ceph/src/tools/rbd_mirror/ImageSync.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3
   4 #include "ImageSync.h"
   5 #include "InstanceWatcher.h"
   6 #include "ProgressContext.h"
   7 #include "common/errno.h"
   8 #include "journal/Journaler.h"
   9 #include "librbd/ExclusiveLock.h"
  10 #include "librbd/ImageCtx.h"
  11 #include "librbd/ObjectMap.h"
  12 #include "librbd/Utils.h"
  13 #include "librbd/journal/Types.h"
  14 #include "tools/rbd_mirror/image_sync/ImageCopyRequest.h"
  15 #include "tools/rbd_mirror/image_sync/SnapshotCopyRequest.h"
  16 #include "tools/rbd_mirror/image_sync/SyncPointCreateRequest.h"
  17 #include "tools/rbd_mirror/image_sync/SyncPointPruneRequest.h"
  18
  19 #define dout_context g_ceph_context
  20 #define dout_subsys ceph_subsys_rbd_mirror
  21 #undef dout_prefix
  22 #define dout_prefix *_dout << "rbd::mirror::ImageSync: " \
  23                            << this << " " << __func__
  24
  25 namespace rbd {
  26 namespace mirror {
  27
  28 using namespace image_sync;
  29 using librbd::util::create_context_callback;
  30 using librbd::util::unique_lock_name;
  31
  32 template <typename I>
  33 ImageSync<I>::ImageSync(I *local_image_ctx, I *remote_image_ctx,
  34                         SafeTimer *timer, Mutex *timer_lock,
  35                         const std::string &mirror_uuid, Journaler *journaler,
  36                         MirrorPeerClientMeta *client_meta,
  37                         ContextWQ *work_queue,
  38                         InstanceWatcher<I> *instance_watcher,
  39                         Context *on_finish, ProgressContext *progress_ctx)
  40   : BaseRequest("rbd::mirror::ImageSync", local_image_ctx->cct, on_finish),
  41     m_local_image_ctx(local_image_ctx), m_remote_image_ctx(remote_image_ctx),
  42     m_timer(timer), m_timer_lock(timer_lock), m_mirror_uuid(mirror_uuid),
  43     m_journaler(journaler), m_client_meta(client_meta),
  44     m_work_queue(work_queue), m_instance_watcher(instance_watcher),
  45     m_progress_ctx(progress_ctx),
  46     m_lock(unique_lock_name("ImageSync::m_lock", this)) {
  47 }
  48
  49 template <typename I>
  50 ImageSync<I>::~ImageSync() {
  51   assert(m_snapshot_copy_request == nullptr);
  52   assert(m_image_copy_request == nullptr);
  53 }
  54
  55 template <typename I>
  56 void ImageSync<I>::send() {
  57   send_notify_sync_request();
  58 }
  59
  60 template <typename I>
  61 void ImageSync<I>::cancel() {
  62   Mutex::Locker locker(m_lock);
  63
  64   dout(20) << dendl;
  65
  66   m_canceled = true;
  67
  68   if (m_instance_watcher->cancel_sync_request(m_local_image_ctx->id)) {
  69     return;
  70   }
  71
  72   if (m_snapshot_copy_request != nullptr) {
  73     m_snapshot_copy_request->cancel();
  74   }
  75
  76   if (m_image_copy_request != nullptr) {
  77     m_image_copy_request->cancel();
  78   }
  79 }
  80
  81 template <typename I>
  82 void ImageSync<I>::send_notify_sync_request() {
  83   update_progress("NOTIFY_SYNC_REQUEST");
  84
  85   dout(20) << dendl;
  86
  87   Context *ctx = create_context_callback<
  88     ImageSync<I>, &ImageSync<I>::handle_notify_sync_request>(this);
  89   m_instance_watcher->notify_sync_request(m_local_image_ctx->id, ctx);
