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 "common/CommandTable.h"
  20 #include "common/Finisher.h"
  21 #include "common/Timer.h"
  22 #include "common/ceph_mutex.h"
  23 #include "common/cmdparse.h"
  24 #include "common/compiler_extensions.h"
  25 #include "include/common_fwd.h"
  26 #include "include/cephfs/ceph_ll_client.h"
  27 #include "include/filepath.h"
  28 #include "include/interval_set.h"
  29 #include "include/lru.h"
  30 #include "include/types.h"
  31 #include "include/unordered_map.h"
  32 #include "include/unordered_set.h"
  33 #include "include/cephfs/metrics/Types.h"
  34 #include "mds/mdstypes.h"
  35 #include "msg/Dispatcher.h"
  36 #include "msg/MessageRef.h"
  37 #include "msg/Messenger.h"
  38 #include "osdc/ObjectCacher.h"
  39
  40 #include "RWRef.h"
  41 #include "InodeRef.h"
  42 #include "MetaSession.h"
  43 #include "UserPerm.h"
  44
  45 #include <fstream>
  46 #include <map>
  47 #include <memory>
  48 #include <set>
  49 #include <string>
  50 #include <thread>
  51
  52 using std::set;
  53 using std::map;
  54 using std::fstream;
  55
  56 class FSMap;
  57 class FSMapUser;
  58 class MonClient;
  59
  60
  61 struct DirStat;
  62 struct LeaseStat;
  63 struct InodeStat;
  64
  65 class Filer;
  66 class Objecter;
  67 class WritebackHandler;
  68
  69 class MDSMap;
  70 class Message;
  71 class destructive_lock_ref_t;
  72
  73 enum {
  74   l_c_first = 20000,
  75   l_c_reply,
  76   l_c_lat,
  77   l_c_wrlat,
  78   l_c_read,
  79   l_c_fsync,
  80   l_c_last,
  81 };
  82
  83
  84 class MDSCommandOp : public CommandOp
  85 {
  86   public:
  87   mds_gid_t     mds_gid;
  88
  89   explicit MDSCommandOp(ceph_tid_t t) : CommandOp(t) {}
  90 };
  91
  92 /* error code for ceph_fuse */
  93 #define CEPH_FUSE_NO_MDS_UP    -((1<<16)+0) /* no mds up deteced in ceph_fuse */
  94 #define CEPH_FUSE_LAST         -((1<<16)+1) /* (unused) */
  95
  96 // ============================================
  97 // types for my local metadata cache
  98 /* basic structure:
  99
 100  - Dentries live in an LRU loop.  they get expired based on last access.
 101       see include/lru.h.  items can be bumped to "mid" or "top" of list, etc.
 102  - Inode has ref count for each Fh, Dir, or Dentry that points to it.
 103  - when Inode ref goes to 0, it's expired.
 104  - when Dir is empty, it's removed (and it's Inode ref--)
 105
 106 */
 107
 108 /* getdir result */
 109 struct DirEntry {
 110   explicit DirEntry(const std::string &s) : d_name(s), stmask(0) {}
 111   DirEntry(const std::string &n, struct stat& s, int stm)
 112     : d_name(n), st(s), stmask(stm) {}
 113
 114   std::string d_name;
 115   struct stat st;
 116   int stmask;
 117 };
 118
 119 struct Cap;
 120 class Dir;
 121 class Dentry;
 122 struct SnapRealm;
 123 struct Fh;
 124 struct CapSnap;
 125
 126 struct MetaRequest;
 127 class ceph_lock_state_t;
 128
 129 // ========================================================
 130 // client interface
 131
 132 struct dir_result_t {
 133   static const int SHIFT = 28;
 134   static const int64_t MASK = (1 << SHIFT) - 1;
 135   static const int64_t HASH = 0xFFULL << (SHIFT + 24); // impossible frag bits
 136   static const loff_t END = 1ULL << (SHIFT + 32);
 137
 138   struct dentry {
 139     int64_t offset;
 140     std::string name;
 141     std::string alternate_name;
 142     InodeRef inode;
 143     explicit dentry(int64_t o) : offset(o) {}
 144     dentry(int64_t o, std::string n, std::string an, InodeRef in) :
 145       offset(o), name(std::move(n)), alternate_name(std::move(an)), inode(std::move(in)) {}
 146   };
 147   struct dentry_off_lt {
 148     bool operator()(const dentry& d, int64_t off) const {
 149       return dir_result_t::fpos_cmp(d.offset, off) < 0;
 150     }
 151   };
 152
 153
 154   explicit dir_result_t(Inode *in, const UserPerm& perms);
 155
 156
 157   static uint64_t make_fpos(unsigned h, unsigned l, bool hash) {
 158     uint64_t v =  ((uint64_t)h<< SHIFT) | (uint64_t)l;
 159     if (hash)
 160       v |= HASH;
 161     else
 162       ceph_assert((v & HASH) != HASH);
 163     return v;
 164   }
 165   static unsigned fpos_high(uint64_t p) {
 166     unsigned v = (p & (END-1)) >> SHIFT;
 167     if ((p & HASH) == HASH)
 168       return ceph_frag_value(v);
 169     return v;
 170   }
 171   static unsigned fpos_low(uint64_t p) {
 172     return p & MASK;
 173   }
 174   static int fpos_cmp(uint64_t l, uint64_t r) {
 175     int c = ceph_frag_compare(fpos_high(l), fpos_high(r));
 176     if (c)
 177       return c;
 178     if (fpos_low(l) == fpos_low(r))
 179       return 0;
 180     return fpos_low(l) < fpos_low(r) ? -1 : 1;
 181   }
 182
 183   unsigned offset_high() { return fpos_high(offset); }
 184   unsigned offset_low() { return fpos_low(offset); }
 185
 186   void set_end() { offset |= END; }
 187   bool at_end() { return (offset & END); }
 188
 189   void set_hash_order() { offset |= HASH; }
 190   bool hash_order() { return (offset & HASH) == HASH; }
 191
 192   bool is_cached() {
 193     if (buffer.empty())
 194       return false;
 195     if (hash_order()) {
 196       return buffer_frag.contains(offset_high());
 197     } else {
 198       return buffer_frag == frag_t(offset_high());
 199     }
 200   }
 201
 202   void reset() {
 203     last_name.clear();
 204     next_offset = 2;
 205     offset = 0;
 206     ordered_count = 0;
 207     cache_index = 0;
 208     buffer.clear();
 209   }
 210
 211   InodeRef inode;
 212   int64_t offset;        // hash order:
 213                          //   (0xff << 52) | ((24 bits hash) << 28) |
 214                          //   (the nth entry has hash collision);
 215                          // frag+name order;
 216                          //   ((frag value) << 28) | (the nth entry in frag);
 217
 218   unsigned next_offset;  // offset of next chunk (last_name's + 1)
 219   std::string last_name;      // last entry in previous chunk
 220
 221   uint64_t release_count;
 222   uint64_t ordered_count;
 223   unsigned cache_index;
 224   int start_shared_gen;  // dir shared_gen at start of readdir
 225   UserPerm perms;
 226
 227   frag_t buffer_frag;
 228
 229   std::vector<dentry> buffer;
 230   struct dirent de;
 231 };
 232
 233 class Client : public Dispatcher, public md_config_obs_t {
 234 public:
 235   friend class C_Block_Sync; // Calls block map and protected helpers
 236   friend class C_Client_CacheInvalidate;  // calls ino_invalidate_cb
 237   friend class C_Client_DentryInvalidate;  // calls dentry_invalidate_cb
 238   friend class C_Client_FlushComplete; // calls put_inode()
 239   friend class C_Client_Remount;
 240   friend class C_Client_RequestInterrupt;
 241   friend class C_Deleg_Timeout; // Asserts on client_lock, called when a delegation is unreturned
 242   friend class C_Client_CacheRelease; // Asserts on client_lock
 243   friend class SyntheticClient;
 244   friend void intrusive_ptr_release(Inode *in);
 245   template <typename T> friend struct RWRefState;
 246   template <typename T> friend class RWRef;
 247
 248   using Dispatcher::cct;
 249   using clock = ceph::coarse_mono_clock;
 250
 251   typedef int (*add_dirent_cb_t)(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in);
 252
 253   struct walk_dentry_result {
 254     InodeRef in;
 255     std::string alternate_name;
 256   };
 257
 258   class CommandHook : public AdminSocketHook {
 259   public:
 260     explicit CommandHook(Client *client);
 261     int call(std::string_view command, const cmdmap_t& cmdmap,
 262              Formatter *f,
 263              std::ostream& errss,
 264              bufferlist& out) override;
 265   private:
 266     Client *m_client;
 267   };
 268
 269   // snapshot info returned via get_snap_info(). nothing to do
 270   // with SnapInfo on the MDS.
