X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=d735be833f4064e88d0cd801a45c6d781d618c55;hb=c119cdb1e4cf3e0fd5080160a5c4286cd7391714;hp=4d8851ef813168a8445ad77ef12c1fdc838101d9;hpb=256a55760810f90bdcbd3d8bc13963677ba49b9d;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 4d8851ef8..d735be833 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -34,6 +34,7 @@ #include #include #include +#include #include #include @@ -54,6 +55,7 @@ #include "shm.h" #include "ust-ctl.h" #include "utils.h" +#include "fd-limit.h" #define CONSUMERD_FILE "lttng-consumerd" @@ -178,6 +180,40 @@ static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN; static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR; static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR; +/* + * Consumer daemon state which is changed when spawning it, killing it or in + * case of a fatal error. + */ +enum consumerd_state { + CONSUMER_STARTED = 1, + CONSUMER_STOPPED = 2, + CONSUMER_ERROR = 3, +}; + +/* + * This consumer daemon state is used to validate if a client command will be + * able to reach the consumer. If not, the client is informed. For instance, + * doing a "lttng start" when the consumer state is set to ERROR will return an + * error to the client. + * + * The following example shows a possible race condition of this scheme: + * + * consumer thread error happens + * client cmd arrives + * client cmd checks state -> still OK + * consumer thread exit, sets error + * client cmd try to talk to consumer + * ... + * + * However, since the consumer is a different daemon, we have no way of making + * sure the command will reach it safely even with this state flag. This is why + * we consider that up to the state validation during command processing, the + * command is safe. After that, we can not guarantee the correctness of the + * client request vis-a-vis the consumer. + */ +static enum consumerd_state ust_consumerd_state; +static enum consumerd_state kernel_consumerd_state; + static void setup_consumerd_path(void) { @@ -449,7 +485,6 @@ static void cleanup(void) if (ret) { PERROR("close"); } - } } for (i = 0; i < 2; i++) { @@ -1089,6 +1124,17 @@ restart_poll: ERR("consumer return code : %s", lttcomm_get_readable_code(-code)); error: + /* Immediately set the consumerd state to stopped */ + if (consumer_data->type == LTTNG_CONSUMER_KERNEL) { + uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR); + } else if (consumer_data->type == LTTNG_CONSUMER64_UST || + consumer_data->type == LTTNG_CONSUMER32_UST) { + uatomic_set(&ust_consumerd_state, CONSUMER_ERROR); + } else { + /* Code flow error... */ + assert(0); + } + if (consumer_data->err_sock >= 0) { ret = close(consumer_data->err_sock); if (ret) { @@ -1424,6 +1470,17 @@ static void *thread_registration_apps(void *data) * Using message-based transmissions to ensure we don't * have to deal with partially received messages. */ + ret = lttng_fd_get(LTTNG_FD_APPS, 1); + if (ret < 0) { + ERR("Exhausted file descriptors allowed for applications."); + free(ust_cmd); + ret = close(sock); + if (ret) { + PERROR("close"); + } + sock = -1; + continue; + } ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg, sizeof(struct ust_register_msg)); if (ret < 0 || ret < sizeof(struct ust_register_msg)) { @@ -1437,6 +1494,7 @@ static void *thread_registration_apps(void *data) if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); sock = -1; continue; } @@ -1482,6 +1540,7 @@ error: if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); } unlink(apps_unix_sock_path); @@ -3167,6 +3226,9 @@ static ssize_t cmd_list_channels(int domain, struct ltt_session *session, nb_chan = session->kernel_session->channel_count; } DBG3("Number of kernel channels %zd", nb_chan); + if (nb_chan <= 0) { + ret = LTTCOMM_KERN_CHAN_NOT_FOUND; + } break; case LTTNG_DOMAIN_UST: if (session->ust_session != NULL) { @@ -3174,6 +3236,9 @@ static ssize_t cmd_list_channels(int domain, struct ltt_session *session, session->ust_session->domain_global.channels); } DBG3("Number of UST global channels %zd", nb_chan); + if (nb_chan <= 0) { + ret = LTTCOMM_UST_CHAN_NOT_FOUND; + } break; default: *channels = NULL; @@ -3191,6 +3256,8 @@ static ssize_t cmd_list_channels(int domain, struct ltt_session *session, list_lttng_channels(domain, session, *channels); } else { *channels = NULL; + /* Ret value was set in the domain switch case */ + goto error; } return nb_chan; @@ -3347,6 +3414,12 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } } + /* Consumer is in an ERROR state. Report back to client */ + if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) { + ret = LTTCOMM_NO_KERNCONSUMERD; + goto error; + } + /* Need a session for kernel command */ if (need_tracing_session) { if (cmd_ctx->session->kernel_session == NULL) { @@ -3367,13 +3440,21 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ret = LTTCOMM_KERN_CONSUMER_FAIL; goto error; } + uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED); } else { pthread_mutex_unlock(&kconsumer_data.pid_mutex); } } + break; case LTTNG_DOMAIN_UST: { + /* Consumer is in an ERROR state. Report back to client */ + if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { + ret = LTTCOMM_NO_USTCONSUMERD; + goto error; + } + if (need_tracing_session) { if (cmd_ctx->session->ust_session == NULL) { ret = create_ust_session(cmd_ctx->session, @@ -3397,6 +3478,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } ust_consumerd64_fd = ustconsumer64_data.cmd_sock; + uatomic_set(&ust_consumerd_state, CONSUMER_STARTED); } else { pthread_mutex_unlock(&ustconsumer64_data.pid_mutex); } @@ -3411,7 +3493,9 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ust_consumerd32_fd = -EINVAL; goto error; } + ust_consumerd32_fd = ustconsumer32_data.cmd_sock; + uatomic_set(&ust_consumerd_state, CONSUMER_STARTED); } else { pthread_mutex_unlock(&ustconsumer32_data.pid_mutex); } @@ -3423,6 +3507,25 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } skip_domain: + /* Validate consumer daemon state when start/stop trace command */ + if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE || + cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) { + switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_UST: + if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) { + ret = LTTCOMM_NO_USTCONSUMERD; + goto error; + } + break; + case LTTNG_DOMAIN_KERNEL: + if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) { + ret = LTTCOMM_NO_KERNCONSUMERD; + goto error; + } + break; + } + } + /* * Check that the UID or GID match that of the tracing session. * The root user can interact with all sessions. @@ -3573,7 +3676,7 @@ skip_domain: } case LTTNG_LIST_CHANNELS: { - size_t nb_chan; + int nb_chan; struct lttng_channel *channels; nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type, @@ -4103,13 +4206,15 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - gid = allowed_group(); - if (gid < 0) { + ret = allowed_group(); + if (ret < 0) { WARN("No tracing group detected"); ret = 0; goto end; } + gid = ret; + /* Set lttng run dir */ ret = chown(rundir, 0, gid); if (ret < 0) { @@ -4388,11 +4493,6 @@ int main(int argc, char **argv) rcu_register_thread(); - /* Create thread quit pipe */ - if ((ret = init_thread_quit_pipe()) < 0) { - goto error; - } - setup_consumerd_path(); /* Parse arguments */ @@ -4403,11 +4503,31 @@ int main(int argc, char **argv) /* Daemonize */ if (opt_daemon) { + int i; + + /* + * fork + * child: setsid, close FD 0, 1, 2, chdir / + * parent: exit (if fork is successful) + */ ret = daemon(0, 0); if (ret < 0) { PERROR("daemon"); goto error; } + /* + * We are in the child. Make sure all other file + * descriptors are closed, in case we are called with + * more opened file descriptors than the standard ones. + */ + for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { + (void) close(i); + } + } + + /* Create thread quit pipe */ + if ((ret = init_thread_quit_pipe()) < 0) { + goto error; } /* Check if daemon is UID = 0 */ @@ -4490,6 +4610,10 @@ int main(int argc, char **argv) } } + /* Set consumer initial state */ + kernel_consumerd_state = CONSUMER_STOPPED; + ust_consumerd_state = CONSUMER_STOPPED; + DBG("Client socket path %s", client_unix_sock_path); DBG("Application socket path %s", apps_unix_sock_path); DBG("LTTng run directory path: %s", rundir); @@ -4556,6 +4680,8 @@ int main(int argc, char **argv) /* Set ulimit for open files */ set_ulimit(); } + /* init lttng_fd tracking must be done after set_ulimit. */ + lttng_fd_init(); ret = set_consumer_sockets(&ustconsumer64_data, rundir); if (ret < 0) {