  90 }
  91
  92 template <typename I>
  93 void ImageSync<I>::handle_notify_sync_request(int r) {
  94   dout(20) << ": r=" << r << dendl;
  95
  96   if (r < 0) {
  97     BaseRequest::finish(r);
  98     return;
  99   }
 100
 101   send_prune_catch_up_sync_point();
 102 }
 103
 104 template <typename I>
 105 void ImageSync<I>::send_prune_catch_up_sync_point() {
 106   update_progress("PRUNE_CATCH_UP_SYNC_POINT");
 107
 108   if (m_client_meta->sync_points.empty()) {
 109     send_create_sync_point();
 110     return;
 111   }
 112
 113   dout(20) << dendl;
 114
 115   // prune will remove sync points with missing snapshots and
 116   // ensure we have a maximum of one sync point (in case we
 117   // restarted)
 118   Context *ctx = create_context_callback<
 119     ImageSync<I>, &ImageSync<I>::handle_prune_catch_up_sync_point>(this);
 120   SyncPointPruneRequest<I> *request = SyncPointPruneRequest<I>::create(
 121     m_remote_image_ctx, false, m_journaler, m_client_meta, ctx);
 122   request->send();
 123 }
 124
 125 template <typename I>
 126 void ImageSync<I>::handle_prune_catch_up_sync_point(int r) {
 127   dout(20) << ": r=" << r << dendl;
 128
 129   if (r < 0) {
 130     derr << ": failed to prune catch-up sync point: "
 131          << cpp_strerror(r) << dendl;
 132     finish(r);
 133     return;
 134   }
 135
 136   send_create_sync_point();
 137 }
 138
 139 template <typename I>
 140 void ImageSync<I>::send_create_sync_point() {
 141   update_progress("CREATE_SYNC_POINT");
 142
 143   // TODO: when support for disconnecting laggy clients is added,
 144   //       re-connect and create catch-up sync point
 145   if (m_client_meta->sync_points.size() > 0) {
 146     send_copy_snapshots();
 147     return;
 148   }
 149
 150   dout(20) << dendl;
 151
 152   Context *ctx = create_context_callback<
 153     ImageSync<I>, &ImageSync<I>::handle_create_sync_point>(this);
 154   SyncPointCreateRequest<I> *request = SyncPointCreateRequest<I>::create(
 155     m_remote_image_ctx, m_mirror_uuid, m_journaler, m_client_meta, ctx);
 156   request->send();
 157 }
 158
 159 template <typename I>
 160 void ImageSync<I>::handle_create_sync_point(int r) {
 161   dout(20) << ": r=" << r << dendl;
 162
 163   if (r < 0) {
 164     derr << ": failed to create sync point: " << cpp_strerror(r)
 165          << dendl;
 166     finish(r);
 167     return;
 168   }
 169
 170   send_copy_snapshots();
 171 }
 172
 173 template <typename I>
 174 void ImageSync<I>::send_copy_snapshots() {
 175   m_lock.Lock();
 176   if (m_canceled) {
 177     m_lock.Unlock();
 178     finish(-ECANCELED);
 179     return;
 180   }
 181
 182   dout(20) << dendl;
 183
 184   Context *ctx = create_context_callback<
 185     ImageSync<I>, &ImageSync<I>::handle_copy_snapshots>(this);
 186   m_snapshot_copy_request = SnapshotCopyRequest<I>::create(
 187     m_local_image_ctx, m_remote_image_ctx, &m_snap_map, m_journaler,
 188     m_client_meta, m_work_queue, ctx);
 189   m_snapshot_copy_request->get();
 190   m_lock.Unlock();
 191
 192   update_progress("COPY_SNAPSHOTS");
 193
 194   m_snapshot_copy_request->send();
 195 }
 196
 197 template <typename I>
 198 void ImageSync<I>::handle_copy_snapshots(int r) {
