Fix: sessiond: ust: deadlock with per-pid buffers

[lttng-tools.git] / src / bin / lttng-sessiond / ust-app.c

diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c
index 48da16b1072a6264949a889c8298958e2c33d4d3..efc82b05241bd6348160a8ab70cafacadb4a37a8 100644 (file)
--- a/src/bin/lttng-sessiond/ust-app.c
+++ b/src/bin/lttng-sessiond/ust-app.c
@@ -27,7 +27,6 @@
 #include <sys/types.h>
 #include <unistd.h>
 #include <urcu/compiler.h>
-#include <lttng/ust-error.h>
 #include <signal.h>
 
 #include <common/common.h>
@@ -38,7 +37,8 @@
 #include "health-sessiond.h"
 #include "ust-app.h"
 #include "ust-consumer.h"
-#include "ust-ctl.h"
+#include "lttng-ust-ctl.h"
+#include "lttng-ust-error.h"
 #include "utils.h"
 #include "session.h"
 #include "lttng-sessiond.h"
@@ -731,6 +731,10 @@ error:
  * nullified. The session lock MUST be held unless the application is
  * in the destroy path.
  *
+ * Do not hold the registry lock while communicating with the consumerd, because
+ * doing so causes inter-process deadlocks between consumerd and sessiond with
+ * the metadata request notification.
+ *
  * Return 0 on success else a negative value.
  */
 static int close_metadata(struct ust_registry_session *registry,
@@ -738,6 +742,8 @@ static int close_metadata(struct ust_registry_session *registry,
 {
        int ret;
        struct consumer_socket *socket;
+       uint64_t metadata_key;
+       bool registry_was_already_closed;
 
        assert(registry);
        assert(consumer);
@@ -745,8 +751,19 @@ static int close_metadata(struct ust_registry_session *registry,
        rcu_read_lock();
 
        pthread_mutex_lock(&registry->lock);
+       metadata_key = registry->metadata_key;
+       registry_was_already_closed = registry->metadata_closed;
+       if (metadata_key != 0) {
+               /*
+                * Metadata closed. Even on error this means that the consumer
+                * is not responding or not found so either way a second close
+                * should NOT be emit for this registry.
+                */
+               registry->metadata_closed = 1;
+       }
+       pthread_mutex_unlock(&registry->lock);
 
-       if (!registry->metadata_key || registry->metadata_closed) {
+       if (metadata_key == 0 || registry_was_already_closed) {
                ret = 0;
                goto end;
        }
@@ -756,23 +773,15 @@ static int close_metadata(struct ust_registry_session *registry,
                        consumer);
        if (!socket) {
                ret = -1;
-               goto error;
+               goto end;
        }
 
-       ret = consumer_close_metadata(socket, registry->metadata_key);
+       ret = consumer_close_metadata(socket, metadata_key);
        if (ret < 0) {
-               goto error;
+               goto end;
        }
 
-error:
-       /*
-        * Metadata closed. Even on error this means that the consumer is not
-        * responding or not found so either way a second close should NOT be emit
-        * for this registry.
-        */
-       registry->metadata_closed = 1;
 end:
-       pthread_mutex_unlock(&registry->lock);
        rcu_read_unlock();
        return ret;
 }
@@ -3012,7 +3021,7 @@ static int create_channel_per_pid(struct ust_app *app,
        if (ret < 0) {
                ERR("Error creating UST channel \"%s\" on the consumer daemon",
                        ua_chan->name);
-               goto error;
+               goto error_remove_from_registry;
        }
 
        ret = send_channel_pid_to_ust(app, ua_sess, ua_chan);
@@ -3020,7 +3029,7 @@ static int create_channel_per_pid(struct ust_app *app,
                if (ret != -ENOTCONN) {
                        ERR("Error sending channel to application");
                }
-               goto error;
+               goto error_remove_from_registry;
        }
 
        session = session_find_by_id(ua_sess->tracing_id);
@@ -3043,9 +3052,13 @@ static int create_channel_per_pid(struct ust_app *app,
        if (cmd_ret != LTTNG_OK) {
                ret = - (int) cmd_ret;
                ERR("Failed to add channel to notification thread");
-               goto error;
+               goto error_remove_from_registry;
        }
 
+error_remove_from_registry:
+       if (ret) {
+               ust_registry_channel_del_free(registry, ua_chan->key, false);
+       }
 error:
        rcu_read_unlock();
        return ret;