Relayd data available command support
[lttng-tools.git] / src / bin / lttng-relayd / main.c
index 81f10b78368c46401085477a51b799f90cf8fb19..0f81d556dd4df73a910471756b75cf6567581a36 100644 (file)
@@ -32,6 +32,7 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <inttypes.h>
 #include <urcu/futex.h>
 #include <urcu/uatomic.h>
 #include <unistd.h>
@@ -56,8 +57,8 @@
 /* command line options */
 static int opt_daemon;
 static char *opt_output_path;
-static struct lttng_uri *control_uri = NULL;
-static struct lttng_uri *data_uri = NULL;
+static struct lttng_uri *control_uri;
+static struct lttng_uri *data_uri;
 
 const char *progname;
 static int is_root;                    /* Set to 1 if the daemon is running as root */
@@ -81,8 +82,8 @@ static pthread_t listener_thread;
 static pthread_t dispatcher_thread;
 static pthread_t worker_thread;
 
-static uint64_t last_relay_stream_id = 0;
-static uint64_t last_relay_session_id = 0;
+static uint64_t last_relay_stream_id;
+static uint64_t last_relay_session_id;
 
 /*
  * Relay command queue.
@@ -93,8 +94,8 @@ static uint64_t last_relay_session_id = 0;
 static struct relay_cmd_queue relay_cmd_queue;
 
 /* buffer allocated at startup, used to store the trace data */
-static char *data_buffer = NULL;
-static unsigned int data_buffer_size = 0;
+static char *data_buffer;
+static unsigned int data_buffer_size;
 
 /*
  * usage function on stderr
@@ -125,7 +126,7 @@ int parse_args(int argc, char **argv)
                { "help", 0, 0, 'h', },
                { "output", 1, 0, 'o', },
                { "verbose", 0, 0, 'v', },
-               { NULL, 0, 0, 0 },
+               { NULL, 0, 0, 0, },
        };
 
        while (1) {
@@ -232,6 +233,9 @@ void cleanup(void)
 {
        DBG("Cleaning up");
 
+       /* free the dynamically allocated opt_output_path */
+       free(opt_output_path);
+
        /* Close thread quit pipes */
        utils_close_pipe(thread_quit_pipe);
 
@@ -443,34 +447,41 @@ error:
        return NULL;
 }
 
+/*
+ * Return nonzero if stream needs to be closed.
+ */
+static
+int close_stream_check(struct relay_stream *stream)
+{
+
+       if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
+               return 1;
+       }
+       return 0;
+}
+
 /*
  * This thread manages the listening for new connections on the network
  */
 static
 void *relay_thread_listener(void *data)
 {
-       int i, ret, pollfd;
+       int i, ret, pollfd, err = -1;
        int val = 1;
        uint32_t revents, nb_fd;
        struct lttng_poll_event events;
        struct lttcomm_sock *control_sock, *data_sock;
 
-       /*
-        * Get allocated in this thread, enqueued to a global queue, dequeued and
-        * freed in the worker thread.
-        */
-       struct relay_command *relay_cmd = NULL;
-
        DBG("[thread] Relay listener started");
 
        control_sock = relay_init_sock(control_uri);
        if (!control_sock) {
-               goto error_sock;
+               goto error_sock_control;
        }
 
        data_sock = relay_init_sock(data_uri);
        if (!data_sock) {
-               goto error_sock;
+               goto error_sock_relay;
        }
 
        /*
@@ -519,14 +530,21 @@ restart:
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
                        if (ret) {
-                               goto error;
+                               err = 0;
+                               goto exit;
                        }
 
                        if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
                                ERR("socket poll error");
                                goto error;
                        } else if (revents & LPOLLIN) {
-                               struct lttcomm_sock *newsock = NULL;
+                               /*
+                                * Get allocated in this thread,
+                                * enqueued to a global queue, dequeued
+                                * and freed in the worker thread.
+                                */
+                               struct relay_command *relay_cmd;
+                               struct lttcomm_sock *newsock;
 
                                relay_cmd = zmalloc(sizeof(struct relay_command));
                                if (relay_cmd == NULL) {
@@ -536,16 +554,19 @@ restart:
 
