ceph/src/include/rbd/librbd.hpp

   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) 2011 New Dream Network
   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 #ifndef __LIBRBD_HPP
  16 #define __LIBRBD_HPP
  17
  18 #include <stdbool.h>
  19 #include <string>
  20 #include <list>
  21 #include <map>
  22 #include <vector>
  23 #include "../rados/buffer.h"
  24 #include "../rados/librados.hpp"
  25 #include "librbd.h"
  26
  27 namespace librbd {
  28
  29   using librados::IoCtx;
  30
  31   class Image;
  32   class ImageOptions;
  33   typedef void *image_ctx_t;
  34   typedef void *completion_t;
  35   typedef void (*callback_t)(completion_t cb, void *arg);
  36
  37   typedef struct {
  38     uint64_t id;
  39     uint64_t size;
  40     std::string name;
  41   } snap_info_t;
  42
  43   typedef struct {
  44     std::string client;
  45     std::string cookie;
  46     std::string address;
  47   } locker_t;
  48
  49   typedef struct {
  50     std::string uuid;
  51     std::string cluster_name;
  52     std::string client_name;
  53   } mirror_peer_t;
  54
  55   typedef rbd_mirror_image_state_t mirror_image_state_t;
  56
  57   typedef struct {
  58     std::string global_id;
  59     mirror_image_state_t state;
  60     bool primary;
  61   } mirror_image_info_t;
  62
  63   typedef rbd_mirror_image_status_state_t mirror_image_status_state_t;
  64
  65   typedef struct {
  66     std::string name;
  67     mirror_image_info_t info;
  68     mirror_image_status_state_t state;
  69     std::string description;
  70     time_t last_update;
  71     bool up;
  72   } mirror_image_status_t;
  73
  74   typedef rbd_image_info_t image_info_t;
  75
  76   class CEPH_RBD_API ProgressContext
  77   {
  78   public:
  79     virtual ~ProgressContext();
  80     virtual int update_progress(uint64_t offset, uint64_t total) = 0;
  81   };
  82
  83   typedef struct {
  84     std::string id;
  85     std::string name;
  86     rbd_trash_image_source_t source;
  87     time_t deletion_time;
  88     time_t deferment_end_time;
  89   } trash_image_info_t;
  90
  91 class CEPH_RBD_API RBD
  92 {
  93 public:
  94   RBD();
  95   ~RBD();
  96
  97   // This must be dynamically allocated with new, and
  98   // must be released with release().
  99   // Do not use delete.
 100   struct AioCompletion {
 101     void *pc;
 102     AioCompletion(void *cb_arg, callback_t complete_cb);
 103     bool is_complete();
 104     int wait_for_complete();
 105     ssize_t get_return_value();
 106     void *get_arg();
 107     void release();
 108   };
 109
 110   void version(int *major, int *minor, int *extra);
 111
 112   int open(IoCtx& io_ctx, Image& image, const char *name);
 113   int open(IoCtx& io_ctx, Image& image, const char *name, const char *snapname);
 114   int open_by_id(IoCtx& io_ctx, Image& image, const char *id);
 115   int open_by_id(IoCtx& io_ctx, Image& image, const char *id, const char *snapname);
 116   int aio_open(IoCtx& io_ctx, Image& image, const char *name,
 117                const char *snapname, RBD::AioCompletion *c);
 118   int aio_open_by_id(IoCtx& io_ctx, Image& image, const char *id,
 119                      const char *snapname, RBD::AioCompletion *c);
 120   // see librbd.h
 121   int open_read_only(IoCtx& io_ctx, Image& image, const char *name,
 122                      const char *snapname);
 123   int open_by_id_read_only(IoCtx& io_ctx, Image& image, const char *id,
 124                            const char *snapname);
 125   int aio_open_read_only(IoCtx& io_ctx, Image& image, const char *name,
 126                          const char *snapname, RBD::AioCompletion *c);
 127   int aio_open_by_id_read_only(IoCtx& io_ctx, Image& image, const char *id,
 128                                const char *snapname, RBD::AioCompletion *c);
 129   int list(IoCtx& io_ctx, std::vector<std::string>& names);
