summaryrefslogtreecommitdiffstats
path: root/mojo/edk/system/dispatcher.h
diff options
context:
space:
mode:
authorjamesr <jamesr@chromium.org>2014-10-02 21:26:48 -0700
committerCommit bot <commit-bot@chromium.org>2014-10-03 04:27:07 +0000
commita03ae49d253d7d1f517e2f92146a6ba201c78aa6 (patch)
treedf4eae1239d49490fa4e85d184b9000224a7a31c /mojo/edk/system/dispatcher.h
parent24b9789375c3c23d0d10e5197c6a7674946ea44b (diff)
downloadchromium_src-a03ae49d253d7d1f517e2f92146a6ba201c78aa6.zip
chromium_src-a03ae49d253d7d1f517e2f92146a6ba201c78aa6.tar.gz
chromium_src-a03ae49d253d7d1f517e2f92146a6ba201c78aa6.tar.bz2
Move mojo edk into mojo/edk
This creates a mojo/edk directory which contains the "embedder developer kit" aka the set of code needed to embed mojo code. mojo/edk/embedder = code from mojo/embedder mojo/edk/system = code from mojo/system mojo/edk/test = code used to test the previous two, from mojo/common/test mojo/edk/ can only depend on mojo/public/, base/ and itself. R=viettrungluu@chromium.org TBR=sky@chromium.org for file renames Committed: https://chromium.googlesource.com/chromium/src/+/ee7ff197a98da4636f33bd713de784948b487bd4 Review URL: https://codereview.chromium.org/621153003 Cr-Commit-Position: refs/heads/master@{#297986}
Diffstat (limited to 'mojo/edk/system/dispatcher.h')
-rw-r--r--mojo/edk/system/dispatcher.h405
1 files changed, 405 insertions, 0 deletions
diff --git a/mojo/edk/system/dispatcher.h b/mojo/edk/system/dispatcher.h
new file mode 100644
index 0000000..3e2230a
--- /dev/null
+++ b/mojo/edk/system/dispatcher.h
@@ -0,0 +1,405 @@
+// Copyright 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef MOJO_EDK_SYSTEM_DISPATCHER_H_
+#define MOJO_EDK_SYSTEM_DISPATCHER_H_
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <vector>
+
+#include "base/macros.h"
+#include "base/memory/ref_counted.h"
+#include "base/memory/scoped_ptr.h"
+#include "base/synchronization/lock.h"
+#include "mojo/edk/embedder/platform_handle.h"
+#include "mojo/edk/embedder/platform_handle_vector.h"
+#include "mojo/edk/system/handle_signals_state.h"
+#include "mojo/edk/system/memory.h"
+#include "mojo/edk/system/system_impl_export.h"
+#include "mojo/public/c/system/buffer.h"
+#include "mojo/public/c/system/data_pipe.h"
+#include "mojo/public/c/system/message_pipe.h"
+#include "mojo/public/c/system/types.h"
+
+namespace mojo {
+
+namespace embedder {
+class PlatformSharedBufferMapping;
+}
+
+namespace system {
+
+class Channel;
+class Core;
+class Dispatcher;
+class DispatcherTransport;
+class HandleTable;
+class LocalMessagePipeEndpoint;
+class ProxyMessagePipeEndpoint;
+class TransportData;
+class Waiter;
+
+typedef std::vector<scoped_refptr<Dispatcher>> DispatcherVector;
+
+namespace test {
+
+// Test helper. We need to declare it here so we can friend it.
+MOJO_SYSTEM_IMPL_EXPORT DispatcherTransport
+ DispatcherTryStartTransport(Dispatcher* dispatcher);
+
+} // namespace test
+
+// A |Dispatcher| implements Mojo primitives that are "attached" to a particular
+// handle. This includes most (all?) primitives except for |MojoWait...()|. This
+// object is thread-safe, with its state being protected by a single lock
+// |lock_|, which is also made available to implementation subclasses (via the
+// |lock()| method).
