ino_invalidate_cb;
 275   client_dentry_callback_t dentry_invalidate_cb;
 276   client_getgroups_callback_t getgroups_cb;
 277   client_umask_callback_t umask_cb;
 278   bool can_invalidate_dentries;
 279   bool require_remount;
 280
 281   Finisher async_ino_invalidator;
 282   Finisher async_dentry_invalidator;
 283   Finisher interrupt_finisher;
 284   Finisher remount_finisher;
 285   Finisher objecter_finisher;
 286
 287   Context *tick_event;
 288   utime_t last_cap_renew;
 289   void renew_caps();
 290   void renew_caps(MetaSession *session);
 291   void flush_cap_releases();
 292 public:
 293   void tick();
 294
 295   UserPerm pick_my_perms() {
 296     uid_t uid = user_id >= 0 ? user_id : -1;
 297     gid_t gid = group_id >= 0 ? group_id : -1;
 298     return UserPerm(uid, gid);
 299   }
 300
 301   static UserPerm pick_my_perms(CephContext *c) {
 302     uid_t uid = c->_conf->client_mount_uid >= 0 ? c->_conf->client_mount_uid : -1;
 303     gid_t gid = c->_conf->client_mount_gid >= 0 ? c->_conf->client_mount_gid : -1;
 304     return UserPerm(uid, gid);
 305   }
 306 protected:
 307   Messenger *messenger;
 308   MonClient *monclient;
 309   Objecter  *objecter;
 310
 311   client_t whoami;
 312
 313   int user_id, group_id;
 314   int acl_type;
 315
 316   void set_cap_epoch_barrier(epoch_t e);
 317   epoch_t cap_epoch_barrier;
 318
 319   // mds sessions
 320   map<mds_rank_t, MetaSession*> mds_sessions;  // mds -> push seq
 321   list<Cond*> waiting_for_mdsmap;
 322
 323   // FSMap, for when using mds_command
 324   list<Cond*> waiting_for_fsmap;
 325   std::unique_ptr<FSMap> fsmap;
 326   std::unique_ptr<FSMapUser> fsmap_user;
 327
 328   // MDS command state
 329   CommandTable<MDSCommandOp> command_table;
 330   void handle_command_reply(MCommandReply *m);
 331   int fetch_fsmap(bool user);
 332   int resolve_mds(
 333       const std::string &mds_spec,
 334       std::vector<mds_gid_t> *targets);
 335
 336   void get_session_metadata(std::map<std::string, std::string> *meta) const;
 337   bool have_open_session(mds_rank_t mds);
 338   void got_mds_push(MetaSession *s);
 339   MetaSession *_get_mds_session(mds_rank_t mds, Connection *con);  ///< return session for mds *and* con; null otherwise
 340   MetaSession *_get_or_open_mds_session(mds_rank_t mds);
 341   MetaSession *_open_mds_session(mds_rank_t mds);
 342   void _close_mds_session(MetaSession *s);
 343   void _closed_mds_session(MetaSession *s);
 344   bool _any_stale_sessions() const;
 345   void _kick_stale_sessions();
 346   void handle_client_session(MClientSession *m);
 347   void send_reconnect(MetaSession *s);
 348   void resend_unsafe_requests(MetaSession *s);
 349   void wait_unsafe_requests();
 350
 351   // mds requests
 352   ceph_tid_t last_tid;
 353   ceph_tid_t oldest_tid; // oldest incomplete mds request, excluding setfilelock requests
 354   map<ceph_tid_t, MetaRequest*> mds_requests;
 355
 356   // cap flushing
 357   ceph_tid_t last_flush_tid;
 358
 359   void dump_mds_requests(Formatter *f);
 360   void dump_mds_sessions(Formatter *f);
 361
 362   int make_request(MetaRequest *req, const UserPerm& perms,
 363                    InodeRef *ptarget = 0, bool *pcreated = 0,
 364                    mds_rank_t use_mds=-1, bufferlist *pdirbl=0);
 365   void put_request(MetaRequest *request);
 366   void unregister_request(MetaRequest *request);
 367
 368   int verify_reply_trace(int r, MetaRequest *request, MClientReply *reply,
 369                          InodeRef *ptarget, bool *pcreated,
 370                          const UserPerm& perms);
 371   void encode_cap_releases(MetaRequest *request, mds_rank_t mds);
 372   int encode_inode_release(Inode *in, MetaRequest *req,
 373                            mds_rank_t mds, int drop,
 374                            int unless,int force=0);
 375   void encode_dentry_release(Dentry *dn, MetaRequest *req,
 376                              mds_rank_t mds, int drop, int unless);
 377   mds_rank_t choose_target_mds(MetaRequest *req, Inode** phash_diri=NULL);
 378   void connect_mds_targets(mds_rank_t mds);
 379   void send_request(MetaRequest *request, MetaSession *session,
 380                     bool drop_cap_releases=false);
 381   MClientRequest *build_client_request(MetaRequest *request);
 382   void kick_requests(MetaSession *session);
 383   void kick_requests_closed(MetaSession *session);
 384   void handle_client_request_forward(MClientRequestForward *reply);
 385   void handle_client_reply(MClientReply *reply);
 386   bool is_dir_operation(MetaRequest *request);
 387
 388   bool   initialized;
 389   bool   mounted;
 390   bool   unmounting;
 391   bool   blacklisted;
 392
 393   // When an MDS has sent us a REJECT, remember that and don't
 394   // contact it again.  Remember which inst rejected us, so that
 395   // when we talk to another inst with the same rank we can
 396   // try again.
