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 = 0, dirty = 0;
  82
  83   uint64_t size = 0;
  84   utime_t ctime, btime, mtime, atime;
  85   version_t time_warp_seq = 0;
  86   uint64_t change_attr = 0;
  87   uint32_t   mode = 0;
  88   uid_t      uid = 0;
  89   gid_t      gid = 0;
  90   std::map<std::string,bufferptr> xattrs;
  91   version_t xattr_version = 0;
  92
  93   bufferlist inline_data;
  94   version_t inline_version = 0;
  95
  96   bool writing = false, dirty_data = false;
  97   uint64_t flush_tid = 0;
  98
  99   int64_t cap_dirtier_uid = -1;
 100   int64_t cap_dirtier_gid = -1;
 101
 102   explicit CapSnap(Inode *i)
 103     : in(i)
 104   {}
 105
 106   void dump(Formatter *f) const;
 107 };
 108
 109 // inode flags
 110 #define I_COMPLETE              (1 << 0)
 111 #define I_DIR_ORDERED           (1 << 1)
 112 #define I_SNAPDIR_OPEN          (1 << 2)
 113 #define I_KICK_FLUSH            (1 << 3)
 114 #define I_CAP_DROPPED           (1 << 4)
 115 #define I_ERROR_FILELOCK        (1 << 5)
 116
 117 struct Inode : RefCountedObject {
 118   ceph::coarse_mono_time hold_caps_until;
 119   Client *client;
 120
 121   // -- the actual inode --
 122   inodeno_t ino; // ORDER DEPENDENCY: oset
 123   snapid_t  snapid;
 124   ino_t faked_ino = 0;
 125
 126   uint32_t   rdev = 0;    // 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 = 0;
 134   uid_t      uid = 0;
 135   gid_t      gid = 0;
 136
 137   // nlink
 138   int32_t    nlink = 0;
 139
 140   // file (data access)
 141   ceph_dir_layout dir_layout{};
 142   file_layout_t layout;
 143   uint64_t   size = 0;        // on directory, # dentries
 144   uint32_t   truncate_seq = 1;
 145   uint64_t   truncate_size = -1;
 146   utime_t    mtime;   // file data modify time.
 147   utime_t    atime;   // file data access time.
 148   uint32_t   time_warp_seq = 0;  // count of (potential) mtime/atime timewarps (i.e., utimes())
 149   uint64_t   change_attr = 0;
 150
 151   uint64_t max_size = 0;  // 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 = 0;           // auth only
 159   version_t xattr_version = 0;
 160   utime_t   snap_btime;        // snapshot creation (birth) time
 161   std::map<std::string, std::string> snap_metadata;
 162
 163   // inline data
 164   version_t  inline_version = 0;
 165   bufferlist inline_data;
 166
 167   bool fscrypt = false; // fscrypt enabled ?
 168
 169   bool is_root()    const { return ino == CEPH_INO_ROOT; }
 170   bool is_symlink() const { return (mode & S_IFMT) == S_IFLNK; }
 171   bool is_dir()     const { return (mode & S_IFMT) == S_IFDIR; }
 172   bool is_file()    const { return (mode & S_IFMT) == S_IFREG; }
 173
 174   bool has_dir_layout() const {
 175     return layout != file_layout_t();
 176   }
 177
 178   __u32 hash_dentry_name(const std::string &dn) {
 179     int which = dir_layout.dl_dir_hash;
 180     if (!which)
 181       which = CEPH_STR_HASH_LINUX;
 182     ceph_assert(ceph_str_hash_valid(which));
 183     return ceph_str_hash(which, dn.data(), dn.length());
 184   }
 185
 186   unsigned flags = 0;
 187
 188   quota_info_t quota;
 189
 190   bool is_complete_and_ordered() {
 191     static const unsigned wants = I_COMPLETE | I_DIR_ORDERED;
 192     return (flags & wants) == wants;
 193   }
 194
 195   // about the dir (if this is one!)
 196   Dir       *dir = 0;     // if i'm a dir.
 197   fragtree_t dirfragtree;
 198   uint64_t dir_release_count = 1;
 199   uint64_t dir_ordered_count = 1;
 200   bool dir_hashed = false;
 201   bool dir_replicated = false;
 202
 203   // per-mds caps
 204   std::map<mds_rank_t, Cap> caps;            // mds -> Cap
 205   Cap *auth_cap = 0;
 206   int64_t cap_dirtier_uid = -1;
 207   int64_t cap_dirtier_gid = -1;
 208   unsigned dirty_caps = 0;
 209   unsigned flushing_caps = 0;
 210   std::map<ceph_tid_t, int> flushing_cap_tids;
 211   int shared_gen = 0;
 212   int cache_gen = 0;
 213   int snap_caps = 0;
 214   int snap_cap_refs = 0;
 215   xlist<Inode*>::item delay_cap_item, dirty_cap_item, flushing_cap_item;
 216
 217   SnapRealm *snaprealm = 0;
 218   xlist<Inode*>::item snaprealm_item;
 219   InodeRef snapdir_parent;  // only if we are a snapdir inode
 220   std::map<snapid_t,CapSnap> cap_snaps;   // pending flush to mds
 221
 222   //int open_by_mode[CEPH_FILE_MODE_NUM];
 223   std::map<int,int> open_by_mode;
 224   std::map<int,int> cap_refs;
 225
 226   ObjectCacher::ObjectSet oset; // ORDER DEPENDENCY: ino
 227
 228   uint64_t reported_size = 0;
 229   uint64_t wanted_max_size = 0;
 230   uint64_t requested_max_size = 0;
 231
 232   uint64_t  ll_ref = 0;   // separate ref count for ll client
 233   xlist<Dentry *> dentries; // if i'm linked to a dentry.
