X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=df4760602b5e4838f90cf124764fbbdad06d5ef4;hp=6bc8b00e914b1d1d6a15c6e82d2e0f520609afcc;hb=025faf73cdb6942ebf437dc4c4f6631f0134b128;hpb=622c9ecfa3a7665e2981aaa4881f58afe07106ce diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 6bc8b00e9..df4760602 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -59,7 +59,6 @@ #include "ust-consumer.h" #include "utils.h" #include "fd-limit.h" -#include "filter.h" #include "health.h" #include "testpoint.h" @@ -228,6 +227,11 @@ struct health_state health_thread_app_manage; struct health_state health_thread_app_reg; struct health_state health_thread_kernel; +/* + * Socket timeout for receiving and sending in seconds. + */ +static int app_socket_timeout; + static void setup_consumerd_path(void) { @@ -392,7 +396,7 @@ static void stop_threads(void) static void cleanup(void) { int ret; - char *cmd; + char *cmd = NULL; struct ltt_session *sess, *stmp; DBG("Cleaning up"); @@ -442,9 +446,6 @@ static void cleanup(void) modprobe_remove_lttng_all(); } - utils_close_pipe(kernel_poll_pipe); - utils_close_pipe(apps_cmd_pipe); - /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", @@ -793,9 +794,13 @@ exit: error: lttng_poll_clean(&events); error_poll_create: + utils_close_pipe(kernel_poll_pipe); + kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1; if (err) { health_error(&health_thread_kernel); ERR("Health error occurred in %s", __func__); + WARN("Kernel thread died unexpectedly. " + "Kernel tracing can continue but CPU hotplug is disabled."); } health_exit(&health_thread_kernel); DBG("Kernel thread dying"); @@ -857,13 +862,6 @@ static void *thread_manage_consumer(void *data) */ health_poll_update(&consumer_data->health); - health_code_update(&consumer_data->health); - - ret = lttcomm_listen_unix_sock(consumer_data->err_sock); - if (ret < 0) { - goto error_listen; - } - /* * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock. * Nothing more will be added to this poll set. @@ -873,6 +871,11 @@ static void *thread_manage_consumer(void *data) goto error_poll; } + /* + * The error socket here is already in a listening state which was done + * just before spawning this thread to avoid a race between the consumer + * daemon exec trying to connect and the listen() call. + */ ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; @@ -928,6 +931,12 @@ restart: goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + health_code_update(&consumer_data->health); DBG2("Receiving code from consumer err_sock"); @@ -1062,7 +1071,6 @@ error: lttng_poll_clean(&events); error_poll: -error_listen: if (err) { health_error(&consumer_data->health); ERR("Health error occurred in %s", __func__); @@ -1202,6 +1210,12 @@ static void *thread_manage_apps(void *data) goto error; } + /* Set socket timeout for both receiving and ending */ + (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock, + app_socket_timeout); + (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock, + app_socket_timeout); + DBG("Apps with sock %d added to poll set", ust_cmd.sock); } @@ -1236,6 +1250,15 @@ exit: error: lttng_poll_clean(&events); error_poll_create: + utils_close_pipe(apps_cmd_pipe); + apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1; + + /* + * We don't clean the UST app hash table here since already registered + * applications can still be controlled so let them be until the session + * daemon dies or the applications stop. + */ + if (err) { health_error(&health_thread_app_manage); ERR("Health error occurred in %s", __func__); @@ -1285,18 +1308,26 @@ static void *thread_dispatch_ust_registration(void *data) * call is blocking so we can be assured that the data will be read * at some point in time or wait to the end of the world :) */ - ret = write(apps_cmd_pipe[1], ust_cmd, - sizeof(struct ust_command)); - if (ret < 0) { - PERROR("write apps cmd pipe"); - if (errno == EBADF) { - /* - * We can't inform the application thread to process - * registration. We will exit or else application - * registration will not occur and tracing will never - * start. - */ - goto error; + if (apps_cmd_pipe[1] >= 0) { + ret = write(apps_cmd_pipe[1], ust_cmd, + sizeof(struct ust_command)); + if (ret < 0) { + PERROR("write apps cmd pipe"); + if (errno == EBADF) { + /* + * We