 397   std::map<mds_rank_t, entity_inst_t> rejected_by_mds;
 398
 399   int local_osd;
 400   epoch_t local_osd_epoch;
 401
 402   int unsafe_sync_write;
 403
 404 public:
 405   entity_name_t get_myname() { return messenger->get_myname(); }
 406   void _sync_write_commit(Inode *in);
 407
 408 protected:
 409   std::unique_ptr<Filer>             filer;
 410   std::unique_ptr<ObjectCacher>      objectcacher;
 411   std::unique_ptr<WritebackHandler>  writeback_handler;
 412
 413   // cache
 414   ceph::unordered_map<vinodeno_t, Inode*> inode_map;
 415
 416   // fake inode number for 32-bits ino_t
 417   ceph::unordered_map<ino_t, vinodeno_t> faked_ino_map;
 418   interval_set<ino_t> free_faked_inos;
 419   ino_t last_used_faked_ino;
 420   void _assign_faked_ino(Inode *in);
 421   void _release_faked_ino(Inode *in);
 422   bool _use_faked_inos;
 423   void _reset_faked_inos();
 424   vinodeno_t _map_faked_ino(ino_t ino);
 425
 426   Inode*                 root;
 427   map<Inode*, InodeRef>  root_parents;
 428   Inode*                 root_ancestor;
 429   LRU                    lru;    // lru list of Dentry's in our local metadata cache.
 430
 431   // all inodes with caps sit on either cap_list or delayed_caps.
 432   xlist<Inode*> delayed_caps, cap_list;
 433   int num_flushing_caps;
 434   ceph::unordered_map<inodeno_t,SnapRealm*> snap_realms;
 435
 436   // Optional extra metadata about me to send to the MDS
 437   std::map<std::string, std::string> metadata;
 438   void populate_metadata(const std::string &mount_root);
 439
 440
 441   /* async block write barrier support */
 442   //map<uint64_t, BarrierContext* > barriers;
 443
 444   SnapRealm *get_snap_realm(inodeno_t r);
 445   SnapRealm *get_snap_realm_maybe(inodeno_t r);
 446   void put_snap_realm(SnapRealm *realm);
 447   bool adjust_realm_parent(SnapRealm *realm, inodeno_t parent);
 448   void update_snap_trace(bufferlist& bl, SnapRealm **realm_ret, bool must_flush=true);
 449   void invalidate_snaprealm_and_children(SnapRealm *realm);
 450
 451   Inode *open_snapdir(Inode *diri);
 452
 453
 454   // file handles, etc.
 455   interval_set<int> free_fd_set;  // unused fds
 456   ceph::unordered_map<int, Fh*> fd_map;
 457   set<Fh*> ll_unclosed_fh_set;
 458   ceph::unordered_set<dir_result_t*> opened_dirs;
 459
 460   int get_fd() {
 461     int fd = free_fd_set.range_start();
 462     free_fd_set.erase(fd, 1);
 463     return fd;
 464   }
 465   void put_fd(int fd) {
 466     free_fd_set.insert(fd, 1);
 467   }
 468
 469   /*
 470    * Resolve file descriptor, or return NULL.
 471    */
 472   Fh *get_filehandle(int fd) {
 473     ceph::unordered_map<int, Fh*>::iterator p = fd_map.find(fd);
 474     if (p == fd_map.end())
 475       return NULL;
 476     return p->second;
 477   }
 478
 479   // global client lock
 480   //  - protects Client and buffer cache both!
 481   Mutex                  client_lock;
 482
 483   // helpers
 484   void wake_inode_waiters(MetaSession *s);
 485   void wait_on_list(list<Cond*>& ls);
 486   void signal_cond_list(list<Cond*>& ls);
 487
 488   void wait_on_context_list(list<Context*>& ls);
 489   void signal_context_list(list<Context*>& ls);
 490
 491   // -- metadata cache stuff
 492
 493   // decrease inode ref.  delete if dangling.
 494   void put_inode(Inode *in, int n=1);
 495   void close_dir(Dir *dir);
 496
 497   friend class C_Client_FlushComplete; // calls put_inode()
 498   friend class C_Client_CacheInvalidate;  // calls ino_invalidate_cb
 499   friend class C_Client_DentryInvalidate;  // calls dentry_invalidate_cb
 500   friend class C_Block_Sync; // Calls block map and protected helpers
 501   friend class C_C_Tick; // Asserts on client_lock
 502   friend class C_Client_RequestInterrupt;
 503   friend class C_Client_Remount;
 504   friend void intrusive_ptr_release(Inode *in);
 505
 506   //int get_cache_size() { return lru.lru_get_size(); }
 507   //void set_cache_size(int m) { lru.lru_set_max(m); }
 508
 509   /**
 510    * Don't call this with in==NULL, use get_or_create for that
 511    * leave dn set to default NULL unless you're trying to add
 512    * a new inode to a pre-created Dentry
 513    */
 514   Dentry* link(Dir *dir, const string& name, Inode *in, Dentry *dn);
 515   void unlink(Dentry *dn, bool keepdir, bool keepdentry);
 516
 517   // path traversal for high-level interface
 518   InodeRef cwd;
 519   int path_walk(const filepath& fp, InodeRef *end, const UserPerm& perms,
 520                 bool followsym=true, int mask=0);
 521
 522   int fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0);
 523   int fill_stat(InodeRef& in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0) {
 524     return fill_stat(in.get(), st, dirstat, rstat);
 525   }
 526
 527   void fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx);
 528   void fill_statx(InodeRef& in, unsigned int mask, struct ceph_statx *stx) {
 529     return fill_statx(in.get(), mask, stx);
 530   }
 531
 532   void touch_dn(Dentry *dn);
 533
 534   // trim cache.