 130   int create(IoCtx& io_ctx, const char *name, uint64_t size, int *order);
 131   int create2(IoCtx& io_ctx, const char *name, uint64_t size,
 132               uint64_t features, int *order);
 133   int create3(IoCtx& io_ctx, const char *name, uint64_t size,
 134               uint64_t features, int *order,
 135               uint64_t stripe_unit, uint64_t stripe_count);
 136   int create4(IoCtx& io_ctx, const char *name, uint64_t size,
 137               ImageOptions& opts);
 138   int clone(IoCtx& p_ioctx, const char *p_name, const char *p_snapname,
 139                IoCtx& c_ioctx, const char *c_name, uint64_t features,
 140                int *c_order);
 141   int clone2(IoCtx& p_ioctx, const char *p_name, const char *p_snapname,
 142              IoCtx& c_ioctx, const char *c_name, uint64_t features,
 143              int *c_order, uint64_t stripe_unit, int stripe_count);
 144   int clone3(IoCtx& p_ioctx, const char *p_name, const char *p_snapname,
 145              IoCtx& c_ioctx, const char *c_name, ImageOptions& opts);
 146   int remove(IoCtx& io_ctx, const char *name);
 147   int remove_with_progress(IoCtx& io_ctx, const char *name, ProgressContext& pctx);
 148   int rename(IoCtx& src_io_ctx, const char *srcname, const char *destname);
 149
 150   int trash_move(IoCtx &io_ctx, const char *name, uint64_t delay);
 151   int trash_get(IoCtx &io_ctx, const char *id, trash_image_info_t *info);
 152   int trash_list(IoCtx &io_ctx, std::vector<trash_image_info_t> &entries);
 153   int trash_remove(IoCtx &io_ctx, const char *image_id, bool force);
 154   int trash_remove_with_progress(IoCtx &io_ctx, const char *image_id,
 155                                  bool force, ProgressContext &pctx);
 156   int trash_restore(IoCtx &io_ctx, const char *id, const char *name);
 157
 158   // RBD pool mirroring support functions
 159   int mirror_mode_get(IoCtx& io_ctx, rbd_mirror_mode_t *mirror_mode);
 160   int mirror_mode_set(IoCtx& io_ctx, rbd_mirror_mode_t mirror_mode);
 161   int mirror_peer_add(IoCtx& io_ctx, std::string *uuid,
 162                       const std::string &cluster_name,
 163                       const std::string &client_name);
 164   int mirror_peer_remove(IoCtx& io_ctx, const std::string &uuid);
 165   int mirror_peer_list(IoCtx& io_ctx, std::vector<mirror_peer_t> *peers);
 166   int mirror_peer_set_client(IoCtx& io_ctx, const std::string &uuid,
 167                              const std::string &client_name);
 168   int mirror_peer_set_cluster(IoCtx& io_ctx, const std::string &uuid,
 169                               const std::string &cluster_name);
 170   int mirror_image_status_list(IoCtx& io_ctx, const std::string &start_id,
 171       size_t max, std::map<std::string, mirror_image_status_t> *images);
 172   int mirror_image_status_summary(IoCtx& io_ctx,
 173       std::map<mirror_image_status_state_t, int> *states);
 174
 175 private:
 176   /* We don't allow assignment or copying */
 177   RBD(const RBD& rhs);
 178   const RBD& operator=(const RBD& rhs);
 179 };
 180
 181 class CEPH_RBD_API ImageOptions {
 182 public:
 183   ImageOptions();
 184   ImageOptions(rbd_image_options_t opts);
 185   ImageOptions(const ImageOptions &imgopts);
 186   ~ImageOptions();
 187
 188   int set(int optname, const std::string& optval);
 189   int set(int optname, uint64_t optval);
 190   int get(int optname, std::string* optval) const;
 191   int get(int optname, uint64_t* optval) const;
 192   int is_set(int optname, bool* is_set);
 193   int unset(int optname);
 194   void clear();
 195   bool empty() const;
 196
 197 private:
 198   friend class RBD;
 199   friend class Image;
 200
 201   rbd_image_options_t opts;
 202 };
 203
 204 class CEPH_RBD_API UpdateWatchCtx {
 205 public:
 206   virtual ~UpdateWatchCtx() {}
 207   /**
 208    * Callback activated when we receive a notify event.