 199   dout(20) << ": r=" << r << dendl;
 200
 201   {
 202     Mutex::Locker locker(m_lock);
 203     m_snapshot_copy_request->put();
 204     m_snapshot_copy_request = nullptr;
 205     if (r == 0 && m_canceled) {
 206       r = -ECANCELED;
 207     }
 208   }
 209
 210   if (r == -ECANCELED) {
 211     dout(10) << ": snapshot copy canceled" << dendl;
 212     finish(r);
 213     return;
 214   } else if (r < 0) {
 215     derr << ": failed to copy snapshot metadata: " << cpp_strerror(r) << dendl;
 216     finish(r);
 217     return;
 218   }
 219
 220   send_copy_image();
 221 }
 222
 223 template <typename I>
 224 void ImageSync<I>::send_copy_image() {
 225   m_lock.Lock();
 226   if (m_canceled) {
 227     m_lock.Unlock();
 228     finish(-ECANCELED);
 229     return;
 230   }
 231
 232   dout(20) << dendl;
 233
 234   Context *ctx = create_context_callback<
 235     ImageSync<I>, &ImageSync<I>::handle_copy_image>(this);
 236   m_image_copy_request = ImageCopyRequest<I>::create(
 237     m_local_image_ctx, m_remote_image_ctx, m_timer, m_timer_lock,
 238     m_journaler, m_client_meta, &m_client_meta->sync_points.front(),
 239     ctx, m_progress_ctx);
 240   m_image_copy_request->get();
 241   m_lock.Unlock();
 242
 243   update_progress("COPY_IMAGE");
 244
 245   m_image_copy_request->send();
 246 }
 247
 248 template <typename I>
 249 void ImageSync<I>::handle_copy_image(int r) {
 250   dout(20) << ": r=" << r << dendl;
 251
 252   {
 253     Mutex::Locker locker(m_lock);
 254     m_image_copy_request->put();
 255     m_image_copy_request = nullptr;
 256     if (r == 0 && m_canceled) {
 257       r = -ECANCELED;
 258     }
 259   }
 260
 261   if (r == -ECANCELED) {
 262     dout(10) << ": image copy canceled" << dendl;
 263     finish(r);
 264     return;
 265   } else if (r < 0) {
 266     derr << ": failed to copy image: " << cpp_strerror(r) << dendl;
 267     finish(r);
 268     return;
 269   }
 270
 271   send_copy_object_map();
 272 }
 273
 274 template <typename I>
 275 void ImageSync<I>::send_copy_object_map() {
 276   update_progress("COPY_OBJECT_MAP");
 277
 278   m_local_image_ctx->owner_lock.get_read();
 279   m_local_image_ctx->snap_lock.get_read();
 280   if (!m_local_image_ctx->test_features(RBD_FEATURE_OBJECT_MAP,
 281                                         m_local_image_ctx->snap_lock)) {
 282     m_local_image_ctx->snap_lock.put_read();
 283     m_local_image_ctx->owner_lock.put_read();
 284     send_prune_sync_points();
 285     return;
 286   }
 287
 288   assert(m_local_image_ctx->object_map != nullptr);
 289
 290   assert(!m_client_meta->sync_points.empty());
 291   librbd::journal::MirrorPeerSyncPoint &sync_point =
 292     m_client_meta->sync_points.front();
 293   auto snap_id_it = m_local_image_ctx->snap_ids.find(
 294     {cls::rbd::UserSnapshotNamespace(), sync_point.snap_name});
 295   assert(snap_id_it != m_local_image_ctx->snap_ids.end());
 296   librados::snap_t snap_id = snap_id_it->second;
 297
 298   dout(20) << ": snap_id=" << snap_id << ", "
 299            << "snap_name=" << sync_point.snap_name << dendl;
 300
 301   Context *finish_op_ctx = nullptr;
 302   if (m_local_image_ctx->exclusive_lock != nullptr) {