 535   void trim_cache(bool trim_kernel_dcache=false);
 536   void trim_cache_for_reconnect(MetaSession *s);
 537   void trim_dentry(Dentry *dn);
 538   void trim_caps(MetaSession *s, int max);
 539   void _invalidate_kernel_dcache();
 540
 541   void dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected);
 542   void dump_cache(Formatter *f);  // debug
 543
 544   // force read-only
 545   void force_session_readonly(MetaSession *s);
 546
 547   void dump_status(Formatter *f);  // debug
 548
 549   // trace generation
 550   ofstream traceout;
 551
 552
 553   Cond mount_cond, sync_cond;
 554
 555
 556   // friends
 557   friend class SyntheticClient;
 558   bool ms_dispatch(Message *m) override;
 559
 560   void ms_handle_connect(Connection *con) override;
 561   bool ms_handle_reset(Connection *con) override;
 562   void ms_handle_remote_reset(Connection *con) override;
 563   bool ms_handle_refused(Connection *con) override;
 564   bool ms_get_authorizer(int dest_type, AuthAuthorizer **authorizer, bool force_new) override;
 565
 566   int authenticate();
 567
 568   Inode* get_quota_root(Inode *in, const UserPerm& perms);
 569   bool check_quota_condition(Inode *in, const UserPerm& perms,
 570                              std::function<bool (const Inode &)> test);
 571   bool is_quota_files_exceeded(Inode *in, const UserPerm& perms);
 572   bool is_quota_bytes_exceeded(Inode *in, int64_t new_bytes,
 573                                const UserPerm& perms);
 574   bool is_quota_bytes_approaching(Inode *in, const UserPerm& perms);
 575
 576   std::map<std::pair<int64_t,std::string>, int> pool_perms;
 577   list<Cond*> waiting_for_pool_perm;
 578   int check_pool_perm(Inode *in, int need);
 579
 580   /**
 581    * Call this when an OSDMap is seen with a full flag (global or per pool)
 582    * set.
 583    *
 584    * @param pool the pool ID affected, or -1 if all.
 585    */
 586   void _handle_full_flag(int64_t pool);
 587
 588   void _close_sessions();
 589
 590   /**
 591    * The basic housekeeping parts of init (perf counters, admin socket)
 592    * that is independent of how objecters/monclient/messengers are
 593    * being set up.
 594    */
 595   void _finish_init();
 596
 597  public:
 598   void set_filer_flags(int flags);
 599   void clear_filer_flags(int flags);
 600
 601   Client(Messenger *m, MonClient *mc, Objecter *objecter_);
 602   ~Client() override;
 603   void tear_down_cache();
 604
 605   void update_metadata(std::string const &k, std::string const &v);
 606
 607   client_t get_nodeid() { return whoami; }
 608
 609   inodeno_t get_root_ino();
 610   Inode *get_root();
 611
 612   virtual int init();
 613   virtual void shutdown();
 614
 615   // messaging
 616   void handle_mds_map(class MMDSMap *m);
 617   void handle_fs_map(class MFSMap *m);
 618   void handle_fs_map_user(class MFSMapUser *m);
 619   void handle_osd_map(class MOSDMap *m);
 620
 621   void handle_lease(MClientLease *m);
 622
 623   // inline data
 624   int uninline_data(Inode *in, Context *onfinish);
 625
 626   // file caps
 627   void check_cap_issue(Inode *in, Cap *cap, unsigned issued);
 628   void add_update_cap(Inode *in, MetaSession *session, uint64_t cap_id,
 629                       unsigned issued, unsigned seq, unsigned mseq, inodeno_t realm,
 630                       int flags, const UserPerm& perms);
 631   void remove_cap(Cap *cap, bool queue_release);
 632   void remove_all_caps(Inode *in);
 633   void remove_session_caps(MetaSession *session);
 634   void mark_caps_dirty(Inode *in, int caps);
 635   int mark_caps_flushing(Inode *in, ceph_tid_t *ptid);
 636   void adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s);
 637   void flush_caps_sync();
 638   void flush_caps(Inode *in, MetaSession *session, bool sync=false);
 639   void kick_flushing_caps(MetaSession *session);
 640   void early_kick_flushing_caps(MetaSession *session);
 641   void kick_maxsize_requests(MetaSession *session);
 642   int get_caps(Inode *in, int need, int want, int *have, loff_t endoff);
 643   int get_caps_used(Inode *in);
 644
 645   void maybe_update_snaprealm(SnapRealm *realm, snapid_t snap_created, snapid_t snap_highwater,
 646                               vector<snapid_t>& snaps);
 647
 648   void handle_quota(struct MClientQuota *m);
 649   void handle_snap(struct MClientSnap *m);
 650   void handle_caps(class MClientCaps *m);
 651   void handle_cap_import(MetaSession *session, Inode *in, class MClientCaps *m);
 652   void handle_cap_export(MetaSession *session, Inode *in, class MClientCaps *m);
 653   void handle_cap_trunc(MetaSession *session, Inode *in, class MClientCaps *m);
 654   void handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, class MClientCaps *m);
 655   void handle_cap_flushsnap_ack(MetaSession *session, Inode *in, class MClientCaps *m);
 656   void handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, class MClientCaps *m);
 657   void cap_delay_requeue(Inode *in);
 658   void send_cap(Inode *in, MetaSession *session, Cap *cap, bool sync,
 659                 int used, int want, int retain, int flush,
 660                 ceph_tid_t flush_tid);
 661
 662   /* Flags for check_caps() */
 663 #define CHECK_CAPS_NODELAY      (0x1)
 664 #define CHECK_CAPS_SYNCHRONOUS  (0x2)
 665
 666   void check_caps(Inode *in, unsigned flags);
 667   void get_cap_ref(Inode *in, int cap);
 668   void put_cap_ref(Inode *in, int cap);
 669   void flush_snaps(Inode *in, bool all_again=false);
 670   void wait_sync_caps(Inode *in, ceph_tid_t want);
 671   void wait_sync_caps(ceph_tid_t want);
 672   void queue_cap_snap(Inode *in, SnapContext &old_snapc);
 673   void finish_cap_snap(Inode *in, CapSnap &capsnap, int used);
 674   void _flushed_cap_snap(Inode *in, snapid_t seq);
 675
 676   void _schedule_invalidate_dentry_callback(Dentry *dn, bool del);
 677   void _async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name);
 678   void _try_to_trim_inode(Inode *in, bool sched_inval);
 679
 680   void _schedule_invalidate_callback(Inode *in, int64_t off, int64_t len);
 681   void _invalidate_inode_cache(Inode *in);
 682   void _invalidate_inode_cache(Inode *in, int64_t off, int64_t len);
 683   void _async_invalidate(vinodeno_t ino, int64_t off, int64_t len);
 684   bool _release(Inode *in);
 685
 686   /**
 687    * Initiate a flush of the data associated with the given inode.