 209    */
 210   virtual void handle_notify() = 0;
 211 };
 212
 213 class CEPH_RBD_API Image
 214 {
 215 public:
 216   Image();
 217   ~Image();
 218
 219   int close();
 220   int aio_close(RBD::AioCompletion *c);
 221
 222   int resize(uint64_t size);
 223   int resize2(uint64_t size, bool allow_shrink, ProgressContext& pctx);
 224   int resize_with_progress(uint64_t size, ProgressContext& pctx);
 225   int stat(image_info_t &info, size_t infosize);
 226   int get_id(std::string *id);
 227   std::string get_block_name_prefix();
 228   int64_t get_data_pool_id();
 229   int parent_info(std::string *parent_poolname, std::string *parent_name,
 230                   std::string *parent_snapname);
 231   int parent_info2(std::string *parent_poolname, std::string *parent_name,
 232                    std::string *parent_id, std::string *parent_snapname);
 233   int old_format(uint8_t *old);
 234   int size(uint64_t *size);
 235   int features(uint64_t *features);
 236   int update_features(uint64_t features, bool enabled);
 237   int overlap(uint64_t *overlap);
 238   int get_flags(uint64_t *flags);
 239   int set_image_notification(int fd, int type);
 240
 241   /* exclusive lock feature */
 242   int is_exclusive_lock_owner(bool *is_owner);
 243   int lock_acquire(rbd_lock_mode_t lock_mode);
 244   int lock_release();
 245   int lock_get_owners(rbd_lock_mode_t *lock_mode,
 246                       std::list<std::string> *lock_owners);
 247   int lock_break(rbd_lock_mode_t lock_mode, const std::string &lock_owner);
 248
 249   /* object map feature */
 250   int rebuild_object_map(ProgressContext &prog_ctx);
 251
 252   int check_object_map(ProgressContext &prog_ctx);
 253
 254   int copy(IoCtx& dest_io_ctx, const char *destname);
 255   int copy2(Image& dest);
 256   int copy3(IoCtx& dest_io_ctx, const char *destname, ImageOptions& opts);
 257   int copy4(IoCtx& dest_io_ctx, const char *destname, ImageOptions& opts,
 258             size_t sparse_size);
 259   int copy_with_progress(IoCtx& dest_io_ctx, const char *destname,
 260                          ProgressContext &prog_ctx);
 261   int copy_with_progress2(Image& dest, ProgressContext &prog_ctx);
 262   int copy_with_progress3(IoCtx& dest_io_ctx, const char *destname,
 263                           ImageOptions& opts, ProgressContext &prog_ctx);
 264   int copy_with_progress4(IoCtx& dest_io_ctx, const char *destname,
 265                           ImageOptions& opts, ProgressContext &prog_ctx,
 266                           size_t sparse_size);
 267
 268   /* striping */
 269   uint64_t get_stripe_unit() const;
 270   uint64_t get_stripe_count() const;
 271
 272   int get_create_timestamp(struct timespec *timestamp);
 273
 274   int flatten();
 275   int flatten_with_progress(ProgressContext &prog_ctx);
 276   /**
 277    * Returns a pair of poolname, imagename for each clone
 278    * of this image at the currently set snapshot.