                                if (pollfd == data_sock->fd) {
                                        newsock = data_sock->ops->accept(data_sock);
-                                       if (newsock < 0) {
+                                       if (!newsock) {
                                                PERROR("accepting data sock");
+                                               free(relay_cmd);
                                                goto error;
                                        }
                                        relay_cmd->type = RELAY_DATA;
                                        DBG("Relay data connection accepted, socket %d", newsock->fd);
-                               } else if (pollfd == control_sock->fd) {
+                               } else {
+                                       assert(pollfd == control_sock->fd);
                                        newsock = control_sock->ops->accept(control_sock);
-                                       if (newsock < 0) {
+                                       if (!newsock) {
                                                PERROR("accepting control sock");
+                                               free(relay_cmd);
                                                goto error;
                                        }
                                        relay_cmd->type = RELAY_CONTROL;
@@ -555,6 +576,8 @@ restart:
                                                &val, sizeof(int));
                                if (ret < 0) {
                                        PERROR("setsockopt inet");
+                                       lttcomm_destroy_sock(newsock);
+                                       free(relay_cmd);
                                        goto error;
                                }
                                relay_cmd->sock = newsock;
@@ -572,28 +595,32 @@ restart:
                }
        }
 
+exit:
 error:
 error_poll_add:
        lttng_poll_clean(&events);
 error_create_poll:
-       if (control_sock->fd >= 0) {
-               ret = control_sock->ops->close(control_sock);
+       if (data_sock->fd >= 0) {
+               ret = data_sock->ops->close(data_sock);
                if (ret) {
                        PERROR("close");
                }
-               lttcomm_destroy_sock(control_sock);
        }
-       if (data_sock->fd >= 0) {
-               ret = data_sock->ops->close(data_sock);
+       lttcomm_destroy_sock(data_sock);
+error_sock_relay:
+       if (control_sock->fd >= 0) {
+               ret = control_sock->ops->close(control_sock);
                if (ret) {
                        PERROR("close");
                }
-               lttcomm_destroy_sock(data_sock);
        }
-
+       lttcomm_destroy_sock(control_sock);
+error_sock_control:
+       if (err) {
+               DBG("Thread exited with error");
+       }
        DBG("Relay listener thread cleanup complete");
        stop_threads();
-error_sock:
        return NULL;
 }
 
@@ -661,7 +688,7 @@ static
 char *expand_full_path(const char *path)
 {
        const char *end_path = path;
-       char *next, *cut_path, *expanded_path;
+       char *next, *cut_path, *expanded_path, *respath;
 
        /* Find last token delimited by '/' */
        while ((next = strpbrk(end_path + 1, "/"))) {
@@ -673,11 +700,12 @@ char *expand_full_path(const char *path)
 
        expanded_path = malloc(PATH_MAX);
        if (expanded_path == NULL) {
-               goto error;
+               respath = NULL;
+               goto end;
        }
 
-       expanded_path = realpath((char *)cut_path, expanded_path);
-       if (expanded_path == NULL) {
+       respath = realpath(cut_path, expanded_path);
+       if (respath == NULL) {
                switch (errno) {
                case ENOENT:
                        ERR("%s: No such file or directory", cut_path);
@@ -686,18 +714,14 @@ char *expand_full_path(const char *path)
                        PERROR("realpath");
                        break;
                }
-               goto error;
+               free(expanded_path);
+       } else {
+               /* Add end part to expanded path */
+               strcat(respath, end_path);
        }
-
-       /* Add end part to expanded path */
-       strcat(expanded_path, end_path);
-
-       free(cut_path);
-       return expanded_path;
-
-error:
+end:
        free(cut_path);
-       return NULL;
+       return respath;
 }
 
 
@@ -749,7 +773,7 @@ int mkdir_recursive(char *path, mode_t mode)
                        if (ret < 0) {
                                ret = mkdir(tmp, mode);
                                if (ret < 0) {
-                                       if (!(errno == EEXIST)) {
+                                       if (errno != EEXIST) {
                                                PERROR("mkdir recursive");
                                                ret = -errno;
                                                goto error;
@@ -762,7 +786,7 @@ int mkdir_recursive(char *path, mode_t mode)
 
        ret = mkdir(tmp, mode);
        if (ret < 0) {
-               if (!(errno == EEXIST)) {
+               if (errno != EEXIST) {
                        PERROR("mkdir recursive last piece");
                        ret = -errno;
                } else {
@@ -774,46 +798,73 @@ error:
        return ret;
 }
 
