mirror of https://github.com/OpenIdentityPlatform/OpenDJ.git

Matthew Swift
11.17.2013 976da4aa6126aefd775b7c223c71e29d1207bf10
opendj-ldap-sdk/src/main/java/com/forgerock/opendj/ldap/LDAPConnection.java
@@ -28,7 +28,9 @@
package com.forgerock.opendj.ldap;
import static com.forgerock.opendj.util.StaticUtils.DEBUG_LOG;
import static org.forgerock.opendj.ldap.CoreMessages.*;
import static org.forgerock.opendj.ldap.ErrorResultException.newErrorResult;
import static org.forgerock.opendj.ldap.requests.Requests.newAbandonRequest;
import java.io.IOException;
import java.net.InetSocketAddress;
@@ -65,7 +67,6 @@
import org.forgerock.opendj.ldap.requests.GenericBindRequest;
import org.forgerock.opendj.ldap.requests.ModifyDNRequest;
import org.forgerock.opendj.ldap.requests.ModifyRequest;
import org.forgerock.opendj.ldap.requests.Requests;
import org.forgerock.opendj.ldap.requests.SearchRequest;
import org.forgerock.opendj.ldap.requests.StartTLSExtendedRequest;
import org.forgerock.opendj.ldap.requests.UnbindRequest;
@@ -128,40 +129,64 @@
    @Override
    public FutureResult<Void> abandonAsync(final AbandonRequest request) {
        final AbstractLDAPFutureResultImpl<?> pendingRequest;
        final int messageID = nextMsgID.getAndIncrement();
        /*
         * Need to be careful here since both abandonAsync and Future.cancel can
         * be called separately by the client application. Therefore
         * future.cancel() should abandon the request, and abandonAsync should
         * cancel the future. In addition, bind or StartTLS requests cannot be
         * abandoned.
         */
        try {
            synchronized (stateLock) {
                checkConnectionIsValid();
                checkBindOrStartTLSInProgress();
                // Remove the future associated with the request to be abandoned.
                pendingRequest = pendingRequests.remove(request.getRequestID());
            }
            if (pendingRequest == null) {
                /*
                 * There has never been a request with the specified message ID
                 * or the response has already been received and handled. We can
                 * ignore this abandon request.
                 * If there is a bind or startTLS in progress then it must be
                 * this request which is being abandoned. The following check
                 * will prevent it from happening.
                 */
                // Message ID will be -1 since no request was sent.
                return new CompletedFutureResult<Void>((Void) null);
            }
            pendingRequest.cancel(false);
            try {
                final ASN1BufferWriter asn1Writer = ASN1BufferWriter.getWriter();
                try {
                    ldapWriter.abandonRequest(asn1Writer, messageID, request);
                    connection.write(asn1Writer.getBuffer(), null);
                    return new CompletedFutureResult<Void>((Void) null, messageID);
                } finally {
                    asn1Writer.recycle();
                }
            } catch (final IOException e) {
                throw adaptRequestIOException(e);
                checkBindOrStartTLSInProgress();
            }
        } catch (final ErrorResultException e) {
            return new CompletedFutureResult<Void>(e, messageID);
            return new CompletedFutureResult<Void>(e);
        }
        // Remove the future associated with the request to be abandoned.
        final AbstractLDAPFutureResultImpl<?> pendingRequest =
                pendingRequests.remove(request.getRequestID());
        if (pendingRequest == null) {
            /*
             * There has never been a request with the specified message ID or
             * the response has already been received and handled. We can ignore
             * this abandon request.
             */
            return new CompletedFutureResult<Void>((Void) null);
        }
        /*
         * This will cancel the future, but will also recursively invoke this
         * method. Since the pending request has been removed, there is no risk
         * of an infinite loop.
         */
        pendingRequest.cancel(false);
        /*
         * FIXME: there's a potential race condition here if a bind or startTLS
         * is initiated just after we removed the pending request.
         */
        return sendAbandonRequest(request);
    }
    private FutureResult<Void> sendAbandonRequest(final AbandonRequest request) {
        final ASN1BufferWriter asn1Writer = ASN1BufferWriter.getWriter();
        try {
            final int messageID = nextMsgID.getAndIncrement();
            ldapWriter.abandonRequest(asn1Writer, messageID, request);
            connection.write(asn1Writer.getBuffer(), null);
            return new CompletedFutureResult<Void>((Void) null, messageID);
        } catch (final IOException e) {
            return new CompletedFutureResult<Void>(adaptRequestIOException(e));
        } finally {
            asn1Writer.recycle();
        }
    }
@@ -557,10 +582,54 @@
                if (future != null && future.checkForTimeout()) {
                    final long diff = (future.getTimestamp() + timeout) - currentTime;
                    if (diff <= 0 && pendingRequests.remove(requestID) != null) {
                        DEBUG_LOG.fine("Cancelling expired future result: " + future);
                        final Result result = Responses.newResult(ResultCode.CLIENT_SIDE_TIMEOUT);
                        future.adaptErrorResult(result);
                        abandonAsync(Requests.newAbandonRequest(future.getRequestID()));
                        if (future.isBindOrStartTLS()) {
                            /*
                             * No other operations can be performed while a bind
                             * or StartTLS request is active, so we cannot time
                             * out the request. We therefore have a choice:
                             * either ignore timeouts for these operations, or
                             * enforce them but doing so requires invalidating
                             * the connection. We'll do the latter, since
                             * ignoring timeouts could cause the application to
                             * hang.
                             */
                            DEBUG_LOG.fine("Failing bind or StartTLS request due to timeout "
                                    + "(connection will be invalidated): " + future);
                            final Result result =
                                    Responses.newResult(ResultCode.CLIENT_SIDE_TIMEOUT)
                                            .setDiagnosticMessage(
                                                    LDAP_CONNECTION_BIND_OR_START_TLS_REQUEST_TIMEOUT
                                                            .get(timeout).toString());
                            future.adaptErrorResult(result);
                            // Fail the connection.
                            final Result errorResult =
                                    Responses.newResult(ResultCode.CLIENT_SIDE_TIMEOUT)
                                            .setDiagnosticMessage(
                                                    LDAP_CONNECTION_BIND_OR_START_TLS_CONNECTION_TIMEOUT
                                                            .get(timeout).toString());
                            connectionErrorOccurred(errorResult);
                        } else {
                            DEBUG_LOG.fine("Failing request due to timeout: " + future);
                            final Result result =
                                    Responses.newResult(ResultCode.CLIENT_SIDE_TIMEOUT)
                                            .setDiagnosticMessage(
                                                    LDAP_CONNECTION_REQUEST_TIMEOUT.get(timeout)
                                                            .toString());
                            future.adaptErrorResult(result);
                            /*
                             * FIXME: there's a potential race condition here if
                             * a bind or startTLS is initiated just after we
                             * check the boolean. It seems potentially even more
                             * dangerous to send the abandon request while
                             * holding the state lock, since a blocking write
                             * could hang the application.
                             */
                            if (!bindOrStartTLSInProgress.get()) {
                                sendAbandonRequest(newAbandonRequest(requestID));
                            }
                        }
                    } else {
                        delay = Math.min(delay, diff);
                    }