Move utils_expand_path and utils_expand_path_keep_symlink to libpath.la
[lttng-tools.git] / src / bin / lttng-relayd / main.c
index 6234e3e170d92515d1255a15087906531369083a..902fcab8e4c93a00933b93477a2ccf1ed54a1401 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
@@ -58,6 +48,7 @@
 #include <common/sessiond-comm/relayd.h>
 #include <common/uri.h>
 #include <common/utils.h>
+#include <common/path.h>
 #include <common/align.h>
 #include <common/config/session-config.h>
 #include <common/dynamic-buffer.h>
@@ -165,7 +156,7 @@ 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 +182,7 @@ 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 +226,22 @@ 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 +466,56 @@ 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 +633,10 @@ 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 +668,23 @@ 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 +708,10 @@ 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 +721,11 @@ 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 +736,6 @@ static void relayd_cleanup(void)
        if (tracing_group_name_override) {
                free((void *) tracing_group_name);
        }
-       fd_tracker_log(the_fd_tracker);
 }
 
 /*
@@ -842,10 +893,22 @@ 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;
 
@@ -854,8 +917,10 @@ 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) {
+       ret = fd_tracker_util_poll_create(the_fd_tracker,
+                       name, events, 1, LTTNG_CLOEXEC);
+       if (ret) {
+               PERROR("Failed to create \"%s\" poll file descriptor", name);
                goto error;
        }
 
@@ -885,13 +950,55 @@ 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 +1006,29 @@ static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
                goto error;
        }
 
-       ret = lttcomm_create_sock(sock);
-       if (ret < 0) {
+       /*
+        * 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;
+               }
+       }
+
+       ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
+                       (const char **) (formated_name ? &formated_name : NULL),
+                       1, create_sock, sock);
+       if (ret) {
+               PERROR("Failed to open \"%s\" relay socket",
+                               formated_name ?: "Unknown");
                goto error;
        }
-       DBG("Listening on sock %d", sock->fd);
+       DBG("Listening on %s socket %d", name, sock->fd);
 
        ret = sock->ops->bind(sock);
        if (ret < 0) {
@@ -917,15 +1042,38 @@ 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
  */
@@ -938,16 +1086,17 @@ 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();
 
-       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 +1105,7 @@ 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 +1173,18 @@ 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 +1237,31 @@ 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:
@@ -1113,6 +1270,7 @@ 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;
@@ -1475,7 +1633,8 @@ 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);
@@ -1590,24 +1749,6 @@ static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
         *        request.
         */
        try_stream_close(stream);
-       if (stream->is_metadata) {
-               struct relay_viewer_stream *vstream;
-
-               vstream = viewer_stream_get_by_id(stream->stream_handle);
-               if (vstream) {
-                       if (stream->no_new_metadata_notified) {
-                               /*
-                                * Since all the metadata has been sent to the
-                                * viewer and that we have a request to close
-                                * its stream, we can safely teardown the
-                                * corresponding metadata viewer stream.
-                                */
-                               viewer_stream_put(vstream);
-                       }
-                       /* Put local reference. */
-                       viewer_stream_put(vstream);
-               }
-       }
        stream_put(stream);
        ret = 0;
 
