Clean-up: relayd: change space to tabs
[lttng-tools.git] / src / bin / lttng-relayd / main.c
index 69f41a31c2a95e58cd345c1e9778f1064adfd62c..10ce291978227acf5d5e35eb5e623cbba80b82fb 100644 (file)
@@ -1,21 +1,11 @@
 /*
- * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
- *                      David Goulet <dgoulet@efficios.com>
- *               2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
- *               2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
+ * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
+ * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
+ * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  *
- * 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.
+ * SPDX-License-Identifier: GPL-2.0-only
  *
- * 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.,
- * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
 #define _LGPL_SOURCE
@@ -165,7 +155,7 @@ 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 +181,7 @@ 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 +225,22 @@ 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);
+                       if (errno != 0 || !isdigit((unsigned char) arg[0])) {
+                               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 +465,56 @@ 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 accommodates 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 +632,10 @@ 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 +667,23 @@ 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
  */
@@ -874,7 +917,11 @@ static int create_named_thread_poll_set(struct lttng_poll_event *events,
        }
 
        ret = fd_tracker_util_poll_create(the_fd_tracker,
-                       name, events, 1, LTTNG_CLOEXEC);
+                       name, events, 1, LTTNG_CLOEXEC);
+       if (ret) {
+               PERROR("Failed to create \"%s\" poll file descriptor", name);
+               goto error;
+       }
 
        /* Add quit pipe */
        ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
@@ -924,6 +971,23 @@ static int close_sock(void *data, int *in_fd)
        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.
  */
@@ -958,7 +1022,11 @@ static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
        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);
+       if (ret) {
+               PERROR("Failed to open \"%s\" relay socket",
+                               formated_name ?: "Unknown");
+               goto error;
+       }
        DBG("Listening on %s socket %d", name, sock->fd);
 
        ret = sock->ops->bind(sock);
@@ -973,15 +1041,38 @@ static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
 
        }
 
+       free(formated_name);
        return sock;
 
 error:
        if (sock) {
                lttcomm_destroy_sock(sock);
        }
+       free(formated_name);
        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
  */
@@ -994,6 +1085,7 @@ static void *relay_thread_listener(void *data)
 
        DBG("[thread] Relay listener started");
 
+       rcu_register_thread();
        health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
 
        health_code_update();
@@ -1080,20 +1172,18 @@ 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");
@@ -1179,6 +1269,7 @@ error_sock_control:
                ERR("Health error occurred in %s", __func__);
        }
        health_unregister(health_relayd);
+       rcu_unregister_thread();
        DBG("Relay listener thread cleanup complete");
        lttng_relay_stop_threads();
        return NULL;
@@ -1541,7 +1632,8 @@ static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
                        group_by_session_path_name =
                                        backward_compat_group_by_session(
                                                        path_name,
-                                                       session->session_name);
+                                                       session->session_name,
+                                                       session->creation_time.value);
                        if (!group_by_session_path_name) {
                                ERR("Failed to apply group by session to stream of session %" PRIu64,
                                                session->id);
@@ -2579,7 +2671,6 @@ 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) {
@@ -2636,6 +2727,7 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                reply_code = LTTNG_ERR_NOMEM;
                goto end;
        }
+       lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
 
        if (msg->override_name_length) {
                const char *name;
@@ -2676,15 +2768,9 @@ 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;
@@ -2743,7 +2829,6 @@ end:
 end_no_reply:
        lttng_trace_chunk_put(chunk);
        lttng_trace_chunk_put(published_chunk);
-       lttng_directory_handle_put(session_output);
        return ret;
 }
 
@@ -2860,7 +2945,7 @@ static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                                new_path);
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ret = -1;
-                       goto end;
+                       goto end_unlock_session;
                }
                session->ongoing_rotation = false;
        }
@@ -2878,7 +2963,7 @@ static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ret = -1;
-                       goto end;
+                       goto end_unlock_session;
                }
        }
        chunk_status = lttng_trace_chunk_set_close_timestamp(
@@ -3677,7 +3762,8 @@ 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);
        }
@@ -4047,11 +4133,6 @@ static int stdio_open(void *data, int *fds)
        return 0;
 }
 
-static int noop_close(void *data, int *fds)
-{
-       return 0;
-}
-
 static int track_stdio(void)
 {
        int fds[2];
@@ -4061,18 +4142,6 @@ static int track_stdio(void)
                        names, 2, stdio_open, NULL);
 }
 
-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);
-}
-
 /*
  * main
  */
@@ -4081,6 +4150,7 @@ 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();
@@ -4171,7 +4241,23 @@ 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;
This page took 0.0292 seconds and 5 git commands to generate.