 688    * If you specify a Context, you are responsible for holding an inode
 689    * reference for the duration of the flush. If not, _flush() will
 690    * take the reference for you.
 691    * @param in The Inode whose data you wish to flush.
 692    * @param c The Context you wish us to complete once the data is
 693    * flushed. If already flushed, this will be called in-line.
 694    *
 695    * @returns true if the data was already flushed, false otherwise.
 696    */
 697   bool _flush(Inode *in, Context *c);
 698   void _flush_range(Inode *in, int64_t off, uint64_t size);
 699   void _flushed(Inode *in);
 700   void flush_set_callback(ObjectCacher::ObjectSet *oset);
 701
 702   void close_release(Inode *in);
 703   void close_safe(Inode *in);
 704
 705   void lock_fh_pos(Fh *f);
 706   void unlock_fh_pos(Fh *f);
 707
 708   // metadata cache
 709   void update_dir_dist(Inode *in, DirStat *st);
 710
 711   void clear_dir_complete_and_ordered(Inode *diri, bool complete);
 712   void insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri);
 713   Inode* insert_trace(MetaRequest *request, MetaSession *session);
 714   void update_inode_file_bits(Inode *in, uint64_t truncate_seq, uint64_t truncate_size, uint64_t size,
 715                               uint64_t change_attr, uint64_t time_warp_seq, utime_t ctime,
 716                               utime_t mtime, utime_t atime, version_t inline_version,
 717                               bufferlist& inline_data, int issued);
 718   Inode *add_update_inode(InodeStat *st, utime_t ttl, MetaSession *session,
 719                           const UserPerm& request_perms);
 720   Dentry *insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease,
 721                               Inode *in, utime_t from, MetaSession *session,
 722                               Dentry *old_dentry = NULL);
 723   void update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session);
 724
 725   bool use_faked_inos() { return _use_faked_inos; }
 726   vinodeno_t map_faked_ino(ino_t ino);
 727
 728   //notify the mds to flush the mdlog
 729   void flush_mdlog_sync();
 730   void flush_mdlog(MetaSession *session);
 731
 732   // ----------------------
 733   // fs ops.
 734 private:
 735
 736   void fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off);
 737
 738   // some readdir helpers
 739   typedef int (*add_dirent_cb_t)(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in);
 740
 741   int _opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms);
 742   void _readdir_drop_dirp_buffer(dir_result_t *dirp);
 743   bool _readdir_have_frag(dir_result_t *dirp);
 744   void _readdir_next_frag(dir_result_t *dirp);
 745   void _readdir_rechoose_frag(dir_result_t *dirp);
 746   int _readdir_get_frag(dir_result_t *dirp);
 747   int _readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, int caps, bool getref);
 748   void _closedir(dir_result_t *dirp);
 749
 750   // other helpers
 751   void _fragmap_remove_non_leaves(Inode *in);
 752   void _fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds);
 753
 754   void _ll_get(Inode *in);
 755   int _ll_put(Inode *in, int num);
 756   void _ll_drop_pins();
 757
 758   Fh *_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms);
 759   int _release_fh(Fh *fh);
 760   void _put_fh(Fh *fh);
 761
 762
 763   struct C_Readahead : public Context {
 764     Client *client;
 765     Fh *f;
 766     C_Readahead(Client *c, Fh *f);
 767     ~C_Readahead() override;
 768     void finish(int r) override;
 769   };
 770
 771   int _read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, bool *checkeof);
 772   int _read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl);
 773
 774   // internal interface
 775   //   call these with client_lock held!
