Fix: report error if consumer can't be spawned

On error of the UST consumer execl(), we now exit the forked process
like the kernel does.

Furthermore, fix an error handling value when timing out when waiting
for the consumer to bootstrap thus returning an error to the client and
not continuing with the command thinking that the consumer was actually
spawned.

So, the expected behavior is to wait for the pthread condition and if a
timeout is seen, the client is informed thus making the client command
hang for the default timeout value of 30 seconds.

Fixes #702

Signed-off-by: David Goulet <dgoulet@efficios.com>

diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c
index f0aaf73b064f4845974876503de4e7bf7f89f2be..f40c52487004f9d28b70a3ce735e1deaf4c94e5f 100644 (file)
--- a/src/bin/lttng-sessiond/main.c
+++ b/src/bin/lttng-sessiond/main.c
@@ -2151,19 +2151,23 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data)
        if (ret != 0) {
                errno = ret;
                if (ret == ETIMEDOUT) {
+                       int pth_ret;
+
                        /*
                         * Call has timed out so we kill the kconsumerd_thread and return
                         * an error.
                         */
                        ERR("Condition timed out. The consumer thread was never ready."
                                        " Killing it");
-                       ret = pthread_cancel(consumer_data->thread);
-                       if (ret < 0) {
+                       pth_ret = pthread_cancel(consumer_data->thread);
+                       if (pth_ret < 0) {
                                PERROR("pthread_cancel consumer thread");
                        }
                } else {
                        PERROR("pthread_cond_wait failed consumer thread");
                }
+               /* Caller is expecting a negative value on failure. */
+               ret = -1;
                goto error;
        }
 
@@ -2249,10 +2253,11 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data)
                                consumer_to_use = consumerd32_bin;
                        } else {
                                DBG("Could not find any valid consumerd executable");
+                               ret = -EINVAL;
                                break;
                        }
                        DBG("Using kernel consumer at: %s",  consumer_to_use);
-                       execl(consumer_to_use,
+                       ret = execl(consumer_to_use,
                                "lttng-consumerd", verbosity, "-k",
                                "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
                                "--consumerd-err-sock", consumer_data->err_unix_sock_path,
@@ -2300,9 +2305,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data)
                        if (consumerd64_libdir[0] != '\0') {
                                free(tmpnew);
                        }
-                       if (ret) {
-                               goto error;
-                       }
                        break;
                }
                case LTTNG_CONSUMER32_UST:
@@ -2346,9 +2348,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data)
                        if (consumerd32_libdir[0] != '\0') {
                                free(tmpnew);
                        }
-                       if (ret) {
-                               goto error;
-                       }
                        break;
                }
                default:
@@ -2356,8 +2355,9 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data)
                        exit(EXIT_FAILURE);
                }
                if (errno != 0) {
-                       PERROR("kernel start consumer exec");
+                       PERROR("Consumer execl()");
                }
+               /* Reaching this point, we got a failure on our execl(). */
                exit(EXIT_FAILURE);
        } else if (pid > 0) {
                ret = pid;