can't inform the application thread to process + * registration. We will exit or else application + * registration will not occur and tracing will never + * start. + */ + goto error; + } + } + } else { + /* Application manager thread is not available. */ + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust_cmd sock"); } } free(ust_cmd); @@ -1402,6 +1433,12 @@ static void *thread_registration_apps(void *data) goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because + * either way, the show must go on. + */ + (void) utils_set_fd_cloexec(sock); + /* Create UST registration command for enqueuing */ ust_cmd = zmalloc(sizeof(struct ust_command)); if (ust_cmd == NULL) { @@ -1814,7 +1851,17 @@ error: */ static int start_consumerd(struct consumer_data *consumer_data) { - int ret; + int ret, err; + + /* + * Set the listen() state on the socket since there is a possible race + * between the exec() of the consumer daemon and this call if place in the + * consumer thread. See bug #366 for more details. + */ + ret = lttcomm_listen_unix_sock(consumer_data->err_sock); + if (ret < 0) { + goto error; + } pthread_mutex_lock(&consumer_data->pid_mutex); if (consumer_data->pid != 0) { @@ -1845,6 +1892,13 @@ end: return 0; error: + /* Cleanup already created socket on error. */ + if (consumer_data->err_sock >= 0) { + err = close(consumer_data->err_sock); + if (err < 0) { + PERROR("close consumer data error socket"); + } + } return ret; } @@ -2407,8 +2461,7 @@ skip_domain: { ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.context.channel_name, - cmd_ctx->lsm->u.context.event_name, - &cmd_ctx->lsm->u.context.ctx); + &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]); break; } case LTTNG_DISABLE_CHANNEL: @@ -2471,7 +2524,7 @@ skip_domain: { ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]); + &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_ALL_EVENT: @@ -2480,7 +2533,7 @@ skip_domain: ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.enable.channel_name, - cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]); + cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]); break; } case LTTNG_LIST_TRACEPOINTS: @@ -2792,15 +2845,15 @@ skip_domain: cmd_ctx->lsm->u.reg.path, cdata); break; } - case LTTNG_SET_FILTER: + case LTTNG_ENABLE_EVENT_WITH_FILTER: { struct lttng_filter_bytecode *bytecode; - if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) { + if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) { ret = LTTNG_ERR_FILTER_INVAL; goto error; } - bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len); + bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len); if (!bytecode) { ret = LTTNG_ERR_FILTER_NOMEM; goto error; @@ -2808,7 +2861,7 @@ skip_domain: /* Receive var. len. data */ DBG("Receiving var len data from client ..."); ret = lttcomm_recv_unix_sock(sock, bytecode, - cmd_ctx->lsm->u.filter.bytecode_len); + cmd_ctx->lsm->u.enable.bytecode_len); if (ret <= 0) { DBG("Nothing recv() from client var len data... continuing"); *sock_error = 1; @@ -2817,16 +2870,15 @@ skip_domain: } if (bytecode->len + sizeof(*bytecode) - != cmd_ctx->lsm->u.filter.bytecode_len) { + != cmd_ctx->lsm->u.enable.bytecode_len) { free(bytecode); ret = LTTNG_ERR_FILTER_INVAL; goto error; } - ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.filter.channel_name, - cmd_ctx->lsm->u.filter.event_name, - bytecode); + ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, + cmd_ctx->lsm->u.enable.channel_name, + &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]); break; } case LTTNG_DATA_PENDING: @@ -2882,6 +2934,12 @@ static void *thread_manage_health(void *data) goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + ret = lttcomm_listen_unix_sock(sock); if (ret < 0) { goto error; @@ -2946,6 +3004,12 @@ restart: goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(new_sock); + DBG("Receiving data from client for health..."); ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg)); if (ret <= 0) { @@ -3062,7 +3126,7 @@ static void *thread_manage_clients(void *data) ret = lttcomm_listen_unix_sock(client_sock); if (ret < 0) { - goto error; + goto error_listen; } /* @@ -3071,7 +3135,7 @@ static void *thread_manage_clients(void *data) */ ret = create_thread_poll_set(&events, 2); if (ret < 