ceph/src/client/Inode.h

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3
   4 #ifndef CEPH_CLIENT_INODE_H
   5 #define CEPH_CLIENT_INODE_H
   6
   7 #include <numeric>
   8
   9 #include "include/compat.h"
  10 #include "include/ceph_assert.h"
  11 #include "include/types.h"
  12 #include "include/xlist.h"
  13
  14 #include "mds/flock.h"
  15 #include "mds/mdstypes.h" // hrm
  16
  17 #include "osdc/ObjectCacher.h"
  18
  19 #include "InodeRef.h"
  20 #include "MetaSession.h"
  21 #include "UserPerm.h"
  22 #include "Delegation.h"
  23
  24 class Client;
  25 class Dentry;
  26 class Dir;
  27 struct SnapRealm;
  28 struct Inode;
  29 class MetaRequest;
  30 class filepath;
  31 class Fh;
  32
  33 class Cap {
  34 public:
  35   Cap() = delete;
  36   Cap(Inode &i, MetaSession *s) : inode(i),
  37                                   session(s),
  38                                   gen(s->cap_gen),
  39                                   cap_item(this)
  40   {
  41     s->caps.push_back(&cap_item);
  42   }
  43   ~Cap() {
  44     cap_item.remove_myself();
  45   }
  46
  47   void touch(void) {
  48     // move to back of LRU
  49     session->caps.push_back(&cap_item);
  50   }
  51
  52   void dump(Formatter *f) const;
  53
  54   Inode &inode;
  55   MetaSession *session;
  56   uint64_t cap_id = 0;
  57   unsigned issued = 0;
  58   unsigned implemented = 0;
  59   unsigned wanted = 0;   // as known to mds.
  60   uint64_t seq = 0;
  61   uint64_t issue_seq = 0;
  62   __u32 mseq = 0;  // migration seq
  63   __u32 gen;
  64   UserPerm latest_perms;
  65
  66 private:
  67   /* Note that this Cap will not move (see Inode::caps):
  68    *
  69    * Section 23.1.2#8
  70    * The insert members shall not affect the validity of iterators and
  71    * references to the container, and the erase members shall invalidate only
  72    * iterators and references to the erased elements.
  73    */
  74   xlist<Cap *>::item cap_item;
  75 };
  76
  77 struct CapSnap {
  78   //snapid_t follows;  // map key
  79   InodeRef in;
  80   SnapContext context;
  81   int issued, dirty;
  82
  83   uint64_t size;
  84   utime_t ctime, btime, mtime, atime;
  85   version_t time_warp_seq;
  86   uint64_t change_attr;
  87   uint32_t   mode;
  88   uid_t      uid;
  89   gid_t      gid;
  90   map<string,bufferptr> xattrs;
  91   version_t xattr_version;
  92
  93   bufferlist inline_data;
  94   version_t inline_version;
  95
  96   bool writing, dirty_data;
  97   uint64_t flush_tid;
  98
  99   int64_t cap_dirtier_uid;
 100   int64_t cap_dirtier_gid;
 101
 102   explicit CapSnap(Inode *i)
 103     : in(i), issued(0), dirty(0), size(0), time_warp_seq(0), change_attr(0),
 104       mode(0), uid(0), gid(0), xattr_version(0), inline_version(0),
 105       writing(false), dirty_data(false), flush_tid(0), cap_dirtier_uid(-1),
 106       cap_dirtier_gid(-1)
 107   {}
 108
 109   void dump(Formatter *f) const;
 110 };
 111
 112 // inode flags
 113 #define I_COMPLETE              (1 << 0)
 114 #define I_DIR_ORDERED           (1 << 1)
 115 #define I_SNAPDIR_OPEN          (1 << 2)
 116 #define I_KICK_FLUSH            (1 << 3)
 117 #define I_CAP_DROPPED           (1 << 4)
 118 #define I_ERROR_FILELOCK        (1 << 5)
 119
 120 struct Inode {
 121   Client *client;
 122
 123   // -- the actual inode --
 124   inodeno_t ino; // ORDER DEPENDENCY: oset
 125   snapid_t  snapid;
 126   ino_t faked_ino;
 127
 128   uint32_t   rdev;    // if special file
 129
 130   // affected by any inode change...
 131   utime_t    ctime;   // inode change time
 132   utime_t    btime;   // birth time
 133
 134   // perm (namespace permissions)
 135   uint32_t   mode;
 136   uid_t      uid;
 137   gid_t      gid;
 138
 139   // nlink
 140   int32_t    nlink;
 141
 142   // file (data access)
 143   ceph_dir_layout dir_layout;
 144   file_layout_t layout;
 145   uint64_t   size;        // on directory, # dentries
 146   uint32_t   truncate_seq;
 147   uint64_t   truncate_size;
 148   utime_t    mtime;   // file data modify time.