 776   int _do_lookup(Inode *dir, const string& name, int mask, InodeRef *target,
 777                  const UserPerm& perms);
 778
 779   int _lookup(Inode *dir, const string& dname, int mask, InodeRef *target,
 780               const UserPerm& perm);
 781
 782   int _link(Inode *in, Inode *dir, const char *name, const UserPerm& perm,
 783             InodeRef *inp = 0);
 784   int _unlink(Inode *dir, const char *name, const UserPerm& perm);
 785   int _rename(Inode *olddir, const char *oname, Inode *ndir, const char *nname, const UserPerm& perm);
 786   int _mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm,
 787              InodeRef *inp = 0);
 788   int _rmdir(Inode *dir, const char *name, const UserPerm& perms);
 789   int _symlink(Inode *dir, const char *name, const char *target,
 790                const UserPerm& perms, InodeRef *inp = 0);
 791   int _mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev,
 792              const UserPerm& perms, InodeRef *inp = 0);
 793   int _do_setattr(Inode *in, struct ceph_statx *stx, int mask,
 794                   const UserPerm& perms, InodeRef *inp);
 795   void stat_to_statx(struct stat *st, struct ceph_statx *stx);
 796   int __setattrx(Inode *in, struct ceph_statx *stx, int mask,
 797                  const UserPerm& perms, InodeRef *inp = 0);
 798   int _setattrx(InodeRef &in, struct ceph_statx *stx, int mask,
 799                 const UserPerm& perms);
 800   int _setattr(InodeRef &in, struct stat *attr, int mask,
 801                const UserPerm& perms);
 802   int _ll_setattrx(Inode *in, struct ceph_statx *stx, int mask,
 803                    const UserPerm& perms, InodeRef *inp = 0);
 804   int _getattr(Inode *in, int mask, const UserPerm& perms, bool force=false);
 805   int _getattr(InodeRef &in, int mask, const UserPerm& perms, bool force=false) {
 806     return _getattr(in.get(), mask, perms, force);
 807   }
 808   int _readlink(Inode *in, char *buf, size_t size);
 809   int _getxattr(Inode *in, const char *name, void *value, size_t len,
 810                 const UserPerm& perms);
 811   int _getxattr(InodeRef &in, const char *name, void *value, size_t len,
 812                 const UserPerm& perms);
 813   int _listxattr(Inode *in, char *names, size_t len, const UserPerm& perms);
 814   int _do_setxattr(Inode *in, const char *name, const void *value, size_t len,
 815                    int flags, const UserPerm& perms);
 816   int _setxattr(Inode *in, const char *name, const void *value, size_t len,
 817                 int flags, const UserPerm& perms);
 818   int _setxattr(InodeRef &in, const char *name, const void *value, size_t len,
 819                 int flags, const UserPerm& perms);
 820   int _setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap);
 821   void _setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t len);
 822   int _removexattr(Inode *in, const char *nm, const UserPerm& perms);
 823   int _removexattr(InodeRef &in, const char *nm, const UserPerm& perms);
 824   int _open(Inode *in, int flags, mode_t mode, Fh **fhp,
 825             const UserPerm& perms);
 826   int _renew_caps(Inode *in);
 827   int _create(Inode *in, const char *name, int flags, mode_t mode, InodeRef *inp,
 828               Fh **fhp, int stripe_unit, int stripe_count, int object_size,
 829               const char *data_pool, bool *created, const UserPerm &perms);
 830
 831   loff_t _lseek(Fh *fh, loff_t offset, int whence);
 832   int _read(Fh *fh, int64_t offset, uint64_t size, bufferlist *bl);
 833   int _write(Fh *fh, int64_t offset, uint64_t size, const char *buf,
 834           const struct iovec *iov, int iovcnt);
 835   int _preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write);
 836   int _flush(Fh *fh);
 837   int _fsync(Fh *fh, bool syncdataonly);
 838   int _fsync(Inode *in, bool syncdataonly);
 839   int _sync_fs();
 840   int _fallocate(Fh *fh, int mode, int64_t offset, int64_t length);
 841   int _getlk(Fh *fh, struct flock *fl, uint64_t owner);
 842   int _setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep);
 843   int _flock(Fh *fh, int cmd, uint64_t owner);
 844
 845   int get_or_create(Inode *dir, const char* name,
 846                     Dentry **pdn, bool expect_null=false);
 847
 848   enum {
 849     NO_ACL = 0,
 850     POSIX_ACL,
 851   };
 852
 853   enum {
 854     MAY_EXEC = 1,
 855     MAY_WRITE = 2,
 856     MAY_READ = 4,
 857   };
 858
 859   void init_groups(UserPerm *groups);
 860
 861   int inode_permission(Inode *in, const UserPerm& perms, unsigned want);
 862   int xattr_permission(Inode *in, const char *name, unsigned want,
 863                        const UserPerm& perms);
 864   int may_setattr(Inode *in, struct ceph_statx *stx, int mask,
 865                   const UserPerm& perms);
 866   int may_open(Inode *in, int flags, const UserPerm& perms);
 867   int may_lookup(Inode *dir, const UserPerm& perms);
 868   int may_create(Inode *dir, const UserPerm& perms);
 869   int may_delete(Inode *dir, const char *name, const UserPerm& perms);
 870   int may_hardlink(Inode *in, const UserPerm& perms);
 871
 872   int _getattr_for_perm(Inode *in, const UserPerm& perms);
 873   int _getgrouplist(gid_t **sgids, uid_t uid, gid_t gid);
 874
 875   vinodeno_t _get_vino(Inode *in);
 876   inodeno_t _get_inodeno(Inode *in);
 877
 878   /*
 879    * These define virtual xattrs exposing the recursive directory
 880    * statistics and layout metadata.
