initial code repo
[stor4nfv.git] / src / ceph / src / librbd / ExclusiveLock.cc
diff --git a/src/ceph/src/librbd/ExclusiveLock.cc b/src/ceph/src/librbd/ExclusiveLock.cc
new file mode 100644 (file)
index 0000000..8436bad
--- /dev/null
@@ -0,0 +1,359 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "librbd/ExclusiveLock.h"
+#include "librbd/ImageCtx.h"
+#include "librbd/ImageWatcher.h"
+#include "librbd/ImageState.h"
+#include "librbd/exclusive_lock/PreAcquireRequest.h"
+#include "librbd/exclusive_lock/PostAcquireRequest.h"
+#include "librbd/exclusive_lock/PreReleaseRequest.h"
+#include "librbd/io/ImageRequestWQ.h"
+#include "librbd/Utils.h"
+#include "common/Mutex.h"
+#include "common/dout.h"
+
+#define dout_subsys ceph_subsys_rbd
+#undef dout_prefix
+#define dout_prefix *_dout << "librbd::ExclusiveLock: " << this << " " \
+                           <<  __func__
+
+namespace librbd {
+
+using namespace exclusive_lock;
+
+template <typename I>
+using ML = ManagedLock<I>;
+
+template <typename I>
+ExclusiveLock<I>::ExclusiveLock(I &image_ctx)
+  : ML<I>(image_ctx.md_ctx, image_ctx.op_work_queue, image_ctx.header_oid,
+          image_ctx.image_watcher, managed_lock::EXCLUSIVE,
+          image_ctx.blacklist_on_break_lock,
+          image_ctx.blacklist_expire_seconds),
+    m_image_ctx(image_ctx) {
+  Mutex::Locker locker(ML<I>::m_lock);
+  ML<I>::set_state_uninitialized();
+}
+
+template <typename I>
+bool ExclusiveLock<I>::accept_requests(int *ret_val) const {
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  bool accept_requests = (!ML<I>::is_state_shutdown() &&
+                          ML<I>::is_state_locked() &&
+                          m_request_blocked_count == 0);
+  if (ret_val != nullptr) {
+    *ret_val = m_request_blocked_ret_val;
+  }
+
+  ldout(m_image_ctx.cct, 20) << "=" << accept_requests << dendl;
+  return accept_requests;
+}
+
+template <typename I>
+bool ExclusiveLock<I>::accept_ops() const {
+  Mutex::Locker locker(ML<I>::m_lock);
+  bool accept = accept_ops(ML<I>::m_lock);
+  ldout(m_image_ctx.cct, 20) << "=" << accept << dendl;
+  return accept;
+}
+
+template <typename I>
+bool ExclusiveLock<I>::accept_ops(const Mutex &lock) const {
+  return (!ML<I>::is_state_shutdown() &&
+          (ML<I>::is_state_locked() || ML<I>::is_state_post_acquiring()));
+}
+
+template <typename I>
+void ExclusiveLock<I>::block_requests(int r) {
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  m_request_blocked_count++;
+  if (m_request_blocked_ret_val == 0) {
+    m_request_blocked_ret_val = r;
+  }
+
+  ldout(m_image_ctx.cct, 20) << dendl;
+}
+
+template <typename I>
+void ExclusiveLock<I>::unblock_requests() {
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  assert(m_request_blocked_count > 0);
+  m_request_blocked_count--;
+  if (m_request_blocked_count == 0) {
+    m_request_blocked_ret_val = 0;
+  }
+
+  ldout(m_image_ctx.cct, 20) << dendl;
+}
+
+template <typename I>
+void ExclusiveLock<I>::init(uint64_t features, Context *on_init) {
+  assert(m_image_ctx.owner_lock.is_locked());
+  ldout(m_image_ctx.cct, 10) << dendl;
+
+  {
+    Mutex::Locker locker(ML<I>::m_lock);
+    ML<I>::set_state_initializing();
+  }
+
+  m_image_ctx.io_work_queue->block_writes(new C_InitComplete(this, features,
+                                                             on_init));
+}
+
+template <typename I>
+void ExclusiveLock<I>::shut_down(Context *on_shut_down) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+
+  ML<I>::shut_down(on_shut_down);
+
+  // if stalled in request state machine -- abort