 279    */
 280   int list_children(std::set<std::pair<std::string, std::string> > *children);
 281
 282   /* advisory locking (see librbd.h for details) */
 283   int list_lockers(std::list<locker_t> *lockers,
 284                    bool *exclusive, std::string *tag);
 285   int lock_exclusive(const std::string& cookie);
 286   int lock_shared(const std::string& cookie, const std::string& tag);
 287   int unlock(const std::string& cookie);
 288   int break_lock(const std::string& client, const std::string& cookie);
 289
 290   /* snapshots */
 291   int snap_list(std::vector<snap_info_t>& snaps);
 292   /* DEPRECATED; use snap_exists2 */
 293   bool snap_exists(const char *snapname) __attribute__ ((deprecated));
 294   int snap_exists2(const char *snapname, bool *exists);
 295   int snap_create(const char *snapname);
 296   int snap_remove(const char *snapname);
 297   int snap_remove2(const char *snapname, uint32_t flags, ProgressContext& pctx);
 298   int snap_rollback(const char *snap_name);
 299   int snap_rollback_with_progress(const char *snap_name, ProgressContext& pctx);
 300   int snap_protect(const char *snap_name);
 301   int snap_unprotect(const char *snap_name);
 302   int snap_is_protected(const char *snap_name, bool *is_protected);
 303   int snap_set(const char *snap_name);
 304   int snap_rename(const char *srcname, const char *dstname);
 305   int snap_get_limit(uint64_t *limit);
 306   int snap_set_limit(uint64_t limit);
 307   int snap_get_timestamp(uint64_t snap_id, struct timespec *timestamp);
 308
 309   /* I/O */
 310   ssize_t read(uint64_t ofs, size_t len, ceph::bufferlist& bl);
 311   /* @param op_flags see librados.h constants beginning with LIBRADOS_OP_FLAG */
 312   ssize_t read2(uint64_t ofs, size_t len, ceph::bufferlist& bl, int op_flags);
 313   int64_t read_iterate(uint64_t ofs, size_t len,
 314                        int (*cb)(uint64_t, size_t, const char *, void *), void *arg);
 315   int read_iterate2(uint64_t ofs, uint64_t len,
 316                     int (*cb)(uint64_t, size_t, const char *, void *), void *arg);
 317   /**
 318    * get difference between two versions of an image
 319    *
 320    * This will return the differences between two versions of an image
 321    * via a callback, which gets the offset and length and a flag
 322    * indicating whether the extent exists (1), or is known/defined to
 323    * be zeros (a hole, 0).  If the source snapshot name is NULL, we
 324    * interpret that as the beginning of time and return all allocated
 325    * regions of the image.  The end version is whatever is currently
 326    * selected for the image handle (either a snapshot or the writeable
 327    * head).
 328    *
 329    * @param fromsnapname start snapshot name, or NULL
 330    * @param ofs start offset
 331    * @param len len in bytes of region to report on
 332    * @param include_parent true if full history diff should include parent
 333    * @param whole_object 1 if diff extents should cover whole object
 334    * @param cb callback to call for each allocated region
 335    * @param arg argument to pass to the callback
 336    * @returns 0 on success, or negative error code on error
 337    */
 338   int diff_iterate(const char *fromsnapname,
 339                    uint64_t ofs, uint64_t len,
 340                    int (*cb)(uint64_t, size_t, int, void *), void *arg);
 341   int diff_iterate2(const char *fromsnapname,
 342                     uint64_t ofs, uint64_t len,
 343                     bool include_parent, bool whole_object,
 344                     int (*cb)(uint64_t, size_t, int, void *), void *arg);
 345
 346   ssize_t write(uint64_t ofs, size_t len, ceph::bufferlist& bl);
 347   /* @param op_flags see librados.h constants beginning with LIBRADOS_OP_FLAG */
 348   ssize_t write2(uint64_t ofs, size_t len, ceph::bufferlist& bl, int op_flags);
 349   int discard(uint64_t ofs, uint64_t len);
 350   ssize_t writesame(uint64_t ofs, size_t len, ceph::bufferlist &bl, int op_flags);
 351   ssize_t compare_and_write(uint64_t ofs, size_t len, ceph::bufferlist &cmp_bl,
 352                             ceph::bufferlist& bl, uint64_t *mismatch_off, int op_flags);
 353
 354   int aio_write(uint64_t off, size_t len, ceph::bufferlist& bl, RBD::AioCompletion *c);
 355   /* @param op_flags see librados.h constants beginning with LIBRADOS_OP_FLAG */
 356   int aio_write2(uint64_t off, size_t len, ceph::bufferlist& bl,
 357                   RBD::AioCompletion *c, int op_flags);
 358   int aio_writesame(uint64_t off, size_t len, ceph::bufferlist& bl,
 359                     RBD::AioCompletion *c, int op_flags);
 360   int aio_compare_and_write(uint64_t off, size_t len, ceph::bufferlist& cmp_bl,
 361                             ceph::bufferlist& bl, RBD::AioCompletion *c,
 362                             uint64_t *mismatch_off, int op_flags);
 363   /**
 364    * read async from image
 365    *
 366    * The target bufferlist is populated with references to buffers
 367    * that contain the data for the given extent of the image.
