relayd: track stdio output fds
[lttng-tools.git] / src / bin / lttng-relayd / main.c
index a04303e4f5814a47a1b7bce3eec973179f6d3866..fbb5f65628f147ebd0861b885abae090b32f52a5 100644 (file)
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <sys/resource.h>
 #include <inttypes.h>
 #include <urcu/futex.h>
 #include <urcu/uatomic.h>
 #include <unistd.h>
 #include <fcntl.h>
+#include <ctype.h>
 
 #include <lttng/lttng.h>
 #include <common/common.h>
@@ -58,6 +60,7 @@
 #include <common/config/session-config.h>
 #include <common/dynamic-buffer.h>
 #include <common/buffer-view.h>
+#include <common/fd-tracker/utils.h>
 #include <urcu/rculist.h>
 
 #include "cmd.h"
@@ -155,6 +158,9 @@ 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;
+
 /* Global relay stream hash table. */
 struct lttng_ht *relay_streams_ht;
 
@@ -167,6 +173,9 @@ struct lttng_ht *sessions_ht;
 /* Relayd health monitoring */
 struct health_app *health_relayd;
 
+/* Global fd tracker. */
+struct fd_tracker *the_fd_tracker;
+
 static struct option long_options[] = {
        { "control-port", 1, 0, 'C', },
        { "data-port", 1, 0, 'D', },
@@ -174,6 +183,7 @@ static struct option long_options[] = {
        { "daemonize", 0, 0, 'd', },
        { "background", 0, 0, 'b', },
        { "group", 1, 0, 'g', },
+       { "fd-cap", 1, 0, '\0', },
        { "help", 0, 0, 'h', },
        { "output", 1, 0, 'o', },
        { "verbose", 0, 0, 'v', },
@@ -196,9 +206,33 @@ static int set_option(int opt, const char *arg, const char *optname)
 
        switch (opt) {
        case 0:
-               fprintf(stderr, "option %s", optname);
-               if (arg) {
-                       fprintf(stderr, " with arg %s\n", arg);
+               if (!strcmp(optname, "fd-cap")) {
+                       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);
+                               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);
+                               ret = -1;
+                               goto end;
+                       }
+                       lttng_opt_fd_cap = (unsigned int) v;
+                       DBG3("File descriptor cap set to %u", lttng_opt_fd_cap);
+
+               } else {
+                       fprintf(stderr, "unknown option %s", optname);
+                       if (arg) {
+                               fprintf(stderr, " with arg %s\n", arg);
+                       }
                }
                break;
        case 'C':
@@ -486,6 +520,18 @@ 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;
+       }
 
 exit:
        free(optstring);
@@ -494,13 +540,9 @@ exit:
 
 static void print_global_objects(void)
 {
-       rcu_register_thread();
-
        print_viewer_streams();
        print_relay_streams();
        print_sessions();
-
-       rcu_unregister_thread();
 }
 
 /*
@@ -523,7 +565,7 @@ static void relayd_cleanup(void)
        free(opt_output_path);
 
        /* Close thread quit pipes */
-       utils_close_pipe(thread_quit_pipe);
+       (void) fd_tracker_util_pipe_close(the_fd_tracker, thread_quit_pipe);
 
        uri_free(control_uri);
        uri_free(data_uri);
@@ -685,17 +727,26 @@ void lttng_relay_notify_ready(void)
  */
 static int init_thread_quit_pipe(void)
 {
-       int ret;
-
-       ret = utils_create_pipe_cloexec(thread_quit_pipe);
+       return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
+                       "Quit pipe", thread_quit_pipe);
+}
 
-       return ret;
+/*
+ * 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;
 
@@ -704,10 +755,8 @@ 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);
@@ -806,7 +855,7 @@ 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;
        }
@@ -948,7 +997,7 @@ 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);
@@ -1198,6 +1247,7 @@ static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
        uint64_t stream_handle = -1ULL;
        char *path_name = NULL, *channel_name = NULL;
        uint64_t tracefile_size = 0, tracefile_count = 0;
+       struct relay_stream_chunk_id stream_chunk_id = { 0 };
 
        if (!session || !conn->version_check_done) {
                ERR("Trying to add a stream before version check");
@@ -1237,7 +1287,8 @@ static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
 
        /* We pass ownership of path_name and channel_name. */
        stream = stream_create(trace, stream_handle, path_name,
-                       channel_name, tracefile_size, tracefile_count);
+               channel_name, tracefile_size, tracefile_count,
+               &stream_chunk_id);
        path_name = NULL;
        channel_name = NULL;
 