 303     finish_op_ctx = m_local_image_ctx->exclusive_lock->start_op();
 304   }
 305   if (finish_op_ctx == nullptr) {
 306     derr << ": lost exclusive lock" << dendl;
 307     m_local_image_ctx->snap_lock.put_read();
 308     m_local_image_ctx->owner_lock.put_read();
 309     finish(-EROFS);
 310     return;
 311   }
 312
 313   // rollback the object map (copy snapshot object map to HEAD)
 314   RWLock::WLocker object_map_locker(m_local_image_ctx->object_map_lock);
 315   auto ctx = new FunctionContext([this, finish_op_ctx](int r) {
 316       handle_copy_object_map(r);
 317       finish_op_ctx->complete(0);
 318     });
 319   m_local_image_ctx->object_map->rollback(snap_id, ctx);
 320   m_local_image_ctx->snap_lock.put_read();
 321   m_local_image_ctx->owner_lock.put_read();
 322 }
 323
 324 template <typename I>
 325 void ImageSync<I>::handle_copy_object_map(int r) {
 326   dout(20) << dendl;
 327
 328   assert(r == 0);
 329   send_refresh_object_map();
 330 }
 331
 332 template <typename I>
 333 void ImageSync<I>::send_refresh_object_map() {
 334   dout(20) << dendl;
 335
 336   update_progress("REFRESH_OBJECT_MAP");
 337
 338   Context *ctx = create_context_callback<
 339     ImageSync<I>, &ImageSync<I>::handle_refresh_object_map>(this);
 340   m_object_map = m_local_image_ctx->create_object_map(CEPH_NOSNAP);
 341   m_object_map->open(ctx);
 342 }
 343
 344 template <typename I>
 345 void ImageSync<I>::handle_refresh_object_map(int r) {
 346   dout(20) << dendl;
 347
 348   assert(r == 0);
 349   {
 350     RWLock::WLocker snap_locker(m_local_image_ctx->snap_lock);
 351     std::swap(m_local_image_ctx->object_map, m_object_map);
 352   }
 353   delete m_object_map;
 354
 355   send_prune_sync_points();
 356 }
 357
 358 template <typename I>
 359 void ImageSync<I>::send_prune_sync_points() {
 360   dout(20) << dendl;
 361
 362   update_progress("PRUNE_SYNC_POINTS");
 363
 364   Context *ctx = create_context_callback<
 365     ImageSync<I>, &ImageSync<I>::handle_prune_sync_points>(this);
 366   SyncPointPruneRequest<I> *request = SyncPointPruneRequest<I>::create(
 367     m_remote_image_ctx, true, m_journaler, m_client_meta, ctx);
 368   request->send();
 369 }
 370
 371 template <typename I>
 372 void ImageSync<I>::handle_prune_sync_points(int r) {
 373   dout(20) << ": r=" << r << dendl;
 374
 375   if (r < 0) {
 376     derr << ": failed to prune sync point: "
 377          << cpp_strerror(r) << dendl;
 378     finish(r);
 379     return;
 380   }
 381
 382   if (!m_client_meta->sync_points.empty()) {
 383     send_copy_image();
 384     return;
 385   }
 386
 387   finish(0);
 388 }
 389
 390 template <typename I>
 391 void ImageSync<I>::update_progress(const std::string &description) {
 392   dout(20) << ": " << description << dendl;
 393
 394   if (m_progress_ctx) {
 395     m_progress_ctx->update_progress("IMAGE_SYNC/" + description);
 396   }
 397 }
 398
 399 template <typename I>
 400 void ImageSync<I>::finish(int r) {
 401   dout(20) << ": r=" << r << dendl;
 402
 403   m_instance_watcher->notify_sync_complete(m_local_image_ctx->id);
 404   BaseRequest::finish(r);
 405 }
 406
 407 } // namespace mirror
 408 } // namespace rbd
 409
 410 template class rbd::mirror::ImageSync<librbd::ImageCtx>;