+static
+char *create_output_path_auto(char *path_name)
+{
+       int ret;
+       char *traces_path = NULL;
+       char *alloc_path = NULL;
+       char *default_path;
+
+       default_path = config_get_default_path();
+       if (default_path == NULL) {
+               ERR("Home path not found.\n \
+                               Please specify an output path using -o, --output PATH");
+               goto exit;
+       }
+       alloc_path = strdup(default_path);
+       if (alloc_path == NULL) {
+               PERROR("Path allocation");
+               goto exit;
+       }
+       ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
+                       "/%s", alloc_path, path_name);
+       if (ret < 0) {
+               PERROR("asprintf trace dir name");
+               goto exit;
+       }
+exit:
+       free(alloc_path);
+       return traces_path;
+}
+
+static
+char *create_output_path_noauto(char *path_name)
+{
+       int ret;
+       char *traces_path = NULL;
+       char *full_path;
+
+       full_path = expand_full_path(opt_output_path);
+       ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
+       if (ret < 0) {
+               PERROR("asprintf trace dir name");
+               goto exit;
+       }
+exit:
+       free(full_path);
+       return traces_path;
+}
+
 /*
  * create_output_path: create the output trace directory
  */
 static
 char *create_output_path(char *path_name)
 {
-       int ret = 0;
-       char *alloc_path = NULL;
-       char *traces_path = NULL;
-       char *full_path = NULL;
-
-       /* Auto output path */
        if (opt_output_path == NULL) {
-               alloc_path = strdup(config_get_default_path());
-               if (alloc_path == NULL) {
-                       ERR("Home path not found.\n \
-                                       Please specify an output path using -o, --output PATH");
-                       ret = -1;
-                       goto exit;
-               }
-
-               ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
-                               "/%s", alloc_path, path_name);
-               if (ret < 0) {
-                       PERROR("asprintf trace dir name");
-                       goto exit;
-               }
+               return create_output_path_auto(path_name);
        } else {
-               full_path = expand_full_path(opt_output_path);
-               ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
-               if (ret < 0) {
-                       PERROR("asprintf trace dir name");
-                       goto exit;
-               }
+               return create_output_path_noauto(path_name);
        }
-       free(alloc_path);
-       free(full_path);
+}
 
-exit:
-       return traces_path;
+static
+void deferred_free_stream(struct rcu_head *head)
+{
+       struct relay_stream *stream =
+               caa_container_of(head, struct relay_stream, rcu_node);
+       free(stream);
 }
 
 /*
@@ -828,12 +879,14 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht
        struct relay_stream *stream;
        int ret;
 
-       if (!cmd->session)
+       if (!cmd->session) {
                return;
+       }
 
-       DBG("Relay deleting session %lu", cmd->session->id);
+       DBG("Relay deleting session %" PRIu64, cmd->session->id);
        free(cmd->session->sock);
 
+       rcu_read_lock();
        cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
                node = lttng_ht_iter_get_node_ulong(&iter);
                if (node) {
@@ -843,10 +896,12 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht
                                close(stream->fd);
                                ret = lttng_ht_del(streams_ht, &iter);
                                assert(!ret);
-                               free(stream);
+                               call_rcu(&stream->rcu_node,
+                                       deferred_free_stream);
                        }
                }
        }
+       rcu_read_unlock();
 }
 
 /*
@@ -884,8 +939,9 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
                goto end_no_session;
        }
 
+       rcu_read_lock();
        stream->stream_handle = ++last_relay_stream_id;
-       stream->seq = 0;
+       stream->prev_seq = -1ULL;
        stream->session = session;
 