@@ -2345,6 +2486,125 @@ end_no_session:
        return ret;
 }
 
+static ssize_t relay_unpack_rotate_streams_header(
+               const struct lttng_buffer_view *payload,
+               struct lttcomm_relayd_rotate_streams *_rotate_streams)
+{
+       struct lttcomm_relayd_rotate_streams rotate_streams;
+       /*
+        * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
+        * This is the smallest version of this structure, but it can be larger;
+        * this variable is updated once the proper size of the structure is known.
+        *
+        * See comment at the declaration of this structure for more information.
+        */
+       ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
+       size_t expected_payload_size_no_padding,
+               expected_payload_size_3_bytes_padding,
+               expected_payload_size_7_bytes_padding;
+
+       if (payload->size < header_len) {
+               ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
+                               header_len, payload->size);
+               goto error;
+       }
+
+       /*
+        * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
+        * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
+        *
+        * We start by "unpacking" `stream_count` to figure out the padding length
+        * emited by our peer.
+        */
+       memcpy(&rotate_streams.stream_count, payload->data,
+                       sizeof(rotate_streams.stream_count));
+       rotate_streams = (typeof(rotate_streams)) {
+               .stream_count = be32toh(rotate_streams.stream_count),
+       };
+
+       /*
+        * Payload size expected given the possible padding lengths in
+        * `struct lttcomm_relayd_rotate_streams`.
+        */
+       expected_payload_size_no_padding = (rotate_streams.stream_count *
+               sizeof(*rotate_streams.rotation_positions)) +
+               sizeof(struct lttcomm_relayd_rotate_streams_packed);
+       expected_payload_size_3_bytes_padding = (rotate_streams.stream_count *
+               sizeof(*rotate_streams.rotation_positions)) +
+               sizeof(struct lttcomm_relayd_rotate_streams_3_bytes_padding);
+       expected_payload_size_7_bytes_padding = (rotate_streams.stream_count *
+               sizeof(*rotate_streams.rotation_positions)) +
+               sizeof(struct lttcomm_relayd_rotate_streams_7_bytes_padding);
+
+       if (payload->size == expected_payload_size_no_padding) {
+               struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
+
+               /*
+                * This handles cases where someone might build with
+                * -fpack-struct or any other toolchain that wouldn't produce
+                * padding to align `value`.
+                */
+               DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
+
+               header_len = sizeof(packed_rotate_streams);
+               memcpy(&packed_rotate_streams, payload->data, header_len);
+
+               /* Unpack the packed structure to the natively-packed version. */
+               *_rotate_streams = (typeof(*_rotate_streams)) {
+                       .stream_count = be32toh(packed_rotate_streams.stream_count),
+                       .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
+                               .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
+                               .value = be64toh(packed_rotate_streams.new_chunk_id.value),
+                       }
+               };
+       } else if (payload->size == expected_payload_size_3_bytes_padding) {
+               struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
+
+               DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
+
+               header_len = sizeof(padded_rotate_streams);
+               memcpy(&padded_rotate_streams, payload->data, header_len);
+
+               /* Unpack the 3-byte padded structure to the natively-packed version. */
+               *_rotate_streams = (typeof(*_rotate_streams)) {
+                       .stream_count = be32toh(padded_rotate_streams.stream_count),
+                       .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
+                               .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
+                               .value = be64toh(padded_rotate_streams.new_chunk_id.value),
+                       }
+               };
+       } else if (payload->size == expected_payload_size_7_bytes_padding) {
+               struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
+
+               DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
+
+               header_len = sizeof(padded_rotate_streams);
+               memcpy(&padded_rotate_streams, payload->data, header_len);
+
+               /* Unpack the 7-byte padded structure to the natively-packed version. */
+               *_rotate_streams = (typeof(*_rotate_streams)) {
+                       .stream_count = be32toh(padded_rotate_streams.stream_count),
+                       .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
+                               .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
+                               .value = be64toh(padded_rotate_streams.new_chunk_id.value),
+                       }
+               };
+
+               header_len = sizeof(padded_rotate_streams);
+       } else {
+               ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
+                               expected_payload_size_no_padding,
+                               expected_payload_size_3_bytes_padding,
+                               expected_payload_size_7_bytes_padding,
+                               payload->size);
+               goto error;
+       }
+
+       return header_len;
+error:
+       return -1;
+}
+
 /*
  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
  * session rotation feature (not the tracefile rotation feature).
@@ -2362,11 +2622,11 @@ static int relay_rotate_session_streams(
        struct lttcomm_relayd_rotate_streams rotate_streams;
        struct lttcomm_relayd_generic_reply reply = {};
        struct relay_stream *stream = NULL;
-       const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
        struct lttng_trace_chunk *next_trace_chunk = NULL;
        struct lttng_buffer_view stream_positions;
        char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
        const char *chunk_id_str = "none";
+       ssize_t header_len;
 
        if (!session || !conn->version_check_done) {
                ERR("Trying to rotate a stream before version check");
@@ -2380,24 +2640,12 @@ static int relay_rotate_session_streams(
                goto end_no_reply;
        }
 
-       if (payload->size < header_len) {
-               ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
-                               header_len, payload->size);
+       header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
+       if (header_len < 0) {
                ret = -1;
                goto end_no_reply;
        }
 