 881    */
 882   struct VXattr {
 883           const string name;
 884           size_t (Client::*getxattr_cb)(Inode *in, char *val, size_t size);
 885           bool readonly, hidden;
 886           bool (Client::*exists_cb)(Inode *in);
 887   };
 888
 889   bool _vxattrcb_quota_exists(Inode *in);
 890   size_t _vxattrcb_quota(Inode *in, char *val, size_t size);
 891   size_t _vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size);
 892   size_t _vxattrcb_quota_max_files(Inode *in, char *val, size_t size);
 893
 894   bool _vxattrcb_layout_exists(Inode *in);
 895   size_t _vxattrcb_layout(Inode *in, char *val, size_t size);
 896   size_t _vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size);
 897   size_t _vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size);
 898   size_t _vxattrcb_layout_object_size(Inode *in, char *val, size_t size);
 899   size_t _vxattrcb_layout_pool(Inode *in, char *val, size_t size);
 900   size_t _vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size);
 901   size_t _vxattrcb_dir_entries(Inode *in, char *val, size_t size);
 902   size_t _vxattrcb_dir_files(Inode *in, char *val, size_t size);
 903   size_t _vxattrcb_dir_subdirs(Inode *in, char *val, size_t size);
 904   size_t _vxattrcb_dir_rentries(Inode *in, char *val, size_t size);
 905   size_t _vxattrcb_dir_rfiles(Inode *in, char *val, size_t size);
 906   size_t _vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size);
 907   size_t _vxattrcb_dir_rbytes(Inode *in, char *val, size_t size);
 908   size_t _vxattrcb_dir_rctime(Inode *in, char *val, size_t size);
 909   size_t _vxattrs_calcu_name_size(const VXattr *vxattrs);
 910
 911   static const VXattr _dir_vxattrs[];
 912   static const VXattr _file_vxattrs[];
 913
 914   static const VXattr *_get_vxattrs(Inode *in);
 915   static const VXattr *_match_vxattr(Inode *in, const char *name);
 916
 917   size_t _file_vxattrs_name_size;
 918   size_t _dir_vxattrs_name_size;
 919   size_t _vxattrs_name_size(const VXattr *vxattrs) {
 920           if (vxattrs == _dir_vxattrs)
 921                   return _dir_vxattrs_name_size;
 922           else if (vxattrs == _file_vxattrs)
 923                   return _file_vxattrs_name_size;
 924           return 0;
 925   }
 926
 927   int _do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep,
 928                    struct flock *fl, uint64_t owner, bool removing=false);
 929   int _interrupt_filelock(MetaRequest *req);
 930   void _encode_filelocks(Inode *in, bufferlist& bl);
 931   void _release_filelocks(Fh *fh);
 932   void _update_lock_state(struct flock *fl, uint64_t owner, ceph_lock_state_t *lock_state);
 933
 934   int _posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl,
 935                         const UserPerm& perms);
 936   int _posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms);
 937   int _posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want);
 938
 939   mds_rank_t _get_random_up_mds() const;
 940
 941   int _ll_getattr(Inode *in, int caps, const UserPerm& perms);
 942
 943 public:
 944   int mount(const std::string &mount_root, const UserPerm& perms,
 945             bool require_mds=false);
 946   void unmount();
 947
 948   int mds_command(
 949     const std::string &mds_spec,
 950     const std::vector<std::string>& cmd,
 951     const bufferlist& inbl,
 952     bufferlist *poutbl, std::string *prs, Context *onfinish);
 953
 954   // these shoud (more or less) mirror the actual system calls.
 955   int statfs(const char *path, struct statvfs *stbuf, const UserPerm& perms);
 956
 957   // crap
 958   int chdir(const char *s, std::string &new_cwd, const UserPerm& perms);
 959   void getcwd(std::string& cwd, const UserPerm& perms);
 960
 961   // namespace ops
 962   int opendir(const char *name, dir_result_t **dirpp, const UserPerm& perms);
 963   int closedir(dir_result_t *dirp);
 964
 965   /**
 966    * Fill a directory listing from dirp, invoking cb for each entry
 967    * with the given pointer, the dirent, the struct stat, the stmask,
 968    * and the offset.
 969    *
 970    * Returns 0 if it reached the end of the directory.
 971    * If @a cb returns a negative error code, stop and return that.
 972    */
 973   int readdir_r_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
 974                    unsigned want=0, unsigned flags=AT_NO_ATTR_SYNC,
 975                    bool getref=false);
 976
 977   struct dirent * readdir(dir_result_t *d);
 978   int readdir_r(dir_result_t *dirp, struct dirent *de);
 979   int readdirplus_r(dir_result_t *dirp, struct dirent *de, struct ceph_statx *stx, unsigned want, unsigned flags, Inode **out);
 980
 981   int getdir(const char *relpath, list<string>& names,
 982              const UserPerm& perms);  // get the whole dir at once.
 983
 984   /**
 985    * Returns the length of the buffer that got filled in, or -errno.
 986    * If it returns -ERANGE you just need to increase the size of the
 987    * buffer and try again.
 988    */
 989   int _getdents(dir_result_t *dirp, char *buf, int buflen, bool ful);  // get a bunch of dentries at once
 990   int getdents(dir_result_t *dirp, char *buf, int buflen) {
 991     return _getdents(dirp, buf, buflen, true);
 992   }
 993   int getdnames(dir_result_t *dirp, char *buf, int buflen) {
 994     return _getdents(dirp, buf, buflen, false);
 995   }
 996
 997   void rewinddir(dir_result_t *dirp);
 998   loff_t telldir(dir_result_t *dirp);
 999   void seekdir(dir_result_t *dirp, loff_t offset);
1000
1001   int link(const char *existing, const char *newname, const UserPerm& perm);
1002   int unlink(const char *path, const UserPerm& perm);
1003   int rename(const char *from, const char *to, const UserPerm& perm);
1004
1005   // dirs
1006   int mkdir(const char *path, mode_t mode, const UserPerm& perm);
1007   int mkdirs(const char *path, mode_t mode, const UserPerm& perms);
1008   int rmdir(const char *path, const UserPerm& perms);
1009
1010   // symlinks
1011   int readlink(const char *path, char *buf, loff_t size, const UserPerm& perms);
1012
1013   int symlink(const char *existing, const char *newname, const UserPerm& perms);
1014
1015   // inode stuff
1016   unsigned statx_to_mask(unsigned int flags, unsigned int want);
1017   int stat(const char *path, struct stat *stbuf, const UserPerm& perms,