 271   struct SnapInfo {
 272     snapid_t id;
 273     std::map<std::string, std::string> metadata;
 274   };
 275
 276   Client(Messenger *m, MonClient *mc, Objecter *objecter_);
 277   Client(const Client&) = delete;
 278   Client(const Client&&) = delete;
 279   virtual ~Client() override;
 280
 281   static UserPerm pick_my_perms(CephContext *c) {
 282     uid_t uid = c->_conf->client_mount_uid >= 0 ? c->_conf->client_mount_uid : -1;
 283     gid_t gid = c->_conf->client_mount_gid >= 0 ? c->_conf->client_mount_gid : -1;
 284     return UserPerm(uid, gid);
 285   }
 286   UserPerm pick_my_perms() {
 287     uid_t uid = user_id >= 0 ? user_id : -1;
 288     gid_t gid = group_id >= 0 ? group_id : -1;
 289     return UserPerm(uid, gid);
 290   }
 291
 292   int mount(const std::string &mount_root, const UserPerm& perms,
 293             bool require_mds=false, const std::string &fs_name="");
 294   void unmount();
 295   bool is_unmounting() const {
 296     return mount_state.check_current_state(CLIENT_UNMOUNTING);
 297   }
 298   bool is_mounted() const {
 299     return mount_state.check_current_state(CLIENT_MOUNTED);
 300   }
 301   bool is_mounting() const {
 302     return mount_state.check_current_state(CLIENT_MOUNTING);
 303   }
 304   bool is_initialized() const {
 305     return initialize_state.check_current_state(CLIENT_INITIALIZED);
 306   }
 307   void abort_conn();
 308
 309   void set_uuid(const std::string& uuid);
 310   void set_session_timeout(unsigned timeout);
 311   int start_reclaim(const std::string& uuid, unsigned flags,
 312                     const std::string& fs_name);
 313   void finish_reclaim();
 314
 315   fs_cluster_id_t get_fs_cid() {
 316     return fscid;
 317   }
 318
 319   int mds_command(
 320     const std::string &mds_spec,
 321     const std::vector<std::string>& cmd,
 322     const bufferlist& inbl,
 323     bufferlist *poutbl, std::string *prs, Context *onfinish);
 324
 325   // these should (more or less) mirror the actual system calls.
 326   int statfs(const char *path, struct statvfs *stbuf, const UserPerm& perms);
 327
 328   // crap
 329   int chdir(const char *s, std::string &new_cwd, const UserPerm& perms);
 330   void _getcwd(std::string& cwd, const UserPerm& perms);
 331   void getcwd(std::string& cwd, const UserPerm& perms);
 332
 333   // namespace ops
 334   int opendir(const char *name, dir_result_t **dirpp, const UserPerm& perms);
 335   int fdopendir(int dirfd, dir_result_t **dirpp, const UserPerm& perms);
 336   int closedir(dir_result_t *dirp);
 337
 338   /**
 339    * Fill a directory listing from dirp, invoking cb for each entry
 340    * with the given pointer, the dirent, the struct stat, the stmask,
 341    * and the offset.
 342    *
 343    * Returns 0 if it reached the end of the directory.
 344    * If @a cb returns a negative error code, stop and return that.
 345    */
 346   int readdir_r_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
 347                    unsigned want=0, unsigned flags=AT_NO_ATTR_SYNC,
 348                    bool getref=false);
 349
 350   struct dirent * readdir(dir_result_t *d);
 351   int readdir_r(dir_result_t *dirp, struct dirent *de);
 352   int readdirplus_r(dir_result_t *dirp, struct dirent *de, struct ceph_statx *stx, unsigned want, unsigned flags, Inode **out);
 353
 354   int getdir(const char *relpath, std::list<std::string>& names,
 355              const UserPerm& perms);  // get the whole dir at once.
 356
 357   /**
 358    * Returns the length of the buffer that got filled in, or -errno.
 359    * If it returns -CEPHFS_ERANGE you just need to increase the size of the
 360    * buffer and try again.
 361    */
 362   int _getdents(dir_result_t *dirp, char *buf, int buflen, bool ful);  // get a bunch of dentries at once
 363   int getdents(dir_result_t *dirp, char *buf, int buflen) {
 364     return _getdents(dirp, buf, buflen, true);
 365   }
 366   int getdnames(dir_result_t *dirp, char *buf, int buflen) {
 367     return _getdents(dirp, buf, buflen, false);
 368   }
 369
 370   void rewinddir(dir_result_t *dirp);
 371   loff_t telldir(dir_result_t *dirp);
 372   void seekdir(dir_result_t *dirp, loff_t offset);
 373
 374   int may_delete(const char *relpath, const UserPerm& perms);
 375   int link(const char *existing, const char *newname, const UserPerm& perm, std::string alternate_name="");
 376   int unlink(const char *path, const UserPerm& perm);
 377   int unlinkat(int dirfd, const char *relpath, int flags, const UserPerm& perm);
 378   int rename(const char *from, const char *to, const UserPerm& perm, std::string alternate_name="");
 379
 380   // dirs
 381   int mkdir(const char *path, mode_t mode, const UserPerm& perm, std::string alternate_name="");
 382   int mkdirat(int dirfd, const char *relpath, mode_t mode, const UserPerm& perm,
 383               std::string alternate_name="");
 384   int mkdirs(const char *path, mode_t mode, const UserPerm& perms);
 385   int rmdir(const char *path, const UserPerm& perms);
 386
 387   // symlinks
 388   int readlink(const char *path, char *buf, loff_t size, const UserPerm& perms);
 389   int readlinkat(int dirfd, const char *relpath, char *buf, loff_t size, const UserPerm& perms);
 390
 391   int symlink(const char *existing, const char *newname, const UserPerm& perms, std::string alternate_name="");
 392   int symlinkat(const char *target, int dirfd, const char *relpath, const UserPerm& perms,
 393                 std::string alternate_name="");
 394
 395   // path traversal for high-level interface
 396   int walk(std::string_view path, struct walk_dentry_result* result, const UserPerm& perms, bool followsym=true);
 397
 398   // inode stuff
 399   unsigned statx_to_mask(unsigned int flags, unsigned int want);
 400   int stat(const char *path, struct stat *stbuf, const UserPerm& perms,
 401            frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL);
 402   int statx(const char *path, struct ceph_statx *stx,
 403             const UserPerm& perms,
 404             unsigned int want, unsigned int flags);
 405   int lstat(const char *path, struct stat *stbuf, const UserPerm& perms,
 406             frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL);
 407
 408   int setattr(const char *relpath, struct stat *attr, int mask,
 409               const UserPerm& perms);
 410   int setattrx(const char *relpath, struct ceph_statx *stx, int mask,
 411                const UserPerm& perms, int flags=0);
 412   int fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms);
 413   int fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms);
 414   int chmod(const char *path, mode_t mode, const UserPerm& perms);
 415   int fchmod(int fd, mode_t mode, const UserPerm& perms);
 416   int chmodat(int dirfd, const char *relpath, mode_t mode, int flags, const UserPerm& perms);
 417   int lchmod(const char *path, mode_t mode, const UserPerm& perms);
 418   int chown(const char *path, uid_t new_uid, gid_t new_gid,
 419             const UserPerm& perms);
 420   int fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms);
 421   int lchown(const char *path, uid_t new_uid, gid_t new_gid,
 422              const UserPerm& perms);
 423   int chownat(int dirfd, const char *relpath, uid_t new_uid, gid_t new_gid,
 424               int flags, const UserPerm& perms);
 425   int utime(const char *path, struct utimbuf *buf, const UserPerm& perms);
 426   int lutime(const char *path, struct utimbuf *buf, const UserPerm& perms);
 427   int futime(int fd, struct utimbuf *buf, const UserPerm& perms);
 428   int utimes(const char *relpath, struct timeval times[2], const UserPerm& perms);
 429   int lutimes(const char *relpath, struct timeval times[2], const UserPerm& perms);
 430   int futimes(int fd, struct timeval times[2], const UserPerm& perms);
 431   int futimens(int fd, struct timespec times[2], const UserPerm& perms);
 432   int utimensat(int dirfd, const char *relpath, struct timespec times[2], int flags,
 433                 const UserPerm& perms);
 434   int flock(int fd, int operation, uint64_t owner);
 435   int truncate(const char *path, loff_t size, const UserPerm& perms);
 436
 437   // file ops
 438   int mknod(const char *path, mode_t mode, const UserPerm& perms, dev_t rdev=0);
 439
 440   int create_and_open(int dirfd, const char *relpath, int flags, const UserPerm& perms,
 441                       mode_t mode, int stripe_unit, int stripe_count, int object_size,
 442                       const char *data_pool, std::string alternate_name);