-       memcpy(&rotate_streams, payload->data, header_len);
-
-       /* Convert header to host endianness. */
-       rotate_streams = (typeof(rotate_streams)) {
-               .stream_count = be32toh(rotate_streams.stream_count),
-               .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
-                       .is_set = !!rotate_streams.new_chunk_id.is_set,
-                       .value = be64toh(rotate_streams.new_chunk_id.value),
-               }
-       };
-
        if (rotate_streams.new_chunk_id.is_set) {
                /*
                 * Retrieve the trace chunk the stream must transition to. As
@@ -2406,7 +2654,10 @@ static int relay_rotate_session_streams(
                 */
                next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
                                sessiond_trace_chunk_registry,
-                               session->sessiond_uuid, session->id,
+                               session->sessiond_uuid,
+                               conn->session->id_sessiond.is_set ?
+                                       conn->session->id_sessiond.value :
+                                       conn->session->id,
                                rotate_streams.new_chunk_id.value);
                if (!next_trace_chunk) {
                        char uuid_str[LTTNG_UUID_STR_LEN];
@@ -2435,7 +2686,7 @@ static int relay_rotate_session_streams(
                        chunk_id_str);
 
        stream_positions = lttng_buffer_view_from_view(payload,
-                       sizeof(rotate_streams), -1);
+                       header_len, -1);
        if (!stream_positions.data ||
                        stream_positions.size <
                                        (rotate_streams.stream_count *
@@ -2494,8 +2745,6 @@ end_no_reply:
        return ret;
 }
 
-
-
 /*
  * relay_create_trace_chunk: create a new trace chunk
  */
@@ -2513,7 +2762,6 @@ 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) {
@@ -2541,6 +2789,8 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        msg->creation_timestamp = be64toh(msg->creation_timestamp);
        msg->override_name_length = be32toh(msg->override_name_length);
 
+       pthread_mutex_lock(&conn->session->lock);
+       session->ongoing_rotation = true;
        if (session->current_trace_chunk &&
                        !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
                chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
@@ -2552,7 +2802,6 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                        goto end;
                }
        }
-       session->ongoing_rotation = true;
        if (!session->current_trace_chunk) {
                if (!session->has_rotated) {
                        new_path = "";
@@ -2570,6 +2819,7 @@ 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;
@@ -2610,15 +2860,9 @@ 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;
@@ -2628,7 +2872,9 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        published_chunk = sessiond_trace_chunk_registry_publish_chunk(
                        sessiond_trace_chunk_registry,
                        conn->session->sessiond_uuid,
-                       conn->session->id,
+                       conn->session->id_sessiond.is_set ?
+                               conn->session->id_sessiond.value :
+                               conn->session->id,
                        chunk);
        if (!published_chunk) {
                char uuid_str[LTTNG_UUID_STR_LEN];
@@ -2643,7 +2889,6 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                goto end;
        }
 
-       pthread_mutex_lock(&conn->session->lock);
        if (conn->session->pending_closure_trace_chunk) {
                /*
                 * Invalid; this means a second create_trace_chunk command was
@@ -2652,7 +2897,7 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
                reply_code = LTTNG_ERR_INVALID_PROTOCOL;
                ret = -1;
-               goto end_unlock_session;
+               goto end;
        }
        conn->session->pending_closure_trace_chunk =
                        conn->session->current_trace_chunk;
@@ -2661,9 +2906,8 @@ static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        if (!conn->session->pending_closure_trace_chunk) {
                session->ongoing_rotation = false;
        }
-end_unlock_session:
-       pthread_mutex_unlock(&conn->session->lock);
 end:
+       pthread_mutex_unlock(&conn->session->lock);
        reply.ret_code = htobe32((uint32_t) reply_code);
        send_ret = conn->sock->ops->sendmsg(conn->sock,
                        &reply,
@@ -2677,7 +2921,6 @@ end:
 end_no_reply:
        lttng_trace_chunk_put(chunk);
        lttng_trace_chunk_put(published_chunk);
-       lttng_directory_handle_put(session_output);
        return ret;
 }
 
@@ -2739,7 +2982,9 @@ static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        chunk = sessiond_trace_chunk_registry_get_chunk(
                        sessiond_trace_chunk_registry,
                        conn->session->sessiond_uuid,
-                       conn->session->id,
+                       conn->session->id_sessiond.is_set ?
+                               conn->session->id_sessiond.value :
+                               conn->session->id,
                        chunk_id);
        if (!chunk) {
                char uuid_str[LTTNG_UUID_STR_LEN];
@@ -2794,7 +3039,7 @@ 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;
        }
@@ -2812,7 +3057,7 @@ 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(
@@ -3611,7 +3856,8 @@ 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 +3919,7 @@ 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 +4188,13 @@ 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 +4216,24 @@ 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);
+}
 
-       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 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);
 }
 
 /*
@@ -3984,6 +4244,7 @@ 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 +4335,34 @@ 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 +4411,7 @@ 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 0.036106 seconds and 5 git commands to generate.