Fix: consumers don't honor protocol on rotate pending relay command

The payload of a command should not be sent to the sessiond if
its execution failed. The sessiond reports an error as soon as
the return code of the command is returned and will never consume
the payload, resulting in a protocol error down the line.

Moreover, the ust-consumer's implementation doesn't check for
errors as it stores the return value of
lttng_consumer_rotate_pending_relay() to an unsigned variable and
checks for negative values.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

diff --git a/src/common/kernel-consumer/kernel-consumer.c b/src/common/kernel-consumer/kernel-consumer.c
index 66ae16bc8f09c224ffd3e1130bf9c1fa0ca0fa13..f1c38ccbfa8c6ec5f18ca64fbe6d58dbb4442dbf 100644 (file)
--- a/src/common/kernel-consumer/kernel-consumer.c
+++ b/src/common/kernel-consumer/kernel-consumer.c
@@ -1183,19 +1183,20 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
        }
        case LTTNG_CONSUMER_ROTATE_PENDING_RELAY:
        {
+               int pending;
                uint32_t pending_reply;
 
                DBG("Consumer rotate pending on relay for session %" PRIu64,
                                msg.u.rotate_pending_relay.session_id);
-               ret = lttng_consumer_rotate_pending_relay(
+               pending = lttng_consumer_rotate_pending_relay(
                                msg.u.rotate_pending_relay.session_id,
                                msg.u.rotate_pending_relay.relayd_id,
                                msg.u.rotate_pending_relay.chunk_id);
-               if (ret < 0) {
+               if (pending < 0) {
                        ERR("Rotate pending relay failed");
                        ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
                } else {
-                       pending_reply = !!ret;
+                       pending_reply = !!pending;
                }
 
                health_code_update();
@@ -1206,6 +1207,15 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        goto end_nosignal;
                }
 
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' flag as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
                /* Send back returned value to session daemon */
                ret = lttcomm_send_unix_sock(sock, &pending_reply,
                                sizeof(pending_reply));

diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c
index cb5743fad626ba7a47e8130487ad94efa2172b07..81c9e72fe7456ebb53dea49f3d5af1a314b2958d 100644 (file)
--- a/src/common/ust-consumer/ust-consumer.c
+++ b/src/common/ust-consumer/ust-consumer.c
@@ -2034,7 +2034,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
        }
        case LTTNG_CONSUMER_ROTATE_PENDING_RELAY:
        {
-               uint32_t pending;
+               int pending;
+               uint32_t pending_reply;
 
                DBG("Consumer rotate pending on relay for session %" PRIu64,
                                msg.u.rotate_pending_relay.session_id);
@@ -2045,18 +2046,31 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                if (pending < 0) {
                        ERR("Rotate pending relay failed");
                        ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
+               } else {
+                       pending_reply = !!pending;
                }
 
                health_code_update();
 
+               /* Send whether the command was successful. */
                ret = consumer_send_status_msg(sock, ret_code);
                if (ret < 0) {
                        /* Somehow, the session daemon is not responding anymore. */
                        goto end_nosignal;
                }
 
-               /* Send back returned value to session daemon */
-               ret = lttcomm_send_unix_sock(sock, &pending, sizeof(pending));
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' reply as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
+               /* Send back the command's payload (pending reply). */
+               ret = lttcomm_send_unix_sock(sock, &pending_reply,
+                               sizeof(pending_reply));
                if (ret < 0) {
                        PERROR("send data pending ret code");
                        goto error_fatal;