 443   int open(const char *path, int flags, const UserPerm& perms, mode_t mode=0, std::string alternate_name="") {
 444     return open(path, flags, perms, mode, 0, 0, 0, NULL, alternate_name);
 445   }
 446   int open(const char *path, int flags, const UserPerm& perms,
 447            mode_t mode, int stripe_unit, int stripe_count, int object_size,
 448            const char *data_pool, std::string alternate_name="");
 449   int openat(int dirfd, const char *relpath, int flags, const UserPerm& perms,
 450              mode_t mode, int stripe_unit, int stripe_count,
 451              int object_size, const char *data_pool, std::string alternate_name);
 452   int openat(int dirfd, const char *path, int flags, const UserPerm& perms, mode_t mode=0,
 453              std::string alternate_name="") {
 454     return openat(dirfd, path, flags, perms, mode, 0, 0, 0, NULL, alternate_name);
 455   }
 456
 457   int lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name,
 458                   const UserPerm& perms);
 459   int lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode=NULL);
 460   int lookup_name(Inode *in, Inode *parent, const UserPerm& perms);
 461   int _close(int fd);
 462   int close(int fd);
 463   loff_t lseek(int fd, loff_t offset, int whence);
 464   int read(int fd, char *buf, loff_t size, loff_t offset=-1);
 465   int preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1);
 466   int write(int fd, const char *buf, loff_t size, loff_t offset=-1);
 467   int pwritev(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1);
 468   int fake_write_size(int fd, loff_t size);
 469   int ftruncate(int fd, loff_t size, const UserPerm& perms);
 470   int fsync(int fd, bool syncdataonly);
 471   int fstat(int fd, struct stat *stbuf, const UserPerm& perms,
 472             int mask=CEPH_STAT_CAP_INODE_ALL);
 473   int fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms,
 474              unsigned int want, unsigned int flags);
 475   int statxat(int dirfd, const char *relpath,
 476               struct ceph_statx *stx, const UserPerm& perms,
 477               unsigned int want, unsigned int flags);
 478   int fallocate(int fd, int mode, loff_t offset, loff_t length);
 479
 480   // full path xattr ops
 481   int getxattr(const char *path, const char *name, void *value, size_t size,
 482                const UserPerm& perms);
 483   int lgetxattr(const char *path, const char *name, void *value, size_t size,
 484                 const UserPerm& perms);
 485   int fgetxattr(int fd, const char *name, void *value, size_t size,
 486                 const UserPerm& perms);
 487   int listxattr(const char *path, char *list, size_t size, const UserPerm& perms);
 488   int llistxattr(const char *path, char *list, size_t size, const UserPerm& perms);
 489   int flistxattr(int fd, char *list, size_t size, const UserPerm& perms);
 490   int removexattr(const char *path, const char *name, const UserPerm& perms);
 491   int lremovexattr(const char *path, const char *name, const UserPerm& perms);
 492   int fremovexattr(int fd, const char *name, const UserPerm& perms);
 493   int setxattr(const char *path, const char *name, const void *value,
 494                size_t size, int flags, const UserPerm& perms);
 495   int lsetxattr(const char *path, const char *name, const void *value,
 496                 size_t size, int flags, const UserPerm& perms);
 497   int fsetxattr(int fd, const char *name, const void *value, size_t size,
 498                 int flags, const UserPerm& perms);
 499
 500   int sync_fs();
 501   int64_t drop_caches();
 502
 503   int get_snap_info(const char *path, const UserPerm &perms, SnapInfo *snap_info);
 504
 505   // hpc lazyio
 506   int lazyio(int fd, int enable);
 507   int lazyio_propagate(int fd, loff_t offset, size_t count);
 508   int lazyio_synchronize(int fd, loff_t offset, size_t count);
 509
 510   // expose file layout
 511   int describe_layout(const char *path, file_layout_t* layout,
 512                       const UserPerm& perms);
 513   int fdescribe_layout(int fd, file_layout_t* layout);
 514   int get_file_stripe_address(int fd, loff_t offset, std::vector<entity_addr_t>& address);
 515   int get_file_extent_osds(int fd, loff_t off, loff_t *len, std::vector<int>& osds);
 516   int get_osd_addr(int osd, entity_addr_t& addr);
 517
 518   // expose mdsmap
 519   int64_t get_default_pool_id();
 520
 521   // expose osdmap
 522   int get_local_osd();
 523   int get_pool_replication(int64_t pool);
 524   int64_t get_pool_id(const char *pool_name);
 525   std::string get_pool_name(int64_t pool);
 526   int get_osd_crush_location(int id, std::vector<std::pair<std::string, std::string> >& path);
 527
 528   int enumerate_layout(int fd, std::vector<ObjectExtent>& result,
 529                        loff_t length, loff_t offset);
 530
 531   int mksnap(const char *path, const char *name, const UserPerm& perm,
 532              mode_t mode=0, const std::map<std::string, std::string> &metadata={});
 533   int rmsnap(const char *path, const char *name, const UserPerm& perm, bool check_perms=false);
 534
 535   // Inode permission checking
 536   int inode_permission(Inode *in, const UserPerm& perms, unsigned want);
 537
 538   // expose caps
 539   int get_caps_issued(int fd);
 540   int get_caps_issued(const char *path, const UserPerm& perms);
 541
 542   snapid_t ll_get_snapid(Inode *in);
 543   vinodeno_t ll_get_vino(Inode *in) {
 544     std::lock_guard lock(client_lock);
 545     return _get_vino(in);
 546   }
 547   // get inode from faked ino
 548   Inode *ll_get_inode(ino_t ino);
 549   Inode *ll_get_inode(vinodeno_t vino);
 550   int ll_lookup(Inode *parent, const char *name, struct stat *attr,
 551                 Inode **out, const UserPerm& perms);
 552   int ll_lookup_inode(struct inodeno_t ino, const UserPerm& perms, Inode **inode);
 553   int ll_lookup_vino(vinodeno_t vino, const UserPerm& perms, Inode **inode);
 554   int ll_lookupx(Inode *parent, const char *name, Inode **out,
 555                         struct ceph_statx *stx, unsigned want, unsigned flags,
 556                         const UserPerm& perms);
 557   bool ll_forget(Inode *in, uint64_t count);
 558   bool ll_put(Inode *in);
 559   int ll_get_snap_ref(snapid_t snap);
 560
 561   int ll_getattr(Inode *in, struct stat *st, const UserPerm& perms);
 562   int ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want,
 563                   unsigned int flags, const UserPerm& perms);
 564   int ll_setattrx(Inode *in, struct ceph_statx *stx, int mask,
 565                   const UserPerm& perms);
 566   int ll_setattr(Inode *in, struct stat *st, int mask,
 567                  const UserPerm& perms);
 568   int ll_getxattr(Inode *in, const char *name, void *value, size_t size,
 569                   const UserPerm& perms);
 570   int ll_setxattr(Inode *in, const char *name, const void *value, size_t size,
 571                   int flags, const UserPerm& perms);
 572   int ll_removexattr(Inode *in, const char *name, const UserPerm& perms);
 573   int ll_listxattr(Inode *in, char *list, size_t size, const UserPerm& perms);
 574   int ll_opendir(Inode *in, int flags, dir_result_t **dirpp,
 575                  const UserPerm& perms);
 576   int ll_releasedir(dir_result_t* dirp);
 577   int ll_fsyncdir(dir_result_t* dirp);
 578   int ll_readlink(Inode *in, char *buf, size_t bufsize, const UserPerm& perms);
 579   int ll_mknod(Inode *in, const char *name, mode_t mode, dev_t rdev,
 580                struct stat *attr, Inode **out, const UserPerm& perms);
 581   int ll_mknodx(Inode *parent, const char *name, mode_t mode, dev_t rdev,
 582                 Inode **out, struct ceph_statx *stx, unsigned want,
 583                 unsigned flags, const UserPerm& perms);
 584   int ll_mkdir(Inode *in, const char *name, mode_t mode, struct stat *attr,
 585                Inode **out, const UserPerm& perm);
 586   int ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out,
 587                 struct ceph_statx *stx, unsigned want, unsigned flags,
 588                 const UserPerm& perms);
 589   int ll_symlink(Inode *in, const char *name, const char *value,
 590                  struct stat *attr, Inode **out, const UserPerm& perms);
 591   int ll_symlinkx(Inode *parent, const char *name, const char *value,
 592                   Inode **out, struct ceph_statx *stx, unsigned want,
 593                   unsigned flags, const UserPerm& perms);
 594   int ll_unlink(Inode *in, const char *name, const UserPerm& perm);