 149   utime_t    atime;   // file data access time.
 150   uint32_t   time_warp_seq;  // count of (potential) mtime/atime timewarps (i.e., utimes())
 151   uint64_t   change_attr;
 152
 153   uint64_t max_size;  // max size we can write to
 154
 155   // dirfrag, recursive accountin
 156   frag_info_t dirstat;
 157   nest_info_t rstat;
 158
 159   // special stuff
 160   version_t version;           // auth only
 161   version_t xattr_version;
 162   utime_t   snap_btime;        // snapshot creation (birth) time
 163   std::map<std::string, std::string> snap_metadata;
 164
 165   // inline data
 166   version_t  inline_version;
 167   bufferlist inline_data;
 168
 169   bool fscrypt = false; // fscrypt enabled ?
 170
 171   bool is_root()    const { return ino == MDS_INO_ROOT; }
 172   bool is_symlink() const { return (mode & S_IFMT) == S_IFLNK; }
 173   bool is_dir()     const { return (mode & S_IFMT) == S_IFDIR; }
 174   bool is_file()    const { return (mode & S_IFMT) == S_IFREG; }
 175
 176   bool has_dir_layout() const {
 177     return layout != file_layout_t();
 178   }
 179
 180   __u32 hash_dentry_name(const string &dn) {
 181     int which = dir_layout.dl_dir_hash;
 182     if (!which)
 183       which = CEPH_STR_HASH_LINUX;
 184     ceph_assert(ceph_str_hash_valid(which));
 185     return ceph_str_hash(which, dn.data(), dn.length());
 186   }
 187
 188   unsigned flags;
 189
 190   quota_info_t quota;
 191
 192   bool is_complete_and_ordered() {
 193     static const unsigned wants = I_COMPLETE | I_DIR_ORDERED;
 194     return (flags & wants) == wants;
 195   }
 196
 197   // about the dir (if this is one!)
 198   Dir       *dir;     // if i'm a dir.
 199   fragtree_t dirfragtree;
 200   uint64_t dir_release_count, dir_ordered_count;
 201   bool dir_hashed, dir_replicated;
 202
 203   // per-mds caps
 204   std::map<mds_rank_t, Cap> caps;            // mds -> Cap
 205   Cap *auth_cap;
 206   int64_t cap_dirtier_uid;
 207   int64_t cap_dirtier_gid;
 208   unsigned dirty_caps, flushing_caps;
 209   std::map<ceph_tid_t, int> flushing_cap_tids;
 210   int shared_gen, cache_gen;
 211   int snap_caps, snap_cap_refs;
 212   utime_t hold_caps_until;
 213   xlist<Inode*>::item delay_cap_item, dirty_cap_item, flushing_cap_item;
 214
 215   SnapRealm *snaprealm;
 216   xlist<Inode*>::item snaprealm_item;
 217   InodeRef snapdir_parent;  // only if we are a snapdir inode
 218   map<snapid_t,CapSnap> cap_snaps;   // pending flush to mds
 219
 220   //int open_by_mode[CEPH_FILE_MODE_NUM];
 221   map<int,int> open_by_mode;
 222   map<int,int> cap_refs;
 223
 224   ObjectCacher::ObjectSet oset; // ORDER DEPENDENCY: ino
 225
 226   uint64_t     reported_size, wanted_max_size, requested_max_size;
 227
 228   int       _ref;      // ref count. 1 for each dentry, fh that links to me.
 229   uint64_t  ll_ref;   // separate ref count for ll client
 230   xlist<Dentry *> dentries; // if i'm linked to a dentry.