        root_path = create_output_path(stream_info.pathname);
@@ -895,7 +951,6 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
        }
        ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
        if (ret < 0) {
-               free(root_path);
                ERR("relay creating output directory");
                goto end;
        }
@@ -927,9 +982,9 @@ end:
        free(root_path);
        /* send the session id to the client or a negative return code on error */
        if (ret < 0) {
-               reply.ret_code = htobe32(LTTCOMM_ERR);
+               reply.ret_code = htobe32(LTTNG_ERR_UNK);
        } else {
-               reply.ret_code = htobe32(LTTCOMM_OK);
+               reply.ret_code = htobe32(LTTNG_OK);
        }
        reply.handle = htobe64(stream->stream_handle);
        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
@@ -937,6 +992,87 @@ end:
        if (send_ret < 0) {
                ERR("Relay sending stream id");
        }
+       rcu_read_unlock();
+
+end_no_session:
+       return ret;
+}
+
+/*
+ * relay_close_stream: close a specific stream
+ */
+static
+int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
+               struct relay_command *cmd, struct lttng_ht *streams_ht)
+{
+       struct relay_session *session = cmd->session;
+       struct lttcomm_relayd_close_stream stream_info;
+       struct lttcomm_relayd_generic_reply reply;
+       struct relay_stream *stream;
+       int ret, send_ret;
+       struct lttng_ht_node_ulong *node;
+       struct lttng_ht_iter iter;
+
+       DBG("Close stream received");
+
+       if (!session || session->version_check_done == 0) {
+               ERR("Trying to close a stream before version check");
+               ret = -1;
+               goto end_no_session;
+       }
+
+       ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
+                       sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL);
+       if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
+               ERR("Relay didn't receive valid add_stream struct size : %d", ret);
+               ret = -1;
+               goto end_no_session;
+       }
+
+       rcu_read_lock();
+       lttng_ht_lookup(streams_ht,
+                       (void *)((unsigned long) be64toh(stream_info.stream_id)),
+                       &iter);
+       node = lttng_ht_iter_get_node_ulong(&iter);
+       if (node == NULL) {
+               DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id));
+               ret = -1;
+               goto end_unlock;
+       }
+
+       stream = caa_container_of(node, struct relay_stream, stream_n);
+       if (!stream) {
+               ret = -1;
+               goto end_unlock;
+       }
+
+       stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
+       stream->close_flag = 1;
+
+       if (close_stream_check(stream)) {
+               int delret;
+
+               close(stream->fd);
+               delret = lttng_ht_del(streams_ht, &iter);
+               assert(!delret);
+               call_rcu(&stream->rcu_node,
+                               deferred_free_stream);
+               DBG("Closed tracefile %d from close stream", stream->fd);
+       }
+
+end_unlock:
+       rcu_read_unlock();
+
+       if (ret < 0) {
+               reply.ret_code = htobe32(LTTNG_ERR_UNK);
+       } else {
+               reply.ret_code = htobe32(LTTNG_OK);
+       }
+       send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
+                       sizeof(struct lttcomm_relayd_generic_reply), 0);
+       if (send_ret < 0) {
+               ERR("Relay sending stream id");
+       }
 
 end_no_session:
        return ret;
@@ -951,7 +1087,7 @@ void relay_unknown_command(struct relay_command *cmd)
        struct lttcomm_relayd_generic_reply reply;
        int ret;
 
-       reply.ret_code = htobe32(LTTCOMM_ERR);
+       reply.ret_code = htobe32(LTTNG_ERR_UNK);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
        if (ret < 0) {
@@ -967,13 +1103,13 @@ static
 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
 {
-       int ret = htobe32(LTTCOMM_OK);
+       int ret = htobe32(LTTNG_OK);
        struct lttcomm_relayd_generic_reply reply;
        struct relay_session *session = cmd->session;
 
        if (!session) {
                DBG("Trying to start the streaming without a session established");
-               ret = htobe32(LTTCOMM_ERR);
+               ret = htobe32(LTTNG_ERR_UNK);
        }
 
        reply.ret_code = ret;
@@ -988,6 +1124,7 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
 
 /*
  * Get stream from stream id.
+ * Need to be called with RCU read-side lock held.
  */
 static
 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
@@ -1002,7 +1139,7 @@ struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
                        &iter);
        node = lttng_ht_iter_get_node_ulong(&iter);
        if (node == NULL) {
-               DBG("Relay stream %lu not found", stream_id);
+               DBG("Relay stream %" PRIu64 " not found", stream_id);
                ret = NULL;
                goto end;
        }
@@ -1013,6 +1150,36 @@ end:
        return ret;
 }
 
+/*
+ * Append padding to the file pointed by the file descriptor fd.
+ */
+static int write_padding_to_file(int fd, uint32_t size)
+{
+       int ret = 0;
+       char *zeros;
+
+       if (size == 0) {
+               goto end;
+       }
+
+       zeros = zmalloc(size);
+       if (zeros == NULL) {
+               PERROR("zmalloc zeros for padding");
+               ret = -1;
+               goto end;
+       }
+
+       do {
+               ret = write(fd, zeros, size);
+       } while (ret < 0 && errno == EINTR);
+       if (ret < 0) {
+               PERROR("write padding to file");
+       }
+
+end:
+       return ret;
+}
+
 /*
  * relay_recv_metadata: receive the metada for the session.
  */
@@ -1020,7 +1187,7 @@ static
 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd, struct lttng_ht *streams_ht)
 {
-       int ret = htobe32(LTTCOMM_OK);
+       int ret = htobe32(LTTNG_OK);
        struct relay_session *session = cmd->session;
        struct lttcomm_relayd_metadata_payload *metadata_struct;
        struct relay_stream *metadata_stream;
@@ -1032,8 +1199,14 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
                goto end;
        }
 