 595   int ll_rmdir(Inode *in, const char *name, const UserPerm& perms);
 596   int ll_rename(Inode *parent, const char *name, Inode *newparent,
 597                 const char *newname, const UserPerm& perm);
 598   int ll_link(Inode *in, Inode *newparent, const char *newname,
 599               const UserPerm& perm);
 600   int ll_open(Inode *in, int flags, Fh **fh, const UserPerm& perms);
 601   int _ll_create(Inode *parent, const char *name, mode_t mode,
 602               int flags, InodeRef *in, int caps, Fh **fhp,
 603               const UserPerm& perms);
 604   int ll_create(Inode *parent, const char *name, mode_t mode, int flags,
 605                 struct stat *attr, Inode **out, Fh **fhp,
 606                 const UserPerm& perms);
 607   int ll_createx(Inode *parent, const char *name, mode_t mode,
 608                 int oflags, Inode **outp, Fh **fhp,
 609                 struct ceph_statx *stx, unsigned want, unsigned lflags,
 610                 const UserPerm& perms);
 611   int ll_read_block(Inode *in, uint64_t blockid, char *buf,  uint64_t offset,
 612                     uint64_t length, file_layout_t* layout);
 613
 614   int ll_write_block(Inode *in, uint64_t blockid,
 615                      char* buf, uint64_t offset,
 616                      uint64_t length, file_layout_t* layout,
 617                      uint64_t snapseq, uint32_t sync);
 618   int ll_commit_blocks(Inode *in, uint64_t offset, uint64_t length);
 619
 620   int ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms);
 621   int ll_walk(const char* name, Inode **i, struct ceph_statx *stx,
 622                unsigned int want, unsigned int flags, const UserPerm& perms);
 623   uint32_t ll_stripe_unit(Inode *in);
 624   int ll_file_layout(Inode *in, file_layout_t *layout);
 625   uint64_t ll_snap_seq(Inode *in);
 626
 627   int ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl);
 628   int ll_write(Fh *fh, loff_t off, loff_t len, const char *data);
 629   int64_t ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off);
 630   int64_t ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off);
 631   loff_t ll_lseek(Fh *fh, loff_t offset, int whence);
 632   int ll_flush(Fh *fh);
 633   int ll_fsync(Fh *fh, bool syncdataonly);
 634   int ll_sync_inode(Inode *in, bool syncdataonly);
 635   int ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length);
 636   int ll_release(Fh *fh);
 637   int ll_getlk(Fh *fh, struct flock *fl, uint64_t owner);
 638   int ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep);
 639   int ll_flock(Fh *fh, int cmd, uint64_t owner);
 640   int ll_lazyio(Fh *fh, int enable);
 641   int ll_file_layout(Fh *fh, file_layout_t *layout);
 642   void ll_interrupt(void *d);
 643   bool ll_handle_umask() {
 644     return acl_type != NO_ACL;
 645   }
 646
 647   int ll_get_stripe_osd(struct Inode *in, uint64_t blockno,
 648                         file_layout_t* layout);
 649   uint64_t ll_get_internal_offset(struct Inode *in, uint64_t blockno);
 650
 651   int ll_num_osds(void);
 652   int ll_osdaddr(int osd, uint32_t *addr);
 653   int ll_osdaddr(int osd, char* buf, size_t size);
 654
 655   void _ll_register_callbacks(struct ceph_client_callback_args *args);
 656   void ll_register_callbacks(struct ceph_client_callback_args *args); // deprecated
 657   int ll_register_callbacks2(struct ceph_client_callback_args *args);
 658   int test_dentry_handling(bool can_invalidate);
 659
 660   const char** get_tracked_conf_keys() const override;
 661   void handle_conf_change(const ConfigProxy& conf,
 662                                   const std::set <std::string> &changed) override;
 663   uint32_t get_deleg_timeout() { return deleg_timeout; }
 664   int set_deleg_timeout(uint32_t timeout);
 665   int ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv);
 666
 667   entity_name_t get_myname() { return messenger->get_myname(); }
 668   void wait_on_list(std::list<ceph::condition_variable*>& ls);
 669   void signal_cond_list(std::list<ceph::condition_variable*>& ls);
 670
 671   void set_filer_flags(int flags);
 672   void clear_filer_flags(int flags);
 673
 674   void tear_down_cache();
 675
 676   void update_metadata(std::string const &k, std::string const &v);
 677
 678   client_t get_nodeid() { return whoami; }
 679
 680   inodeno_t get_root_ino();
 681   Inode *get_root();
 682
 683   virtual int init();
 684   virtual void shutdown();
 685
 686   // messaging
 687   void cancel_commands(const MDSMap& newmap);
 688   void handle_mds_map(const MConstRef<MMDSMap>& m);
 689   void handle_fs_map(const MConstRef<MFSMap>& m);
 690   void handle_fs_map_user(const MConstRef<MFSMapUser>& m);
 691   void handle_osd_map(const MConstRef<MOSDMap>& m);
 692
 693   void handle_lease(const MConstRef<MClientLease>& m);
 694
 695   // inline data
 696   int uninline_data(Inode *in, Context *onfinish);
 697
 698   // file caps
 699   void check_cap_issue(Inode *in, unsigned issued);
 700   void add_update_cap(Inode *in, MetaSession *session, uint64_t cap_id,
 701                       unsigned issued, unsigned wanted, unsigned seq, unsigned mseq,
 702                       inodeno_t realm, int flags, const UserPerm& perms);
 703   void remove_cap(Cap *cap, bool queue_release);
 704   void remove_all_caps(Inode *in);
 705   void remove_session_caps(MetaSession *session, int err);
 706   int mark_caps_flushing(Inode *in, ceph_tid_t *ptid);
 707   void adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s);
 708   void flush_caps_sync();
 709   void kick_flushing_caps(Inode *in, MetaSession *session);
 710   void kick_flushing_caps(MetaSession *session);
 711   void early_kick_flushing_caps(MetaSession *session);
 712   int get_caps(Fh *fh, int need, int want, int *have, loff_t endoff);
 713   int get_caps_used(Inode *in);
 714
 715   void maybe_update_snaprealm(SnapRealm *realm, snapid_t snap_created, snapid_t snap_highwater,
 716                               std::vector<snapid_t>& snaps);
 717
 718   void handle_quota(const MConstRef<MClientQuota>& m);
 719   void handle_snap(const MConstRef<MClientSnap>& m);
 720   void handle_caps(const MConstRef<MClientCaps>& m);
 721   void handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m);
 722   void handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m);
 723   void handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m);
 724   void handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m);
 725   void handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m);
 726   void handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m);
 727   void cap_delay_requeue(Inode *in);
 728
 729   void send_cap(Inode *in, MetaSession *session, Cap *cap, int flags,
 730                 int used, int want, int retain, int flush,
 731                 ceph_tid_t flush_tid);
 732
 733   void send_flush_snap(Inode *in, MetaSession *session, snapid_t follows, CapSnap& capsnap);
 734
 735   void flush_snaps(Inode *in);
 736   void get_cap_ref(Inode *in, int cap);
 737   void put_cap_ref(Inode *in, int cap);
 738   void wait_sync_caps(Inode *in, ceph_tid_t want);
 739   void wait_sync_caps(ceph_tid_t want);
 740   void queue_cap_snap(Inode *in, SnapContext &old_snapc);
 741   void finish_cap_snap(Inode *in, CapSnap &capsnap, int used);
 742
 743   void _schedule_invalidate_dentry_callback(Dentry *dn, bool del);
 744   void _async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, std::string& name);
 745   void _try_to_trim_inode(Inode *in, bool sched_inval);
 746
 747   void _schedule_invalidate_callback(Inode *in, int64_t off, int64_t len);
 748   void _invalidate_inode_cache(Inode *in);
 749   void _invalidate_inode_cache(Inode *in, int64_t off, int64_t len);
 750   void _async_invalidate(vinodeno_t ino, int64_t off, int64_t len);
 751
 752   void _schedule_ino_release_callback(Inode *in);
 753   void _async_inode_release(vinodeno_t ino);
 754
 755   bool _release(Inode *in);
 756
 757   /**
 758    * Initiate a flush of the data associated with the given inode.
 759    * If you specify a Context, you are responsible for holding an inode
 760    * reference for the duration of the flush. If not, _flush() will
 761    * take the reference for you.
 762    * @param in The Inode whose data you wish to flush.
 763    * @param c The Context you wish us to complete once the data is
 764    * flushed. If already flushed, this will be called in-line.
 765    *
 766    * @returns true if the data was already flushed, false otherwise.