+  handle_peer_notification(0);
+}
+
+template <typename I>
+void ExclusiveLock<I>::handle_peer_notification(int r) {
+  Mutex::Locker locker(ML<I>::m_lock);
+  if (!ML<I>::is_state_waiting_for_lock()) {
+    return;
+  }
+
+  ldout(m_image_ctx.cct, 10) << dendl;
+  assert(ML<I>::is_action_acquire_lock());
+
+  m_acquire_lock_peer_ret_val = r;
+  ML<I>::execute_next_action();
+}
+
+template <typename I>
+Context *ExclusiveLock<I>::start_op() {
+  assert(m_image_ctx.owner_lock.is_locked());
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  if (!accept_ops(ML<I>::m_lock)) {
+    return nullptr;
+  }
+
+  m_async_op_tracker.start_op();
+  return new FunctionContext([this](int r) {
+      m_async_op_tracker.finish_op();
+    });
+}
+
+template <typename I>
+void ExclusiveLock<I>::handle_init_complete(uint64_t features) {
+  ldout(m_image_ctx.cct, 10) << ": features=" << features << dendl;
+
+  {
+    RWLock::RLocker owner_locker(m_image_ctx.owner_lock);
+    if (m_image_ctx.clone_copy_on_read ||
+        (features & RBD_FEATURE_JOURNALING) != 0) {
+      m_image_ctx.io_work_queue->set_require_lock(io::DIRECTION_BOTH, true);
+    } else {
+      m_image_ctx.io_work_queue->set_require_lock(io::DIRECTION_WRITE, true);
+    }
+  }
+
+  Mutex::Locker locker(ML<I>::m_lock);
+  ML<I>::set_state_unlocked();
+}
+
+template <typename I>
+void ExclusiveLock<I>::shutdown_handler(int r, Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+
+  {
+    RWLock::WLocker owner_locker(m_image_ctx.owner_lock);
+    m_image_ctx.io_work_queue->set_require_lock(io::DIRECTION_BOTH, false);
+    m_image_ctx.exclusive_lock = nullptr;
+  }
+
+  m_image_ctx.io_work_queue->unblock_writes();
+  m_image_ctx.image_watcher->flush(on_finish);
+}
+
+template <typename I>
+void ExclusiveLock<I>::pre_acquire_lock_handler(Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+
+  int acquire_lock_peer_ret_val = 0;
+  {
+    Mutex::Locker locker(ML<I>::m_lock);
+    std::swap(acquire_lock_peer_ret_val, m_acquire_lock_peer_ret_val);
+  }
+
+  if (acquire_lock_peer_ret_val == -EROFS) {
+    ldout(m_image_ctx.cct, 10) << ": peer nacked lock request" << dendl;
+    on_finish->complete(acquire_lock_peer_ret_val);
+    return;
+  }
+
+  PreAcquireRequest<I> *req = PreAcquireRequest<I>::create(m_image_ctx,
+                                                           on_finish);
+  m_image_ctx.op_work_queue->queue(new FunctionContext([req](int r) {
+    req->send();
+  }));
+}
+
+template <typename I>
+void ExclusiveLock<I>::post_acquire_lock_handler(int r, Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << ": r=" << r << dendl;
+
+  if (r == -EROFS) {
+    // peer refused to release the exclusive lock
+    on_finish->complete(r);
+    return;
+  } else if (r < 0) {
+    ML<I>::m_lock.Lock();
+    assert(ML<I>::is_state_acquiring());
+
+    // PostAcquire state machine will not run, so we need complete prepare
+    m_image_ctx.state->handle_prepare_lock_complete();
+
+    // if lock is in-use by another client, request the lock
+    if (ML<I>::is_action_acquire_lock() && (r == -EBUSY || r == -EAGAIN)) {
+      ML<I>::set_state_waiting_for_lock();
+      ML<I>::m_lock.Unlock();
+
+      // request the lock from a peer
+      m_image_ctx.image_watcher->notify_request_lock();
+
+      // inform manage lock that we have interrupted the state machine
+      r = -ECANCELED;
+    } else {
+      ML<I>::m_lock.Unlock();
+
+      // clear error if peer owns lock
+      if (r == -EAGAIN) {
+        r = 0;
+      }
+    }
+
+    on_finish->complete(r);