-       data_size = be64toh(recv_hdr->data_size);
-       payload_size = data_size - sizeof(uint64_t);
+       data_size = payload_size = be64toh(recv_hdr->data_size);
+       if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
+               ERR("Incorrect data size");
+               ret = -1;
+               goto end;
+       }
+       payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
+
        if (data_buffer_size < data_size) {
                data_buffer = realloc(data_buffer, data_size);
                if (!data_buffer) {
@@ -1044,32 +1217,44 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
                data_buffer_size = data_size;
        }
        memset(data_buffer, 0, data_size);
-       DBG2("Relay receiving metadata, waiting for %lu bytes", data_size);
-       ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
+       DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
+       ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size,
+                       MSG_WAITALL);
        if (ret < 0 || ret != data_size) {
                ret = -1;
                ERR("Relay didn't receive the whole metadata");
                goto end;
        }
        metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
+
+       rcu_read_lock();
        metadata_stream = relay_stream_from_stream_id(
                        be64toh(metadata_struct->stream_id), streams_ht);
        if (!metadata_stream) {
                ret = -1;
-               goto end;
+               goto end_unlock;
        }
 
        do {
                ret = write(metadata_stream->fd, metadata_struct->payload,
                                payload_size);
        } while (ret < 0 && errno == EINTR);
-       if (ret < (payload_size)) {
+       if (ret < payload_size) {
                ERR("Relay error writing metadata on file");
                ret = -1;
-               goto end;
+               goto end_unlock;
        }
+
+       ret = write_padding_to_file(metadata_stream->fd,
+                       be32toh(metadata_struct->padding_size));
+       if (ret < 0) {
+               goto end_unlock;
+       }
+
        DBG2("Relay metadata written");
 
+end_unlock:
+       rcu_read_unlock();
 end:
        return ret;
 }
@@ -1081,9 +1266,9 @@ static
 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
 {
-       int ret = htobe32(LTTCOMM_OK);
+       int ret;
        struct lttcomm_relayd_version reply;
-       struct relay_session *session = NULL;
+       struct relay_session *session;
 
        if (cmd->session == NULL) {
                session = zmalloc(sizeof(struct relay_session));
@@ -1093,12 +1278,19 @@ int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
                        goto end;
                }
                session->id = ++last_relay_session_id;
-               DBG("Created session %lu", session->id);
+               DBG("Created session %" PRIu64, session->id);
                cmd->session = session;
+       } else {
+               session = cmd->session;
        }
        session->version_check_done = 1;
 
-       sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
+       ret = sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
+       if (ret < 2) {
+               ERR("Error in scanning version");
+               ret = -1;
+               goto end;
+       }
        reply.major = htobe32(reply.major);
        reply.minor = htobe32(reply.minor);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
@@ -1106,12 +1298,109 @@ int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
        if (ret < 0) {
                ERR("Relay sending version");
        }
-       DBG("Version check done");
+       DBG("Version check done (%u.%u)", be32toh(reply.major),
+                       be32toh(reply.minor));
 