+class MOJO_SYSTEM_IMPL_EXPORT Dispatcher
+ : public base::RefCountedThreadSafe<Dispatcher> {
+ public:
+ enum Type {
+ kTypeUnknown = 0,
+ kTypeMessagePipe,
+ kTypeDataPipeProducer,
+ kTypeDataPipeConsumer,
+ kTypeSharedBuffer,
+
+ // "Private" types (not exposed via the public interface):
+ kTypePlatformHandle = -1
+ };
+ virtual Type GetType() const = 0;
+
+ // These methods implement the various primitives named |Mojo...()|. These
+ // take |lock_| and handle races with |Close()|. Then they call out to
+ // subclasses' |...ImplNoLock()| methods (still under |lock_|), which actually
+ // implement the primitives.
+ // NOTE(vtl): This puts a big lock around each dispatcher (i.e., handle), and
+ // prevents the various |...ImplNoLock()|s from releasing the lock as soon as
+ // possible. If this becomes an issue, we can rethink this.
+ MojoResult Close();
+
+ // |transports| may be non-null if and only if there are handles to be
+ // written; not that |this| must not be in |transports|. On success, all the
+ // dispatchers in |transports| must have been moved to a closed state; on
+ // failure, they should remain in their original state.
+ MojoResult WriteMessage(UserPointer<const void> bytes,
+ uint32_t num_bytes,
+ std::vector<DispatcherTransport>* transports,
+ MojoWriteMessageFlags flags);
+ // |dispatchers| must be non-null but empty, if |num_dispatchers| is non-null
+ // and nonzero. On success, it will be set to the dispatchers to be received
+ // (and assigned handles) as part of the message.
+ MojoResult ReadMessage(UserPointer<void> bytes,
+ UserPointer<uint32_t> num_bytes,
+ DispatcherVector* dispatchers,
+ uint32_t* num_dispatchers,
+ MojoReadMessageFlags flags);
+ MojoResult WriteData(UserPointer<const void> elements,
+ UserPointer<uint32_t> elements_num_bytes,
+ MojoWriteDataFlags flags);
+ MojoResult BeginWriteData(UserPointer<void*> buffer,
+ UserPointer<uint32_t> buffer_num_bytes,
+ MojoWriteDataFlags flags);
+ MojoResult EndWriteData(uint32_t num_bytes_written);
+ MojoResult ReadData(UserPointer<void> elements,
+ UserPointer<uint32_t> num_bytes,
+ MojoReadDataFlags flags);
+ MojoResult BeginReadData(UserPointer<const void*> buffer,
+ UserPointer<uint32_t> buffer_num_bytes,
+ MojoReadDataFlags flags);
+ MojoResult EndReadData(uint32_t num_bytes_read);
+ // |options| may be null. |new_dispatcher| must not be null, but
+ // |*new_dispatcher| should be null (and will contain the dispatcher for the
+ // new handle on success).
+ MojoResult DuplicateBufferHandle(
+ UserPointer<const MojoDuplicateBufferHandleOptions> options,
+ scoped_refptr<Dispatcher>* new_dispatcher);
+ MojoResult MapBuffer(
+ uint64_t offset,
+ uint64_t num_bytes,
+ MojoMapBufferFlags flags,
+ scoped_ptr<embedder::PlatformSharedBufferMapping>* mapping);
+
+ // Gets the current handle signals state. (The default implementation simply
+ // returns a default-constructed |HandleSignalsState|, i.e., no signals
+ // satisfied or satisfiable.) Note: The state is subject to change from other
+ // threads.
+ HandleSignalsState GetHandleSignalsState() const;
+
+ // Adds a waiter to this dispatcher. The waiter will be woken up when this
+ // object changes state to satisfy |signals| with context |context|. It will
+ // also be woken up when it becomes impossible for the object to ever satisfy
+ // |signals| with a suitable error status.
+ //
+ // If |signals_state| is non-null, on *failure* |*signals_state| will be set
+ // to the current handle signals state (on success, it is left untouched).
+ //
+ // Returns:
+ // - |MOJO_RESULT_OK| if the waiter was added;
+ // - |MOJO_RESULT_ALREADY_EXISTS| if |signals| is already satisfied;
+ // - |MOJO_RESULT_INVALID_ARGUMENT| if the dispatcher has been closed; and
+ // - |MOJO_RESULT_FAILED_PRECONDITION| if it is not (or no longer) possible
+ // that |signals| will ever be satisfied.
