relayd: track relayd control connection sockets
[lttng-tools.git] / src / bin / lttng-relayd / main.c
index 6234e3e170d92515d1255a15087906531369083a..00292a02d6545db9d5b94cb4e08939e9b739a723 100644 (file)
@@ -165,7 +165,7 @@ static uint64_t last_relay_stream_id;
 static struct relay_conn_queue relay_conn_queue;
 
 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
-static unsigned int lttng_opt_fd_cap;
+static unsigned int lttng_opt_fd_pool_size = -1;
 
 /* Global relay stream hash table. */
 struct lttng_ht *relay_streams_ht;
@@ -191,7 +191,7 @@ static struct option long_options[] = {
        { "daemonize", 0, 0, 'd', },
        { "background", 0, 0, 'b', },
        { "group", 1, 0, 'g', },
-       { "fd-cap", 1, 0, '\0', },
+       { "fd-pool-size", 1, 0, '\0', },
        { "help", 0, 0, 'h', },
        { "output", 1, 0, 'o', },
        { "verbose", 0, 0, 'v', },
@@ -235,29 +235,22 @@ static int set_option(int opt, const char *arg, const char *optname)
 
        switch (opt) {
        case 0:
-               if (!strcmp(optname, "fd-cap")) {
+               if (!strcmp(optname, "fd-pool-size")) {
                        unsigned long v;
 
                        errno = 0;
                        v = strtoul(arg, NULL, 0);
                        if (errno != 0 || !isdigit(arg[0])) {
-                               ERR("Wrong value in --fd-cap parameter: %s",
-                                               arg);
+                               ERR("Wrong value in --fd-pool-size parameter: %s", arg);
                                ret = -1;
                                goto end;
                        }
-                       if (v < DEFAULT_RELAYD_MINIMAL_FD_CAP) {
-                               ERR("File descriptor cap must be set to at least %d",
-                                               DEFAULT_RELAYD_MINIMAL_FD_CAP);
-                       }
                        if (v >= UINT_MAX) {
-                               ERR("File descriptor cap overflow in --fd-cap parameter: %s",
-                                               arg);
+                               ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
                                ret = -1;
                                goto end;
                        }
-                       lttng_opt_fd_cap = (unsigned int) v;
-                       DBG3("File descriptor cap set to %u", lttng_opt_fd_cap);
+                       lttng_opt_fd_pool_size = (unsigned int) v;
                } else {
                        fprintf(stderr, "unknown option %s", optname);
                        if (arg) {
@@ -482,6 +475,56 @@ static int parse_env_options(void)
        return ret;
 }
 
+static int set_fd_pool_size(void)
+{
+       int ret = 0;
+       struct rlimit rlimit;
+
+       ret = getrlimit(RLIMIT_NOFILE, &rlimit);
+       if (ret) {
+               PERROR("Failed to get file descriptor limit");
+               ret = -1;
+               goto end;
+       }
+
+       DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
+                       (uint64_t) rlimit.rlim_cur,
+                       (uint64_t) rlimit.rlim_max);
+       if (lttng_opt_fd_pool_size == -1) {
+               /* Use default value (soft limit - reserve). */
+               if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+                       ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
+                                       (uint64_t) rlimit.rlim_cur, DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+                       ret = -1;
+                       goto end;
+               }
+               lttng_opt_fd_pool_size = rlimit.rlim_cur -
+                               DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+               goto end;
+       }
+
+       if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+               ERR("File descriptor pool size must be set to at least %d",
+                               DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+               ret = -1;
+               goto end;
+       }
+
+       if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
+               ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64 ").",
+                               lttng_opt_fd_pool_size, (uint64_t) rlimit.rlim_cur);
+               ret = -1;
+               goto end;
+       }
+
+       DBG("File descriptor pool size argument (%u) adjusted to %u to accomodate transient fd uses",
+                       lttng_opt_fd_pool_size,
+                       lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
+       lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+end:
+       return ret;
+}
+
 static int set_options(int argc, char **argv)
 {
        int c, ret = 0, option_index = 0, retval = 0;
@@ -599,17 +642,10 @@ static int set_options(int argc, char **argv)
                        goto exit;
                }
        }
-       if (lttng_opt_fd_cap == 0) {
-               int ret;
-               struct rlimit rlimit;
-
-               ret = getrlimit(RLIMIT_NOFILE, &rlimit);
-               if (ret) {
-                       PERROR("Failed to get file descriptor limit");
-                       retval = -1;
-               }
-
-               lttng_opt_fd_cap = rlimit.rlim_cur;
+       ret = set_fd_pool_size();
+       if (ret) {
+               retval = -1;
+               goto exit;
        }
 
        if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
@@ -641,6 +677,23 @@ static void print_global_objects(void)
        print_sessions();
 }
 
