X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=236341d2737fc21e06d74509c4e13a7d945455d2;hb=bcccc1bfd7e1586259ff1c516a458e04321d0e78;hp=12322837e7f681038e0ed6fe54d2ac72e035668d;hpb=ca2eb7f43cf00d12f563905d741a6789c3d130ee;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 12322837e..236341d27 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -2,18 +2,18 @@ * Copyright (C) 2011 - David Goulet * Mathieu Desnoyers * - * This program is free software; you can redistribute it and/or modify it - * under the terms of the GNU General Public License as published by the Free - * Software Foundation; only version 2 of the License. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License, version 2 only, + * as published by the Free Software Foundation. * - * This program is distributed in the hope that it will be useful, but WITHOUT - * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for - * more details. + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. * - * You should have received a copy of the GNU General Public License along with - * this program; if not, write to the Free Software Foundation, Inc., 59 Temple - * Place - Suite 330, Boston, MA 02111-1307, USA. + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #define _GNU_SOURCE @@ -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" @@ -81,14 +83,10 @@ const char default_tracing_group[] = DEFAULT_TRACING_GROUP; const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR; const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE; -/* Variables */ -int opt_verbose; /* Not static for lttngerr.h */ -int opt_verbose_consumer; /* Not static for lttngerr.h */ -int opt_quiet; /* Not static for lttngerr.h */ - const char *progname; const char *opt_tracing_group; static int opt_sig_parent; +static int opt_verbose_consumer; static int opt_daemon; static int opt_no_kernel; static int is_root; /* Set to 1 if the daemon is running as root */ @@ -182,6 +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) { @@ -453,7 +485,6 @@ static void cleanup(void) if (ret) { PERROR("close"); } - } } for (i = 0; i < 2; i++) { @@ -599,7 +630,7 @@ static int send_kconsumer_session_streams(struct consumer_data *consumer_data, lkm.u.channel.channel_key = session->metadata->fd; lkm.u.channel.max_sb_size = session->metadata->conf->attr.subbuf_size; lkm.u.channel.mmap_len = 0; /* for kernel */ - DBG("Sending metadata channel %d to consumer", lkm.u.stream.stream_key); + DBG("Sending metadata channel %d to consumer", lkm.u.channel.channel_key); ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm)); if (ret < 0) { PERROR("send consumer channel"); @@ -1093,6 +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) { @@ -1245,16 +1287,28 @@ static void *thread_manage_apps(void *data) * At this point, we know that a registered application made * the event at poll_wait. */ - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - /* Removing from the poll set */ - ret = lttng_poll_del(&events, pollfd); - if (ret < 0) { - goto error; - } - /* Socket closed on remote end. */ - ust_app_unregister(pollfd); - break; + if (revents & (LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ssize_t readlen; + char dummy; + + do { + readlen = recv(pollfd, &dummy, 1, MSG_PEEK); + } while (readlen == -1 && errno == EINTR); + + /* Peer has performed an orderly shutdown */ + if (readlen == 0) { + /* Removing from the poll set */ + ret = lttng_poll_del(&events, pollfd); + if (ret < 0) { + goto error; + } + + /* Socket closed on remote end. */ + ust_app_unregister(pollfd); + break; + + } } } } @@ -1428,6 +1482,17 @@ 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)) { @@ -1441,6 +1506,7 @@ static void *thread_registration_apps(void *data) if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); sock = -1; continue; } @@ -1486,6 +1552,7 @@ error: if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); } unlink(apps_unix_sock_path); @@ -2850,7 +2917,8 @@ static int cmd_start_trace(struct ltt_session *session) usess = session->ust_session; if (session->enabled) { - ret = LTTCOMM_UST_START_FAIL; + /* Already started. */ + ret = LTTCOMM_TRACE_ALREADY_STARTED; goto error; } @@ -2942,7 +3010,7 @@ static int cmd_stop_trace(struct ltt_session *session) usess = session->ust_session; if (!session->enabled) { - ret = LTTCOMM_UST_STOP_FAIL; + ret = LTTCOMM_TRACE_ALREADY_STOPPED; goto error; } @@ -3170,6 +3238,9 @@ 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) { @@ -3177,6 +3248,9 @@ 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; @@ -3194,6 +3268,8 @@ 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; @@ -3306,9 +3382,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx) break; default: DBG("Getting session %s by name", cmd_ctx->lsm->session.name); + /* + * We keep the session list lock across _all_ commands + * for now, because the per-session lock does not + * handle teardown properly. + */ session_lock_list(); cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name); - session_unlock_list(); if (cmd_ctx->session == NULL) { if (cmd_ctx->lsm->session.name != NULL) { ret = LTTCOMM_SESS_NOT_FOUND; @@ -3346,6 +3426,12 @@ 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) { @@ -3366,13 +3452,21 @@ 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, @@ -3396,6 +3490,7 @@ 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); } @@ -3410,7 +3505,9 @@ 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); } @@ -3422,6 +3519,25 @@ 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. @@ -3538,6 +3654,11 @@ skip_domain: { ret = cmd_destroy_session(cmd_ctx->session, cmd_ctx->lsm->session.name); + /* + * Set session to NULL so we do not unlock it after + * free. + */ + cmd_ctx->session = NULL; break; } case LTTNG_LIST_DOMAINS: @@ -3567,7 +3688,7 @@ 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, @@ -3672,6 +3793,9 @@ setup_error: if (cmd_ctx->session) { session_unlock(cmd_ctx->session); } + if (need_tracing_session) { + session_unlock_list(); + } init_setup_error: return ret; } @@ -3801,7 +3925,7 @@ static void *thread_manage_clients(void *data) PERROR("close"); } sock = -1; - free(cmd_ctx); + clean_command_ctx(&cmd_ctx); continue; } @@ -3988,11 +4112,11 @@ static int parse_args(int argc, char **argv) opt_no_kernel = 1; break; case 'q': - opt_quiet = 1; + lttng_opt_quiet = 1; break; case 'v': /* Verbose level can increase using multiple -v */ - opt_verbose += 1; + lttng_opt_verbose += 1; break; case 'Z': opt_verbose_consumer += 1; @@ -4094,13 +4218,15 @@ 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) { @@ -4255,10 +4381,11 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, ret = mkdir(path, S_IRWXU); if (ret < 0) { if (errno != EEXIST) { + PERROR("mkdir"); ERR("Failed to create %s", path); goto error; } - ret = 0; + ret = -1; } /* Create the kconsumerd error unix socket */ @@ -4378,11 +4505,6 @@ 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 */ @@ -4393,11 +4515,31 @@ 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 */ @@ -4480,6 +4622,10 @@ 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); @@ -4518,6 +4664,12 @@ int main(int argc, char **argv) goto error; } + /* + * Init UST app hash table. Alloc hash table before this point since + * cleanup() can get called after that point. + */ + ust_app_ht_alloc(); + /* After this point, we can safely call cleanup() with "goto exit" */ /* @@ -4540,6 +4692,8 @@ 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) { @@ -4583,9 +4737,6 @@ int main(int argc, char **argv) /* Init UST command queue. */ cds_wfq_init(&ust_cmd_queue.queue); - /* Init UST app hash table */ - ust_app_ht_alloc(); - /* * Get session list pointer. This pointer MUST NOT be free(). This list is * statically declared in session.c