1018            frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL);
1019   int statx(const char *path, struct ceph_statx *stx,
1020             const UserPerm& perms,
1021             unsigned int want, unsigned int flags);
1022   int lstat(const char *path, struct stat *stbuf, const UserPerm& perms,
1023             frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL);
1024
1025   int setattr(const char *relpath, struct stat *attr, int mask,
1026               const UserPerm& perms);
1027   int setattrx(const char *relpath, struct ceph_statx *stx, int mask,
1028                const UserPerm& perms, int flags=0);
1029   int fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms);
1030   int fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms);
1031   int chmod(const char *path, mode_t mode, const UserPerm& perms);
1032   int fchmod(int fd, mode_t mode, const UserPerm& perms);
1033   int lchmod(const char *path, mode_t mode, const UserPerm& perms);
1034   int chown(const char *path, uid_t new_uid, gid_t new_gid,
1035             const UserPerm& perms);
1036   int fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms);
1037   int lchown(const char *path, uid_t new_uid, gid_t new_gid,
1038              const UserPerm& perms);
1039   int utime(const char *path, struct utimbuf *buf, const UserPerm& perms);
1040   int lutime(const char *path, struct utimbuf *buf, const UserPerm& perms);
1041   int flock(int fd, int operation, uint64_t owner);
1042   int truncate(const char *path, loff_t size, const UserPerm& perms);
1043
1044   // file ops
1045   int mknod(const char *path, mode_t mode, const UserPerm& perms, dev_t rdev=0);
1046   int open(const char *path, int flags, const UserPerm& perms, mode_t mode=0);
1047   int open(const char *path, int flags, const UserPerm& perms,
1048            mode_t mode, int stripe_unit, int stripe_count, int object_size,
1049            const char *data_pool);
1050   int lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name,
1051                   const UserPerm& perms);
1052   int lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode=NULL);
1053   int lookup_parent(Inode *in, const UserPerm& perms, Inode **parent=NULL);
1054   int lookup_name(Inode *in, Inode *parent, const UserPerm& perms);
1055   int close(int fd);
1056   loff_t lseek(int fd, loff_t offset, int whence);
1057   int read(int fd, char *buf, loff_t size, loff_t offset=-1);
1058   int preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1);
1059   int write(int fd, const char *buf, loff_t size, loff_t offset=-1);
1060   int pwritev(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1);
1061   int fake_write_size(int fd, loff_t size);
1062   int ftruncate(int fd, loff_t size, const UserPerm& perms);
1063   int fsync(int fd, bool syncdataonly);
1064   int fstat(int fd, struct stat *stbuf, const UserPerm& perms,
1065             int mask=CEPH_STAT_CAP_INODE_ALL);
1066   int fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms,
1067              unsigned int want, unsigned int flags);
1068   int fallocate(int fd, int mode, loff_t offset, loff_t length);
1069
1070   // full path xattr ops
1071   int getxattr(const char *path, const char *name, void *value, size_t size,
1072                const UserPerm& perms);
1073   int lgetxattr(const char *path, const char *name, void *value, size_t size,
1074                 const UserPerm& perms);
1075   int fgetxattr(int fd, const char *name, void *value, size_t size,
1076                 const UserPerm& perms);
1077   int listxattr(const char *path, char *list, size_t size, const UserPerm& perms);
1078   int llistxattr(const char *path, char *list, size_t size, const UserPerm& perms);
1079   int flistxattr(int fd, char *list, size_t size, const UserPerm& perms);
1080   int removexattr(const char *path, const char *name, const UserPerm& perms);
1081   int lremovexattr(const char *path, const char *name, const UserPerm& perms);
1082   int fremovexattr(int fd, const char *name, const UserPerm& perms);
1083   int setxattr(const char *path, const char *name, const void *value,
1084                size_t size, int flags, const UserPerm& perms);
1085   int lsetxattr(const char *path, const char *name, const void *value,
1086                 size_t size, int flags, const UserPerm& perms);
1087   int fsetxattr(int fd, const char *name, const void *value, size_t size,
1088                 int flags, const UserPerm& perms);
1089
1090   int sync_fs();
1091   int64_t drop_caches();
1092
1093   // hpc lazyio
1094   int lazyio_propogate(int fd, loff_t offset, size_t count);
1095   int lazyio_synchronize(int fd, loff_t offset, size_t count);
1096
1097   // expose file layout
1098   int describe_layout(const char *path, file_layout_t* layout,
1099                       const UserPerm& perms);
1100   int fdescribe_layout(int fd, file_layout_t* layout);
1101   int get_file_stripe_address(int fd, loff_t offset, vector<entity_addr_t>& address);
1102   int get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds);
1103   int get_osd_addr(int osd, entity_addr_t& addr);
1104
1105   // expose mdsmap
1106   int64_t get_default_pool_id();
1107
1108   // expose osdmap
1109   int get_local_osd();
1110   int get_pool_replication(int64_t pool);
1111   int64_t get_pool_id(const char *pool_name);
1112   string get_pool_name(int64_t pool);
1113   int get_osd_crush_location(int id, vector<pair<string, string> >& path);
1114
1115   int enumerate_layout(int fd, vector<ObjectExtent>& result,
1116                        loff_t length, loff_t offset);
1117
1118   int mksnap(const char *path, const char *name, const UserPerm& perm);
1119   int rmsnap(const char *path, const char *name, const UserPerm& perm);
1120
1121   // expose caps
1122   int get_caps_issued(int fd);
1123   int get_caps_issued(const char *path, const UserPerm& perms);
1124
1125   // low-level interface v2
1126   inodeno_t ll_get_inodeno(Inode *in) {
1127     Mutex::Locker lock(client_lock);
1128     return _get_inodeno(in);
1129   }
1130   snapid_t ll_get_snapid(Inode *in);
1131   vinodeno_t ll_get_vino(Inode *in) {
1132     Mutex::Locker lock(client_lock);
1133     return _get_vino(in);
1134   }
1135   // get inode from faked ino
1136   Inode *ll_get_inode(ino_t ino);
1137   Inode *ll_get_inode(vinodeno_t vino);