@@ -2483,7 +2534,8 @@ static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr
                goto end_stream_unlock;
        }
 
-       stream->chunk_id = stream_info.new_chunk_id;
+       assert(stream->current_chunk_id.is_set);
+       stream->current_chunk_id.value = stream_info.new_chunk_id;
 
        if (stream->is_metadata) {
                /*
@@ -2817,7 +2869,8 @@ int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr,
 
        chunk_id = be64toh(msg.chunk_id);
 
-       DBG("Evaluating rotate pending for chunk id %" PRIu64, chunk_id);
+       DBG("Evaluating rotate pending for session \"%s\" and  chunk id %" PRIu64,
+                       session->session_name, chunk_id);
 
        /*
         * Iterate over all the streams in the session and check if they are
@@ -2839,7 +2892,7 @@ int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr,
                        rotate_pending = true;
                        DBG("Stream %" PRIu64 " is still rotating",
                                        stream->stream_handle);
-               } else if (stream->chunk_id < chunk_id) {
+               } else if (stream->current_chunk_id.value < chunk_id) {
                        /*
                         * Stream closed on the consumer but still active on the
                         * relay.
@@ -3596,7 +3649,7 @@ 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;
        }
@@ -3868,12 +3921,13 @@ 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");
        }
@@ -3895,11 +3949,45 @@ 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);
+}
 
-       ret = utils_create_pipe_cloexec(relay_conn_pipe);
+static
+int stdio_open(void *data, int *fds)
+{
+       fds[0] = fileno(stdout);
+       fds[1] = fileno(stderr);
+       return 0;
+}
 
-       return ret;
+static
+int noop_close(void *data, int *fds)
+{
+       return 0;
+}
+
+static
+int track_stdio(void)
+{
+       int fds[2];
+       const char *names[] = { "stdout", "stderr" };
+
+       return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
+                       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);
 }
 
 /*
@@ -3922,6 +4010,11 @@ int main(int argc, char **argv)
                goto exit_options;
        }
 
+       ret = fclose(stdin);
+       if (ret) {
+               PERROR("Failed to close stdin");
+               goto exit_options;
+       }
        /* Try to create directory if -o, --output is specified. */
        if (opt_output_path) {
                if (*opt_output_path != '/') {
@@ -3941,23 +4034,32 @@ int main(int argc, char **argv)
 
        /* Daemonize */
        if (opt_daemon || opt_background) {
-               int i;
-
                ret = lttng_daemonize(&child_ppid, &recv_child_signal,
                        !opt_background);
                if (ret < 0) {
                        retval = -1;
                        goto exit_options;
                }
+       }
 
-               /*
-                * 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);
-               }
+       /*
+        * The RCU thread registration (and use, through the fd-tracker's
+        * creation) is done after the daemonization to allow us to not
+        * deal with liburcu's fork() management as the call RCU needs to
+        * be restored.
+        */
+       rcu_register_thread();
+
+       the_fd_tracker = fd_tracker_create(lttng_opt_fd_cap);
+       if (!the_fd_tracker) {
+               retval = -1;
+               goto exit_options;
+       }
+
+       ret = track_stdio();
+       if (ret) {
+               retval = -1;
+               goto exit_options;
        }
 
        /* Initialize thread health monitoring */
@@ -4008,7 +4110,7 @@ int main(int argc, char **argv)
                goto exit_init_data;
        }
 
-       ret = utils_create_pipe(health_quit_pipe);
+       ret = init_health_quit_pipe();
        if (ret) {
                retval = -1;
                goto exit_health_quit_pipe;
@@ -4104,7 +4206,7 @@ exit_dispatcher_thread:
        }
 exit_health_thread:
 
-       utils_close_pipe(health_quit_pipe);
+       (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe);
 exit_health_quit_pipe:
 
 exit_init_data:
@@ -4122,6 +4224,14 @@ exit_options:
        /* Ensure all prior call_rcu are done. */
        rcu_barrier();
 
+       untrack_stdio();
+       /*
+        * fd_tracker_destroy() will log the contents of the fd-tracker
+        * if a leak is detected.
+        */
+       fd_tracker_destroy(the_fd_tracker);
+       rcu_unregister_thread();
+
        if (!retval) {
                exit(EXIT_SUCCESS);
        } else {
This page took 0.029054 seconds and 5 git commands to generate.