+static int noop_close(void *data, int *fds)
+{
+       return 0;
+}
+
+static void untrack_stdio(void)
+{
+       int fds[] = { fileno(stdout), fileno(stderr) };
+
+       /*
+        * noop_close is used since we don't really want to close
+        * the stdio output fds; we merely want to stop tracking them.
+        */
+       (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+                       fds, 2, noop_close, NULL);
+}
+
 /*
  * Cleanup the daemon
  */
@@ -664,7 +717,10 @@ static void relayd_cleanup(void)
                health_app_destroy(health_relayd);
        }
        /* Close thread quit pipes */
-       utils_close_pipe(health_quit_pipe);
+       if (health_quit_pipe[0] != -1) {
+               (void) fd_tracker_util_pipe_close(
+                               the_fd_tracker, health_quit_pipe);
+       }
        if (thread_quit_pipe[0] != -1) {
                (void) fd_tracker_util_pipe_close(
                                the_fd_tracker, thread_quit_pipe);
@@ -674,6 +730,11 @@ static void relayd_cleanup(void)
                                sessiond_trace_chunk_registry);
        }
        if (the_fd_tracker) {
+               untrack_stdio();
+               /*
+                * fd_tracker_destroy() will log the contents of the fd-tracker
+                * if a leak is detected.
+                */
                fd_tracker_destroy(the_fd_tracker);
        }
 
@@ -684,7 +745,6 @@ static void relayd_cleanup(void)
        if (tracing_group_name_override) {
                free((void *) tracing_group_name);
        }
-       fd_tracker_log(the_fd_tracker);
 }
 
 /*
@@ -842,10 +902,22 @@ static int init_thread_quit_pipe(void)
                        the_fd_tracker, "Quit pipe", thread_quit_pipe);
 }
 
+/*
+ * Init health quit pipe.
+ *
+ * Return -1 on error or 0 if all pipes are created.
+ */
+static int init_health_quit_pipe(void)
+{
+       return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
+                       "Health quit pipe", health_quit_pipe);
+}
+
 /*
  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
  */
-static int create_thread_poll_set(struct lttng_poll_event *events, int size)
+static int create_named_thread_poll_set(struct lttng_poll_event *events,
+               int size, const char *name)
 {
        int ret;
 
@@ -854,10 +926,8 @@ static int create_thread_poll_set(struct lttng_poll_event *events, int size)
                goto error;
        }
 
-       ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
-       if (ret < 0) {
-               goto error;
-       }
+       ret = fd_tracker_util_poll_create(the_fd_tracker,
+                       name, events, 1, LTTNG_CLOEXEC);
 
        /* Add quit pipe */
        ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
@@ -885,13 +955,55 @@ static int check_thread_quit_pipe(int fd, uint32_t events)
        return 0;
 }
 
+static int create_sock(void *data, int *out_fd)
+{
+       int ret;
+       struct lttcomm_sock *sock = data;
+
+       ret = lttcomm_create_sock(sock);
+       if (ret < 0) {
+               goto end;
+       }
+
+       *out_fd = sock->fd;
+end:
+       return ret;
+}
+
+static int close_sock(void *data, int *in_fd)
+{
+       struct lttcomm_sock *sock = data;
+
+       return sock->ops->close(sock);
+}
+
+static int accept_sock(void *data, int *out_fd)
+{
+       int ret = 0;
+       /* Socks is an array of in_sock, out_sock. */
+       struct lttcomm_sock **socks = data;
+       struct lttcomm_sock *in_sock = socks[0];
+
+        socks[1] = in_sock->ops->accept(in_sock);
+       if (!socks[1]) {
+               ret = -1;
+               goto end;
+       }
+       *out_fd = socks[1]->fd;
+end:
+       return ret;
+}
+
 /*
  * Create and init socket from uri.
  */
-static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
+static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
+               const char *name)
 {
-       int ret;
+       int ret, sock_fd;
        struct lttcomm_sock *sock = NULL;
+       char uri_str[PATH_MAX];
+       char *formated_name = NULL;
 
        sock = lttcomm_alloc_sock_from_uri(uri);
        if (sock == NULL) {
@@ -899,11 +1011,25 @@ static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
                goto error;
        }
 