 234   std::string    symlink;  // symlink content, if it's a symlink
 235   std::map<std::string,bufferptr> xattrs;
 236   std::map<frag_t,int> fragmap;  // known frag -> mds mappings
 237   std::map<frag_t, std::vector<mds_rank_t>> frag_repmap; // non-auth mds mappings
 238
 239   std::list<ceph::condition_variable*> waitfor_caps;
 240   std::list<ceph::condition_variable*> waitfor_commit;
 241   std::list<ceph::condition_variable*> waitfor_deleg;
 242
 243   Dentry *get_first_parent() {
 244     ceph_assert(!dentries.empty());
 245     return *dentries.begin();
 246   }
 247
 248   void make_long_path(filepath& p);
 249   void make_short_path(filepath& p);
 250   void make_nosnap_relative_path(filepath& p);
 251
 252   // The ref count. 1 for each dentry, fh, inode_map,
 253   // cwd that links to me.
 254   void iget() { get(); }
 255   void iput(int n=1) { ceph_assert(n >= 0); while (n--) put(); }
 256
 257   void ll_get() {
 258     ll_ref++;
 259   }
 260   void ll_put(uint64_t n=1) {
 261     ceph_assert(ll_ref >= n);
 262     ll_ref -= n;
 263   }
 264
 265   // file locks
 266   std::unique_ptr<ceph_lock_state_t> fcntl_locks;
 267   std::unique_ptr<ceph_lock_state_t> flock_locks;
 268
 269   bool has_any_filelocks() {
 270     return
 271       (fcntl_locks && !fcntl_locks->empty()) ||
 272       (flock_locks && !flock_locks->empty());
 273   }
 274
 275   std::list<Delegation> delegations;
 276
 277   xlist<MetaRequest*> unsafe_ops;
 278
 279   std::set<Fh*> fhs;
 280
 281   mds_rank_t dir_pin = MDS_RANK_NONE;
 282
 283   Inode() = delete;
 284   Inode(Client *c, vinodeno_t vino, file_layout_t *newlayout)
 285     : client(c), ino(vino.ino), snapid(vino.snapid), delay_cap_item(this),
 286       dirty_cap_item(this), flushing_cap_item(this), snaprealm_item(this),
 287       oset((void *)this, newlayout->pool_id, this->ino) {}
 288   ~Inode();
 289
 290   vinodeno_t vino() const { return vinodeno_t(ino, snapid); }
 291
 292   struct Compare {
 293     bool operator() (Inode* const & left, Inode* const & right) {
 294       if (left->ino.val < right->ino.val) {
 295         return (left->snapid.val < right->snapid.val);
 296       }
 297       return false;
 298     }
 299   };
 300
 301   bool check_mode(const UserPerm& perms, unsigned want);
 302
 303   // CAPS --------
 304   void get_open_ref(int mode);
 305   bool put_open_ref(int mode);
 306
 307   void get_cap_ref(int cap);
 308   int put_cap_ref(int cap);
 309   bool is_any_caps();
 310   bool cap_is_valid(const Cap &cap) const;
 311   int caps_issued(int *implemented = 0) const;
 312   void try_touch_cap(mds_rank_t mds);
 313   bool caps_issued_mask(unsigned mask, bool allow_impl=false);
 314   int caps_used();
 315   int caps_file_wanted();
 316   int caps_wanted();
 317   int caps_mds_wanted();
 318   int caps_dirty();
 319   const UserPerm *get_best_perms();
 320
 321   bool have_valid_size();
 322   Dir *open_dir();
 323
 324   void add_fh(Fh *f) {fhs.insert(f);}
 325   void rm_fh(Fh *f) {fhs.erase(f);}
 326   void set_async_err(int r);
 327   void dump(Formatter *f) const;
 328
 329   void break_all_delegs() { break_deleg(false); };
 330
 331   void recall_deleg(bool skip_read);
 332   bool has_recalled_deleg();
 333   int set_deleg(Fh *fh, unsigned type, ceph_deleg_cb_t cb, void *priv);
 334   void unset_deleg(Fh *fh);
 335
 336   void mark_caps_dirty(int caps);
 337   void mark_caps_clean();
 338 private:
 339   // how many opens for write on this Inode?
 340   long open_count_for_write()
 341   {
 342     return (long)(open_by_mode[CEPH_FILE_MODE_RDWR] +
 343                   open_by_mode[CEPH_FILE_MODE_WR]);
 344   };
 345
 346   // how many opens of any sort on this inode?
 347   long open_count()
 348   {
 349     return (long) std::accumulate(open_by_mode.begin(), open_by_mode.end(), 0,
 350                                   [] (int value, const std::map<int, int>::value_type& p)
 351                    { return value + p.second; });
 352   };
 353
 354   void break_deleg(bool skip_read);
 355   bool delegations_broken(bool skip_read);
 356
 357 };
 358
 359 std::ostream& operator<<(std::ostream &out, const Inode &in);
 360
 361 #endif