 767    */
 768   bool _flush(Inode *in, Context *c);
 769   void _flush_range(Inode *in, int64_t off, uint64_t size);
 770   void _flushed(Inode *in);
 771   void flush_set_callback(ObjectCacher::ObjectSet *oset);
 772
 773   void close_release(Inode *in);
 774   void close_safe(Inode *in);
 775
 776   void lock_fh_pos(Fh *f);
 777   void unlock_fh_pos(Fh *f);
 778
 779   // metadata cache
 780   void update_dir_dist(Inode *in, DirStat *st, mds_rank_t from);
 781
 782   void clear_dir_complete_and_ordered(Inode *diri, bool complete);
 783   void insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri);
 784   Inode* insert_trace(MetaRequest *request, MetaSession *session);
 785   void update_inode_file_size(Inode *in, int issued, uint64_t size,
 786                               uint64_t truncate_seq, uint64_t truncate_size);
 787   void update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq,
 788                               utime_t ctime, utime_t mtime, utime_t atime);
 789
 790   Inode *add_update_inode(InodeStat *st, utime_t ttl, MetaSession *session,
 791                           const UserPerm& request_perms);
 792   Dentry *insert_dentry_inode(Dir *dir, const std::string& dname, LeaseStat *dlease,
 793                               Inode *in, utime_t from, MetaSession *session,
 794                               Dentry *old_dentry = NULL);
 795   void update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session);
 796
 797   bool use_faked_inos() { return _use_faked_inos; }
 798   vinodeno_t map_faked_ino(ino_t ino);
 799
 800   //notify the mds to flush the mdlog
 801   void flush_mdlog_sync(Inode *in);
 802   void flush_mdlog_sync();
 803   void flush_mdlog(MetaSession *session);
 804
 805   void renew_caps();
 806   void renew_caps(MetaSession *session);
 807   void flush_cap_releases();
 808   void renew_and_flush_cap_releases();
 809   void tick();
 810   void start_tick_thread();
 811
 812   void update_read_io_size(size_t size) {
 813     total_read_ops++;
 814     total_read_size += size;
 815   }
 816
 817   void update_write_io_size(size_t size) {
 818     total_write_ops++;
 819     total_write_size += size;
 820   }
 821
 822   void inc_dentry_nr() {
 823     ++dentry_nr;
 824   }
 825   void dec_dentry_nr() {
 826     --dentry_nr;
 827   }
 828   void dlease_hit() {
 829     ++dlease_hits;
 830   }
 831   void dlease_miss() {
 832     ++dlease_misses;
 833   }
 834   std::tuple<uint64_t, uint64_t, uint64_t> get_dlease_hit_rates() {
 835     return std::make_tuple(dlease_hits, dlease_misses, dentry_nr);
 836   }
 837
 838   void cap_hit() {
 839     ++cap_hits;
 840   }
 841   void cap_miss() {
 842     ++cap_misses;
 843   }
 844   std::pair<uint64_t, uint64_t> get_cap_hit_rates() {
 845     return std::make_pair(cap_hits, cap_misses);
 846   }
 847
 848   void inc_opened_files() {
 849     ++opened_files;
 850   }
 851   void dec_opened_files() {
 852     --opened_files;
 853   }
 854   std::pair<uint64_t, uint64_t> get_opened_files_rates() {
 855     return std::make_pair(opened_files, inode_map.size());
 856   }
 857
 858   void inc_pinned_icaps() {
 859     ++pinned_icaps;
 860   }
 861   void dec_pinned_icaps(uint64_t nr=1) {
 862     pinned_icaps -= nr;
 863   }
 864   std::pair<uint64_t, uint64_t> get_pinned_icaps_rates() {
 865     return std::make_pair(pinned_icaps, inode_map.size());
 866   }
 867
 868   void inc_opened_inodes() {
 869     ++opened_inodes;
 870   }
 871   void dec_opened_inodes() {
 872     --opened_inodes;
 873   }
 874   std::pair<uint64_t, uint64_t> get_opened_inodes_rates() {
 875     return std::make_pair(opened_inodes, inode_map.size());
 876   }
 877
 878   /* timer_lock for 'timer' */
 879   ceph::mutex timer_lock = ceph::make_mutex("Client::timer_lock");
 880   SafeTimer timer;
 881
 882   /* tick thread */
 883   std::thread upkeeper;
 884   ceph::condition_variable upkeep_cond;
 885   bool tick_thread_stopped = false;
 886
 887   std::unique_ptr<PerfCounters> logger;
 888   std::unique_ptr<MDSMap> mdsmap;
 889
 890   bool fuse_default_permissions;
 891
 892 protected:
 893   /* Flags for check_caps() */
 894   static const unsigned CHECK_CAPS_NODELAY = 0x1;
 895   static const unsigned CHECK_CAPS_SYNCHRONOUS = 0x2;
 896
 897   void check_caps(Inode *in, unsigned flags);
 898
 899   void set_cap_epoch_barrier(epoch_t e);
 900
 901   void handle_command_reply(const MConstRef<MCommandReply>& m);
 902   int fetch_fsmap(bool user);
 903   int resolve_mds(
 904       const std::string &mds_spec,
 905       std::vector<mds_gid_t> *targets);
 906
 907   void get_session_metadata(std::map<std::string, std::string> *meta) const;
 908   bool have_open_session(mds_rank_t mds);
 909   void got_mds_push(MetaSession *s);
 910   MetaSessionRef _get_mds_session(mds_rank_t mds, Connection *con);  ///< return session for mds *and* con; null otherwise
 911   MetaSessionRef _get_or_open_mds_session(mds_rank_t mds);
 912   MetaSessionRef _open_mds_session(mds_rank_t mds);
 913   void _close_mds_session(MetaSession *s);
 914   void _closed_mds_session(MetaSession *s, int err=0, bool rejected=false);
 915   bool _any_stale_sessions() const;
 916   void _kick_stale_sessions();
 917   void handle_client_session(const MConstRef<MClientSession>& m);
 918   void send_reconnect(MetaSession *s);
 919   void resend_unsafe_requests(MetaSession *s);
 920   void wait_unsafe_requests();
 921
 922   void dump_mds_requests(Formatter *f);
 923   void dump_mds_sessions(Formatter *f, bool cap_dump=false);
 924
 925   int make_request(MetaRequest *req, const UserPerm& perms,
 926                    InodeRef *ptarget = 0, bool *pcreated = 0,
 927                    mds_rank_t use_mds=-1, bufferlist *pdirbl=0);
 928   void put_request(MetaRequest *request);
 929   void unregister_request(MetaRequest *request);
 930
 931   int verify_reply_trace(int r, MetaSession *session, MetaRequest *request,
 932                          const MConstRef<MClientReply>& reply,
 933                          InodeRef *ptarget, bool *pcreated,
 934                          const UserPerm& perms);
 935   void encode_cap_releases(MetaRequest *request, mds_rank_t mds);
 936   int encode_inode_release(Inode *in, MetaRequest *req,
 937                            mds_rank_t mds, int drop,
 938                            int unless,int force=0);
 939   void encode_dentry_release(Dentry *dn, MetaRequest *req,
 940                              mds_rank_t mds, int drop, int unless);
 941   mds_rank_t choose_target_mds(MetaRequest *req, Inode** phash_diri=NULL);
 942   void connect_mds_targets(mds_rank_t mds);
 943   void send_request(MetaRequest *request, MetaSession *session,
 944                     bool drop_cap_releases=false);
 945   MRef<MClientRequest> build_client_request(MetaRequest *request);
 946   void kick_requests(MetaSession *session);
 947   void kick_requests_closed(MetaSession *session);
 948   void handle_client_request_forward(const MConstRef<MClientRequestForward>& reply);
 949   void handle_client_reply(const MConstRef<MClientReply>& reply);
 950   bool is_dir_operation(MetaRequest *request);
 951
 952   int path_walk(const filepath& fp, struct walk_dentry_result* result, const UserPerm& perms, bool followsym=true, int mask=0,
 953                 InodeRef dirinode=nullptr);
 954   int path_walk(const filepath& fp, InodeRef *end, const UserPerm& perms,
 955                 bool followsym=true, int mask=0, InodeRef dirinode=nullptr);
 956
 957   // fake inode number for 32-bits ino_t
 958   void _assign_faked_ino(Inode *in);
 959   void _assign_faked_root(Inode *in);
 960   void _release_faked_ino(Inode *in);
 961   void _reset_faked_inos();
 962   vinodeno_t _map_faked_ino(ino_t ino);
 963
 964   // Optional extra metadata about me to send to the MDS
 965   void populate_metadata(const std::string &mount_root);
 966
 967   SnapRealm *get_snap_realm(inodeno_t r);
 968   SnapRealm *get_snap_realm_maybe(inodeno_t r);
 969   void put_snap_realm(SnapRealm *realm);
 970   bool adjust_realm_parent(SnapRealm *realm, inodeno_t parent);
 971   void update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool must_flush=true);
 972   void invalidate_snaprealm_and_children(SnapRealm *realm);
 973
 974   Inode *open_snapdir(Inode *diri);
 975
 976   int get_fd() {
 977     int fd = free_fd_set.range_start();
 978     free_fd_set.erase(fd, 1);
 979     return fd;
 980   }
 981   void put_fd(int fd) {
 982     free_fd_set.insert(fd, 1);
 983   }
 984
 985   /*
 986    * Resolve file descriptor, or return NULL.
