X-Git-Url: http://git.efficios.com/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=5ec39649969f9fef523e10b7e94896149d5812c1;hb=bd11e201a9d7e2f2f71a841a6617aa53074b030a;hp=9d0d01f14662bee7b023eb0c621c431d651c3add;hpb=4ccd8c8fb76d95ecf81a599903e1ff94203e899f;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 9d0d01f14..5ec396499 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -60,17 +60,14 @@ #include "ust-consumer.h" #include "utils.h" #include "fd-limit.h" -#include "health.h" +#include "health-sessiond.h" #include "testpoint.h" #include "ust-thread.h" #define CONSUMERD_FILE "lttng-consumerd" -/* Const values */ -const char default_tracing_group[] = DEFAULT_TRACING_GROUP; - const char *progname; -const char *opt_tracing_group; +static const char *tracing_group_name = DEFAULT_TRACING_GROUP; static const char *opt_pidfile; static int opt_sig_parent; static int opt_verbose_consumer; @@ -233,6 +230,9 @@ static int app_socket_timeout; /* Set in main() with the current page size. */ long page_size; +/* Application health monitoring */ +struct health_app *health_sessiond; + static void setup_consumerd_path(void) { @@ -322,25 +322,6 @@ int sessiond_check_thread_quit_pipe(int fd, uint32_t events) return 0; } -/* - * Return group ID of the tracing group or -1 if not found. - */ -static gid_t allowed_group(void) -{ - struct group *grp; - - if (opt_tracing_group) { - grp = getgrnam(opt_tracing_group); - } else { - grp = getgrnam(default_tracing_group); - } - if (!grp) { - return -1; - } else { - return grp->gr_gid; - } -} - /* * Init thread quit pipe. * @@ -755,7 +736,7 @@ static void *thread_manage_kernel(void *data) DBG("[thread] Thread manage kernel started"); - health_register(HEALTH_TYPE_KERNEL); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL); /* * This first step of the while is to clean this structure which could free @@ -880,7 +861,7 @@ error_testpoint: WARN("Kernel thread died unexpectedly. " "Kernel tracing can continue but CPU hotplug is disabled."); } - health_unregister(); + health_unregister(health_sessiond); DBG("Kernel thread dying"); return NULL; } @@ -921,7 +902,7 @@ static void *thread_manage_consumer(void *data) DBG("[thread] Manage consumer started"); - health_register(HEALTH_TYPE_CONSUMER); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); health_code_update(); @@ -1200,7 +1181,7 @@ error_poll: health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); return NULL; @@ -1220,7 +1201,7 @@ static void *thread_manage_apps(void *data) rcu_register_thread(); rcu_thread_online(); - health_register(HEALTH_TYPE_APP_MANAGE); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE); if (testpoint(thread_manage_apps)) { goto error_testpoint; @@ -1366,7 +1347,7 @@ error_testpoint: health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Application communication apps thread cleanup complete"); rcu_thread_offline(); rcu_unregister_thread(); @@ -1512,7 +1493,7 @@ static void *thread_dispatch_ust_registration(void *data) .count = 0, }; - health_register(HEALTH_TYPE_APP_REG_DISPATCH); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); health_code_update(); @@ -1723,7 +1704,7 @@ error: health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); return NULL; } @@ -1743,7 +1724,7 @@ static void *thread_registration_apps(void *data) DBG("[thread] Manage application registration started"); - health_register(HEALTH_TYPE_APP_REG); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG); if (testpoint(thread_registration_apps)) { goto error_testpoint; @@ -1923,7 +1904,7 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); - health_unregister(); + health_unregister(health_sessiond); return NULL; } @@ -2124,6 +2105,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) "lttng-consumerd", verbosity, "-k", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); break; case LTTNG_CONSUMER64_UST: @@ -2162,6 +2144,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); if (consumerd64_libdir[0] != '\0') { free(tmpnew); @@ -2207,6 +2190,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); if (consumerd32_libdir[0] != '\0') { free(tmpnew); @@ -2292,21 +2276,6 @@ error: return ret; } -/* - * Compute health status of each consumer. If one of them is zero (bad - * state), we return 0. - */ -static int check_consumer_health(void) -{ - int ret; - - ret = health_check_state(HEALTH_TYPE_CONSUMER); - - DBG3("Health consumer check %d", ret); - - return ret; -} - /* * Setup necessary data for kernel tracer action. */ @@ -3454,8 +3423,8 @@ static void *thread_manage_health(void *data) int sock = -1, new_sock = -1, ret, i, pollfd, err = -1; uint32_t revents, nb_fd; struct lttng_poll_event events; - struct lttcomm_health_msg msg; - struct lttcomm_health_data reply; + struct health_comm_msg msg; + struct health_comm_reply reply; DBG("[thread] Manage health check started"); @@ -3472,6 +3441,27 @@ static void *thread_manage_health(void *data) goto error; } + if (is_root) { + /* lttng health client socket path permissions */ + ret = chown(health_unix_sock_path, 0, + utils_get_group_id(tracing_group_name)); + if (ret < 0) { + ERR("Unable to set group on %s", health_unix_sock_path); + PERROR("chown"); + ret = -1; + goto error; + } + + ret = chmod(health_unix_sock_path, + S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); + if (ret < 0) { + ERR("Unable to set permissions on %s", health_unix_sock_path); + PERROR("chmod"); + ret = -1; + goto error; + } + } + /* * Set the CLOEXEC flag. Return code is useless because either way, the * show must go on. @@ -3562,58 +3552,18 @@ restart: rcu_thread_online(); - switch (msg.component) { - case LTTNG_HEALTH_CMD: - reply.ret_code = health_check_state(HEALTH_TYPE_CMD); - break; - case LTTNG_HEALTH_APP_MANAGE: - reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE); - break; - case LTTNG_HEALTH_APP_REG: - reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG); - break; - case LTTNG_HEALTH_KERNEL: - reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL); - break; - case LTTNG_HEALTH_CONSUMER: - reply.ret_code = check_consumer_health(); - break; - case LTTNG_HEALTH_HT_CLEANUP: - reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP); - break; - case LTTNG_HEALTH_APP_MANAGE_NOTIFY: - reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY); - break; - case LTTNG_HEALTH_APP_REG_DISPATCH: - reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH); - break; - case LTTNG_HEALTH_ALL: - reply.ret_code = - health_check_state(HEALTH_TYPE_APP_MANAGE) && - health_check_state(HEALTH_TYPE_APP_REG) && - health_check_state(HEALTH_TYPE_CMD) && - health_check_state(HEALTH_TYPE_KERNEL) && - check_consumer_health() && - health_check_state(HEALTH_TYPE_HT_CLEANUP) && - health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) && - health_check_state(HEALTH_TYPE_APP_REG_DISPATCH); - break; - default: - reply.ret_code = LTTNG_ERR_UND; - break; - } - - /* - * Flip ret value since 0 is a success and 1 indicates a bad health for - * the client where in the sessiond it is the opposite. Again, this is - * just to make things easier for us poor developer which enjoy a lot - * lazyness. - */ - if (reply.ret_code == 0 || reply.ret_code == 1) { - reply.ret_code = !reply.ret_code; + reply.ret_code = 0; + for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) { + /* + * health_check_state returns 0 if health is + * bad. + */ + if (!health_check_state(health_sessiond, i)) { + reply.ret_code |= 1ULL << i; + } } - DBG2("Health check return value %d", reply.ret_code); + DBG2("Health check return value %" PRIx64, reply.ret_code); ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply)); if (ret < 0) { @@ -3664,7 +3614,7 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); - health_register(HEALTH_TYPE_CMD); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD); if (testpoint(thread_manage_clients)) { goto error_testpoint; @@ -3888,7 +3838,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Client thread dying"); @@ -3985,7 +3935,7 @@ static int parse_args(int argc, char **argv) opt_daemon = 1; break; case 'g': - opt_tracing_group = optarg; + tracing_group_name = optarg; break; case 'h': usage(); @@ -4146,14 +4096,7 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - ret = allowed_group(); - if (ret < 0) { - WARN("No tracing group detected"); - /* Setting gid to 0 if no tracing group is found */ - gid = 0; - } else { - gid = ret; - } + gid = utils_get_group_id(tracing_group_name); /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -4162,8 +4105,12 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure all applications and tracing group can search the run dir */ - ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH); + /* + * Ensure all applications and tracing group can search the run + * dir. Allow everyone to read the directory, since it does not + * buy us anything to hide its content. + */ + ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH); if (ret < 0) { ERR("Unable to set permissions on %s", rundir); PERROR("chmod"); @@ -4177,21 +4124,21 @@ static int set_permissions(char *rundir) } /* kconsumer error socket path */ - ret = chown(kconsumer_data.err_unix_sock_path, 0, gid); + ret = chown(kconsumer_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path); PERROR("chown"); } /* 64-bit ustconsumer error socket path */ - ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid); + ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path); PERROR("chown"); } /* 32-bit ustconsumer compat32 error socket path */ - ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid); + ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path); PERROR("chown"); @@ -4235,7 +4182,7 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, int ret; char path[PATH_MAX]; - switch (consumer_data->type) { + switch (consumer_data->type) { case LTTNG_CONSUMER_KERNEL: snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir); break; @@ -4253,7 +4200,7 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, DBG2("Creating consumer directory: %s", path); - ret = mkdir(path, S_IRWXU); + ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP); if (ret < 0) { if (errno != EEXIST) { PERROR("mkdir"); @@ -4262,6 +4209,14 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, } ret = -1; } + if (is_root) { + ret = chown(path, 0, utils_get_group_id(tracing_group_name)); + if (ret < 0) { + ERR("Unable to set group on %s", path); + PERROR("chown"); + goto error; + } + } /* Create the kconsumerd error unix socket */ consumer_data->err_sock = @@ -4719,7 +4674,11 @@ int main(int argc, char **argv) * Initialize the health check subsystem. This call should set the * appropriate time values. */ - health_init(); + health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES); + if (!health_sessiond) { + PERROR("health_app_create error"); + goto exit_health_sessiond_cleanup; + } /* Create thread to manage the client socket */ ret = pthread_create(&ht_cleanup_thread, NULL, @@ -4862,6 +4821,8 @@ exit_health: goto error; /* join error, exit without cleanup */ } exit_ht_cleanup: + health_app_destroy(health_sessiond); +exit_health_sessiond_cleanup: exit: /* * cleanup() is called when no other thread is running.