 end:
        return ret;
 }
 
+/*
+ * Check for data availability for a given stream id from the session daemon.
+ */
+static
+int relay_data_available(struct lttcomm_relayd_hdr *recv_hdr,
+               struct relay_command *cmd, struct lttng_ht *streams_ht)
+{
+       struct relay_session *session = cmd->session;
+       struct lttcomm_relayd_data_available msg;
+       struct lttcomm_relayd_generic_reply reply;
+       struct relay_stream *stream;
+       int ret;
+       struct lttng_ht_node_ulong *node;
+       struct lttng_ht_iter iter;
+       uint64_t last_net_seq_num, stream_id;
+
+       DBG("Data available command received");
+
+       if (!session || session->version_check_done == 0) {
+               ERR("Trying to check for data before version check");
+               ret = -1;
+               goto end_no_session;
+       }
+
+       ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), MSG_WAITALL);
+       if (ret < sizeof(msg)) {
+               ERR("Relay didn't receive valid data_available struct size : %d", ret);
+               ret = -1;
+               goto end_no_session;
+       }
+
+       stream_id = be64toh(msg.stream_id);
+       last_net_seq_num = be64toh(msg.last_net_seq_num);
+
+       rcu_read_lock();
+       lttng_ht_lookup(streams_ht, (void *)((unsigned long) stream_id), &iter);
+       node = lttng_ht_iter_get_node_ulong(&iter);
+       if (node == NULL) {
+               DBG("Relay stream %" PRIu64 " not found", stream_id);
+               ret = -1;
+               goto end_unlock;
+       }
+
+       stream = caa_container_of(node, struct relay_stream, stream_n);
+       assert(stream);
+
+       DBG("Data available for stream id %" PRIu64 " prev_seq %" PRIu64
+                       " and last_seq %" PRIu64, stream_id, stream->prev_seq,
+                       last_net_seq_num);
+
+       if (stream->prev_seq == -1UL || stream->prev_seq <= last_net_seq_num) {
+               /* Data has in fact been written and is available */
+               ret = 1;
+       } else {
+               /* Data still being streamed. */
+               ret = 0;
+       }
+
+end_unlock:
+       rcu_read_unlock();
+
+       reply.ret_code = htobe32(ret);
+       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+       if (ret < 0) {
+               ERR("Relay data available ret code failed");
+       }
+
+end_no_session:
+       return ret;
+}
+
+/*
+ * Wait for the control socket to reach a quiescent state.
+ *
+ * Note that for now, when receiving this command from the session daemon, this
+ * means that every subsequent commands or data received on the control socket
+ * has been handled. So, this is why we simply return OK here.
+ */
+static
+int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
+               struct relay_command *cmd)
+{
+       int ret;
+       struct lttcomm_relayd_generic_reply reply;
+
+       DBG("Checking quiescent state on control socket");
+
+       reply.ret_code = htobe32(LTTNG_OK);
+       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+       if (ret < 0) {
+               ERR("Relay data available ret code failed");
+       }
+
+       return ret;
+}
+
 /*
  * relay_process_control: Process the commands received on the control socket
  */
@@ -1139,6 +1428,15 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
        case RELAYD_VERSION:
                ret = relay_send_version(recv_hdr, cmd);
                break;
+       case RELAYD_CLOSE_STREAM:
+               ret = relay_close_stream(recv_hdr, cmd, streams_ht);
+               break;
+       case RELAYD_DATA_AVAILABLE:
+               ret = relay_data_available(recv_hdr, cmd, streams_ht);
+               break;
+       case RELAYD_QUIESCENT_CONTROL:
+               ret = relay_quiescent_control(recv_hdr, cmd);
+               break;
        case RELAYD_UPDATE_SYNC_INFO:
        default:
                ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
@@ -1161,6 +1459,7 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
        struct relay_stream *stream;
        struct lttcomm_relayd_data_hdr data_hdr;
        uint64_t stream_id;
+       uint64_t net_seq_num;
        uint32_t data_size;
 
        ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
@@ -1172,10 +1471,12 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
        }
 
        stream_id = be64toh(data_hdr.stream_id);
+
+       rcu_read_lock();
        stream = relay_stream_from_stream_id(stream_id, streams_ht);
        if (!stream) {
                ret = -1;
-               goto end;
+               goto end_unlock;
        }
 
        data_size = be32toh(data_hdr.data_size);
@@ -1184,16 +1485,20 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
                if (!data_buffer) {
                        ERR("Allocating data buffer");
                        ret = -1;
-                       goto end;
+                       goto end_unlock;
                }
                data_buffer_size = data_size;
        }
        memset(data_buffer, 0, data_size);
 
+       net_seq_num = be64toh(data_hdr.net_seq_num);
+
+       DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
+               data_size, stream_id, net_seq_num);
        ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
        if (ret <= 0) {
                ret = -1;
-               goto end;
+               goto end_unlock;
        }
 
        do {
@@ -1202,22 +1507,43 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
        if (ret < data_size) {
                ERR("Relay error writing data to file");
                ret = -1;
-               goto end;
+               goto end_unlock;
+       }
+
+       DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
+                       ret, stream->stream_handle);
+
+       ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
+       if (ret < 0) {
+               goto end_unlock;
+       }
+
+       stream->prev_seq = net_seq_num;
+
+       /* Check if we need to close the FD */
+       if (close_stream_check(stream)) {
+               struct lttng_ht_iter iter;
+
+               close(stream->fd);
+               iter.iter.node = &stream->stream_n.node;
+               ret = lttng_ht_del(streams_ht, &iter);
+               assert(!ret);
+               call_rcu(&stream->rcu_node,
+                       deferred_free_stream);
+               DBG("Closed tracefile %d after recv data", stream->fd);
        }
-       DBG2("Relay wrote %d bytes to tracefile for stream id %lu", ret, stream->stream_handle);
 
+end_unlock:
+       rcu_read_unlock();
 end:
        return ret;
 }
 
 static
-void relay_cleanup_connection(struct lttng_ht *relay_connections_ht, struct lttng_poll_event *events,
-               struct lttng_ht *streams_ht, int pollfd, struct lttng_ht_iter *iter)
+void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
 {
        int ret;
 
