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