+ MojoResult AddWaiter(Waiter* waiter,
+ MojoHandleSignals signals,
+ uint32_t context,
+ HandleSignalsState* signals_state);
+ // Removes a waiter from this dispatcher. (It is valid to call this multiple
+ // times for the same |waiter| on the same object, so long as |AddWaiter()|
+ // was called at most once.) If |signals_state| is non-null, |*signals_state|
+ // will be set to the current handle signals state.
+ void RemoveWaiter(Waiter* waiter, HandleSignalsState* signals_state);
+
+ // A dispatcher must be put into a special state in order to be sent across a
+ // message pipe. Outside of tests, only |HandleTableAccess| is allowed to do
+ // this, since there are requirements on the handle table (see below).
+ //
+ // In this special state, only a restricted set of operations is allowed.
+ // These are the ones available as |DispatcherTransport| methods. Other
+ // |Dispatcher| methods must not be called until |DispatcherTransport::End()|
+ // has been called.
+ class HandleTableAccess {
+ private:
+ friend class Core;
+ friend class HandleTable;
+ // Tests also need this, to avoid needing |Core|.
+ friend DispatcherTransport test::DispatcherTryStartTransport(Dispatcher*);
+
+ // This must be called under the handle table lock and only if the handle
+ // table entry is not marked busy. The caller must maintain a reference to
+ // |dispatcher| until |DispatcherTransport::End()| is called.
+ static DispatcherTransport TryStartTransport(Dispatcher* dispatcher);
+ };
+
+ // A |TransportData| may serialize dispatchers that are given to it (and which
+ // were previously attached to the |MessageInTransit| that is creating it) to
+ // a given |Channel| and then (probably in a different process) deserialize.
+ // Note that the |MessageInTransit| "owns" (i.e., has the only ref to) these
+ // dispatchers, so there are no locking issues. (There's no lock ordering
+ // issue, and in fact no need to take dispatcher locks at all.)
+ // TODO(vtl): Consider making another wrapper similar to |DispatcherTransport|
+ // (but with an owning, unique reference), and having
+ // |CreateEquivalentDispatcherAndCloseImplNoLock()| return that wrapper (and
+ // |MessageInTransit|, etc. only holding on to such wrappers).
+ class TransportDataAccess {
+ private:
+ friend class TransportData;
+
+ // Serialization API. These functions may only be called on such
+ // dispatchers. (|channel| is the |Channel| to which the dispatcher is to be
+ // serialized.) See the |Dispatcher| methods of the same names for more
+ // details.
+ static void StartSerialize(Dispatcher* dispatcher,
+ Channel* channel,
+ size_t* max_size,
+ size_t* max_platform_handles);
+ static bool EndSerializeAndClose(
+ Dispatcher* dispatcher,
+ Channel* channel,
+ void* destination,
+ size_t* actual_size,
+ embedder::PlatformHandleVector* platform_handles);
+
+ // Deserialization API.
+ // Note: This "clears" (i.e., reset to the invalid handle) any platform
+ // handles that it takes ownership of.
+ static scoped_refptr<Dispatcher> Deserialize(
+ Channel* channel,
+ int32_t type,
+ const void* source,
+ size_t size,
+ embedder::PlatformHandleVector* platform_handles);
+ };
+
+ protected:
+ friend class base::RefCountedThreadSafe<Dispatcher>;
+
+ Dispatcher();
+ virtual ~Dispatcher();
+
+ // These are to be overridden by subclasses (if necessary). They are called
+ // exactly once -- first |CancelAllWaitersNoLock()|, then |CloseImplNoLock()|,
+ // when the dispatcher is being closed. They are called under |lock_|.
+ virtual void CancelAllWaitersNoLock();
+ virtual void CloseImplNoLock();
+ virtual scoped_refptr<Dispatcher>
+ CreateEquivalentDispatcherAndCloseImplNoLock() = 0;
+
+ // These are to be overridden by subclasses (if necessary). They are never
+ // called after the dispatcher has been closed. They are called under |lock_|.
+ // See the descriptions of the methods without the "ImplNoLock" for more
+ // information.