-       ret = lttng_ht_del(relay_connections_ht, iter);
-       assert(!ret);
        lttng_poll_del(events, pollfd);
 
        ret = close(pollfd);
@@ -1230,32 +1556,58 @@ static
 int relay_add_connection(int fd, struct lttng_poll_event *events,
                struct lttng_ht *relay_connections_ht)
 {
-       int ret;
        struct relay_command *relay_connection;
+       int ret;
 
        relay_connection = zmalloc(sizeof(struct relay_command));
        if (relay_connection == NULL) {
                PERROR("Relay command zmalloc");
-               ret = -1;
-               goto end;
+               goto error;
        }
        ret = read(fd, relay_connection, sizeof(struct relay_command));
-       if (ret < 0 || ret < sizeof(relay_connection)) {
+       if (ret < 0 || ret < sizeof(struct relay_command)) {
                PERROR("read relay cmd pipe");
-               ret = -1;
-               goto end;
+               goto error_read;
        }
 
        lttng_ht_node_init_ulong(&relay_connection->sock_n,
                        (unsigned long) relay_connection->sock->fd);
+       rcu_read_lock();
        lttng_ht_add_unique_ulong(relay_connections_ht,
                        &relay_connection->sock_n);
-       ret = lttng_poll_add(events,
+       rcu_read_unlock();
+       return lttng_poll_add(events,
                        relay_connection->sock->fd,
                        LPOLLIN | LPOLLRDHUP);
 
-end:
-       return ret;
+error_read:
+       free(relay_connection);
+error:
+       return -1;
+}
+
+static
+void deferred_free_connection(struct rcu_head *head)
+{
+       struct relay_command *relay_connection =
+               caa_container_of(head, struct relay_command, rcu_node);
+       free(relay_connection);
+}
+
+static
+void relay_del_connection(struct lttng_ht *relay_connections_ht,
+               struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
+               struct relay_command *relay_connection)
+{
+       int ret;
+
+       ret = lttng_ht_del(relay_connections_ht, iter);
+       assert(!ret);
+       if (relay_connection->type == RELAY_CONTROL) {
+               relay_delete_session(relay_connection, streams_ht);
+       }
+       call_rcu(&relay_connection->rcu_node,
+               deferred_free_connection);
 }
 
 /*
@@ -1264,7 +1616,7 @@ end:
 static
 void *relay_thread_worker(void *data)
 {
-       int i, ret, pollfd;
+       int i, ret, pollfd, err = -1;
        uint32_t revents, nb_fd;
        struct relay_command *relay_connection;
        struct lttng_poll_event events;
@@ -1276,11 +1628,19 @@ void *relay_thread_worker(void *data)
 
        DBG("[thread] Relay worker started");
 
+       rcu_register_thread();
+
        /* table of connections indexed on socket */
        relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
+       if (!relay_connections_ht) {
+               goto relay_connections_ht_error;
+       }
 
        /* tables of streams indexed by stream ID */
        streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
+       if (!streams_ht) {
+               goto streams_ht_error;
+       }
 
        ret = create_thread_poll_set(&events, 2);
        if (ret < 0) {
@@ -1300,6 +1660,7 @@ void *relay_thread_worker(void *data)
 
                /* Infinite blocking call, waiting for transmission */
        restart:
+               DBG3("Relayd worker thread polling...");
                ret = lttng_poll_wait(&events, -1);
                if (ret < 0) {
                        /*
@@ -1319,7 +1680,8 @@ void *relay_thread_worker(void *data)
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
                        if (ret) {
-                               goto error;
+                               err = 0;
+                               goto exit;
                        }
 
                        /* Inspect the relay cmd pipe for new connection */
@@ -1336,12 +1698,14 @@ void *relay_thread_worker(void *data)
                                        }
                                }
                        } else if (revents > 0) {
+                               rcu_read_lock();
                                lttng_ht_lookup(relay_connections_ht,
                                                (void *)((unsigned long) pollfd),
                                                &iter);
                                node = lttng_ht_iter_get_node_ulong(&iter);
                                if (node == NULL) {
                                        DBG2("Relay sock %d not found", pollfd);
+                                       rcu_read_unlock();
                                        goto error;
                                }
                                relay_connection = caa_container_of(node,
@@ -1349,17 +1713,16 @@ void *relay_thread_worker(void *data)
 