-       ret = lttcomm_create_sock(sock);
-       if (ret < 0) {
-               goto error;
+       /*
+        * Don't fail to create the socket if the name can't be built as it is
+        * only used for debugging purposes.
+        */
+       ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
+       uri_str[sizeof(uri_str) - 1] = '\0';
+       if (ret >= 0) {
+               ret = asprintf(&formated_name, "%s socket @ %s", name,
+                               uri_str);
+               if (ret < 0) {
+                       formated_name = NULL;
+               }
        }
-       DBG("Listening on sock %d", sock->fd);
+
+       ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
+                       (const char **) (formated_name ? &formated_name : NULL),
+                       1, create_sock, sock);
+       free(formated_name);
+       DBG("Listening on %s socket %d", name, sock->fd);
 
        ret = sock->ops->bind(sock);
        if (ret < 0) {
@@ -926,6 +1052,27 @@ error:
        return NULL;
 }
 
+static
+struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
+               const char *name)
+{
+       int out_fd, ret;
+       struct lttcomm_sock *socks[2] = { listening_sock, NULL };
+       struct lttcomm_sock *new_sock = NULL;
+
+        ret = fd_tracker_open_unsuspendable_fd(
+                       the_fd_tracker, &out_fd,
+                       (const char **) &name,
+                       1, accept_sock, &socks);
+       if (ret) {
+               goto end;
+       }
+       new_sock = socks[1];
+       DBG("%s accepted, socket %d", name, new_sock->fd);
+end:
+       return new_sock;
+}
+
 /*
  * This thread manages the listening for new connections on the network
  */
@@ -942,12 +1089,12 @@ static void *relay_thread_listener(void *data)
 
        health_code_update();
 
-       control_sock = relay_socket_create(control_uri);
+       control_sock = relay_socket_create(control_uri, "Control listener");
        if (!control_sock) {
                goto error_sock_control;
        }
 
-       data_sock = relay_socket_create(data_uri);
+       data_sock = relay_socket_create(data_uri, "Data listener");
        if (!data_sock) {
                goto error_sock_relay;
        }
@@ -956,7 +1103,7 @@ static void *relay_thread_listener(void *data)
         * Pass 3 as size here for the thread quit pipe, control and
         * data socket.
         */
-       ret = create_thread_poll_set(&events, 3);
+       ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
        if (ret < 0) {
                goto error_create_poll;
        }
@@ -1024,20 +1171,18 @@ restart:
                                 */
                                int val = 1;
                                struct relay_connection *new_conn;
-                               struct lttcomm_sock *newsock;
+                               struct lttcomm_sock *newsock = NULL;
                                enum connection_type type;
 
                                if (pollfd == data_sock->fd) {
                                        type = RELAY_DATA;
-                                       newsock = data_sock->ops->accept(data_sock);
-                                       DBG("Relay data connection accepted, socket %d",
-                                                       newsock->fd);
+                                       newsock = accept_relayd_sock(data_sock,
+                                                       "Data socket to relayd");
                                } else {
                                        assert(pollfd == control_sock->fd);
                                        type = RELAY_CONTROL;
-                                       newsock = control_sock->ops->accept(control_sock);
-                                       DBG("Relay control connection accepted, socket %d",
-                                                       newsock->fd);
+                                       newsock = accept_relayd_sock(control_sock,
+                                                       "Control socket to relayd");
                                }
                                if (!newsock) {
                                        PERROR("accepting sock");
@@ -1090,21 +1235,31 @@ exit:
 error:
 error_poll_add:
 error_testpoint:
-       lttng_poll_clean(&events);
+       (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
 error_create_poll:
        if (data_sock->fd >= 0) {
-               ret = data_sock->ops->close(data_sock);
+               int data_sock_fd = data_sock->fd;
+
+               ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+                               &data_sock_fd, 1, close_sock,
+                               data_sock);
                if (ret) {
-                       PERROR("close");
+                       PERROR("Failed to close the data listener socket file descriptor");
                }
+               data_sock->fd = -1;
        }
        lttcomm_destroy_sock(data_sock);
 error_sock_relay:
        if (control_sock->fd >= 0) {
-               ret = control_sock->ops->close(control_sock);
+               int control_sock_fd = control_sock->fd;
+
+               ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+                               &control_sock_fd, 1, close_sock,
+                               control_sock);
                if (ret) {
-                       PERROR("close");
+                       PERROR("Failed to close the control listener socket file descriptor");
                }
+               control_sock->fd = -1;
        }
        lttcomm_destroy_sock(control_sock);
 error_sock_control:
@@ -2513,7 +2668,6 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
        enum lttng_error_code reply_code = LTTNG_OK;
        enum lttng_trace_chunk_status chunk_status;
-       struct lttng_directory_handle *session_output = NULL;
        const char *new_path;
 
        if (!session || !conn->version_check_done) {
@@ -2610,15 +2764,9 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                goto end;
        }
 