1138   int ll_lookup(Inode *parent, const char *name, struct stat *attr,
1139                 Inode **out, const UserPerm& perms);
1140   int ll_lookupx(Inode *parent, const char *name, Inode **out,
1141                         struct ceph_statx *stx, unsigned want, unsigned flags,
1142                         const UserPerm& perms);
1143   bool ll_forget(Inode *in, int count);
1144   bool ll_put(Inode *in);
1145   int ll_getattr(Inode *in, struct stat *st, const UserPerm& perms);
1146   int ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want,
1147                   unsigned int flags, const UserPerm& perms);
1148   int ll_setattrx(Inode *in, struct ceph_statx *stx, int mask,
1149                   const UserPerm& perms);
1150   int ll_setattr(Inode *in, struct stat *st, int mask,
1151                  const UserPerm& perms);
1152   int ll_getxattr(Inode *in, const char *name, void *value, size_t size,
1153                   const UserPerm& perms);
1154   int ll_setxattr(Inode *in, const char *name, const void *value, size_t size,
1155                   int flags, const UserPerm& perms);
1156   int ll_removexattr(Inode *in, const char *name, const UserPerm& perms);
1157   int ll_listxattr(Inode *in, char *list, size_t size, const UserPerm& perms);
1158   int ll_opendir(Inode *in, int flags, dir_result_t **dirpp,
1159                  const UserPerm& perms);
1160   int ll_releasedir(dir_result_t* dirp);
1161   int ll_fsyncdir(dir_result_t* dirp);
1162   int ll_readlink(Inode *in, char *buf, size_t bufsize, const UserPerm& perms);
1163   int ll_mknod(Inode *in, const char *name, mode_t mode, dev_t rdev,
1164                struct stat *attr, Inode **out, const UserPerm& perms);
1165   int ll_mknodx(Inode *parent, const char *name, mode_t mode, dev_t rdev,
1166                 Inode **out, struct ceph_statx *stx, unsigned want,
1167                 unsigned flags, const UserPerm& perms);
1168   int ll_mkdir(Inode *in, const char *name, mode_t mode, struct stat *attr,
1169                Inode **out, const UserPerm& perm);
1170   int ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out,
1171                 struct ceph_statx *stx, unsigned want, unsigned flags,
1172                 const UserPerm& perms);
1173   int ll_symlink(Inode *in, const char *name, const char *value,
1174                  struct stat *attr, Inode **out, const UserPerm& perms);
1175   int ll_symlinkx(Inode *parent, const char *name, const char *value,
1176                   Inode **out, struct ceph_statx *stx, unsigned want,
1177                   unsigned flags, const UserPerm& perms);
1178   int ll_unlink(Inode *in, const char *name, const UserPerm& perm);
1179   int ll_rmdir(Inode *in, const char *name, const UserPerm& perms);
1180   int ll_rename(Inode *parent, const char *name, Inode *newparent,
1181                 const char *newname, const UserPerm& perm);
1182   int ll_link(Inode *in, Inode *newparent, const char *newname,
1183               const UserPerm& perm);
1184   int ll_open(Inode *in, int flags, Fh **fh, const UserPerm& perms);
1185   int _ll_create(Inode *parent, const char *name, mode_t mode,
1186               int flags, InodeRef *in, int caps, Fh **fhp,
1187               const UserPerm& perms);
1188   int ll_create(Inode *parent, const char *name, mode_t mode, int flags,
1189                 struct stat *attr, Inode **out, Fh **fhp,
1190                 const UserPerm& perms);
1191   int ll_createx(Inode *parent, const char *name, mode_t mode,
1192                 int oflags, Inode **outp, Fh **fhp,
1193                 struct ceph_statx *stx, unsigned want, unsigned lflags,
1194                 const UserPerm& perms);
1195   int ll_read_block(Inode *in, uint64_t blockid, char *buf,  uint64_t offset,
1196                     uint64_t length, file_layout_t* layout);
1197
1198   int ll_write_block(Inode *in, uint64_t blockid,
1199                      char* buf, uint64_t offset,
1200                      uint64_t length, file_layout_t* layout,
1201                      uint64_t snapseq, uint32_t sync);
1202   int ll_commit_blocks(Inode *in, uint64_t offset, uint64_t length);
1203
1204   int ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms);
1205   int ll_walk(const char* name, Inode **i, struct ceph_statx *stx,
1206                unsigned int want, unsigned int flags, const UserPerm& perms);
1207   uint32_t ll_stripe_unit(Inode *in);
1208   int ll_file_layout(Inode *in, file_layout_t *layout);
1209   uint64_t ll_snap_seq(Inode *in);
1210
1211   int ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl);
1212   int ll_write(Fh *fh, loff_t off, loff_t len, const char *data);
1213   loff_t ll_lseek(Fh *fh, loff_t offset, int whence);
1214   int ll_flush(Fh *fh);
1215   int ll_fsync(Fh *fh, bool syncdataonly);
1216   int ll_fallocate(Fh *fh, int mode, loff_t offset, loff_t length);
1217   int ll_release(Fh *fh);
1218   int ll_getlk(Fh *fh, struct flock *fl, uint64_t owner);
1219   int ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep);
1220   int ll_flock(Fh *fh, int cmd, uint64_t owner);
1221   int ll_file_layout(Fh *fh, file_layout_t *layout);
1222   void ll_interrupt(void *d);
1223   bool ll_handle_umask() {
1224     return acl_type != NO_ACL;
1225   }
1226
1227   int ll_get_stripe_osd(struct Inode *in, uint64_t blockno,
1228                         file_layout_t* layout);
1229   uint64_t ll_get_internal_offset(struct Inode *in, uint64_t blockno);
1230
1231   int ll_num_osds(void);
1232   int ll_osdaddr(int osd, uint32_t *addr);
1233   int ll_osdaddr(int osd, char* buf, size_t size);
1234
1235   void ll_register_callbacks(struct client_callback_args *args);
1236   int test_dentry_handling(bool can_invalidate);
1237
1238   const char** get_tracked_conf_keys() const override;
1239   void handle_conf_change(const struct md_config_t *conf,
1240                                   const std::set <std::string> &changed) override;
1241 };
1242
1243 /**
1244  * Specialization of Client that manages its own Objecter instance
1245  * and handles init/shutdown of messenger/monclient
1246  */
1247 class StandaloneClient : public Client
1248 {
1249   public:
1250   StandaloneClient(Messenger *m, MonClient *mc);
1251
1252   ~StandaloneClient() override;
1253
1254   int init() override;
1255   void shutdown() override;
1256 };
1257
1258 #endif