+ virtual MojoResult WriteMessageImplNoLock(
+ UserPointer<const void> bytes,
+ uint32_t num_bytes,
+ std::vector<DispatcherTransport>* transports,
+ MojoWriteMessageFlags flags);
+ virtual MojoResult ReadMessageImplNoLock(UserPointer<void> bytes,
+ UserPointer<uint32_t> num_bytes,
+ DispatcherVector* dispatchers,
+ uint32_t* num_dispatchers,
+ MojoReadMessageFlags flags);
+ virtual MojoResult WriteDataImplNoLock(UserPointer<const void> elements,
+ UserPointer<uint32_t> num_bytes,
+ MojoWriteDataFlags flags);
+ virtual MojoResult BeginWriteDataImplNoLock(
+ UserPointer<void*> buffer,
+ UserPointer<uint32_t> buffer_num_bytes,
+ MojoWriteDataFlags flags);
+ virtual MojoResult EndWriteDataImplNoLock(uint32_t num_bytes_written);
+ virtual MojoResult ReadDataImplNoLock(UserPointer<void> elements,
+ UserPointer<uint32_t> num_bytes,
+ MojoReadDataFlags flags);
+ virtual MojoResult BeginReadDataImplNoLock(
+ UserPointer<const void*> buffer,
+ UserPointer<uint32_t> buffer_num_bytes,
+ MojoReadDataFlags flags);
+ virtual MojoResult EndReadDataImplNoLock(uint32_t num_bytes_read);
+ virtual MojoResult DuplicateBufferHandleImplNoLock(
+ UserPointer<const MojoDuplicateBufferHandleOptions> options,
+ scoped_refptr<Dispatcher>* new_dispatcher);
+ virtual MojoResult MapBufferImplNoLock(
+ uint64_t offset,
+ uint64_t num_bytes,
+ MojoMapBufferFlags flags,
+ scoped_ptr<embedder::PlatformSharedBufferMapping>* mapping);
+ virtual HandleSignalsState GetHandleSignalsStateImplNoLock() const;
+ virtual MojoResult AddWaiterImplNoLock(Waiter* waiter,
+ MojoHandleSignals signals,
+ uint32_t context,
+ HandleSignalsState* signals_state);
+ virtual void RemoveWaiterImplNoLock(Waiter* waiter,
+ HandleSignalsState* signals_state);
+
+ // These implement the API used to serialize dispatchers to a |Channel|
+ // (described below). They will only be called on a dispatcher that's attached
+ // to and "owned" by a |MessageInTransit|. See the non-"impl" versions for
+ // more information.
+ //
+ // Note: |StartSerializeImplNoLock()| is actually called with |lock_| NOT
+ // held, since the dispatcher should only be accessible to the calling thread.
+ // On Debug builds, |EndSerializeAndCloseImplNoLock()| is called with |lock_|
+ // held, to satisfy any |lock_.AssertAcquired()| (e.g., in |CloseImplNoLock()|
+ // -- and anything it calls); disentangling those assertions is
+ // difficult/fragile, and would weaken our general checking of invariants.
+ //
+ // TODO(vtl): Consider making these pure virtual once most things support
+ // being passed over a message pipe.
+ virtual void StartSerializeImplNoLock(Channel* channel,
+ size_t* max_size,
+ size_t* max_platform_handles);
+ virtual bool EndSerializeAndCloseImplNoLock(
+ Channel* channel,
+ void* destination,
+ size_t* actual_size,
+ embedder::PlatformHandleVector* platform_handles);
+
+ // Available to subclasses. (Note: Returns a non-const reference, just like
+ // |base::AutoLock|'s constructor takes a non-const reference.)
+ base::Lock& lock() const { return lock_; }
+
+ private:
+ friend class DispatcherTransport;
+
+ // This should be overridden to return true if/when there's an ongoing
+ // operation (e.g., two-phase read/writes on data pipes) that should prevent a
+ // handle from being sent over a message pipe (with status "busy").
+ virtual bool IsBusyNoLock() const;
+
+ // Closes the dispatcher. This must be done under lock, and unlike |Close()|,
+ // the dispatcher must not be closed already. (This is the "equivalent" of
+ // |CreateEquivalentDispatcherAndCloseNoLock()|, for situations where the
+ // dispatcher must be disposed of instead of "transferred".)