                                if (revents & (LPOLLERR)) {
                                        ERR("POLL ERROR");
-                                       relay_cleanup_connection(relay_connections_ht,
-                                                       &events, streams_ht, pollfd, &iter);
-                                       free(relay_connection);
+                                       relay_cleanup_poll_connection(&events, pollfd);
+                                       relay_del_connection(relay_connections_ht,
+                                                       streams_ht, &iter,
+                                                       relay_connection);
                                } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
                                        DBG("Socket %d hung up", pollfd);
-                                       relay_cleanup_connection(relay_connections_ht,
-                                                       &events, streams_ht, pollfd, &iter);
-                                       if (relay_connection->type == RELAY_CONTROL) {
-                                               relay_delete_session(relay_connection, streams_ht);
-                                       }
-                                       free(relay_connection);
+                                       relay_cleanup_poll_connection(&events, pollfd);
+                                       relay_del_connection(relay_connections_ht,
+                                                       streams_ht, &iter,
+                                                       relay_connection);
                                } else if (revents & LPOLLIN) {
                                        /* control socket */
                                        if (relay_connection->type == RELAY_CONTROL) {
@@ -1368,14 +1731,14 @@ void *relay_thread_worker(void *data)
                                                                sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL);
                                                /* connection closed */
                                                if (ret <= 0) {
-                                                       relay_cleanup_connection(relay_connections_ht,
-                                                                       &events, streams_ht, pollfd, &iter);
-                                                       relay_delete_session(relay_connection, streams_ht);
-                                                       free(relay_connection);
+                                                       relay_cleanup_poll_connection(&events, pollfd);
+                                                       relay_del_connection(relay_connections_ht,
+                                                                       streams_ht, &iter,
+                                                                       relay_connection);
                                                        DBG("Control connection closed with %d", pollfd);
                                                } else {
                                                        if (relay_connection->session) {
-                                                               DBG2("Relay worker receiving data for session : %lu",
+                                                               DBG2("Relay worker receiving data for session : %" PRIu64,
                                                                                relay_connection->session->id);
                                                        }
                                                        ret = relay_process_control(&recv_hdr,
@@ -1386,9 +1749,10 @@ void *relay_thread_worker(void *data)
                                                         * command: clear the session
                                                         * */
                                                        if (ret < 0) {
-                                                               relay_cleanup_connection(relay_connections_ht,
-                                                                               &events, streams_ht, pollfd, &iter);
-                                                               free(relay_connection);
+                                                               relay_cleanup_poll_connection(&events, pollfd);
+                                                               relay_del_connection(relay_connections_ht,
+                                                                               streams_ht, &iter,
+                                                                               relay_connection);
                                                                DBG("Connection closed with %d", pollfd);
                                                        }
                                                }
@@ -1397,36 +1761,48 @@ void *relay_thread_worker(void *data)
                                                ret = relay_process_data(relay_connection, streams_ht);
                                                /* connection closed */
                                                if (ret < 0) {
-                                                       relay_cleanup_connection(relay_connections_ht,
-                                                                       &events, streams_ht, pollfd, &iter);
-                                                       relay_delete_session(relay_connection, streams_ht);
+                                                       relay_cleanup_poll_connection(&events, pollfd);
+                                                       relay_del_connection(relay_connections_ht,
+                                                                       streams_ht, &iter,
+                                                                       relay_connection);
                                                        DBG("Data connection closed with %d", pollfd);
                                                }
                                        }
                                }
+                               rcu_read_unlock();
                        }
                }
        }
 
+exit:
 error:
        lttng_poll_clean(&events);
 
        /* empty the hash table and free the memory */
+       rcu_read_lock();
        cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
                node = lttng_ht_iter_get_node_ulong(&iter);
                if (node) {
                        relay_connection = caa_container_of(node,
                                        struct relay_command, sock_n);
-                       free(relay_connection);
+                       relay_del_connection(relay_connections_ht,
+                                       streams_ht, &iter,
+                                       relay_connection);
                }
-               ret = lttng_ht_del(relay_connections_ht, &iter);
-               assert(!ret);
        }
+       rcu_read_unlock();
 error_poll_create:
-       free(data_buffer);
+       lttng_ht_destroy(streams_ht);
+streams_ht_error:
        lttng_ht_destroy(relay_connections_ht);
+relay_connections_ht_error:
+       if (err) {
+               DBG("Thread exited with error");
+       }
        DBG("Worker thread cleanup complete");
+       free(data_buffer);
        stop_threads();
+       rcu_unregister_thread();
        return NULL;
 }
 
This page took 0.037801 seconds and 4 git commands to generate.