-       session_output = session_create_output_directory_handle(
-                       conn->session);
-       if (!session_output) {
-               reply_code = LTTNG_ERR_CREATE_DIR_FAIL;
-               goto end;
-       }
-       chunk_status = lttng_trace_chunk_set_as_owner(chunk, session_output);
-       lttng_directory_handle_put(session_output);
-       session_output = NULL;
+       assert(conn->session->output_directory);
+       chunk_status = lttng_trace_chunk_set_as_owner(chunk,
+                       conn->session->output_directory);
        if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                reply_code = LTTNG_ERR_UNK;
                ret = -1;
@@ -2677,7 +2825,6 @@ end:
 end_no_reply:
        lttng_trace_chunk_put(chunk);
        lttng_trace_chunk_put(published_chunk);
-       lttng_directory_handle_put(session_output);
        return ret;
 }
 
@@ -3611,7 +3758,8 @@ static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollf
 
        (void) lttng_poll_del(events, pollfd);
 
-       ret = close(pollfd);
+       ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
+                       fd_tracker_util_close_fd, NULL);
        if (ret < 0) {
                ERR("Closing pollfd %d", pollfd);
        }
@@ -3673,7 +3821,7 @@ static void *relay_thread_worker(void *data)
                goto relay_connections_ht_error;
        }
 
-       ret = create_thread_poll_set(&events, 2);
+       ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
        if (ret < 0) {
                goto error_poll_create;
        }
@@ -3942,12 +4090,13 @@ error:
        }
        rcu_read_unlock();
 
-       lttng_poll_clean(&events);
+       (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
 error_poll_create:
        lttng_ht_destroy(relay_connections_ht);
 relay_connections_ht_error:
        /* Close relay conn pipes */
-       utils_close_pipe(relay_conn_pipe);
+       (void) fd_tracker_util_pipe_close(the_fd_tracker,
+                       relay_conn_pipe);
        if (err) {
                DBG("Thread exited with error");
        }
@@ -3969,11 +4118,24 @@ error_testpoint:
  */
 static int create_relay_conn_pipe(void)
 {
-       int ret;
+       return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
+                       "Relayd connection pipe", relay_conn_pipe);
+}
+
+static int stdio_open(void *data, int *fds)
+{
+       fds[0] = fileno(stdout);
+       fds[1] = fileno(stderr);
+       return 0;
+}
 
-       ret = utils_create_pipe_cloexec(relay_conn_pipe);
+static int track_stdio(void)
+{
+       int fds[2];
+       const char *names[] = { "stdout", "stderr" };
 
-       return ret;
+       return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
+                       names, 2, stdio_open, NULL);
 }
 
 /*
@@ -3984,6 +4146,7 @@ int main(int argc, char **argv)
        bool thread_is_rcu_registered = false;
        int ret = 0, retval = 0;
        void *status;
+       char *unlinked_file_directory_path = NULL, *output_path = NULL;
 
        /* Parse environment variables */
        ret = parse_env_options();
@@ -4074,12 +4237,34 @@ int main(int argc, char **argv)
        rcu_register_thread();
        thread_is_rcu_registered = true;
 
-       the_fd_tracker = fd_tracker_create(lttng_opt_fd_cap);
+       output_path = create_output_path("");
+       if (!output_path) {
+               ERR("Failed to get output path");
+               retval = -1;
+               goto exit_options;
+       }
+       ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
+                       DEFAULT_UNLINKED_FILES_DIRECTORY);
+       free(output_path);
+       if (ret < 0) {
+               ERR("Failed to format unlinked file directory path");
+               retval = -1;
+               goto exit_options;
+       }
+       the_fd_tracker = fd_tracker_create(
+                       unlinked_file_directory_path, lttng_opt_fd_pool_size);
+       free(unlinked_file_directory_path);
        if (!the_fd_tracker) {
                retval = -1;
                goto exit_options;
        }
 
+       ret = track_stdio();
+       if (ret) {
+               retval = -1;
+               goto exit_options;
+       }
+
        /* Initialize thread health monitoring */
        health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
        if (!health_relayd) {
@@ -4128,7 +4313,7 @@ int main(int argc, char **argv)
                goto exit_options;
        }
 
-       ret = utils_create_pipe(health_quit_pipe);
+       ret = init_health_quit_pipe();
        if (ret) {
                retval = -1;
                goto exit_options;
This page took 0.031678 seconds and 5 git commands to generate.