+    return;
+  }
+
+  Mutex::Locker locker(ML<I>::m_lock);
+  m_pre_post_callback = on_finish;
+  using EL = ExclusiveLock<I>;
+  PostAcquireRequest<I> *req = PostAcquireRequest<I>::create(m_image_ctx,
+      util::create_context_callback<EL, &EL::handle_post_acquiring_lock>(this),
+      util::create_context_callback<EL, &EL::handle_post_acquired_lock>(this));
+
+  m_image_ctx.op_work_queue->queue(new FunctionContext([req](int r) {
+    req->send();
+  }));
+}
+
+template <typename I>
+void ExclusiveLock<I>::handle_post_acquiring_lock(int r) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  assert(r == 0);
+
+  // lock is owned at this point
+  ML<I>::set_state_post_acquiring();
+}
+
+template <typename I>
+void ExclusiveLock<I>::handle_post_acquired_lock(int r) {
+  ldout(m_image_ctx.cct, 10) << ": r=" << r << dendl;
+
+  Context *on_finish = nullptr;
+  {
+    Mutex::Locker locker(ML<I>::m_lock);
+    assert(ML<I>::is_state_acquiring() || ML<I>::is_state_post_acquiring());
+
+    assert (m_pre_post_callback != nullptr);
+    std::swap(m_pre_post_callback, on_finish);
+  }
+
+  if (r >= 0) {
+    m_image_ctx.image_watcher->notify_acquired_lock();
+    m_image_ctx.io_work_queue->set_require_lock(io::DIRECTION_BOTH, false);
+    m_image_ctx.io_work_queue->unblock_writes();
+  }
+
+  on_finish->complete(r);
+}
+
+template <typename I>
+void ExclusiveLock<I>::pre_release_lock_handler(bool shutting_down,
+                                                Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+  Mutex::Locker locker(ML<I>::m_lock);
+
+  PreReleaseRequest<I> *req = PreReleaseRequest<I>::create(
+    m_image_ctx, shutting_down, m_async_op_tracker, on_finish);
+  m_image_ctx.op_work_queue->queue(new FunctionContext([req](int r) {
+    req->send();
+  }));
+}
+
+template <typename I>
+void ExclusiveLock<I>::post_release_lock_handler(bool shutting_down, int r,
+                                                 Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << ": r=" << r << " shutting_down="
+                             << shutting_down << dendl;
+  if (!shutting_down) {
+    {
+      Mutex::Locker locker(ML<I>::m_lock);
+      assert(ML<I>::is_state_pre_releasing() || ML<I>::is_state_releasing());
+    }
+
+    if (r >= 0) {
+      m_image_ctx.image_watcher->notify_released_lock();
+    }
+  } else {
+    {
+      RWLock::WLocker owner_locker(m_image_ctx.owner_lock);
+      m_image_ctx.io_work_queue->set_require_lock(io::DIRECTION_BOTH, false);
+      m_image_ctx.exclusive_lock = nullptr;
+    }
+
+    if (r >= 0) {
+      m_image_ctx.io_work_queue->unblock_writes();
+    }
+
+    m_image_ctx.image_watcher->notify_released_lock();
+  }
+
+  on_finish->complete(r);
+}
+
+template <typename I>
+void ExclusiveLock<I>::post_reacquire_lock_handler(int r, Context *on_finish) {
+  ldout(m_image_ctx.cct, 10) << dendl;
+  if (r >= 0) {
+    m_image_ctx.image_watcher->notify_acquired_lock();
+  }
+
+  on_finish->complete(r);
+}
+
+template <typename I>
+struct ExclusiveLock<I>::C_InitComplete : public Context {
+  ExclusiveLock *exclusive_lock;
+  uint64_t features;
+  Context *on_init;
+
+  C_InitComplete(ExclusiveLock *exclusive_lock, uint64_t features,
+                 Context *on_init)
+    : exclusive_lock(exclusive_lock), features(features), on_init(on_init) {
+  }
+  void finish(int r) override {
+    if (r == 0) {
+      exclusive_lock->handle_init_complete(features);
+    }
+    on_init->complete(r);
+  }
+};
+
+} // namespace librbd
+
+template class librbd::ExclusiveLock<librbd::ImageCtx>;