+ void CloseNoLock();
+
+ // Creates an equivalent dispatcher -- representing the same resource as this
+ // dispatcher -- and close (i.e., disable) this dispatcher. I.e., this
+ // dispatcher will look as though it was closed, but the resource it
+ // represents will be assigned to the new dispatcher. This must be called
+ // under the dispatcher's lock.
+ scoped_refptr<Dispatcher> CreateEquivalentDispatcherAndCloseNoLock();
+
+ // API to serialize dispatchers to a |Channel|, exposed to only
+ // |TransportData| (via |TransportData|). They may only be called on a
+ // dispatcher attached to a |MessageInTransit| (and in particular not in
+ // |CoreImpl|'s handle table).
+ //
+ // Starts the serialization. Returns (via the two "out" parameters) the
+ // maximum amount of space that may be needed to serialize this dispatcher to
+ // the given |Channel| (no more than
+ // |TransportData::kMaxSerializedDispatcherSize|) and the maximum number of
+ // |PlatformHandle|s that may need to be attached (no more than
+ // |TransportData::kMaxSerializedDispatcherPlatformHandles|). If this
+ // dispatcher cannot be serialized to the given |Channel|, |*max_size| and
+ // |*max_platform_handles| should be set to zero. A call to this method will
+ // ALWAYS be followed by a call to |EndSerializeAndClose()| (even if this
+ // dispatcher cannot be serialized to the given |Channel|).
+ void StartSerialize(Channel* channel,
+ size_t* max_size,
+ size_t* max_platform_handles);
+ // Completes the serialization of this dispatcher to the given |Channel| and
+ // closes it. (This call will always follow an earlier call to
+ // |StartSerialize()|, with the same |Channel|.) This does so by writing to
+ // |destination| and appending any |PlatformHandle|s needed to
+ // |platform_handles| (which may be null if no platform handles were indicated
+ // to be required to |StartSerialize()|). This may write no more than the
+ // amount indicated by |StartSerialize()|. (WARNING: Beware of races, e.g., if
+ // something can be mutated between the two calls!) Returns true on success,
+ // in which case |*actual_size| is set to the amount it actually wrote to
+ // |destination|. On failure, |*actual_size| should not be modified; however,
+ // the dispatcher will still be closed.
+ bool EndSerializeAndClose(Channel* channel,
+ void* destination,
+ size_t* actual_size,
+ embedder::PlatformHandleVector* platform_handles);
+
+ // This protects the following members as well as any state added by
+ // subclasses.
+ mutable base::Lock lock_;
+ bool is_closed_;
+
+ DISALLOW_COPY_AND_ASSIGN(Dispatcher);
+};
+
+// Wrapper around a |Dispatcher| pointer, while it's being processed to be
+// passed in a message pipe. See the comment about
+// |Dispatcher::HandleTableAccess| for more details.
+//
+// Note: This class is deliberately "thin" -- no more expensive than a
+// |Dispatcher*|.
+class MOJO_SYSTEM_IMPL_EXPORT DispatcherTransport {
+ public:
+ DispatcherTransport() : dispatcher_(nullptr) {}
+
+ void End();
+
+ Dispatcher::Type GetType() const { return dispatcher_->GetType(); }
+ bool IsBusy() const { return dispatcher_->IsBusyNoLock(); }
+ void Close() { dispatcher_->CloseNoLock(); }
+ scoped_refptr<Dispatcher> CreateEquivalentDispatcherAndClose() {
+ return dispatcher_->CreateEquivalentDispatcherAndCloseNoLock();
+ }
+
+ bool is_valid() const { return !!dispatcher_; }
+
+ protected:
+ Dispatcher* dispatcher() { return dispatcher_; }
+
+ private:
+ friend class Dispatcher::HandleTableAccess;
+
+ explicit DispatcherTransport(Dispatcher* dispatcher)
+ : dispatcher_(dispatcher) {}
+
+ Dispatcher* dispatcher_;
+
+ // Copy and assign allowed.
+};
+
+} // namespace system
+} // namespace mojo
+
+#endif // MOJO_EDK_SYSTEM_DISPATCHER_H_