 987    */
 988   Fh *get_filehandle(int fd) {
 989     auto it = fd_map.find(fd);
 990     if (it == fd_map.end())
 991       return NULL;
 992     return it->second;
 993   }
 994   int get_fd_inode(int fd, InodeRef *in);
 995
 996   // helpers
 997   void wake_up_session_caps(MetaSession *s, bool reconnect);
 998
 999   void wait_on_context_list(std::list<Context*>& ls);
1000   void signal_context_list(std::list<Context*>& ls);
1001
1002   // -- metadata cache stuff
1003
1004   // decrease inode ref.  delete if dangling.
1005   void _put_inode(Inode *in, int n);
1006   void delay_put_inodes(bool wakeup=false);
1007   void put_inode(Inode *in, int n=1);
1008   void close_dir(Dir *dir);
1009
1010   int subscribe_mdsmap(const std::string &fs_name="");
1011
1012   void _abort_mds_sessions(int err);
1013
1014   // same as unmount() but for when the client_lock is already held
1015   void _unmount(bool abort);
1016
1017   //int get_cache_size() { return lru.lru_get_size(); }
1018
1019   /**
1020    * Don't call this with in==NULL, use get_or_create for that
1021    * leave dn set to default NULL unless you're trying to add
1022    * a new inode to a pre-created Dentry
1023    */
1024   Dentry* link(Dir *dir, const std::string& name, Inode *in, Dentry *dn);
1025   void unlink(Dentry *dn, bool keepdir, bool keepdentry);
1026
1027   int fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0);
1028   int fill_stat(InodeRef& in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0) {
1029     return fill_stat(in.get(), st, dirstat, rstat);
1030   }
1031
1032   void fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx);
1033   void fill_statx(InodeRef& in, unsigned int mask, struct ceph_statx *stx) {
1034     return fill_statx(in.get(), mask, stx);
1035   }
1036
1037   void touch_dn(Dentry *dn);
1038
1039   // trim cache.
1040   void trim_cache(bool trim_kernel_dcache=false);
1041   void trim_cache_for_reconnect(MetaSession *s);
1042   void trim_dentry(Dentry *dn);
1043   void trim_caps(MetaSession *s, uint64_t max);
1044   void _invalidate_kernel_dcache();
1045   void _trim_negative_child_dentries(InodeRef& in);
1046
1047   void dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected);
1048   void dump_cache(Formatter *f);  // debug
1049
1050   // force read-only
1051   void force_session_readonly(MetaSession *s);
1052
1053   void dump_status(Formatter *f);  // debug
1054
1055   bool ms_dispatch2(const MessageRef& m) override;
1056
1057   void ms_handle_connect(Connection *con) override;
1058   bool ms_handle_reset(Connection *con) override;
1059   void ms_handle_remote_reset(Connection *con) override;
1060   bool ms_handle_refused(Connection *con) override;
1061
1062   int authenticate();
1063
1064   Inode* get_quota_root(Inode *in, const UserPerm& perms);
1065   bool check_quota_condition(Inode *in, const UserPerm& perms,
1066                              std::function<bool (const Inode &)> test);
1067   bool is_quota_files_exceeded(Inode *in, const UserPerm& perms);
1068   bool is_quota_bytes_exceeded(Inode *in, int64_t new_bytes,
1069                                const UserPerm& perms);
1070   bool is_quota_bytes_approaching(Inode *in, const UserPerm& perms);
1071
1072   int check_pool_perm(Inode *in, int need);
1073
1074   void handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply);
1075
1076   /**
1077    * Call this when an OSDMap is seen with a full flag (global or per pool)
1078    * set.
1079    *
1080    * @param pool the pool ID affected, or -1 if all.
1081    */
1082   void _handle_full_flag(int64_t pool);
1083
1084   void _close_sessions();
1085
1086   void _pre_init();
1087
1088   /**
1089    * The basic housekeeping parts of init (perf counters, admin socket)
1090    * that is independent of how objecters/monclient/messengers are
1091    * being set up.
1092    */
1093   void _finish_init();
1094
1095   // global client lock
1096   //  - protects Client and buffer cache both!
1097   ceph::mutex client_lock = ceph::make_mutex("Client::client_lock");
1098
1099   std::map<snapid_t, int> ll_snap_ref;
1100
1101   InodeRef               root = nullptr;
1102   map<Inode*, InodeRef>  root_parents;
1103   Inode*                 root_ancestor = nullptr;
1104   LRU                    lru;    // lru list of Dentry's in our local metadata cache.
1105
1106   InodeRef cwd;
1107
1108   std::unique_ptr<Filer>             filer;
1109   std::unique_ptr<ObjectCacher>      objectcacher;
1110   std::unique_ptr<WritebackHandler>  writeback_handler;
1111
1112   Messenger *messenger;
1113   MonClient *monclient;
1114   Objecter  *objecter;
1115
1116   client_t whoami;
1117
1118   /* The state migration mechanism */
1119   enum _state {
1120     /* For the initialize_state */
1121     CLIENT_NEW, // The initial state for the initialize_state or after Client::shutdown()
1122     CLIENT_INITIALIZING, // At the beginning of the Client::init()
1123     CLIENT_INITIALIZED, // At the end of CLient::init()
1124
1125     /* For the mount_state */
1126     CLIENT_UNMOUNTED, // The initial state for the mount_state or after unmounted
1127     CLIENT_MOUNTING, // At the beginning of Client::mount()
1128     CLIENT_MOUNTED, // At the end of Client::mount()
1129     CLIENT_UNMOUNTING, // At the beginning of the Client::_unmout()
1130   };
1131
1132   typedef enum _state state_t;
1133   using RWRef_t = RWRef<state_t>;
1134
1135   struct mount_state_t : public RWRefState<state_t> {
1136     public:
1137       bool is_valid_state(state_t state) const override {
1138         switch (state) {
1139           case Client::CLIENT_MOUNTING:
1140           case Client::CLIENT_MOUNTED:
1141           case Client::CLIENT_UNMOUNTING:
1142           case Client::CLIENT_UNMOUNTED:
1143             return true;
1144           default:
1145             return false;
1146         }
1147       }
1148
1149       int check_reader_state(state_t require) const override {
1150         if (require == Client::CLIENT_MOUNTING &&
1151             (state == Client::CLIENT_MOUNTING || state == Client::CLIENT_MOUNTED))
1152           return true;
1153         else
1154           return false;
1155       }
1156
1157       /* The state migration check */
1158       int check_writer_state(state_t require) const override {
1159         if (require == Client::CLIENT_MOUNTING &&
1160             state == Client::CLIENT_UNMOUNTED)
1161           return true;
1162         else if (require == Client::CLIENT_MOUNTED &&
1163             state == Client::CLIENT_MOUNTING)
1164           return true;
1165         else if (require == Client::CLIENT_UNMOUNTING &&
1166             state == Client::CLIENT_MOUNTED)
1167           return true;
1168         else if (require == Client::CLIENT_UNMOUNTED &&
1169             state == Client::CLIENT_UNMOUNTING)
1170           return true;
1171         else
1172           return false;
1173       }
1174
1175       mount_state_t(state_t state, const char *lockname, uint64_t reader_cnt=0)
1176         : RWRefState (state, lockname, reader_cnt) {}
1177       ~mount_state_t() {}
1178   };
1179
1180   struct initialize_state_t : public RWRefState<state_t> {
1181     public:
1182       bool is_valid_state(state_t state) const override {
1183         switch (state) {
1184           case Client::CLIENT_NEW:
1185           case Client::CLIENT_INITIALIZING:
1186           case Client::CLIENT_INITIALIZED:
1187             return true;
1188           default:
1189             return false;
1190         }
1191       }
1192
1193       int check_reader_state(state_t require) const override {
1194         if (require == Client::CLIENT_INITIALIZED &&
1195             state >= Client::CLIENT_INITIALIZED)
1196           return true;
1197         else
1198           return false;
1199       }
1200
1201       /* The state migration check */
1202       int check_writer_state(state_t require) const override {
1203         if (require == Client::CLIENT_INITIALIZING &&
1204             (state == Client::CLIENT_NEW))
1205           return true;
1206         else if (require == Client::CLIENT_INITIALIZED &&
1207             (state == Client::CLIENT_INITIALIZING))
1208           return true;
1209         else if (require == Client::CLIENT_NEW &&
1210             (state == Client::CLIENT_INITIALIZED))
1211           return true;
1212         else
1213           return false;
1214       }
1215
1216       initialize_state_t(state_t state, const char *lockname, uint64_t reader_cnt=0)
1217         : RWRefState (state, lockname, reader_cnt) {}
1218       ~initialize_state_t() {}
1219   };
1220
1221   struct mount_state_t mount_state;
1222   struct initialize_state_t initialize_state;
1223
1224 private:
1225   struct C_Readahead : public Context {
1226     C_Readahead(Client *c, Fh *f);
1227     ~C_Readahead() override;
1228     void finish(int r) override;
1229
1230     Client *client;
1231     Fh *f;
1232   };
1233
1234   /*
1235    * These define virtual xattrs exposing the recursive directory
1236    * statistics and layout metadata.