 231   string    symlink;  // symlink content, if it's a symlink
 232   map<string,bufferptr> xattrs;
 233   map<frag_t,int> fragmap;  // known frag -> mds mappings
 234   map<frag_t, std::vector<mds_rank_t>> frag_repmap; // non-auth mds mappings
 235
 236   std::list<ceph::condition_variable*> waitfor_caps;
 237   std::list<ceph::condition_variable*> waitfor_commit;
 238   std::list<ceph::condition_variable*> waitfor_deleg;
 239
 240   Dentry *get_first_parent() {
 241     ceph_assert(!dentries.empty());
 242     return *dentries.begin();
 243   }
 244
 245   void make_long_path(filepath& p);
 246   void make_short_path(filepath& p);
 247   void make_nosnap_relative_path(filepath& p);
 248
 249   void get();
 250   int _put(int n=1);
 251
 252   int get_num_ref() {
 253     return _ref;
 254   }
 255
 256   void ll_get() {
 257     ll_ref++;
 258   }
 259   void ll_put(uint64_t n=1) {
 260     ceph_assert(ll_ref >= n);
 261     ll_ref -= n;
 262   }
 263
 264   // file locks
 265   std::unique_ptr<ceph_lock_state_t> fcntl_locks;
 266   std::unique_ptr<ceph_lock_state_t> flock_locks;
 267
 268   bool has_any_filelocks() {
 269     return
 270       (fcntl_locks && !fcntl_locks->empty()) ||
 271       (flock_locks && !flock_locks->empty());
 272   }
 273
 274   list<Delegation> delegations;
 275
 276   xlist<MetaRequest*> unsafe_ops;
 277
 278   std::set<Fh*> fhs;
 279
 280   mds_rank_t dir_pin;
 281
 282   Inode(Client *c, vinodeno_t vino, file_layout_t *newlayout)
 283     : client(c), ino(vino.ino), snapid(vino.snapid), faked_ino(0),
 284       rdev(0), mode(0), uid(0), gid(0), nlink(0),
 285       size(0), truncate_seq(1), truncate_size(-1),
 286       time_warp_seq(0), change_attr(0), max_size(0), version(0),
 287       xattr_version(0), inline_version(0), flags(0),
 288       dir(0), dir_release_count(1), dir_ordered_count(1),
 289       dir_hashed(false), dir_replicated(false), auth_cap(NULL),
 290       cap_dirtier_uid(-1), cap_dirtier_gid(-1),
 291       dirty_caps(0), flushing_caps(0), shared_gen(0), cache_gen(0),
 292       snap_caps(0), snap_cap_refs(0),
 293       delay_cap_item(this), dirty_cap_item(this), flushing_cap_item(this),
 294       snaprealm(0), snaprealm_item(this),
 295       oset((void *)this, newlayout->pool_id, this->ino),
 296       reported_size(0), wanted_max_size(0), requested_max_size(0),
 297       _ref(0), ll_ref(0), dir_pin(MDS_RANK_NONE)
 298   {
 299     memset(&dir_layout, 0, sizeof(dir_layout));
 300   }
 301   ~Inode();
 302
 303   vinodeno_t vino() const { return vinodeno_t(ino, snapid); }
 304
 305   struct Compare {
 306     bool operator() (Inode* const & left, Inode* const & right) {
 307       if (left->ino.val < right->ino.val) {
 308         return (left->snapid.val < right->snapid.val);
 309       }
 310       return false;
 311     }
 312   };
 313
 314   bool check_mode(const UserPerm& perms, unsigned want);
 315
 316   // CAPS --------
 317   void get_open_ref(int mode);
 318   bool put_open_ref(int mode);
 319
 320   void get_cap_ref(int cap);
 321   int put_cap_ref(int cap);
 322   bool is_any_caps();
 323   bool cap_is_valid(const Cap &cap) const;
 324   int caps_issued(int *implemented = 0) const;
 325   void try_touch_cap(mds_rank_t mds);
 326   bool caps_issued_mask(unsigned mask, bool allow_impl=false);
 327   int caps_used();
 328   int caps_file_wanted();
 329   int caps_wanted();
 330   int caps_mds_wanted();
 331   int caps_dirty();
 332   const UserPerm *get_best_perms();
 333
 334   bool have_valid_size();
 335   Dir *open_dir();
 336
 337   void add_fh(Fh *f) {fhs.insert(f);}
 338   void rm_fh(Fh *f) {fhs.erase(f);}
 339   void set_async_err(int r);
 340   void dump(Formatter *f) const;
 341
 342   void break_all_delegs() { break_deleg(false); };
 343
 344   void recall_deleg(bool skip_read);
 345   bool has_recalled_deleg();
 346   int set_deleg(Fh *fh, unsigned type, ceph_deleg_cb_t cb, void *priv);
 347   void unset_deleg(Fh *fh);
 348
 349   void mark_caps_dirty(int caps);
 350   void mark_caps_clean();
 351 private:
 352   // how many opens for write on this Inode?
 353   long open_count_for_write()
 354   {
 355     return (long)(open_by_mode[CEPH_FILE_MODE_RDWR] +
 356                   open_by_mode[CEPH_FILE_MODE_WR]);
 357   };
 358
 359   // how many opens of any sort on this inode?
 360   long open_count()
 361   {
 362     return (long) std::accumulate(open_by_mode.begin(), open_by_mode.end(), 0,
 363                                   [] (int value, const std::map<int, int>::value_type& p)
 364                    { return value + p.second; });
 365   };
 366
 367   void break_deleg(bool skip_read);
 368   bool delegations_broken(bool skip_read);
 369
 370 };
 371
 372 ostream& operator<<(ostream &out, const Inode &in);
 373
 374 #endif