 368    *
 369    * NOTE: If caching is enabled, the bufferlist will directly
 370    * reference buffers in the cache to avoid an unnecessary data copy.
 371    * As a result, if the user intends to modify the buffer contents
 372    * directly, they should make a copy first (unconditionally, or when
 373    * the reference count on ther underlying buffer is more than 1).
 374    *
 375    * @param off offset in image
 376    * @param len length of read
 377    * @param bl bufferlist to read into
 378    * @param c aio completion to notify when read is complete
 379    */
 380   int aio_read(uint64_t off, size_t len, ceph::bufferlist& bl, RBD::AioCompletion *c);
 381   /* @param op_flags see librados.h constants beginning with LIBRADOS_OP_FLAG */
 382   int aio_read2(uint64_t off, size_t len, ceph::bufferlist& bl,
 383                   RBD::AioCompletion *c, int op_flags);
 384   int aio_discard(uint64_t off, uint64_t len, RBD::AioCompletion *c);
 385
 386   int flush();
 387   /**
 388    * Start a flush if caching is enabled. Get a callback when
 389    * the currently pending writes are on disk.
 390    *
 391    * @param image the image to flush writes to
 392    * @param c what to call when flushing is complete
 393    * @returns 0 on success, negative error code on failure
 394    */
 395   int aio_flush(RBD::AioCompletion *c);
 396
 397   /**
 398    * Drop any cached data for this image
 399    *
 400    * @returns 0 on success, negative error code on failure
 401    */
 402   int invalidate_cache();
 403
 404   int poll_io_events(RBD::AioCompletion **comps, int numcomp);
 405
 406   int metadata_get(const std::string &key, std::string *value);
 407   int metadata_set(const std::string &key, const std::string &value);
 408   int metadata_remove(const std::string &key);
 409   /**
 410    * Returns a pair of key/value for this image
 411    */
 412   int metadata_list(const std::string &start, uint64_t max, std::map<std::string, ceph::bufferlist> *pairs);
 413
 414   // RBD image mirroring support functions
 415   int mirror_image_enable();
 416   int mirror_image_disable(bool force);
 417   int mirror_image_promote(bool force);
 418   int mirror_image_demote();
 419   int mirror_image_resync();
 420   int mirror_image_get_info(mirror_image_info_t *mirror_image_info,
 421                             size_t info_size);
 422   int mirror_image_get_status(mirror_image_status_t *mirror_image_status,
 423                               size_t status_size);
 424   int aio_mirror_image_promote(bool force, RBD::AioCompletion *c);
 425   int aio_mirror_image_demote(RBD::AioCompletion *c);
 426   int aio_mirror_image_get_info(mirror_image_info_t *mirror_image_info,
 427                                 size_t info_size, RBD::AioCompletion *c);
 428   int aio_mirror_image_get_status(mirror_image_status_t *mirror_image_status,
 429                                   size_t status_size, RBD::AioCompletion *c);
 430
 431   int update_watch(UpdateWatchCtx *ctx, uint64_t *handle);
 432   int update_unwatch(uint64_t handle);
 433
 434 private:
 435   friend class RBD;
 436
 437   Image(const Image& rhs);
 438   const Image& operator=(const Image& rhs);
 439
 440   image_ctx_t ctx;
 441 };
 442
 443 }
 444
 445 #endif