1237    */
1238   struct VXattr {
1239     const std::string name;
1240     size_t (Client::*getxattr_cb)(Inode *in, char *val, size_t size);
1241     bool readonly;
1242     bool (Client::*exists_cb)(Inode *in);
1243     unsigned int flags;
1244   };
1245
1246   enum {
1247     NO_ACL = 0,
1248     POSIX_ACL,
1249   };
1250
1251   enum {
1252     MAY_EXEC = 1,
1253     MAY_WRITE = 2,
1254     MAY_READ = 4,
1255   };
1256
1257   std::unique_ptr<CephContext, std::function<void(CephContext*)>> cct_deleter;
1258
1259   /* Flags for VXattr */
1260   static const unsigned VXATTR_RSTAT = 0x1;
1261   static const unsigned VXATTR_DIRSTAT = 0x2;
1262
1263   static const VXattr _dir_vxattrs[];
1264   static const VXattr _file_vxattrs[];
1265   static const VXattr _common_vxattrs[];
1266
1267
1268   bool is_reserved_vino(vinodeno_t &vino);
1269
1270   void fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off);
1271
1272   int _opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms);
1273   void _readdir_drop_dirp_buffer(dir_result_t *dirp);
1274   bool _readdir_have_frag(dir_result_t *dirp);
1275   void _readdir_next_frag(dir_result_t *dirp);
1276   void _readdir_rechoose_frag(dir_result_t *dirp);
1277   int _readdir_get_frag(dir_result_t *dirp);
1278   int _readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, int caps, bool getref);
1279   void _closedir(dir_result_t *dirp);
1280
1281   // other helpers
1282   void _fragmap_remove_non_leaves(Inode *in);
1283   void _fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds);
1284
1285   void _ll_get(Inode *in);
1286   int _ll_put(Inode *in, uint64_t num);
1287   void _ll_drop_pins();
1288
1289   Fh *_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms);
1290   int _release_fh(Fh *fh);
1291   void _put_fh(Fh *fh);
1292
1293   int _do_remount(bool retry_on_error);
1294
1295   int _read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, bool *checkeof);
1296   int _read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl);
1297
1298   bool _dentry_valid(const Dentry *dn);
1299
1300   // internal interface
1301   //   call these with client_lock held!
1302   int _do_lookup(Inode *dir, const std::string& name, int mask, InodeRef *target,
1303                  const UserPerm& perms);
1304
1305   int _lookup(Inode *dir, const std::string& dname, int mask, InodeRef *target,
1306               const UserPerm& perm, std::string* alternate_name=nullptr);
1307
1308   int _link(Inode *in, Inode *dir, const char *name, const UserPerm& perm, std::string alternate_name,
1309             InodeRef *inp = 0);
1310   int _unlink(Inode *dir, const char *name, const UserPerm& perm);
1311   int _rename(Inode *olddir, const char *oname, Inode *ndir, const char *nname, const UserPerm& perm, std::string alternate_name);
1312   int _mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm,
1313              InodeRef *inp = 0, const std::map<std::string, std::string> &metadata={},
1314              std::string alternate_name="");
1315   int _rmdir(Inode *dir, const char *name, const UserPerm& perms);
1316   int _symlink(Inode *dir, const char *name, const char *target,
1317                const UserPerm& perms, std::string alternate_name, InodeRef *inp = 0);
1318   int _mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev,
1319              const UserPerm& perms, InodeRef *inp = 0);
1320   int _do_setattr(Inode *in, struct ceph_statx *stx, int mask,
1321                   const UserPerm& perms, InodeRef *inp);
1322   void stat_to_statx(struct stat *st, struct ceph_statx *stx);
1323   int __setattrx(Inode *in, struct ceph_statx *stx, int mask,
1324                  const UserPerm& perms, InodeRef *inp = 0);
1325   int _setattrx(InodeRef &in, struct ceph_statx *stx, int mask,
1326                 const UserPerm& perms);
1327   int _setattr(InodeRef &in, struct stat *attr, int mask,
1328                const UserPerm& perms);
1329   int _ll_setattrx(Inode *in, struct ceph_statx *stx, int mask,
1330                    const UserPerm& perms, InodeRef *inp = 0);
1331   int _getattr(Inode *in, int mask, const UserPerm& perms, bool force=false);
1332   int _getattr(InodeRef &in, int mask, const UserPerm& perms, bool force=false) {
1333     return _getattr(in.get(), mask, perms, force);
1334   }
1335   int _readlink(Inode *in, char *buf, size_t size);
1336   int _getxattr(Inode *in, const char *name, void *value, size_t len,
1337                 const UserPerm& perms);
1338   int _getxattr(InodeRef &in, const char *name, void *value, size_t len,
1339                 const UserPerm& perms);
1340   int _listxattr(Inode *in, char *names, size_t len, const UserPerm& perms);
1341   int _do_setxattr(Inode *in, const char *name, const void *value, size_t len,
1342                    int flags, const UserPerm& perms);
1343   int _setxattr(Inode *in, const char *name, const void *value, size_t len,
1344                 int flags, const UserPerm& perms);
1345   int _setxattr(InodeRef &in, const char *name, const void *value, size_t len,
1346                 int flags, const UserPerm& perms);
1347   int _setxattr_check_data_pool(std::string& name, std::string& value, const OSDMap *osdmap);
1348   void _setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t len);
1349   int _removexattr(Inode *in, const char *nm, const UserPerm& perms);
1350   int _removexattr(InodeRef &in, const char *nm, const UserPerm& perms);
1351   int _open(Inode *in, int flags, mode_t mode, Fh **fhp,
1352             const UserPerm& perms);
1353   int _renew_caps(Inode *in);
1354   int _create(Inode *in, const char *name, int flags, mode_t mode, InodeRef *inp,
1355               Fh **fhp, int stripe_unit, int stripe_count, int object_size,
1356               const char *data_pool, bool *created, const UserPerm &perms,
1357               std::string alternate_name);
1358
1359   loff_t _lseek(Fh *fh, loff_t offset, int whence);
1360   int64_t _read(Fh *fh, int64_t offset, uint64_t size, bufferlist *bl);
1361   int64_t _write(Fh *fh, int64_t offset, uint64_t size, const char *buf,
1362           const struct iovec *iov, int iovcnt);
1363   int64_t _preadv_pwritev_locked(Fh *fh, const struct iovec *iov,
1364                                  unsigned iovcnt, int64_t offset,
1365                                  bool write, bool clamp_to_int);
1366   int _preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt,
1367                       int64_t offset, bool write);
1368   int _flush(Fh *fh);
1369   int _fsync(Fh *fh, bool syncdataonly);
1370   int _fsync(Inode *in, bool syncdataonly);
1371   int _sync_fs();
1372   int _fallocate(Fh *fh, int mode, int64_t offset, int64_t length);
1373   int _getlk(Fh *fh, struct flock *fl, uint64_t owner);
1374   int _setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep);
1375   int _flock(Fh *fh, int cmd, uint64_t owner);
1376   int _lazyio(Fh *fh, int enable);
1377
1378   int get_or_create(Inode *dir, const char* name,
1379                     Dentry **pdn, bool expect_null=false);
1380
1381   int xattr_permission(Inode *in, const char *name, unsigned want,
1382                        const UserPerm& perms);
1383   int may_setattr(Inode *in, struct ceph_statx *stx, int mask,
1384                   const UserPerm& perms);
1385   int may_open(Inode *in, int flags, const UserPerm& perms);
1386   int may_lookup(Inode *dir, const UserPerm& perms);
1387   int may_create(Inode *dir, const UserPerm& perms);
1388   int may_delete(Inode *dir, const char *name, const UserPerm& perms);
1389   int may_hardlink(Inode *in, const UserPerm& perms);
1390
1391   int _getattr_for_perm(Inode *in, const UserPerm& perms);
1392
1393   vinodeno_t _get_vino(Inode *in);
1394
1395   bool _vxattrcb_quota_exists(Inode *in);
1396   size_t _vxattrcb_quota(Inode *in, char *val, size_t size);
1397   size_t _vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size);
1398   size_t _vxattrcb_quota_max_files(Inode *in, char *val, size_t size);
1399
1400   bool _vxattrcb_layout_exists(Inode *in);
1401   size_t _vxattrcb_layout(Inode *in, char *val, size_t size);
1402   size_t _vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size);
1403   size_t _vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size);
1404   size_t _vxattrcb_layout_object_size(Inode *in, char *val, size_t size);
1405   size_t _vxattrcb_layout_pool(Inode *in, char *val, size_t size);
1406   size_t _vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size);
1407   size_t _vxattrcb_dir_entries(Inode *in, char *val, size_t size);
1408   size_t _vxattrcb_dir_files(Inode *in, char *val, size_t size);
1409   size_t _vxattrcb_dir_subdirs(Inode *in, char *val, size_t size);
1410   size_t _vxattrcb_dir_rentries(Inode *in, char *val, size_t size);
1411   size_t _vxattrcb_dir_rfiles(Inode *in, char *val, size_t size);
1412   size_t _vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size);
1413   size_t _vxattrcb_dir_rsnaps(Inode *in, char *val, size_t size);
1414   size_t _vxattrcb_dir_rbytes(Inode *in, char *val, size_t size);
1415   size_t _vxattrcb_dir_rctime(Inode *in, char *val, size_t size);
1416
1417   bool _vxattrcb_dir_pin_exists(Inode *in);
1418   size_t _vxattrcb_dir_pin(Inode *in, char *val, size_t size);
1419
1420   bool _vxattrcb_snap_btime_exists(Inode *in);
1421   size_t _vxattrcb_snap_btime(Inode *in, char *val, size_t size);
1422
1423   size_t _vxattrcb_caps(Inode *in, char *val, size_t size);
1424
1425   bool _vxattrcb_mirror_info_exists(Inode *in);
1426   size_t _vxattrcb_mirror_info(Inode *in, char *val, size_t size);
1427
1428   size_t _vxattrcb_cluster_fsid(Inode *in, char *val, size_t size);
1429   size_t _vxattrcb_client_id(Inode *in, char *val, size_t size);
1430
1431   static const VXattr *_get_vxattrs(Inode *in);
1432   static const VXattr *_match_vxattr(Inode *in, const char *name);
1433
1434   int _do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep,
1435                    struct flock *fl, uint64_t owner, bool removing=false);
1436   int _interrupt_filelock(MetaRequest *req);
1437   void _encode_filelocks(Inode *in, bufferlist& bl);
1438   void _release_filelocks(Fh *fh);
1439   void _update_lock_state(struct flock *fl, uint64_t owner, ceph_lock_state_t *lock_state);
1440
1441   int _posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl,
1442                         const UserPerm& perms);
1443   int _posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms);
1444   int _posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want);
1445
1446   mds_rank_t _get_random_up_mds() const;
1447
1448   int _ll_getattr(Inode *in, int caps, const UserPerm& perms);
1449   int _lookup_parent(Inode *in, const UserPerm& perms, Inode **parent=NULL);
1450   int _lookup_name(Inode *in, Inode *parent, const UserPerm& perms);
1451   int _lookup_vino(vinodeno_t ino, const UserPerm& perms, Inode **inode=NULL);
1452   bool _ll_forget(Inode *in, uint64_t count);
1453
1454   void collect_and_send_metrics();
1455   void collect_and_send_global_metrics();
1456
1457   uint32_t deleg_timeout = 0;
1458
1459   client_switch_interrupt_callback_t switch_interrupt_cb = nullptr;
1460   client_remount_callback_t remount_cb = nullptr;
1461   client_ino_callback_t ino_invalidate_cb = nullptr;
1462   client_dentry_callback_t dentry_invalidate_cb = nullptr;
1463   client_umask_callback_t umask_cb = nullptr;
1464   client_ino_release_t ino_release_cb = nullptr;
1465   void *callback_handle = nullptr;
1466   bool can_invalidate_dentries = false;
1467
1468   Finisher async_ino_invalidator;
1469   Finisher async_dentry_invalidator;
1470   Finisher interrupt_finisher;
1471   Finisher remount_finisher;
1472   Finisher async_ino_releasor;
1473   Finisher objecter_finisher;
1474
1475   utime_t last_cap_renew;
1476
1477   CommandHook m_command_hook;
1478
1479   int user_id, group_id;
1480   int acl_type = NO_ACL;
1481
1482   epoch_t cap_epoch_barrier = 0;
1483
1484   // mds sessions
1485   map<mds_rank_t, MetaSessionRef> mds_sessions;  // mds -> push seq
1486   std::set<mds_rank_t> mds_ranks_closing;  // mds ranks currently tearing down sessions
1487   std::list<ceph::condition_variable*> waiting_for_mdsmap;
1488
1489   // FSMap, for when using mds_command
1490   std::list<ceph::condition_variable*> waiting_for_fsmap;
1491   std::unique_ptr<FSMap> fsmap;
1492   std::unique_ptr<FSMapUser> fsmap_user;
1493
1494   // This mutex only protects command_table
1495   ceph::mutex command_lock = ceph::make_mutex("Client::command_lock");
1496   // MDS command state
1497   CommandTable<MDSCommandOp> command_table;
1498
1499   bool _use_faked_inos;
1500
1501   // Cluster fsid
1502   fs_cluster_id_t fscid;
1503
1504   // file handles, etc.
1505   interval_set<int> free_fd_set;  // unused fds
1506   ceph::unordered_map<int, Fh*> fd_map;
1507   set<Fh*> ll_unclosed_fh_set;
1508   ceph::unordered_set<dir_result_t*> opened_dirs;
1509   uint64_t fd_gen = 1;
1510
1511   bool   mount_aborted = false;
1512   bool   blocklisted = false;
1513
1514   ceph::unordered_map<vinodeno_t, Inode*> inode_map;
1515   ceph::unordered_map<ino_t, vinodeno_t> faked_ino_map;
1516   interval_set<ino_t> free_faked_inos;
1517   ino_t last_used_faked_ino;
1518   ino_t last_used_faked_root;
1519
1520   int local_osd = -CEPHFS_ENXIO;
1521   epoch_t local_osd_epoch = 0;
1522
1523   // mds requests
1524   ceph_tid_t last_tid = 0;
1525   ceph_tid_t oldest_tid = 0; // oldest incomplete mds request, excluding setfilelock requests
1526   map<ceph_tid_t, MetaRequest*> mds_requests;
1527
1528   // cap flushing
1529   ceph_tid_t last_flush_tid = 1;
1530
1531   xlist<Inode*> delayed_list;
1532   int num_flushing_caps = 0;
1533   ceph::unordered_map<inodeno_t,SnapRealm*> snap_realms;
1534   std::map<std::string, std::string> metadata;
1535
1536   utime_t last_auto_reconnect;
1537
1538   // trace generation
1539   std::ofstream traceout;
1540
1541   ceph::condition_variable mount_cond, sync_cond;
1542
1543   std::map<std::pair<int64_t,std::string>, int> pool_perms;
1544   std::list<ceph::condition_variable*> waiting_for_pool_perm;
1545
1546   uint64_t retries_on_invalidate = 0;
1547
1548   // state reclaim
1549   std::list<ceph::condition_variable*> waiting_for_reclaim;
1550   int reclaim_errno = 0;
1551   epoch_t reclaim_osd_epoch = 0;
1552   entity_addrvec_t reclaim_target_addrs;
1553
1554   // dentry lease metrics
1555   uint64_t dentry_nr = 0;
1556   uint64_t dlease_hits = 0;
1557   uint64_t dlease_misses = 0;
1558
1559   uint64_t cap_hits = 0;
1560   uint64_t cap_misses = 0;
1561
1562   uint64_t opened_files = 0;
1563   uint64_t pinned_icaps = 0;
1564   uint64_t opened_inodes = 0;
1565
1566   uint64_t total_read_ops = 0;
1567   uint64_t total_read_size = 0;
1568
1569   uint64_t total_write_ops = 0;
1570   uint64_t total_write_size = 0;
1571
1572   ceph::spinlock delay_i_lock;
1573   std::map<Inode*,int> delay_i_release;
1574 };
1575
1576 /**
1577  * Specialization of Client that manages its own Objecter instance
1578  * and handles init/shutdown of messenger/monclient
1579  */
1580 class StandaloneClient : public Client
1581 {
1582 public:
1583   StandaloneClient(Messenger *m, MonClient *mc, boost::asio::io_context& ictx);
1584
1585   ~StandaloneClient() override;
1586
1587   int init() override;
1588   void shutdown() override;
1589 };
1590
1591 #endif