X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=5cfd57971f7423fe21f69c1d58b2cfafca99a05b;hp=9e2b21b2dd417d7fc217f75de2293ff932f8b43c;hb=ceed52b545103258e84f1c7040700be9dbbbaec6;hpb=531d29f9e514ad5a4880600d5c7e373f2f254ead diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 9e2b21b2d..5cfd57971 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -2,22 +2,21 @@ * 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 -#include #include #include #include @@ -34,26 +33,28 @@ #include #include #include -#include +#include #include #include #include #include +#include #include #include #include +#include #include "lttng-sessiond.h" #include "channel.h" #include "context.h" #include "event.h" -#include "futex.h" #include "kernel.h" #include "modprobe.h" #include "shm.h" #include "ust-ctl.h" #include "utils.h" +#include "fd-limit.h" #define CONSUMERD_FILE "lttng-consumerd" @@ -81,14 +82,10 @@ 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 +179,40 @@ 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) { @@ -301,14 +332,22 @@ static gid_t allowed_group(void) */ static int init_thread_quit_pipe(void) { - int ret; + int ret, i; - ret = pipe2(thread_quit_pipe, O_CLOEXEC); + ret = pipe(thread_quit_pipe); if (ret < 0) { PERROR("thread quit pipe"); goto error; } + for (i = 0; i < 2; i++) { + ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC); + if (ret < 0) { + PERROR("fcntl"); + goto error; + } + } + error: return ret; } @@ -445,7 +484,6 @@ static void cleanup(void) if (ret) { PERROR("close"); } - } } for (i = 0; i < 2; i++) { @@ -591,7 +629,7 @@ 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"); @@ -1085,6 +1123,17 @@ 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) { @@ -1420,6 +1469,17 @@ 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)) { @@ -1433,6 +1493,7 @@ static void *thread_registration_apps(void *data) if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); sock = -1; continue; } @@ -1478,6 +1539,7 @@ error: if (ret) { PERROR("close"); } + lttng_fd_put(LTTNG_FD_APPS, 1); } unlink(apps_unix_sock_path); @@ -2828,6 +2890,36 @@ error: return -ret; } +/* + * Command LTTNG_LIST_TRACEPOINT_FIELDS processed by the client thread. + */ +static ssize_t cmd_list_tracepoint_fields(int domain, + struct lttng_event_field **fields) +{ + int ret; + ssize_t nb_fields = 0; + + switch (domain) { + case LTTNG_DOMAIN_UST: + nb_fields = ust_app_list_event_fields(fields); + if (nb_fields < 0) { + ret = LTTCOMM_UST_LIST_FAIL; + goto error; + } + break; + case LTTNG_DOMAIN_KERNEL: + default: /* fall-through */ + ret = LTTCOMM_UND; + goto error; + } + + return nb_fields; + +error: + /* Return negative value to differentiate return code */ + return -ret; +} + /* * Command LTTNG_START_TRACE processed by the client thread. */ @@ -2842,7 +2934,8 @@ 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; } @@ -2934,7 +3027,7 @@ 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; } @@ -2985,11 +3078,12 @@ error: /* * Command LTTNG_CREATE_SESSION processed by the client thread. */ -static int cmd_create_session(char *name, char *path, struct ucred *creds) +static int cmd_create_session(char *name, char *path, lttng_sock_cred *creds) { int ret; - ret = session_create(name, path, creds->uid, creds->gid); + ret = session_create(name, path, LTTNG_SOCK_GET_UID_CRED(creds), + LTTNG_SOCK_GET_GID_CRED(creds)); if (ret != LTTCOMM_OK) { goto error; } @@ -3274,6 +3368,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch(cmd_ctx->lsm->cmd_type) { case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: + case LTTNG_LIST_TRACEPOINT_FIELDS: case LTTNG_LIST_DOMAINS: case LTTNG_LIST_CHANNELS: case LTTNG_LIST_EVENTS: @@ -3293,13 +3388,18 @@ static int process_client_msg(struct command_ctx *cmd_ctx) case LTTNG_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: + case LTTNG_LIST_TRACEPOINT_FIELDS: need_tracing_session = 0; 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; @@ -3337,6 +3437,12 @@ 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) { @@ -3357,13 +3463,21 @@ 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, @@ -3387,6 +3501,7 @@ 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); } @@ -3401,7 +3516,9 @@ 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); } @@ -3413,13 +3530,33 @@ 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. */ if (need_tracing_session) { if (!session_access_ok(cmd_ctx->session, - cmd_ctx->creds.uid, cmd_ctx->creds.gid)) { + LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), + LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) { ret = LTTCOMM_EPERM; goto error; } @@ -3508,6 +3645,37 @@ skip_domain: ret = LTTCOMM_OK; break; } + case LTTNG_LIST_TRACEPOINT_FIELDS: + { + struct lttng_event_field *fields; + ssize_t nb_fields; + + nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type, &fields); + if (nb_fields < 0) { + ret = -nb_fields; + goto error; + } + + /* + * Setup lttng message with payload size set to the event list size in + * bytes and then copy list into the llm payload. + */ + ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event_field) * nb_fields); + if (ret < 0) { + free(fields); + goto setup_error; + } + + /* Copy event list into message payload */ + memcpy(cmd_ctx->llm->payload, fields, + sizeof(struct lttng_event_field) * nb_fields); + + free(fields); + + ret = LTTCOMM_OK; + break; + } + case LTTNG_START_TRACE: { ret = cmd_start_trace(cmd_ctx->session); @@ -3528,6 +3696,11 @@ 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: @@ -3557,7 +3730,7 @@ 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, @@ -3612,7 +3785,9 @@ skip_domain: unsigned int nr_sessions; session_lock_list(); - nr_sessions = lttng_sessions_count(cmd_ctx->creds.uid, cmd_ctx->creds.gid); + nr_sessions = lttng_sessions_count( + LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), + LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions); if (ret < 0) { @@ -3622,7 +3797,8 @@ skip_domain: /* Filled the session array */ list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload), - cmd_ctx->creds.uid, cmd_ctx->creds.gid); + LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), + LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); session_unlock_list(); @@ -3659,6 +3835,9 @@ setup_error: if (cmd_ctx->session) { session_unlock(cmd_ctx->session); } + if (need_tracing_session) { + session_unlock_list(); + } init_setup_error: return ret; } @@ -3788,7 +3967,7 @@ static void *thread_manage_clients(void *data) PERROR("close"); } sock = -1; - free(cmd_ctx); + clean_command_ctx(&cmd_ctx); continue; } @@ -3975,11 +4154,11 @@ 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; @@ -4081,13 +4260,15 @@ 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) { @@ -4142,7 +4323,24 @@ end: */ static int create_kernel_poll_pipe(void) { - return pipe2(kernel_poll_pipe, O_CLOEXEC); + int ret, i; + + ret = pipe(kernel_poll_pipe); + if (ret < 0) { + PERROR("kernel poll pipe"); + goto error; + } + + for (i = 0; i < 2; i++) { + ret = fcntl(kernel_poll_pipe[i], F_SETFD, FD_CLOEXEC); + if (ret < 0) { + PERROR("fcntl kernel_poll_pipe"); + goto error; + } + } + +error: + return ret; } /* @@ -4151,7 +4349,24 @@ static int create_kernel_poll_pipe(void) */ static int create_apps_cmd_pipe(void) { - return pipe2(apps_cmd_pipe, O_CLOEXEC); + int ret, i; + + ret = pipe(apps_cmd_pipe); + if (ret < 0) { + PERROR("apps cmd pipe"); + goto error; + } + + for (i = 0; i < 2; i++) { + ret = fcntl(apps_cmd_pipe[i], F_SETFD, FD_CLOEXEC); + if (ret < 0) { + PERROR("fcntl apps_cmd_pipe"); + goto error; + } + } + +error: + return ret; } /* @@ -4208,10 +4423,11 @@ 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 */ @@ -4331,11 +4547,6 @@ 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 */ @@ -4346,11 +4557,31 @@ 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 */ @@ -4433,6 +4664,10 @@ 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); @@ -4471,6 +4706,12 @@ 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" */ /* @@ -4493,6 +4734,8 @@ 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) { @@ -4536,9 +4779,6 @@ 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