0) { - goto error; + goto error_create_poll; } /* Add the application registration socket */ @@ -3143,6 +3207,12 @@ static void *thread_manage_clients(void *data) goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + /* Set socket option for credentials retrieval */ ret = lttcomm_setsockopt_creds_unix_sock(sock); if (ret < 0) { @@ -3244,13 +3314,18 @@ static void *thread_manage_clients(void *data) exit: error: - if (err) { - health_error(&health_thread_cmd); - ERR("Health error occurred in %s", __func__); + if (sock >= 0) { + ret = close(sock); + if (ret) { + PERROR("close"); + } } - health_exit(&health_thread_cmd); - DBG("Client thread dying"); + lttng_poll_clean(&events); + clean_command_ctx(&cmd_ctx); + +error_listen: +error_create_poll: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3258,15 +3333,15 @@ error: PERROR("close"); } } - if (sock >= 0) { - ret = close(sock); - if (ret) { - PERROR("close"); - } + + if (err) { + health_error(&health_thread_cmd); + ERR("Health error occurred in %s", __func__); } - lttng_poll_clean(&events); - clean_command_ctx(&cmd_ctx); + health_exit(&health_thread_cmd); + + DBG("Client thread dying"); rcu_unregister_thread(); return NULL; @@ -3443,6 +3518,14 @@ static int init_daemon_socket(void) goto end; } + /* Set the cloexec flag */ + ret = utils_set_fd_cloexec(client_sock); + if (ret < 0) { + ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). " + "Continuing but note that the consumer daemon will have a " + "reference to this socket on exec()", client_sock); + } + /* File permission MUST be 660 */ ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); if (ret < 0) { @@ -3459,6 +3542,14 @@ static int init_daemon_socket(void) goto end; } + /* Set the cloexec flag */ + ret = utils_set_fd_cloexec(apps_sock); + if (ret < 0) { + ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). " + "Continuing but note that the consumer daemon will have a " + "reference to this socket on exec()", apps_sock); + } + /* File permission MUST be 666 */ ret = chmod(apps_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH); @@ -3468,6 +3559,9 @@ static int init_daemon_socket(void) goto end; } + DBG3("Session daemon client socket %d and application socket %d created", + client_sock, apps_sock); + end: umask(old_umask); return ret; @@ -3727,7 +3821,7 @@ int main(int argc, char **argv) { int ret = 0; void *status; - const char *home_path; + const char *home_path, *env_app_timeout; init_kernel_workarounds(); @@ -3964,8 +4058,10 @@ int main(int argc, char **argv) } /* Setup the kernel pipe for waking up the kernel thread */ - if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) { - goto exit; + if (is_root && !opt_no_kernel) { + if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) { + goto exit; + } } /* Setup the thread apps communication pipe. */ @@ -4007,6 +4103,14 @@ int main(int argc, char **argv) health_init(&ustconsumer64_data.health); health_poll_update(&ustconsumer64_data.health); + /* Check for the application socket timeout env variable. */ + env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV); + if (env_app_timeout) { + app_socket_timeout = atoi(env_app_timeout); + } else { + app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT; + } + /* Create thread to manage the client socket */ ret = pthread_create(&health_thread, NULL, thread_manage_health, (void *) NULL); @@ -4047,18 +4151,21 @@ int main(int argc, char **argv) goto exit_apps; } - /* Create kernel thread to manage kernel event */ - ret = pthread_create(&kernel_thread, NULL, - thread_manage_kernel, (void *) NULL); - if (ret != 0) { - PERROR("pthread_create kernel"); - goto exit_kernel; - } + /* Don't start this thread if kernel tracing is not requested nor root */ + if (is_root && !opt_no_kernel) { + /* Create kernel thread to manage kernel event */ + ret = pthread_create(&kernel_thread, NULL, + thread_manage_kernel, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create kernel"); + goto exit_kernel; + } - ret = pthread_join(kernel_thread, &status); - if (ret != 0) { - PERROR("pthread_join"); - goto error; /* join error, exit without cleanup */ + ret = pthread_join(kernel_thread, &status); + if (ret != 0) { + PERROR("pthread_join"); + goto error; /* join error, exit without cleanup */ + } } exit_kernel: