/*
* Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
* Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
*
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License as published by the Free
- * Software Foundation; only version 2 of the License.
+ * 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.
*
- * 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.
+ * 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., 59 Temple
- * Place - Suite 330, Boston, MA 02111-1307, USA.
+ * 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 _GNU_SOURCE
#include <getopt.h>
#include <grp.h>
#include <limits.h>
+#include <paths.h>
#include <pthread.h>
-#include <semaphore.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <inttypes.h>
#include <sys/mman.h>
#include <sys/mount.h>
#include <sys/resource.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
-#include <urcu/futex.h>
+#include <urcu/uatomic.h>
#include <unistd.h>
#include <config.h>
#include <common/common.h>
-#include <common/compat/poll.h>
#include <common/compat/socket.h>
#include <common/defaults.h>
#include <common/kernel-consumer/kernel-consumer.h>
-#include <common/ust-consumer/ust-consumer.h>
+#include <common/futex.h>
+#include <common/relayd/relayd.h>
+#include <common/utils.h>
+#include <common/config/config.h>
#include "lttng-sessiond.h"
+#include "buffer-registry.h"
#include "channel.h"
+#include "cmd.h"
+#include "consumer.h"
#include "context.h"
#include "event.h"
-#include "futex.h"
#include "kernel.h"
+#include "kernel-consumer.h"
#include "modprobe.h"
#include "shm.h"
#include "ust-ctl.h"
+#include "ust-consumer.h"
#include "utils.h"
+#include "fd-limit.h"
+#include "health-sessiond.h"
+#include "testpoint.h"
+#include "ust-thread.h"
+#include "jul-thread.h"
#define CONSUMERD_FILE "lttng-consumerd"
-struct consumer_data {
- enum lttng_consumer_type type;
-
- pthread_t thread; /* Worker thread interacting with the consumer */
- sem_t sem;
-
- /* Mutex to control consumerd pid assignation */
- pthread_mutex_t pid_mutex;
- pid_t pid;
-
- int err_sock;
- int cmd_sock;
-
- /* consumer error and command Unix socket path */
- char err_unix_sock_path[PATH_MAX];
- char cmd_unix_sock_path[PATH_MAX];
-};
-
-/* Const values */
-const char default_home_dir[] = DEFAULT_HOME_DIR;
-const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
-const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
-const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
-
-/* Variables */
-int opt_verbose; /* Not static for lttngerr.h */
-int opt_verbose_consumer; /* Not static for lttngerr.h */
-int opt_quiet; /* Not static for lttngerr.h */
-
const char *progname;
-const char *opt_tracing_group;
+static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
+static int tracing_group_name_override;
+static char *opt_pidfile;
static int opt_sig_parent;
+static int opt_verbose_consumer;
static int opt_daemon;
static int opt_no_kernel;
static int is_root; /* Set to 1 if the daemon is running as root */
static pid_t ppid; /* Parent PID for --sig-parent option */
+static pid_t child_ppid; /* Internal parent PID use with daemonize. */
static char *rundir;
-/* Consumer daemon specific control data */
+/* Set to 1 when a SIGUSR1 signal is received. */
+static int recv_child_signal;
+
+/*
+ * Consumer daemon specific control data. Every value not initialized here is
+ * set to 0 by the static definition.
+ */
static struct consumer_data kconsumer_data = {
.type = LTTNG_CONSUMER_KERNEL,
.err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
.cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
+ .err_sock = -1,
+ .cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer64_data = {
.type = LTTNG_CONSUMER64_UST,
.err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
+ .err_sock = -1,
+ .cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer32_data = {
.type = LTTNG_CONSUMER32_UST,
.err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
+ .err_sock = -1,
+ .cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
+/* Command line options */
+static const struct option long_options[] = {
+ { "client-sock", 1, 0, 'c' },
+ { "apps-sock", 1, 0, 'a' },
+ { "kconsumerd-cmd-sock", 1, 0, 'C' },
+ { "kconsumerd-err-sock", 1, 0, 'E' },
+ { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
+ { "ustconsumerd32-err-sock", 1, 0, 'H' },
+ { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
+ { "ustconsumerd64-err-sock", 1, 0, 'F' },
+ { "consumerd32-path", 1, 0, 'u' },
+ { "consumerd32-libdir", 1, 0, 'U' },
+ { "consumerd64-path", 1, 0, 't' },
+ { "consumerd64-libdir", 1, 0, 'T' },
+ { "daemonize", 0, 0, 'd' },
+ { "sig-parent", 0, 0, 'S' },
+ { "help", 0, 0, 'h' },
+ { "group", 1, 0, 'g' },
+ { "version", 0, 0, 'V' },
+ { "quiet", 0, 0, 'q' },
+ { "verbose", 0, 0, 'v' },
+ { "verbose-consumer", 0, 0, 'Z' },
+ { "no-kernel", 0, 0, 'N' },
+ { "pidfile", 1, 0, 'p' },
+ { "jul-tcp-port", 1, 0, 'J' },
+ { "config", 1, 0, 'f' },
+ { NULL, 0, 0, 0 }
+};
+
+/* Command line options to ignore from configuration file */
+static const char *config_ignore_options[] = { "help", "version", "config" };
+
+/* Shared between threads */
static int dispatch_thread_exit;
/* Global application Unix socket path */
static char client_unix_sock_path[PATH_MAX];
/* global wait shm path for UST */
static char wait_shm_path[PATH_MAX];
+/* Global health check unix path */
+static char health_unix_sock_path[PATH_MAX];
/* Sockets and FDs */
static int client_sock = -1;
static int apps_sock = -1;
-static int kernel_tracer_fd = -1;
+int kernel_tracer_fd = -1;
static int kernel_poll_pipe[2] = { -1, -1 };
/*
*/
static int apps_cmd_pipe[2] = { -1, -1 };
+int apps_cmd_notify_pipe[2] = { -1, -1 };
+
/* Pthread, Mutexes and Semaphores */
static pthread_t apps_thread;
+static pthread_t apps_notify_thread;
static pthread_t reg_apps_thread;
static pthread_t client_thread;
static pthread_t kernel_thread;
static pthread_t dispatch_thread;
-
+static pthread_t health_thread;
+static pthread_t ht_cleanup_thread;
+static pthread_t jul_reg_thread;
/*
* UST registration command queue. This queue is tied with a futex and uses a N
static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
+static int consumerd32_bin_override;
+static int consumerd64_bin_override;
+static int consumerd32_libdir_override;
+static int consumerd64_libdir_override;
+
+static const char *module_proc_lttng = "/proc/lttng";
+
+/*
+ * Consumer daemon state which is changed when spawning it, killing it or in
+ * case of a fatal error.
+ */
+enum consumerd_state {
+ CONSUMER_STARTED = 1,
+ CONSUMER_STOPPED = 2,
+ CONSUMER_ERROR = 3,
+};
+
+/*
+ * This consumer daemon state is used to validate if a client command will be
+ * able to reach the consumer. If not, the client is informed. For instance,
+ * doing a "lttng start" when the consumer state is set to ERROR will return an
+ * error to the client.
+ *
+ * The following example shows a possible race condition of this scheme:
+ *
+ * consumer thread error happens
+ * client cmd arrives
+ * client cmd checks state -> still OK
+ * consumer thread exit, sets error
+ * client cmd try to talk to consumer
+ * ...
+ *
+ * However, since the consumer is a different daemon, we have no way of making
+ * sure the command will reach it safely even with this state flag. This is why
+ * we consider that up to the state validation during command processing, the
+ * command is safe. After that, we can not guarantee the correctness of the
+ * client request vis-a-vis the consumer.
+ */
+static enum consumerd_state ust_consumerd_state;
+static enum consumerd_state kernel_consumerd_state;
+
+/*
+ * Socket timeout for receiving and sending in seconds.
+ */
+static int app_socket_timeout;
+
+/* Set in main() with the current page size. */
+long page_size;
+
+/* Application health monitoring */
+struct health_app *health_sessiond;
+
+/* JUL TCP port for registration. Used by the JUL thread. */
+unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
+
+const char * const config_section_name = "sessiond";
static
void setup_consumerd_path(void)
/*
* 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,
- unsigned int size)
+int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
{
int ret;
- if (events == NULL || size == 0) {
- ret = -1;
- goto error;
- }
+ assert(events);
ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
if (ret < 0) {
}
/* Add quit pipe */
- ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
+ ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
if (ret < 0) {
goto error;
}
*
* Return 1 if it was triggered else 0;
*/
-static int check_thread_quit_pipe(int fd, uint32_t events)
+int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
{
if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
return 1;
return 0;
}
-/*
- * Return group ID of the tracing group or -1 if not found.
- */
-static gid_t allowed_group(void)
-{
- struct group *grp;
-
- if (opt_tracing_group) {
- grp = getgrnam(opt_tracing_group);
- } else {
- grp = getgrnam(default_tracing_group);
- }
- if (!grp) {
- return -1;
- } else {
- return grp->gr_gid;
- }
-}
-
/*
* Init thread quit pipe.
*
return ret;
}
-/*
- * Complete teardown of a kernel session. This free all data structure related
- * to a kernel session and update counter.
- */
-static void teardown_kernel_session(struct ltt_session *session)
-{
- if (!session->kernel_session) {
- DBG3("No kernel session when tearing down session");
- return;
- }
-
- DBG("Tearing down kernel session");
-
- /*
- * If a custom kernel consumer was registered, close the socket before
- * tearing down the complete kernel session structure
- */
- if (session->kernel_session->consumer_fd != kconsumer_data.cmd_sock) {
- lttcomm_close_unix_sock(session->kernel_session->consumer_fd);
- }
-
- trace_kernel_destroy_session(session->kernel_session);
-}
-
-/*
- * Complete teardown of all UST sessions. This will free everything on his path
- * and destroy the core essence of all ust sessions :)
- */
-static void teardown_ust_session(struct ltt_session *session)
-{
- int ret;
-
- if (!session->ust_session) {
- DBG3("No UST session when tearing down session");
- return;
- }
-
- DBG("Tearing down UST session(s)");
-
- ret = ust_app_destroy_trace_all(session->ust_session);
- if (ret) {
- ERR("Error in ust_app_destroy_trace_all");
- }
-
- trace_ust_destroy_session(session->ust_session);
-}
-
/*
* Stop all threads by closing the thread quit pipe.
*/
}
/* Dispatch thread */
- dispatch_thread_exit = 1;
+ CMM_STORE_SHARED(dispatch_thread_exit, 1);
futex_nto1_wake(&ust_cmd_queue.futex);
}
+/*
+ * Close every consumer sockets.
+ */
+static void close_consumer_sockets(void)
+{
+ int ret;
+
+ if (kconsumer_data.err_sock >= 0) {
+ ret = close(kconsumer_data.err_sock);
+ if (ret < 0) {
+ PERROR("kernel consumer err_sock close");
+ }
+ }
+ if (ustconsumer32_data.err_sock >= 0) {
+ ret = close(ustconsumer32_data.err_sock);
+ if (ret < 0) {
+ PERROR("UST consumerd32 err_sock close");
+ }
+ }
+ if (ustconsumer64_data.err_sock >= 0) {
+ ret = close(ustconsumer64_data.err_sock);
+ if (ret < 0) {
+ PERROR("UST consumerd64 err_sock close");
+ }
+ }
+ if (kconsumer_data.cmd_sock >= 0) {
+ ret = close(kconsumer_data.cmd_sock);
+ if (ret < 0) {
+ PERROR("kernel consumer cmd_sock close");
+ }
+ }
+ if (ustconsumer32_data.cmd_sock >= 0) {
+ ret = close(ustconsumer32_data.cmd_sock);
+ if (ret < 0) {
+ PERROR("UST consumerd32 cmd_sock close");
+ }
+ }
+ if (ustconsumer64_data.cmd_sock >= 0) {
+ ret = close(ustconsumer64_data.cmd_sock);
+ if (ret < 0) {
+ PERROR("UST consumerd64 cmd_sock close");
+ }
+ }
+}
+
/*
* Cleanup the daemon
*/
static void cleanup(void)
{
- int ret, i;
- char *cmd;
+ int ret;
struct ltt_session *sess, *stmp;
+ char path[PATH_MAX];
DBG("Cleaning up");
- DBG("Removing %s directory", rundir);
- ret = asprintf(&cmd, "rm -rf %s", rundir);
- if (ret < 0) {
- ERR("asprintf failed. Something is really wrong!");
- }
+ /*
+ * Close the thread quit pipe. It has already done its job,
+ * since we are now called.
+ */
+ utils_close_pipe(thread_quit_pipe);
- /* Remove lttng run directory */
- ret = system(cmd);
- if (ret < 0) {
- ERR("Unable to clean %s", rundir);
- }
- free(cmd);
+ /*
+ * If opt_pidfile is undefined, the default file will be wiped when
+ * removing the rundir.
+ */
+ if (opt_pidfile) {
+ ret = remove(opt_pidfile);
+ if (ret < 0) {
+ PERROR("remove pidfile %s", opt_pidfile);
+ }
+ }
+
+ DBG("Removing sessiond and consumerd content of directory %s", rundir);
+
+ /* sessiond */
+ snprintf(path, PATH_MAX,
+ "%s/%s",
+ rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX, "%s/%s", rundir,
+ DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ /* kconsumerd */
+ snprintf(path, PATH_MAX,
+ DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_KCONSUMERD_PATH,
+ rundir);
+ DBG("Removing directory %s", path);
+ (void) rmdir(path);
+
+ /* ust consumerd 32 */
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD32_PATH,
+ rundir);
+ DBG("Removing directory %s", path);
+ (void) rmdir(path);
+
+ /* ust consumerd 64 */
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD64_PATH,
+ rundir);
+ DBG("Removing directory %s", path);
+ (void) rmdir(path);
+
+ /*
+ * We do NOT rmdir rundir because there are other processes
+ * using it, for instance lttng-relayd, which can start in
+ * parallel with this teardown.
+ */
+
+ free(rundir);
DBG("Cleaning up all sessions");
/* Cleanup ALL session */
cds_list_for_each_entry_safe(sess, stmp,
&session_list_ptr->head, list) {
- teardown_kernel_session(sess);
- teardown_ust_session(sess);
- free(sess);
+ cmd_destroy_session(sess, kernel_poll_pipe[1]);
}
}
DBG("Closing all UST sockets");
ust_app_clean_list();
-
- pthread_mutex_destroy(&kconsumer_data.pid_mutex);
+ buffer_reg_destroy_registries();
if (is_root && !opt_no_kernel) {
DBG2("Closing kernel fd");
modprobe_remove_lttng_all();
}
+ close_consumer_sockets();
+
/*
- * Closing all pipes used for communication between threads.
+ * If the override option is set, the pointer points to a *non* const thus
+ * freeing it even though the variable type is set to const.
*/
- for (i = 0; i < 2; i++) {
- if (kernel_poll_pipe[i] >= 0) {
- ret = close(kernel_poll_pipe[i]);
- if (ret) {
- PERROR("close");
- }
-
- }
+ if (tracing_group_name_override) {
+ free((void *) tracing_group_name);
}
- for (i = 0; i < 2; i++) {
- if (thread_quit_pipe[i] >= 0) {
- ret = close(thread_quit_pipe[i]);
- if (ret) {
- PERROR("close");
- }
- }
+ if (consumerd32_bin_override) {
+ free((void *) consumerd32_bin);
}
- for (i = 0; i < 2; i++) {
- if (apps_cmd_pipe[i] >= 0) {
- ret = close(apps_cmd_pipe[i]);
- if (ret) {
- PERROR("close");
- }
- }
+ if (consumerd64_bin_override) {
+ free((void *) consumerd64_bin);
+ }
+ if (consumerd32_libdir_override) {
+ free((void *) consumerd32_libdir);
+ }
+ if (consumerd64_libdir_override) {
+ free((void *) consumerd64_libdir);
+ }
+
+ if (opt_pidfile) {
+ free(opt_pidfile);
}
/* <fun> */
static int send_unix_sock(int sock, void *buf, size_t len)
{
/* Check valid length */
- if (len <= 0) {
+ if (len == 0) {
return -1;
}
}
}
-/*
- * Send all stream fds of kernel channel to the consumer.
- */
-static int send_kconsumer_channel_streams(struct consumer_data *consumer_data,
- int sock, struct ltt_kernel_channel *channel,
- uid_t uid, gid_t gid)
-{
- int ret;
- struct ltt_kernel_stream *stream;
- struct lttcomm_consumer_msg lkm;
-
- DBG("Sending streams of channel %s to kernel consumer",
- channel->channel->name);
-
- /* Send channel */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
- lkm.u.channel.channel_key = channel->fd;
- lkm.u.channel.max_sb_size = channel->channel->attr.subbuf_size;
- lkm.u.channel.mmap_len = 0; /* for kernel */
- DBG("Sending channel %d to consumer", lkm.u.channel.channel_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer channel");
- goto error;
- }
-
- /* Send streams */
- cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
- if (!stream->fd) {
- continue;
- }
- lkm.cmd_type = LTTNG_CONSUMER_ADD_STREAM;
- lkm.u.stream.channel_key = channel->fd;
- lkm.u.stream.stream_key = stream->fd;
- lkm.u.stream.state = stream->state;
- lkm.u.stream.output = channel->channel->attr.output;
- lkm.u.stream.mmap_len = 0; /* for kernel */
- lkm.u.stream.uid = uid;
- lkm.u.stream.gid = gid;
- strncpy(lkm.u.stream.path_name, stream->pathname, PATH_MAX - 1);
- lkm.u.stream.path_name[PATH_MAX - 1] = '\0';
- DBG("Sending stream %d to consumer", lkm.u.stream.stream_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- ret = lttcomm_send_fds_unix_sock(sock, &stream->fd, 1);
- if (ret < 0) {
- PERROR("send consumer stream ancillary data");
- goto error;
- }
- }
-
- DBG("consumer channel streams sent");
-
- return 0;
-
-error:
- return ret;
-}
-
-/*
- * Send all stream fds of the kernel session to the consumer.
- */
-static int send_kconsumer_session_streams(struct consumer_data *consumer_data,
- struct ltt_kernel_session *session)
-{
- int ret;
- struct ltt_kernel_channel *chan;
- struct lttcomm_consumer_msg lkm;
- int sock = session->consumer_fd;
-
- DBG("Sending metadata stream fd");
-
- /* Extra protection. It's NOT supposed to be set to 0 at this point */
- if (session->consumer_fd == 0) {
- session->consumer_fd = consumer_data->cmd_sock;
- }
-
- if (session->metadata_stream_fd != 0) {
- /* Send metadata channel fd */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
- lkm.u.channel.channel_key = session->metadata->fd;
- lkm.u.channel.max_sb_size = session->metadata->conf->attr.subbuf_size;
- lkm.u.channel.mmap_len = 0; /* for kernel */
- DBG("Sending metadata channel %d to consumer", lkm.u.stream.stream_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer channel");
- goto error;
- }
-
- /* Send metadata stream fd */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_STREAM;
- lkm.u.stream.channel_key = session->metadata->fd;
- lkm.u.stream.stream_key = session->metadata_stream_fd;
- lkm.u.stream.state = LTTNG_CONSUMER_ACTIVE_STREAM;
- lkm.u.stream.output = DEFAULT_KERNEL_CHANNEL_OUTPUT;
- lkm.u.stream.mmap_len = 0; /* for kernel */
- lkm.u.stream.uid = session->uid;
- lkm.u.stream.gid = session->gid;
- strncpy(lkm.u.stream.path_name, session->metadata->pathname, PATH_MAX - 1);
- lkm.u.stream.path_name[PATH_MAX - 1] = '\0';
- DBG("Sending metadata stream %d to consumer", lkm.u.stream.stream_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- ret = lttcomm_send_fds_unix_sock(sock, &session->metadata_stream_fd, 1);
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- }
-
- cds_list_for_each_entry(chan, &session->channel_list.head, list) {
- ret = send_kconsumer_channel_streams(consumer_data, sock, chan,
- session->uid, session->gid);
- if (ret < 0) {
- goto error;
- }
- }
-
- DBG("consumer fds (metadata and channel streams) sent");
-
- return 0;
-
-error:
- return ret;
-}
-
/*
* Notify UST applications using the shm mmap futex.
*/
{
int ret = 0;
struct ltt_session *session;
+ struct ltt_kernel_session *ksess;
struct ltt_kernel_channel *channel;
DBG("Updating kernel streams for channel fd %d", fd);
session_unlock(session);
continue;
}
+ ksess = session->kernel_session;
- /* This is not suppose to be 0 but this is an extra security check */
- if (session->kernel_session->consumer_fd == 0) {
- session->kernel_session->consumer_fd = consumer_data->cmd_sock;
- }
-
- cds_list_for_each_entry(channel,
- &session->kernel_session->channel_list.head, list) {
+ cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
if (channel->fd == fd) {
DBG("Channel found, updating kernel streams");
ret = kernel_open_channel_stream(channel);
if (ret < 0) {
goto error;
}
+ /* Update the stream global counter */
+ ksess->stream_count_global += ret;
/*
* Have we already sent fds to the consumer? If yes, it means
* that tracing is started so it is safe to send our updated
* stream fds.
*/
- if (session->kernel_session->consumer_fds_sent == 1) {
- ret = send_kconsumer_channel_streams(consumer_data,
- session->kernel_session->consumer_fd, channel,
- session->uid, session->gid);
- if (ret < 0) {
- goto error;
+ if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
+ struct lttng_ht_iter iter;
+ struct consumer_socket *socket;
+
+ rcu_read_lock();
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht,
+ &iter.iter, socket, node.node) {
+ pthread_mutex_lock(socket->lock);
+ ret = kernel_consumer_send_channel_stream(socket,
+ channel, ksess,
+ session->output_traces ? 1 : 0);
+ pthread_mutex_unlock(socket->lock);
+ if (ret < 0) {
+ rcu_read_unlock();
+ goto error;
+ }
}
+ rcu_read_unlock();
}
goto error;
}
}
/*
- * For each tracing session, update newly registered apps.
+ * For each tracing session, update newly registered apps. The session list
+ * lock MUST be acquired before calling this.
*/
static void update_ust_app(int app_sock)
{
struct ltt_session *sess, *stmp;
- session_lock_list();
+ /* Consumer is in an ERROR state. Stop any application update. */
+ if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
+ /* Stop the update process since the consumer is dead. */
+ return;
+ }
/* For all tracing session(s) */
cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
}
session_unlock(sess);
}
-
- session_unlock_list();
}
/*
*/
static void *thread_manage_kernel(void *data)
{
- int ret, i, pollfd, update_poll_flag = 1;
+ int ret, i, pollfd, update_poll_flag = 1, err = -1;
uint32_t revents, nb_fd;
char tmp;
struct lttng_poll_event events;
- DBG("Thread manage kernel started");
+ DBG("[thread] Thread manage kernel started");
- ret = create_thread_poll_set(&events, 2);
- if (ret < 0) {
- goto error_poll_create;
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
+
+ /*
+ * This first step of the while is to clean this structure which could free
+ * non NULL pointers so initialize it before the loop.
+ */
+ lttng_poll_init(&events);
+
+ if (testpoint(thread_manage_kernel)) {
+ goto error_testpoint;
}
- ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
- if (ret < 0) {
- goto error;
+ health_code_update();
+
+ if (testpoint(thread_manage_kernel_before_loop)) {
+ goto error_testpoint;
}
while (1) {
+ health_code_update();
+
if (update_poll_flag == 1) {
- /*
- * Reset number of fd in the poll set. Always 2 since there is the thread
- * quit pipe and the kernel pipe.
- */
- events.nb_fd = 2;
+ /* Clean events object. We are about to populate it again. */
+ lttng_poll_clean(&events);
+ ret = sessiond_set_thread_pollset(&events, 2);
+ if (ret < 0) {
+ goto error_poll_create;
+ }
+
+ ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* This will add the available kernel channel if any. */
ret = update_kernel_poll(&events);
if (ret < 0) {
goto error;
update_poll_flag = 0;
}
- nb_fd = LTTNG_POLL_GETNB(&events);
-
- DBG("Thread kernel polling on %d fds", nb_fd);
-
- /* Zeroed the poll events */
- lttng_poll_reset(&events);
+ DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
/* Poll infinite value of time */
restart:
+ health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
continue;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update();
+
/* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Check for data on kernel pipe */
if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
- ret = read(kernel_poll_pipe[0], &tmp, 1);
+ (void) lttng_read(kernel_poll_pipe[0],
+ &tmp, 1);
+ /*
+ * Ret value is useless here, if this pipe gets any actions an
+ * update is required anyway.
+ */
update_poll_flag = 1;
continue;
} else {
}
}
+exit:
error:
lttng_poll_clean(&events);
error_poll_create:
+error_testpoint:
+ utils_close_pipe(kernel_poll_pipe);
+ kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ WARN("Kernel thread died unexpectedly. "
+ "Kernel tracing can continue but CPU hotplug is disabled.");
+ }
+ health_unregister(health_sessiond);
DBG("Kernel thread dying");
return NULL;
}
+/*
+ * Signal pthread condition of the consumer data that the thread.
+ */
+static void signal_consumer_condition(struct consumer_data *data, int state)
+{
+ pthread_mutex_lock(&data->cond_mutex);
+
+ /*
+ * The state is set before signaling. It can be any value, it's the waiter
+ * job to correctly interpret this condition variable associated to the
+ * consumer pthread_cond.
+ *
+ * A value of 0 means that the corresponding thread of the consumer data
+ * was not started. 1 indicates that the thread has started and is ready
+ * for action. A negative value means that there was an error during the
+ * thread bootstrap.
+ */
+ data->consumer_thread_is_ready = state;
+ (void) pthread_cond_signal(&data->cond);
+
+ pthread_mutex_unlock(&data->cond_mutex);
+}
+
/*
* This thread manage the consumer error sent back to the session daemon.
*/
static void *thread_manage_consumer(void *data)
{
- int sock = -1, i, ret, pollfd;
+ int sock = -1, i, ret, pollfd, err = -1;
uint32_t revents, nb_fd;
enum lttcomm_return_code code;
struct lttng_poll_event events;
DBG("[thread] Manage consumer started");
- ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
- if (ret < 0) {
- goto error_listen;
- }
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
+
+ health_code_update();
/*
- * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
- * Nothing more will be added to this poll set.
+ * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
+ * metadata_sock. Nothing more will be added to this poll set.
*/
- ret = create_thread_poll_set(&events, 2);
+ ret = sessiond_set_thread_pollset(&events, 3);
if (ret < 0) {
goto error_poll;
}
+ /*
+ * The error socket here is already in a listening state which was done
+ * just before spawning this thread to avoid a race between the consumer
+ * daemon exec trying to connect and the listen() call.
+ */
ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
if (ret < 0) {
goto error;
}
- nb_fd = LTTNG_POLL_GETNB(&events);
+ health_code_update();
- /* Inifinite blocking call, waiting for transmission */
+ /* Infinite blocking call, waiting for transmission */
restart:
- ret = lttng_poll_wait(&events, -1);
+ health_poll_entry();
+
+ if (testpoint(thread_manage_consumer)) {
+ goto error;
+ }
+
+ ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
goto error;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update();
+
/* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
goto error;
}
+ /*
+ * Set the CLOEXEC flag. Return code is useless because either way, the
+ * show must go on.
+ */
+ (void) utils_set_fd_cloexec(sock);
+
+ health_code_update();
+
DBG2("Receiving code from consumer err_sock");
/* Getting status code from kconsumerd */
goto error;
}
- if (code == CONSUMERD_COMMAND_SOCK_READY) {
+ health_code_update();
+
+ if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
+ /* Connect both socket, command and metadata. */
consumer_data->cmd_sock =
lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
- if (consumer_data->cmd_sock < 0) {
- sem_post(&consumer_data->sem);
- PERROR("consumer connect");
+ consumer_data->metadata_fd =
+ lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
+ if (consumer_data->cmd_sock < 0
+ || consumer_data->metadata_fd < 0) {
+ PERROR("consumer connect cmd socket");
+ /* On error, signal condition and quit. */
+ signal_consumer_condition(consumer_data, -1);
+ goto error;
+ }
+ consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
+ /* Create metadata socket lock. */
+ consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
+ if (consumer_data->metadata_sock.lock == NULL) {
+ PERROR("zmalloc pthread mutex");
+ ret = -1;
goto error;
}
- /* Signal condition to tell that the kconsumerd is ready */
- sem_post(&consumer_data->sem);
- DBG("consumer command socket ready");
+ pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
+
+ signal_consumer_condition(consumer_data, 1);
+ DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
+ DBG("Consumer metadata socket ready (fd: %d)",
+ consumer_data->metadata_fd);
} else {
ERR("consumer error when waiting for SOCK_READY : %s",
lttcomm_get_readable_code(-code));
goto error;
}
- /* Remove the kconsumerd error sock since we've established a connexion */
+ /* Remove the consumerd error sock since we've established a connexion */
ret = lttng_poll_del(&events, consumer_data->err_sock);
if (ret < 0) {
goto error;
}
+ /* Add new accepted error socket. */
ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
if (ret < 0) {
goto error;
}
- /* Update number of fd */
- nb_fd = LTTNG_POLL_GETNB(&events);
-
- /* Inifinite blocking call, waiting for transmission */
-restart_poll:
- ret = lttng_poll_wait(&events, -1);
+ /* Add metadata socket that is successfully connected. */
+ ret = lttng_poll_add(&events, consumer_data->metadata_fd,
+ LPOLLIN | LPOLLRDHUP);
if (ret < 0) {
- /*
- * Restart interrupted system call.
- */
- if (errno == EINTR) {
- goto restart_poll;
- }
goto error;
}
- for (i = 0; i < nb_fd; i++) {
- /* Fetch once the poll data */
- revents = LTTNG_POLL_GETEV(&events, i);
- pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update();
- /* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
- if (ret) {
+ /* Infinite blocking call, waiting for transmission */
+restart_poll:
+ while (1) {
+ health_poll_entry();
+ ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
+ if (ret < 0) {
+ /*
+ * Restart interrupted system call.
+ */
+ if (errno == EINTR) {
+ goto restart_poll;
+ }
goto error;
}
- /* Event on the kconsumerd socket */
- if (pollfd == sock) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
- ERR("consumer err socket second poll error");
+ nb_fd = ret;
+
+ for (i = 0; i < nb_fd; i++) {
+ /* Fetch once the poll data */
+ revents = LTTNG_POLL_GETEV(&events, i);
+ pollfd = LTTNG_POLL_GETFD(&events, i);
+
+ health_code_update();
+
+ /* Thread quit pipe has been closed. Killing thread. */
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
+ if (ret) {
+ err = 0;
+ goto exit;
+ }
+
+ if (pollfd == sock) {
+ /* Event on the consumerd socket */
+ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ ERR("consumer err socket second poll error");
+ goto error;
+ }
+ health_code_update();
+ /* Wait for any kconsumerd error */
+ ret = lttcomm_recv_unix_sock(sock, &code,
+ sizeof(enum lttcomm_return_code));
+ if (ret <= 0) {
+ ERR("consumer closed the command socket");
+ goto error;
+ }
+
+ ERR("consumer return code : %s",
+ lttcomm_get_readable_code(-code));
+
+ goto exit;
+ } else if (pollfd == consumer_data->metadata_fd) {
+ /* UST metadata requests */
+ ret = ust_consumer_metadata_request(
+ &consumer_data->metadata_sock);
+ if (ret < 0) {
+ ERR("Handling metadata request");
+ goto error;
+ }
+ break;
+ } else {
+ ERR("Unknown pollfd");
goto error;
}
}
+ health_code_update();
}
- /* Wait for any kconsumerd error */
- ret = lttcomm_recv_unix_sock(sock, &code,
- sizeof(enum lttcomm_return_code));
- if (ret <= 0) {
- ERR("consumer closed the command socket");
- goto error;
+exit:
+error:
+ /*
+ * We lock here because we are about to close the sockets and some other
+ * thread might be using them so get exclusive access which will abort all
+ * other consumer command by other threads.
+ */
+ pthread_mutex_lock(&consumer_data->lock);
+
+ /* Immediately set the consumerd state to stopped */
+ if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
+ uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
+ } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
+ consumer_data->type == LTTNG_CONSUMER32_UST) {
+ uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
+ } else {
+ /* Code flow error... */
+ assert(0);
}
- ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
-
-error:
if (consumer_data->err_sock >= 0) {
ret = close(consumer_data->err_sock);
if (ret) {
PERROR("close");
}
+ consumer_data->err_sock = -1;
}
if (consumer_data->cmd_sock >= 0) {
ret = close(consumer_data->cmd_sock);
if (ret) {
PERROR("close");
}
+ consumer_data->cmd_sock = -1;
+ }
+ if (*consumer_data->metadata_sock.fd_ptr >= 0) {
+ ret = close(*consumer_data->metadata_sock.fd_ptr);
+ if (ret) {
+ PERROR("close");
+ }
}
+
if (sock >= 0) {
ret = close(sock);
if (ret) {
unlink(consumer_data->err_unix_sock_path);
unlink(consumer_data->cmd_unix_sock_path);
consumer_data->pid = 0;
+ pthread_mutex_unlock(&consumer_data->lock);
+
+ /* Cleanup metadata socket mutex. */
+ pthread_mutex_destroy(consumer_data->metadata_sock.lock);
+ free(consumer_data->metadata_sock.lock);
lttng_poll_clean(&events);
error_poll:
-error_listen:
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_unregister(health_sessiond);
DBG("consumer thread cleanup completed");
return NULL;
*/
static void *thread_manage_apps(void *data)
{
- int i, ret, pollfd;
+ int i, ret, pollfd, err = -1;
+ ssize_t size_ret;
uint32_t revents, nb_fd;
- struct ust_command ust_cmd;
struct lttng_poll_event events;
DBG("[thread] Manage application started");
rcu_register_thread();
rcu_thread_online();
- ret = create_thread_poll_set(&events, 2);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
+
+ if (testpoint(thread_manage_apps)) {
+ goto error_testpoint;
+ }
+
+ health_code_update();
+
+ ret = sessiond_set_thread_pollset(&events, 2);
if (ret < 0) {
goto error_poll_create;
}
goto error;
}
- while (1) {
- /* Zeroed the events structure */
- lttng_poll_reset(&events);
+ if (testpoint(thread_manage_apps_before_loop)) {
+ goto error;
+ }
- nb_fd = LTTNG_POLL_GETNB(&events);
+ health_code_update();
- DBG("Apps thread polling on %d fds", nb_fd);
+ while (1) {
+ DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
goto error;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update();
+
/* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Inspect the apps cmd pipe */
ERR("Apps command pipe error");
goto error;
} else if (revents & LPOLLIN) {
+ int sock;
+
/* Empty pipe */
- ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
- if (ret < 0 || ret < sizeof(ust_cmd)) {
+ size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
+ if (size_ret < sizeof(sock)) {
PERROR("read apps cmd pipe");
goto error;
}
- /* Register applicaton to the session daemon */
- ret = ust_app_register(&ust_cmd.reg_msg,
- ust_cmd.sock);
- if (ret == -ENOMEM) {
- goto error;
- } else if (ret < 0) {
- break;
- }
+ health_code_update();
/*
- * Validate UST version compatibility.
+ * We only monitor the error events of the socket. This
+ * thread does not handle any incoming data from UST
+ * (POLLIN).
*/
- ret = ust_app_validate_version(ust_cmd.sock);
- if (ret >= 0) {
- /*
- * Add channel(s) and event(s) to newly registered apps
- * from lttng global UST domain.
- */
- update_ust_app(ust_cmd.sock);
- }
-
- ret = ust_app_register_done(ust_cmd.sock);
+ ret = lttng_poll_add(&events, sock,
+ LPOLLERR | LPOLLHUP | LPOLLRDHUP);
if (ret < 0) {
- /*
- * If the registration is not possible, we simply
- * unregister the apps and continue
- */
- ust_app_unregister(ust_cmd.sock);
- } else {
- /*
- * We just need here to monitor the close of the UST
- * socket and poll set monitor those by default.
- * Listen on POLLIN (even if we never expect any
- * data) to ensure that hangup wakes us.
- */
- ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
- if (ret < 0) {
- goto error;
- }
-
- DBG("Apps with sock %d added to poll set",
- ust_cmd.sock);
+ goto error;
}
- break;
+ DBG("Apps with sock %d added to poll set", sock);
}
} else {
/*
/* Socket closed on remote end. */
ust_app_unregister(pollfd);
- break;
}
}
+
+ health_code_update();
}
}
+exit:
error:
lttng_poll_clean(&events);
error_poll_create:
+error_testpoint:
+ utils_close_pipe(apps_cmd_pipe);
+ apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
+
+ /*
+ * We don't clean the UST app hash table here since already registered
+ * applications can still be controlled so let them be until the session
+ * daemon dies or the applications stop.
+ */
+
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_unregister(health_sessiond);
DBG("Application communication apps thread cleanup complete");
rcu_thread_offline();
rcu_unregister_thread();
return NULL;
}
+/*
+ * Send a socket to a thread This is called from the dispatch UST registration
+ * thread once all sockets are set for the application.
+ *
+ * The sock value can be invalid, we don't really care, the thread will handle
+ * it and make the necessary cleanup if so.
+ *
+ * On success, return 0 else a negative value being the errno message of the
+ * write().
+ */
+static int send_socket_to_thread(int fd, int sock)
+{
+ ssize_t ret;
+
+ /*
+ * It's possible that the FD is set as invalid with -1 concurrently just
+ * before calling this function being a shutdown state of the thread.
+ */
+ if (fd < 0) {
+ ret = -EBADF;
+ goto error;
+ }
+
+ ret = lttng_write(fd, &sock, sizeof(sock));
+ if (ret < sizeof(sock)) {
+ PERROR("write apps pipe %d", fd);
+ if (ret < 0) {
+ ret = -errno;
+ }
+ goto error;
+ }
+
+ /* All good. Don't send back the write positive ret value. */
+ ret = 0;
+error:
+ return (int) ret;
+}
+
+/*
+ * Sanitize the wait queue of the dispatch registration thread meaning removing
+ * invalid nodes from it. This is to avoid memory leaks for the case the UST
+ * notify socket is never received.
+ */
+static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
+{
+ int ret, nb_fd = 0, i;
+ unsigned int fd_added = 0;
+ struct lttng_poll_event events;
+ struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
+
+ assert(wait_queue);
+
+ lttng_poll_init(&events);
+
+ /* Just skip everything for an empty queue. */
+ if (!wait_queue->count) {
+ goto end;
+ }
+
+ ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
+ if (ret < 0) {
+ goto error_create;
+ }
+
+ cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+ &wait_queue->head, head) {
+ assert(wait_node->app);
+ ret = lttng_poll_add(&events, wait_node->app->sock,
+ LPOLLHUP | LPOLLERR);
+ if (ret < 0) {
+ goto error;
+ }
+
+ fd_added = 1;
+ }
+
+ if (!fd_added) {
+ goto end;
+ }
+
+ /*
+ * Poll but don't block so we can quickly identify the faulty events and
+ * clean them afterwards from the wait queue.
+ */
+ ret = lttng_poll_wait(&events, 0);
+ if (ret < 0) {
+ goto error;
+ }
+ nb_fd = ret;
+
+ for (i = 0; i < nb_fd; i++) {
+ /* Get faulty FD. */
+ uint32_t revents = LTTNG_POLL_GETEV(&events, i);
+ int pollfd = LTTNG_POLL_GETFD(&events, i);
+
+ cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+ &wait_queue->head, head) {
+ if (pollfd == wait_node->app->sock &&
+ (revents & (LPOLLHUP | LPOLLERR))) {
+ cds_list_del(&wait_node->head);
+ wait_queue->count--;
+ ust_app_destroy(wait_node->app);
+ free(wait_node);
+ break;
+ }
+ }
+ }
+
+ if (nb_fd > 0) {
+ DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
+ }
+
+end:
+ lttng_poll_clean(&events);
+ return;
+
+error:
+ lttng_poll_clean(&events);
+error_create:
+ ERR("Unable to sanitize wait queue");
+ return;
+}
+
/*
* Dispatch request from the registration threads to the application
* communication thread.
*/
static void *thread_dispatch_ust_registration(void *data)
{
- int ret;
+ int ret, err = -1;
struct cds_wfq_node *node;
struct ust_command *ust_cmd = NULL;
+ struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
+ struct ust_reg_wait_queue wait_queue = {
+ .count = 0,
+ };
+
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
+
+ health_code_update();
+
+ CDS_INIT_LIST_HEAD(&wait_queue.head);
DBG("[thread] Dispatch UST command started");
- while (!dispatch_thread_exit) {
+ while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ health_code_update();
+
/* Atomically prepare the queue futex */
futex_nto1_prepare(&ust_cmd_queue.futex);
do {
+ struct ust_app *app = NULL;
+ ust_cmd = NULL;
+
+ /*
+ * Make sure we don't have node(s) that have hung up before receiving
+ * the notify socket. This is to clean the list in order to avoid
+ * memory leaks from notify socket that are never seen.
+ */
+ sanitize_wait_queue(&wait_queue);
+
+ health_code_update();
/* Dequeue command for registration */
node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
if (node == NULL) {
ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
ust_cmd->sock, ust_cmd->reg_msg.name,
ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
- /*
- * Inform apps thread of the new application registration. This
- * call is blocking so we can be assured that the data will be read
- * at some point in time or wait to the end of the world :)
- */
- ret = write(apps_cmd_pipe[1], ust_cmd,
- sizeof(struct ust_command));
- if (ret < 0) {
- PERROR("write apps cmd pipe");
- if (errno == EBADF) {
+
+ if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
+ wait_node = zmalloc(sizeof(*wait_node));
+ if (!wait_node) {
+ PERROR("zmalloc wait_node dispatch");
+ ret = close(ust_cmd->sock);
+ if (ret < 0) {
+ PERROR("close ust sock dispatch %d", ust_cmd->sock);
+ }
+ lttng_fd_put(LTTNG_FD_APPS, 1);
+ free(ust_cmd);
+ goto error;
+ }
+ CDS_INIT_LIST_HEAD(&wait_node->head);
+
+ /* Create application object if socket is CMD. */
+ wait_node->app = ust_app_create(&ust_cmd->reg_msg,
+ ust_cmd->sock);
+ if (!wait_node->app) {
+ ret = close(ust_cmd->sock);
+ if (ret < 0) {
+ PERROR("close ust sock dispatch %d", ust_cmd->sock);
+ }
+ lttng_fd_put(LTTNG_FD_APPS, 1);
+ free(wait_node);
+ free(ust_cmd);
+ continue;
+ }
+ /*
+ * Add application to the wait queue so we can set the notify
+ * socket before putting this object in the global ht.
+ */
+ cds_list_add(&wait_node->head, &wait_queue.head);
+ wait_queue.count++;
+
+ free(ust_cmd);
+ /*
+ * We have to continue here since we don't have the notify
+ * socket and the application MUST be added to the hash table
+ * only at that moment.
+ */
+ continue;
+ } else {
+ /*
+ * Look for the application in the local wait queue and set the
+ * notify socket if found.
+ */
+ cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+ &wait_queue.head, head) {
+ health_code_update();
+ if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
+ wait_node->app->notify_sock = ust_cmd->sock;
+ cds_list_del(&wait_node->head);
+ wait_queue.count--;
+ app = wait_node->app;
+ free(wait_node);
+ DBG3("UST app notify socket %d is set", ust_cmd->sock);
+ break;
+ }
+ }
+
+ /*
+ * With no application at this stage the received socket is
+ * basically useless so close it before we free the cmd data
+ * structure for good.
+ */
+ if (!app) {
+ ret = close(ust_cmd->sock);
+ if (ret < 0) {
+ PERROR("close ust sock dispatch %d", ust_cmd->sock);
+ }
+ lttng_fd_put(LTTNG_FD_APPS, 1);
+ }
+ free(ust_cmd);
+ }
+
+ if (app) {
+ /*
+ * @session_lock_list
+ *
+ * Lock the global session list so from the register up to the
+ * registration done message, no thread can see the application
+ * and change its state.
+ */
+ session_lock_list();
+ rcu_read_lock();
+
+ /*
+ * Add application to the global hash table. This needs to be
+ * done before the update to the UST registry can locate the
+ * application.
+ */
+ ust_app_add(app);
+
+ /* Set app version. This call will print an error if needed. */
+ (void) ust_app_version(app);
+
+ /* Send notify socket through the notify pipe. */
+ ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
+ app->notify_sock);
+ if (ret < 0) {
+ rcu_read_unlock();
+ session_unlock_list();
+ /*
+ * No notify thread, stop the UST tracing. However, this is
+ * not an internal error of the this thread thus setting
+ * the health error code to a normal exit.
+ */
+ err = 0;
+ goto error;
+ }
+
+ /*
+ * Update newly registered application with the tracing
+ * registry info already enabled information.
+ */
+ update_ust_app(app->sock);
+
+ /*
+ * Don't care about return value. Let the manage apps threads
+ * handle app unregistration upon socket close.
+ */
+ (void) ust_app_register_done(app->sock);
+
+ /*
+ * Even if the application socket has been closed, send the app
+ * to the thread and unregistration will take place at that
+ * place.
+ */
+ ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
+ if (ret < 0) {
+ rcu_read_unlock();
+ session_unlock_list();
/*
- * We can't inform the application thread to process
- * registration. We will exit or else application
- * registration will not occur and tracing will never
- * start.
+ * No apps. thread, stop the UST tracing. However, this is
+ * not an internal error of the this thread thus setting
+ * the health error code to a normal exit.
*/
+ err = 0;
goto error;
}
+
+ rcu_read_unlock();
+ session_unlock_list();
}
- free(ust_cmd);
} while (node != NULL);
+ health_poll_entry();
/* Futex wait on queue. Blocking call on futex() */
futex_nto1_wait(&ust_cmd_queue.futex);
+ health_poll_exit();
}
+ /* Normal exit, no error */
+ err = 0;
error:
+ /* Clean up wait queue. */
+ cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+ &wait_queue.head, head) {
+ cds_list_del(&wait_node->head);
+ wait_queue.count--;
+ free(wait_node);
+ }
+
DBG("Dispatch thread dying");
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_unregister(health_sessiond);
return NULL;
}
*/
static void *thread_registration_apps(void *data)
{
- int sock = -1, i, ret, pollfd;
+ int sock = -1, i, ret, pollfd, err = -1;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
/*
DBG("[thread] Manage application registration started");
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
+
+ if (testpoint(thread_registration_apps)) {
+ goto error_testpoint;
+ }
+
ret = lttcomm_listen_unix_sock(apps_sock);
if (ret < 0) {
goto error_listen;
* Pass 2 as size here for the thread quit pipe and apps socket. Nothing
* more will be added to this poll set.
*/
- ret = create_thread_poll_set(&events, 2);
+ ret = sessiond_set_thread_pollset(&events, 2);
if (ret < 0) {
goto error_create_poll;
}
while (1) {
DBG("Accepting application registration");
- nb_fd = LTTNG_POLL_GETNB(&events);
-
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
goto error;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
+ health_code_update();
+
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
/* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
goto error;
}
- /* Create UST registration command for enqueuing */
- ust_cmd = zmalloc(sizeof(struct ust_command));
- if (ust_cmd == NULL) {
- PERROR("ust command zmalloc");
+ /*
+ * Set socket timeout for both receiving and ending.
+ * app_socket_timeout is in seconds, whereas
+ * lttcomm_setsockopt_rcv_timeout and
+ * lttcomm_setsockopt_snd_timeout expect msec as
+ * parameter.
+ */
+ (void) lttcomm_setsockopt_rcv_timeout(sock,
+ app_socket_timeout * 1000);
+ (void) lttcomm_setsockopt_snd_timeout(sock,
+ app_socket_timeout * 1000);
+
+ /*
+ * Set the CLOEXEC flag. Return code is useless because
+ * either way, the show must go on.
+ */
+ (void) utils_set_fd_cloexec(sock);
+
+ /* Create UST registration command for enqueuing */
+ ust_cmd = zmalloc(sizeof(struct ust_command));
+ if (ust_cmd == NULL) {
+ PERROR("ust command zmalloc");
goto error;
}
* Using message-based transmissions to ensure we don't
* have to deal with partially received messages.
*/
- ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
- sizeof(struct ust_register_msg));
- if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
- if (ret < 0) {
- PERROR("lttcomm_recv_unix_sock register apps");
- } else {
- ERR("Wrong size received on apps register");
+ ret = lttng_fd_get(LTTNG_FD_APPS, 1);
+ if (ret < 0) {
+ ERR("Exhausted file descriptors allowed for applications.");
+ free(ust_cmd);
+ ret = close(sock);
+ if (ret) {
+ PERROR("close");
}
+ sock = -1;
+ continue;
+ }
+
+ health_code_update();
+ ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
+ if (ret < 0) {
free(ust_cmd);
+ /* Close socket of the application. */
ret = close(sock);
if (ret) {
PERROR("close");
}
+ lttng_fd_put(LTTNG_FD_APPS, 1);
sock = -1;
continue;
}
+ health_code_update();
ust_cmd->sock = sock;
+ sock = -1;
DBG("UST registration received with pid:%d ppid:%d uid:%d"
" gid:%d sock:%d name:%s (version %d.%d)",
}
}
+exit:
error:
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+
/* Notify that the registration thread is gone */
notify_ust_apps(0);
PERROR("close");
}
}
- if (clock >= 0) {
+ if (sock >= 0) {
ret = close(sock);
if (ret) {
PERROR("close");
}
+ lttng_fd_put(LTTNG_FD_APPS, 1);
}
unlink(apps_unix_sock_path);
lttng_poll_clean(&events);
error_listen:
error_create_poll:
+error_testpoint:
DBG("UST Registration thread cleanup complete");
+ health_unregister(health_sessiond);
return NULL;
}
*/
static int spawn_consumer_thread(struct consumer_data *consumer_data)
{
- int ret;
+ int ret, clock_ret;
struct timespec timeout;
- timeout.tv_sec = DEFAULT_SEM_WAIT_TIMEOUT;
- timeout.tv_nsec = 0;
+ /* Make sure we set the readiness flag to 0 because we are NOT ready */
+ consumer_data->consumer_thread_is_ready = 0;
- /* Setup semaphore */
- ret = sem_init(&consumer_data->sem, 0, 0);
- if (ret < 0) {
- PERROR("sem_init consumer semaphore");
+ /* Setup pthread condition */
+ ret = pthread_condattr_init(&consumer_data->condattr);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_condattr_init consumer data");
+ goto error;
+ }
+
+ /*
+ * Set the monotonic clock in order to make sure we DO NOT jump in time
+ * between the clock_gettime() call and the timedwait call. See bug #324
+ * for a more details and how we noticed it.
+ */
+ ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_condattr_setclock consumer data");
+ goto error;
+ }
+
+ ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_cond_init consumer data");
goto error;
}
- ret = pthread_create(&consumer_data->thread, NULL,
- thread_manage_consumer, consumer_data);
+ ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
+ consumer_data);
if (ret != 0) {
PERROR("pthread_create consumer");
ret = -1;
goto error;
}
+ /* We are about to wait on a pthread condition */
+ pthread_mutex_lock(&consumer_data->cond_mutex);
+
/* Get time for sem_timedwait absolute timeout */
- ret = clock_gettime(CLOCK_REALTIME, &timeout);
- if (ret < 0) {
- PERROR("clock_gettime spawn consumer");
- /* Infinite wait for the kconsumerd thread to be ready */
- ret = sem_wait(&consumer_data->sem);
- } else {
- /* Normal timeout if the gettime was successful */
- timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
- ret = sem_timedwait(&consumer_data->sem, &timeout);
+ clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
+ /*
+ * Set the timeout for the condition timed wait even if the clock gettime
+ * call fails since we might loop on that call and we want to avoid to
+ * increment the timeout too many times.
+ */
+ timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
+
+ /*
+ * The following loop COULD be skipped in some conditions so this is why we
+ * set ret to 0 in order to make sure at least one round of the loop is
+ * done.
+ */
+ ret = 0;
+
+ /*
+ * Loop until the condition is reached or when a timeout is reached. Note
+ * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
+ * be returned but the pthread_cond(3), from the glibc-doc, says that it is
+ * possible. This loop does not take any chances and works with both of
+ * them.
+ */
+ while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
+ if (clock_ret < 0) {
+ PERROR("clock_gettime spawn consumer");
+ /* Infinite wait for the consumerd thread to be ready */
+ ret = pthread_cond_wait(&consumer_data->cond,
+ &consumer_data->cond_mutex);
+ } else {
+ ret = pthread_cond_timedwait(&consumer_data->cond,
+ &consumer_data->cond_mutex, &timeout);
+ }
}
- if (ret < 0) {
- if (errno == ETIMEDOUT) {
+ /* Release the pthread condition */
+ pthread_mutex_unlock(&consumer_data->cond_mutex);
+
+ if (ret != 0) {
+ errno = ret;
+ if (ret == ETIMEDOUT) {
/*
* Call has timed out so we kill the kconsumerd_thread and return
* an error.
*/
- ERR("The consumer thread was never ready. Killing it");
+ ERR("Condition timed out. The consumer thread was never ready."
+ " Killing it");
ret = pthread_cancel(consumer_data->thread);
if (ret < 0) {
PERROR("pthread_cancel consumer thread");
}
} else {
- PERROR("semaphore wait failed consumer thread");
+ PERROR("pthread_cond_wait failed consumer thread");
}
goto error;
}
pthread_mutex_lock(&consumer_data->pid_mutex);
if (consumer_data->pid == 0) {
- ERR("Kconsumerd did not start");
+ ERR("Consumerd did not start");
pthread_mutex_unlock(&consumer_data->pid_mutex);
goto error;
}
static int join_consumer_thread(struct consumer_data *consumer_data)
{
void *status;
- int ret;
- if (consumer_data->pid != 0) {
+ /* Consumer pid must be a real one. */
+ if (consumer_data->pid > 0) {
+ int ret;
ret = kill(consumer_data->pid, SIGTERM);
if (ret) {
ERR("Error killing consumer daemon");
"lttng-consumerd", verbosity, "-k",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
break;
case LTTNG_CONSUMER64_UST:
ret = putenv(tmpnew);
if (ret) {
ret = -errno;
+ free(tmpnew);
goto error;
}
}
ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
if (consumerd64_libdir[0] != '\0') {
free(tmpnew);
ret = putenv(tmpnew);
if (ret) {
ret = -errno;
+ free(tmpnew);
goto error;
}
}
ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
if (consumerd32_libdir[0] != '\0') {
free(tmpnew);
{
int ret;
+ /*
+ * Set the listen() state on the socket since there is a possible race
+ * between the exec() of the consumer daemon and this call if place in the
+ * consumer thread. See bug #366 for more details.
+ */
+ ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
+ if (ret < 0) {
+ goto error;
+ }
+
pthread_mutex_lock(&consumer_data->pid_mutex);
if (consumer_data->pid != 0) {
pthread_mutex_unlock(&consumer_data->pid_mutex);
return 0;
error:
- return ret;
-}
+ /* Cleanup already created sockets on error. */
+ if (consumer_data->err_sock >= 0) {
+ int err;
-/*
- * Check version of the lttng-modules.
- */
-static int validate_lttng_modules_version(void)
-{
- return kernel_validate_version(kernel_tracer_fd);
+ err = close(consumer_data->err_sock);
+ if (err < 0) {
+ PERROR("close consumer data error socket");
+ }
+ }
+ return ret;
}
/*
}
/* Validate kernel version */
- ret = validate_lttng_modules_version();
+ ret = kernel_validate_version(kernel_tracer_fd);
if (ret < 0) {
goto error_version;
}
PERROR("close");
}
kernel_tracer_fd = -1;
- return LTTCOMM_KERN_VERSION;
+ return LTTNG_ERR_KERN_VERSION;
error_modules:
ret = close(kernel_tracer_fd);
error:
WARN("No kernel tracer available");
kernel_tracer_fd = -1;
- return LTTCOMM_KERN_NA;
+ if (!is_root) {
+ return LTTNG_ERR_NEED_ROOT_SESSIOND;
+ } else {
+ return LTTNG_ERR_KERN_NA;
+ }
}
+
/*
- * Init tracing by creating trace directory and sending fds kernel consumer.
+ * Copy consumer output from the tracing session to the domain session. The
+ * function also applies the right modification on a per domain basis for the
+ * trace files destination directory.
+ *
+ * Should *NOT* be called with RCU read-side lock held.
*/
-static int init_kernel_tracing(struct ltt_kernel_session *session)
+static int copy_session_consumer(int domain, struct ltt_session *session)
{
- int ret = 0;
+ int ret;
+ const char *dir_name;
+ struct consumer_output *consumer;
- if (session->consumer_fds_sent == 0) {
+ assert(session);
+ assert(session->consumer);
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ DBG3("Copying tracing session consumer output in kernel session");
/*
- * Assign default kernel consumer socket if no consumer assigned to the
- * kernel session. At this point, it's NOT suppose to be 0 but this is
- * an extra security check.
+ * XXX: We should audit the session creation and what this function
+ * does "extra" in order to avoid a destroy since this function is used
+ * in the domain session creation (kernel and ust) only. Same for UST
+ * domain.
*/
- if (session->consumer_fd == 0) {
- session->consumer_fd = kconsumer_data.cmd_sock;
- }
+ if (session->kernel_session->consumer) {
+ consumer_destroy_output(session->kernel_session->consumer);
+ }
+ session->kernel_session->consumer =
+ consumer_copy_output(session->consumer);
+ /* Ease our life a bit for the next part */
+ consumer = session->kernel_session->consumer;
+ dir_name = DEFAULT_KERNEL_TRACE_DIR;
+ break;
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_UST:
+ DBG3("Copying tracing session consumer output in UST session");
+ if (session->ust_session->consumer) {
+ consumer_destroy_output(session->ust_session->consumer);
+ }
+ session->ust_session->consumer =
+ consumer_copy_output(session->consumer);
+ /* Ease our life a bit for the next part */
+ consumer = session->ust_session->consumer;
+ dir_name = DEFAULT_UST_TRACE_DIR;
+ break;
+ default:
+ ret = LTTNG_ERR_UNKNOWN_DOMAIN;
+ goto error;
+ }
- ret = send_kconsumer_session_streams(&kconsumer_data, session);
- if (ret < 0) {
- ret = LTTCOMM_KERN_CONSUMER_FAIL;
- goto error;
- }
+ /* Append correct directory to subdir */
+ strncat(consumer->subdir, dir_name,
+ sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
+ DBG3("Copy session consumer subdir %s", consumer->subdir);
- session->consumer_fds_sent = 1;
- }
+ ret = LTTNG_OK;
error:
return ret;
/*
* Create an UST session and add it to the session ust list.
+ *
+ * Should *NOT* be called with RCU read-side lock held.
*/
static int create_ust_session(struct ltt_session *session,
struct lttng_domain *domain)
{
- struct ltt_ust_session *lus = NULL;
int ret;
+ struct ltt_ust_session *lus = NULL;
+
+ assert(session);
+ assert(domain);
+ assert(session->consumer);
switch (domain->type) {
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
break;
default:
- ret = LTTCOMM_UNKNOWN_DOMAIN;
+ ERR("Unknown UST domain on create session %d", domain->type);
+ ret = LTTNG_ERR_UNKNOWN_DOMAIN;
goto error;
}
DBG("Creating UST session");
- lus = trace_ust_create_session(session->path, session->id, domain);
+ lus = trace_ust_create_session(session->id);
if (lus == NULL) {
- ret = LTTCOMM_UST_SESS_FAIL;
+ ret = LTTNG_ERR_UST_SESS_FAIL;
goto error;
}
- ret = run_as_mkdir_recursive(lus->pathname, S_IRWXU | S_IRWXG,
- session->uid, session->gid);
- if (ret < 0) {
- if (ret != -EEXIST) {
- ERR("Trace directory creation error");
- ret = LTTCOMM_UST_SESS_FAIL;
- goto error;
- }
- }
-
- /* The domain type dictate different actions on session creation */
- switch (domain->type) {
- case LTTNG_DOMAIN_UST:
- /* No ustctl for the global UST domain */
- break;
- default:
- ERR("Unknown UST domain on create session %d", domain->type);
- goto error;
- }
lus->uid = session->uid;
lus->gid = session->gid;
+ lus->output_traces = session->output_traces;
+ lus->snapshot_mode = session->snapshot_mode;
+ lus->live_timer_interval = session->live_timer;
session->ust_session = lus;
- return LTTCOMM_OK;
+ /* Copy session output to the newly created UST session */
+ ret = copy_session_consumer(domain->type, session);
+ if (ret != LTTNG_OK) {
+ goto error;
+ }
+
+ return LTTNG_OK;
error:
free(lus);
+ session->ust_session = NULL;
return ret;
}
ret = kernel_create_session(session, kernel_tracer_fd);
if (ret < 0) {
- ret = LTTCOMM_KERN_SESS_FAIL;
+ ret = LTTNG_ERR_KERN_SESS_FAIL;
goto error;
}
- /* Set kernel consumer socket fd */
- if (kconsumer_data.cmd_sock) {
- session->kernel_session->consumer_fd = kconsumer_data.cmd_sock;
+ /* Code flow safety */
+ assert(session->kernel_session);
+
+ /* Copy session output to the newly created Kernel session */
+ ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
+ if (ret != LTTNG_OK) {
+ goto error;
}
- ret = run_as_mkdir_recursive(session->kernel_session->trace_path,
- S_IRWXU | S_IRWXG, session->uid, session->gid);
- if (ret < 0) {
- if (ret != -EEXIST) {
- ERR("Trace directory creation error");
- goto error;
+ /* Create directory(ies) on local filesystem. */
+ if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
+ strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
+ ret = run_as_mkdir_recursive(
+ session->kernel_session->consumer->dst.trace_path,
+ S_IRWXU | S_IRWXG, session->uid, session->gid);
+ if (ret < 0) {
+ if (ret != -EEXIST) {
+ ERR("Trace directory creation error");
+ goto error;
+ }
}
}
+
session->kernel_session->uid = session->uid;
session->kernel_session->gid = session->gid;
+ session->kernel_session->output_traces = session->output_traces;
+ session->kernel_session->snapshot_mode = session->snapshot_mode;
+
+ return LTTNG_OK;
error:
+ trace_kernel_destroy_session(session->kernel_session);
+ session->kernel_session = NULL;
return ret;
}
/*
- * Check if the UID or GID match the session. Root user has access to all
- * sessions.
+ * Count number of session permitted by uid/gid.
*/
-static int session_access_ok(struct ltt_session *session, uid_t uid, gid_t gid)
-{
- if (uid != session->uid && gid != session->gid && uid != 0) {
- return 0;
- } else {
- return 1;
- }
-}
-
static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
{
unsigned int i = 0;
struct ltt_session *session;
DBG("Counting number of available session for UID %d GID %d",
- uid, gid);
+ uid, gid);
cds_list_for_each_entry(session, &session_list_ptr->head, list) {
/*
* Only list the sessions the user can control.
}
/*
- * Using the session list, filled a lttng_session array to send back to the
- * client for session listing.
+ * Process the command requested by the lttng client within the command
+ * context structure. This function make sure that the return structure (llm)
+ * is set and ready for transmission before returning.
*
- * The session list lock MUST be acquired before calling this function. Use
- * session_lock_list() and session_unlock_list().
+ * Return any error encountered or 0 for success.
+ *
+ * "sock" is only used for special-case var. len data.
+ *
+ * Should *NOT* be called with RCU read-side lock held.
*/
-static void list_lttng_sessions(struct lttng_session *sessions, uid_t uid,
- gid_t gid)
+static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
+ int *sock_error)
{
- unsigned int i = 0;
- struct ltt_session *session;
-
- DBG("Getting all available session for UID %d GID %d",
- uid, gid);
- /*
- * Iterate over session list and append data after the control struct in
- * the buffer.
- */
- cds_list_for_each_entry(session, &session_list_ptr->head, list) {
- /*
- * Only list the sessions the user can control.
- */
- if (!session_access_ok(session, uid, gid)) {
- continue;
- }
- strncpy(sessions[i].path, session->path, PATH_MAX);
- sessions[i].path[PATH_MAX - 1] = '\0';
- strncpy(sessions[i].name, session->name, NAME_MAX);
- sessions[i].name[NAME_MAX - 1] = '\0';
- sessions[i].enabled = session->enabled;
- i++;
- }
-}
+ int ret = LTTNG_OK;
+ int need_tracing_session = 1;
+ int need_domain;
-/*
- * Fill lttng_channel array of all channels.
- */
-static void list_lttng_channels(int domain, struct ltt_session *session,
- struct lttng_channel *channels)
-{
- int i = 0;
- struct ltt_kernel_channel *kchan;
+ DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
- DBG("Listing channels for session %s", session->name);
+ *sock_error = 0;
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- /* Kernel channels */
- if (session->kernel_session != NULL) {
- cds_list_for_each_entry(kchan,
- &session->kernel_session->channel_list.head, list) {
- /* Copy lttng_channel struct to array */
- memcpy(&channels[i], kchan->channel, sizeof(struct lttng_channel));
- channels[i].enabled = kchan->enabled;
- i++;
- }
- }
- break;
- case LTTNG_DOMAIN_UST:
- {
- struct lttng_ht_iter iter;
- struct ltt_ust_channel *uchan;
-
- cds_lfht_for_each_entry(session->ust_session->domain_global.channels->ht,
- &iter.iter, uchan, node.node) {
- strncpy(channels[i].name, uchan->name, LTTNG_SYMBOL_NAME_LEN);
- channels[i].attr.overwrite = uchan->attr.overwrite;
- channels[i].attr.subbuf_size = uchan->attr.subbuf_size;
- channels[i].attr.num_subbuf = uchan->attr.num_subbuf;
- channels[i].attr.switch_timer_interval =
- uchan->attr.switch_timer_interval;
- channels[i].attr.read_timer_interval =
- uchan->attr.read_timer_interval;
- channels[i].enabled = uchan->enabled;
- switch (uchan->attr.output) {
- case LTTNG_UST_MMAP:
- default:
- channels[i].attr.output = LTTNG_EVENT_MMAP;
- break;
- }
- i++;
- }
+ switch (cmd_ctx->lsm->cmd_type) {
+ case LTTNG_CREATE_SESSION:
+ case LTTNG_CREATE_SESSION_SNAPSHOT:
+ case LTTNG_CREATE_SESSION_LIVE:
+ case LTTNG_DESTROY_SESSION:
+ case LTTNG_LIST_SESSIONS:
+ case LTTNG_LIST_DOMAINS:
+ case LTTNG_START_TRACE:
+ case LTTNG_STOP_TRACE:
+ case LTTNG_DATA_PENDING:
+ case LTTNG_SNAPSHOT_ADD_OUTPUT:
+ case LTTNG_SNAPSHOT_DEL_OUTPUT:
+ case LTTNG_SNAPSHOT_LIST_OUTPUT:
+ case LTTNG_SNAPSHOT_RECORD:
+ need_domain = 0;
break;
- }
default:
- break;
- }
-}
-
-/*
- * Create a list of ust global domain events.
- */
-static int list_lttng_ust_global_events(char *channel_name,
- struct ltt_ust_domain_global *ust_global, struct lttng_event **events)
-{
- int i = 0, ret = 0;
- unsigned int nb_event = 0;
- struct lttng_ht_iter iter;
- struct lttng_ht_node_str *node;
- struct ltt_ust_channel *uchan;
- struct ltt_ust_event *uevent;
- struct lttng_event *tmp;
-
- DBG("Listing UST global events for channel %s", channel_name);
-
- rcu_read_lock();
-
- lttng_ht_lookup(ust_global->channels, (void *)channel_name, &iter);
- node = lttng_ht_iter_get_node_str(&iter);
- if (node == NULL) {
- ret = -LTTCOMM_UST_CHAN_NOT_FOUND;
- goto error;
+ need_domain = 1;
}
- uchan = caa_container_of(&node->node, struct ltt_ust_channel, node.node);
-
- nb_event += lttng_ht_get_count(uchan->events);
-
- if (nb_event == 0) {
- ret = nb_event;
+ if (opt_no_kernel && need_domain
+ && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
+ if (!is_root) {
+ ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
+ } else {
+ ret = LTTNG_ERR_KERN_NA;
+ }
goto error;
}
- DBG3("Listing UST global %d events", nb_event);
-
- tmp = zmalloc(nb_event * sizeof(struct lttng_event));
- if (tmp == NULL) {
- ret = -LTTCOMM_FATAL;
- goto error;
+ /* Deny register consumer if we already have a spawned consumer. */
+ if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
+ pthread_mutex_lock(&kconsumer_data.pid_mutex);
+ if (kconsumer_data.pid > 0) {
+ ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ pthread_mutex_unlock(&kconsumer_data.pid_mutex);
+ goto error;
+ }
+ pthread_mutex_unlock(&kconsumer_data.pid_mutex);
}
- cds_lfht_for_each_entry(uchan->events->ht, &iter.iter, uevent, node.node) {
- strncpy(tmp[i].name, uevent->attr.name, LTTNG_SYMBOL_NAME_LEN);
- tmp[i].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
- tmp[i].enabled = uevent->enabled;
- switch (uevent->attr.instrumentation) {
- case LTTNG_UST_TRACEPOINT:
- tmp[i].type = LTTNG_EVENT_TRACEPOINT;
- break;
- case LTTNG_UST_PROBE:
- tmp[i].type = LTTNG_EVENT_PROBE;
- break;
- case LTTNG_UST_FUNCTION:
- tmp[i].type = LTTNG_EVENT_FUNCTION;
- break;
- }
- tmp[i].loglevel = uevent->attr.loglevel;
- switch (uevent->attr.loglevel_type) {
- case LTTNG_UST_LOGLEVEL_ALL:
- tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_ALL;
- break;
- case LTTNG_UST_LOGLEVEL_RANGE:
- tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_RANGE;
- break;
- case LTTNG_UST_LOGLEVEL_SINGLE:
- tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_SINGLE;
- break;
+ /*
+ * Check for command that don't needs to allocate a returned payload. We do
+ * this here so we don't have to make the call for no payload at each
+ * command.
+ */
+ switch(cmd_ctx->lsm->cmd_type) {
+ case LTTNG_LIST_SESSIONS:
+ case LTTNG_LIST_TRACEPOINTS:
+ case LTTNG_LIST_TRACEPOINT_FIELDS:
+ case LTTNG_LIST_DOMAINS:
+ case LTTNG_LIST_CHANNELS:
+ case LTTNG_LIST_EVENTS:
+ break;
+ default:
+ /* Setup lttng message with no payload */
+ ret = setup_lttng_msg(cmd_ctx, 0);
+ if (ret < 0) {
+ /* This label does not try to unlock the session */
+ goto init_setup_error;
}
- i++;
}
- ret = nb_event;
- *events = tmp;
-
-error:
- rcu_read_unlock();
- return ret;
-}
-
-/*
- * Fill lttng_event array of all kernel events in the channel.
- */
-static int list_lttng_kernel_events(char *channel_name,
- struct ltt_kernel_session *kernel_session, struct lttng_event **events)
-{
- int i = 0, ret;
- unsigned int nb_event;
- struct ltt_kernel_event *event;
- struct ltt_kernel_channel *kchan;
-
- kchan = trace_kernel_get_channel_by_name(channel_name, kernel_session);
- if (kchan == NULL) {
- ret = LTTCOMM_KERN_CHAN_NOT_FOUND;
- goto error;
+ /* Commands that DO NOT need a session. */
+ switch (cmd_ctx->lsm->cmd_type) {
+ case LTTNG_CREATE_SESSION:
+ case LTTNG_CREATE_SESSION_SNAPSHOT:
+ case LTTNG_CREATE_SESSION_LIVE:
+ case LTTNG_CALIBRATE:
+ case LTTNG_LIST_SESSIONS:
+ case LTTNG_LIST_TRACEPOINTS:
+ case LTTNG_LIST_TRACEPOINT_FIELDS:
+ need_tracing_session = 0;
+ break;
+ default:
+ DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
+ /*
+ * We keep the session list lock across _all_ commands
+ * for now, because the per-session lock does not
+ * handle teardown properly.
+ */
+ session_lock_list();
+ cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
+ if (cmd_ctx->session == NULL) {
+ ret = LTTNG_ERR_SESS_NOT_FOUND;
+ goto error;
+ } else {
+ /* Acquire lock for the session */
+ session_lock(cmd_ctx->session);
+ }
+ break;
}
- nb_event = kchan->event_count;
-
- DBG("Listing events for channel %s", kchan->channel->name);
-
- if (nb_event == 0) {
- ret = nb_event;
- goto error;
+ if (!need_domain) {
+ goto skip_domain;
}
- *events = zmalloc(nb_event * sizeof(struct lttng_event));
- if (*events == NULL) {
- ret = LTTCOMM_FATAL;
- goto error;
- }
-
- /* Kernel channels */
- cds_list_for_each_entry(event, &kchan->events_list.head , list) {
- strncpy((*events)[i].name, event->event->name, LTTNG_SYMBOL_NAME_LEN);
- (*events)[i].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
- (*events)[i].enabled = event->enabled;
- switch (event->event->instrumentation) {
- case LTTNG_KERNEL_TRACEPOINT:
- (*events)[i].type = LTTNG_EVENT_TRACEPOINT;
- break;
- case LTTNG_KERNEL_KPROBE:
- case LTTNG_KERNEL_KRETPROBE:
- (*events)[i].type = LTTNG_EVENT_PROBE;
- memcpy(&(*events)[i].attr.probe, &event->event->u.kprobe,
- sizeof(struct lttng_kernel_kprobe));
- break;
- case LTTNG_KERNEL_FUNCTION:
- (*events)[i].type = LTTNG_EVENT_FUNCTION;
- memcpy(&((*events)[i].attr.ftrace), &event->event->u.ftrace,
- sizeof(struct lttng_kernel_function));
- break;
- case LTTNG_KERNEL_NOOP:
- (*events)[i].type = LTTNG_EVENT_NOOP;
- break;
- case LTTNG_KERNEL_SYSCALL:
- (*events)[i].type = LTTNG_EVENT_SYSCALL;
- break;
- case LTTNG_KERNEL_ALL:
- assert(0);
- break;
- }
- i++;
- }
-
- return nb_event;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_DISABLE_CHANNEL processed by the client thread.
- */
-static int cmd_disable_channel(struct ltt_session *session,
- int domain, char *channel_name)
-{
- int ret;
- struct ltt_ust_session *usess;
-
- usess = session->ust_session;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- ret = channel_kernel_disable(session->kernel_session,
- channel_name);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct ltt_ust_channel *uchan;
- struct lttng_ht *chan_ht;
-
- chan_ht = usess->domain_global.channels;
-
- uchan = trace_ust_find_channel_by_name(chan_ht, channel_name);
- if (uchan == NULL) {
- ret = LTTCOMM_UST_CHAN_NOT_FOUND;
- goto error;
- }
-
- ret = channel_ust_disable(usess, domain, uchan);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
-#endif
- default:
- ret = LTTCOMM_UNKNOWN_DOMAIN;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_ENABLE_CHANNEL processed by the client thread.
- */
-static int cmd_enable_channel(struct ltt_session *session,
- int domain, struct lttng_channel *attr)
-{
- int ret;
- struct ltt_ust_session *usess = session->ust_session;
- struct lttng_ht *chan_ht;
-
- DBG("Enabling channel %s for session %s", attr->name, session->name);
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- struct ltt_kernel_channel *kchan;
-
- kchan = trace_kernel_get_channel_by_name(attr->name,
- session->kernel_session);
- if (kchan == NULL) {
- ret = channel_kernel_create(session->kernel_session,
- attr, kernel_poll_pipe[1]);
- } else {
- ret = channel_kernel_enable(session->kernel_session, kchan);
- }
-
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct ltt_ust_channel *uchan;
-
- chan_ht = usess->domain_global.channels;
-
- uchan = trace_ust_find_channel_by_name(chan_ht, attr->name);
- if (uchan == NULL) {
- ret = channel_ust_create(usess, domain, attr);
- } else {
- ret = channel_ust_enable(usess, domain, uchan);
- }
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
-#endif
- default:
- ret = LTTCOMM_UNKNOWN_DOMAIN;
- goto error;
- }
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_DISABLE_EVENT processed by the client thread.
- */
-static int cmd_disable_event(struct ltt_session *session, int domain,
- char *channel_name, char *event_name)
-{
- int ret;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- struct ltt_kernel_channel *kchan;
- struct ltt_kernel_session *ksess;
-
- ksess = session->kernel_session;
-
- kchan = trace_kernel_get_channel_by_name(channel_name, ksess);
- if (kchan == NULL) {
- ret = LTTCOMM_KERN_CHAN_NOT_FOUND;
- goto error;
- }
-
- ret = event_kernel_disable_tracepoint(ksess, kchan, event_name);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct ltt_ust_channel *uchan;
- struct ltt_ust_session *usess;
-
- usess = session->ust_session;
-
- uchan = trace_ust_find_channel_by_name(usess->domain_global.channels,
- channel_name);
- if (uchan == NULL) {
- ret = LTTCOMM_UST_CHAN_NOT_FOUND;
- goto error;
- }
-
- ret = event_ust_disable_tracepoint(usess, domain, uchan, event_name);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- DBG3("Disable UST event %s in channel %s completed", event_name,
- channel_name);
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
-#endif
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_DISABLE_ALL_EVENT processed by the client thread.
- */
-static int cmd_disable_event_all(struct ltt_session *session, int domain,
- char *channel_name)
-{
- int ret;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- struct ltt_kernel_session *ksess;
- struct ltt_kernel_channel *kchan;
-
- ksess = session->kernel_session;
-
- kchan = trace_kernel_get_channel_by_name(channel_name, ksess);
- if (kchan == NULL) {
- ret = LTTCOMM_KERN_CHAN_NOT_FOUND;
- goto error;
- }
-
- ret = event_kernel_disable_all(ksess, kchan);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct ltt_ust_session *usess;
- struct ltt_ust_channel *uchan;
-
- usess = session->ust_session;
-
- uchan = trace_ust_find_channel_by_name(usess->domain_global.channels,
- channel_name);
- if (uchan == NULL) {
- ret = LTTCOMM_UST_CHAN_NOT_FOUND;
- goto error;
- }
-
- ret = event_ust_disable_all_tracepoints(usess, domain, uchan);
- if (ret != 0) {
- goto error;
- }
-
- DBG3("Disable all UST events in channel %s completed", channel_name);
-
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
-#endif
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_ADD_CONTEXT processed by the client thread.
- */
-static int cmd_add_context(struct ltt_session *session, int domain,
- char *channel_name, char *event_name, struct lttng_event_context *ctx)
-{
- int ret;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- /* Add kernel context to kernel tracer */
- ret = context_kernel_add(session->kernel_session, ctx,
- event_name, channel_name);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
- break;
- case LTTNG_DOMAIN_UST:
- {
- struct ltt_ust_session *usess = session->ust_session;
-
- ret = context_ust_add(usess, domain, ctx, event_name, channel_name);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
-#endif
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_ENABLE_EVENT processed by the client thread.
- */
-static int cmd_enable_event(struct ltt_session *session, int domain,
- char *channel_name, struct lttng_event *event)
-{
- int ret;
- struct lttng_channel *attr;
- struct ltt_ust_session *usess = session->ust_session;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- struct ltt_kernel_channel *kchan;
-
- kchan = trace_kernel_get_channel_by_name(channel_name,
- session->kernel_session);
- if (kchan == NULL) {
- attr = channel_new_default_attr(domain);
- if (attr == NULL) {
- ret = LTTCOMM_FATAL;
- goto error;
- }
- snprintf(attr->name, NAME_MAX, "%s", channel_name);
-
- /* This call will notify the kernel thread */
- ret = channel_kernel_create(session->kernel_session,
- attr, kernel_poll_pipe[1]);
- if (ret != LTTCOMM_OK) {
- free(attr);
- goto error;
- }
- free(attr);
- }
-
- /* Get the newly created kernel channel pointer */
- kchan = trace_kernel_get_channel_by_name(channel_name,
- session->kernel_session);
- if (kchan == NULL) {
- /* This sould not happen... */
- ret = LTTCOMM_FATAL;
- goto error;
- }
-
- ret = event_kernel_enable_tracepoint(session->kernel_session, kchan,
- event);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct lttng_channel *attr;
- struct ltt_ust_channel *uchan;
-
- /* Get channel from global UST domain */
- uchan = trace_ust_find_channel_by_name(usess->domain_global.channels,
- channel_name);
- if (uchan == NULL) {
- /* Create default channel */
- attr = channel_new_default_attr(domain);
- if (attr == NULL) {
- ret = LTTCOMM_FATAL;
- goto error;
- }
- snprintf(attr->name, NAME_MAX, "%s", channel_name);
- attr->name[NAME_MAX - 1] = '\0';
-
- ret = channel_ust_create(usess, domain, attr);
- if (ret != LTTCOMM_OK) {
- free(attr);
- goto error;
- }
- free(attr);
-
- /* Get the newly created channel reference back */
- uchan = trace_ust_find_channel_by_name(
- usess->domain_global.channels, channel_name);
- if (uchan == NULL) {
- /* Something is really wrong */
- ret = LTTCOMM_FATAL;
- goto error;
- }
- }
-
- /* At this point, the session and channel exist on the tracer */
- ret = event_ust_enable_tracepoint(usess, domain, uchan, event);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
-#endif
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_ENABLE_ALL_EVENT processed by the client thread.
- */
-static int cmd_enable_event_all(struct ltt_session *session, int domain,
- char *channel_name, int event_type)
-{
- int ret;
- struct ltt_kernel_channel *kchan;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- kchan = trace_kernel_get_channel_by_name(channel_name,
- session->kernel_session);
- if (kchan == NULL) {
- /* This call will notify the kernel thread */
- ret = channel_kernel_create(session->kernel_session, NULL,
- kernel_poll_pipe[1]);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- /* Get the newly created kernel channel pointer */
- kchan = trace_kernel_get_channel_by_name(channel_name,
- session->kernel_session);
- if (kchan == NULL) {
- /* This sould not happen... */
- ret = LTTCOMM_FATAL;
- goto error;
- }
-
- }
-
- switch (event_type) {
- case LTTNG_EVENT_SYSCALL:
- ret = event_kernel_enable_all_syscalls(session->kernel_session,
- kchan, kernel_tracer_fd);
- break;
- case LTTNG_EVENT_TRACEPOINT:
- /*
- * This call enables all LTTNG_KERNEL_TRACEPOINTS and
- * events already registered to the channel.
- */
- ret = event_kernel_enable_all_tracepoints(session->kernel_session,
- kchan, kernel_tracer_fd);
- break;
- case LTTNG_EVENT_ALL:
- /* Enable syscalls and tracepoints */
- ret = event_kernel_enable_all(session->kernel_session,
- kchan, kernel_tracer_fd);
- break;
- default:
- ret = LTTCOMM_KERN_ENABLE_FAIL;
- goto error;
- }
-
- /* Manage return value */
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- break;
- case LTTNG_DOMAIN_UST:
- {
- struct lttng_channel *attr;
- struct ltt_ust_channel *uchan;
- struct ltt_ust_session *usess = session->ust_session;
-
- /* Get channel from global UST domain */
- uchan = trace_ust_find_channel_by_name(usess->domain_global.channels,
- channel_name);
- if (uchan == NULL) {
- /* Create default channel */
- attr = channel_new_default_attr(domain);
- if (attr == NULL) {
- ret = LTTCOMM_FATAL;
- goto error;
- }
- snprintf(attr->name, NAME_MAX, "%s", channel_name);
- attr->name[NAME_MAX - 1] = '\0';
-
- /* Use the internal command enable channel */
- ret = channel_ust_create(usess, domain, attr);
- if (ret != LTTCOMM_OK) {
- free(attr);
- goto error;
- }
- free(attr);
-
- /* Get the newly created channel reference back */
- uchan = trace_ust_find_channel_by_name(
- usess->domain_global.channels, channel_name);
- if (uchan == NULL) {
- /* Something is really wrong */
- ret = LTTCOMM_FATAL;
- goto error;
- }
- }
-
- /* At this point, the session and channel exist on the tracer */
-
- switch (event_type) {
- case LTTNG_EVENT_ALL:
- case LTTNG_EVENT_TRACEPOINT:
- ret = event_ust_enable_all_tracepoints(usess, domain, uchan);
- if (ret != LTTCOMM_OK) {
- goto error;
- }
- break;
- default:
- ret = LTTCOMM_UST_ENABLE_FAIL;
- goto error;
- }
-
- /* Manage return value */
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- break;
- }
-#if 0
- case LTTNG_DOMAIN_UST_EXEC_NAME:
- case LTTNG_DOMAIN_UST_PID:
- case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
-#endif
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_LIST_TRACEPOINTS processed by the client thread.
- */
-static ssize_t cmd_list_tracepoints(int domain, struct lttng_event **events)
-{
- int ret;
- ssize_t nb_events = 0;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- nb_events = kernel_list_events(kernel_tracer_fd, events);
- if (nb_events < 0) {
- ret = LTTCOMM_KERN_LIST_FAIL;
- goto error;
- }
- break;
- case LTTNG_DOMAIN_UST:
- nb_events = ust_app_list_events(events);
- if (nb_events < 0) {
- ret = LTTCOMM_UST_LIST_FAIL;
- goto error;
- }
- break;
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- return nb_events;
-
-error:
- /* Return negative value to differentiate return code */
- return -ret;
-}
-
-/*
- * Command LTTNG_START_TRACE processed by the client thread.
- */
-static int cmd_start_trace(struct ltt_session *session)
-{
- int ret;
- struct ltt_kernel_session *ksession;
- struct ltt_ust_session *usess;
-
- /* Short cut */
- ksession = session->kernel_session;
- usess = session->ust_session;
-
- if (session->enabled) {
- ret = LTTCOMM_UST_START_FAIL;
- goto error;
- }
-
- session->enabled = 1;
-
- /* Kernel tracing */
- if (ksession != NULL) {
- struct ltt_kernel_channel *kchan;
-
- /* Open kernel metadata */
- if (ksession->metadata == NULL) {
- ret = kernel_open_metadata(ksession, ksession->trace_path);
- if (ret < 0) {
- ret = LTTCOMM_KERN_META_FAIL;
- goto error;
- }
- }
-
- /* Open kernel metadata stream */
- if (ksession->metadata_stream_fd == 0) {
- ret = kernel_open_metadata_stream(ksession);
- if (ret < 0) {
- ERR("Kernel create metadata stream failed");
- ret = LTTCOMM_KERN_STREAM_FAIL;
- goto error;
- }
- }
-
- /* For each channel */
- cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) {
- if (kchan->stream_count == 0) {
- ret = kernel_open_channel_stream(kchan);
- if (ret < 0) {
- ret = LTTCOMM_KERN_STREAM_FAIL;
- goto error;
- }
- /* Update the stream global counter */
- ksession->stream_count_global += ret;
- }
- }
-
- /* Setup kernel consumer socket and send fds to it */
- ret = init_kernel_tracing(ksession);
- if (ret < 0) {
- ret = LTTCOMM_KERN_START_FAIL;
- goto error;
- }
-
- /* This start the kernel tracing */
- ret = kernel_start_session(ksession);
- if (ret < 0) {
- ret = LTTCOMM_KERN_START_FAIL;
- goto error;
- }
-
- /* Quiescent wait after starting trace */
- kernel_wait_quiescent(kernel_tracer_fd);
- }
-
- /* Flag session that trace should start automatically */
- if (usess) {
- usess->start_trace = 1;
-
- ret = ust_app_start_trace_all(usess);
- if (ret < 0) {
- ret = LTTCOMM_UST_START_FAIL;
- goto error;
- }
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_STOP_TRACE processed by the client thread.
- */
-static int cmd_stop_trace(struct ltt_session *session)
-{
- int ret;
- struct ltt_kernel_channel *kchan;
- struct ltt_kernel_session *ksession;
- struct ltt_ust_session *usess;
-
- /* Short cut */
- ksession = session->kernel_session;
- usess = session->ust_session;
-
- if (!session->enabled) {
- ret = LTTCOMM_UST_STOP_FAIL;
- goto error;
- }
-
- session->enabled = 0;
-
- /* Kernel tracer */
- if (ksession != NULL) {
- DBG("Stop kernel tracing");
-
- /* Flush all buffers before stopping */
- ret = kernel_metadata_flush_buffer(ksession->metadata_stream_fd);
- if (ret < 0) {
- ERR("Kernel metadata flush failed");
- }
-
- cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) {
- ret = kernel_flush_buffer(kchan);
- if (ret < 0) {
- ERR("Kernel flush buffer error");
- }
- }
-
- ret = kernel_stop_session(ksession);
- if (ret < 0) {
- ret = LTTCOMM_KERN_STOP_FAIL;
- goto error;
- }
-
- kernel_wait_quiescent(kernel_tracer_fd);
- }
-
- if (usess) {
- usess->start_trace = 0;
-
- ret = ust_app_stop_trace_all(usess);
- if (ret < 0) {
- ret = LTTCOMM_UST_STOP_FAIL;
- goto error;
- }
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_CREATE_SESSION processed by the client thread.
- */
-static int cmd_create_session(char *name, char *path, lttng_sock_cred *creds)
-{
- int ret;
-
- ret = session_create(name, path, LTTNG_SOCK_GET_UID_CRED(creds),
- LTTNG_SOCK_GET_GID_CRED(creds));
- if (ret != LTTCOMM_OK) {
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_DESTROY_SESSION processed by the client thread.
- */
-static int cmd_destroy_session(struct ltt_session *session, char *name)
-{
- int ret;
-
- /* Clean kernel session teardown */
- teardown_kernel_session(session);
- /* UST session teardown */
- teardown_ust_session(session);
-
- /*
- * Must notify the kernel thread here to update it's poll setin order
- * to remove the channel(s)' fd just destroyed.
- */
- ret = notify_thread_pipe(kernel_poll_pipe[1]);
- if (ret < 0) {
- PERROR("write kernel poll pipe");
- }
-
- ret = session_destroy(session);
-
- return ret;
-}
-
-/*
- * Command LTTNG_CALIBRATE processed by the client thread.
- */
-static int cmd_calibrate(int domain, struct lttng_calibrate *calibrate)
-{
- int ret;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- {
- struct lttng_kernel_calibrate kcalibrate;
-
- kcalibrate.type = calibrate->type;
- ret = kernel_calibrate(kernel_tracer_fd, &kcalibrate);
- if (ret < 0) {
- ret = LTTCOMM_KERN_ENABLE_FAIL;
- goto error;
- }
- break;
- }
- case LTTNG_DOMAIN_UST:
- {
- struct lttng_ust_calibrate ucalibrate;
-
- ucalibrate.type = calibrate->type;
- ret = ust_app_calibrate_glb(&ucalibrate);
- if (ret < 0) {
- ret = LTTCOMM_UST_CALIBRATE_FAIL;
- goto error;
- }
- break;
- }
- default:
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_REGISTER_CONSUMER processed by the client thread.
- */
-static int cmd_register_consumer(struct ltt_session *session, int domain,
- char *sock_path)
-{
- int ret, sock;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- /* Can't register a consumer if there is already one */
- if (session->kernel_session->consumer_fds_sent != 0) {
- ret = LTTCOMM_KERN_CONSUMER_FAIL;
- goto error;
- }
-
- sock = lttcomm_connect_unix_sock(sock_path);
- if (sock < 0) {
- ret = LTTCOMM_CONNECT_FAIL;
- goto error;
- }
-
- session->kernel_session->consumer_fd = sock;
- break;
- default:
- /* TODO: Userspace tracing */
- ret = LTTCOMM_UND;
- goto error;
- }
-
- ret = LTTCOMM_OK;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_LIST_DOMAINS processed by the client thread.
- */
-static ssize_t cmd_list_domains(struct ltt_session *session,
- struct lttng_domain **domains)
-{
- int ret, index = 0;
- ssize_t nb_dom = 0;
-
- if (session->kernel_session != NULL) {
- DBG3("Listing domains found kernel domain");
- nb_dom++;
- }
-
- if (session->ust_session != NULL) {
- DBG3("Listing domains found UST global domain");
- nb_dom++;
- }
-
- *domains = zmalloc(nb_dom * sizeof(struct lttng_domain));
- if (*domains == NULL) {
- ret = -LTTCOMM_FATAL;
- goto error;
- }
-
- if (session->kernel_session != NULL) {
- (*domains)[index].type = LTTNG_DOMAIN_KERNEL;
- index++;
- }
-
- if (session->ust_session != NULL) {
- (*domains)[index].type = LTTNG_DOMAIN_UST;
- index++;
- }
-
- return nb_dom;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_LIST_CHANNELS processed by the client thread.
- */
-static ssize_t cmd_list_channels(int domain, struct ltt_session *session,
- struct lttng_channel **channels)
-{
- int ret;
- ssize_t nb_chan = 0;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- if (session->kernel_session != NULL) {
- nb_chan = session->kernel_session->channel_count;
- }
- DBG3("Number of kernel channels %zd", nb_chan);
- break;
- case LTTNG_DOMAIN_UST:
- if (session->ust_session != NULL) {
- nb_chan = lttng_ht_get_count(
- session->ust_session->domain_global.channels);
- }
- DBG3("Number of UST global channels %zd", nb_chan);
- break;
- default:
- *channels = NULL;
- ret = -LTTCOMM_UND;
- goto error;
- }
-
- if (nb_chan > 0) {
- *channels = zmalloc(nb_chan * sizeof(struct lttng_channel));
- if (*channels == NULL) {
- ret = -LTTCOMM_FATAL;
- goto error;
- }
-
- list_lttng_channels(domain, session, *channels);
- } else {
- *channels = NULL;
- }
-
- return nb_chan;
-
-error:
- return ret;
-}
-
-/*
- * Command LTTNG_LIST_EVENTS processed by the client thread.
- */
-static ssize_t cmd_list_events(int domain, struct ltt_session *session,
- char *channel_name, struct lttng_event **events)
-{
- int ret = 0;
- ssize_t nb_event = 0;
-
- switch (domain) {
- case LTTNG_DOMAIN_KERNEL:
- if (session->kernel_session != NULL) {
- nb_event = list_lttng_kernel_events(channel_name,
- session->kernel_session, events);
- }
- break;
- case LTTNG_DOMAIN_UST:
- {
- if (session->ust_session != NULL) {
- nb_event = list_lttng_ust_global_events(channel_name,
- &session->ust_session->domain_global, events);
- }
- break;
- }
- default:
- ret = -LTTCOMM_UND;
- goto error;
- }
-
- ret = nb_event;
-
-error:
- return ret;
-}
-
-/*
- * Process the command requested by the lttng client within the command
- * context structure. This function make sure that the return structure (llm)
- * is set and ready for transmission before returning.
- *
- * Return any error encountered or 0 for success.
- */
-static int process_client_msg(struct command_ctx *cmd_ctx)
-{
- int ret = LTTCOMM_OK;
- int need_tracing_session = 1;
- int need_domain;
-
- DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
-
- switch (cmd_ctx->lsm->cmd_type) {
- case LTTNG_CREATE_SESSION:
- case LTTNG_DESTROY_SESSION:
- case LTTNG_LIST_SESSIONS:
- case LTTNG_LIST_DOMAINS:
- case LTTNG_START_TRACE:
- case LTTNG_STOP_TRACE:
- need_domain = 0;
- break;
- default:
- need_domain = 1;
- }
-
- if (opt_no_kernel && need_domain
- && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
- ret = LTTCOMM_KERN_NA;
- goto error;
- }
-
- /*
- * Check for command that don't needs to allocate a returned payload. We do
- * this here so we don't have to make the call for no payload at each
- * command.
- */
- switch(cmd_ctx->lsm->cmd_type) {
- case LTTNG_LIST_SESSIONS:
- case LTTNG_LIST_TRACEPOINTS:
- case LTTNG_LIST_DOMAINS:
- case LTTNG_LIST_CHANNELS:
- case LTTNG_LIST_EVENTS:
- break;
- default:
- /* Setup lttng message with no payload */
- ret = setup_lttng_msg(cmd_ctx, 0);
- if (ret < 0) {
- /* This label does not try to unlock the session */
- goto init_setup_error;
- }
- }
-
- /* Commands that DO NOT need a session. */
- switch (cmd_ctx->lsm->cmd_type) {
- case LTTNG_CREATE_SESSION:
- case LTTNG_CALIBRATE:
- case LTTNG_LIST_SESSIONS:
- case LTTNG_LIST_TRACEPOINTS:
- need_tracing_session = 0;
- break;
- default:
- DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
- session_lock_list();
- cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
- session_unlock_list();
- if (cmd_ctx->session == NULL) {
- if (cmd_ctx->lsm->session.name != NULL) {
- ret = LTTCOMM_SESS_NOT_FOUND;
- } else {
- /* If no session name specified */
- ret = LTTCOMM_SELECT_SESS;
- }
- goto error;
- } else {
- /* Acquire lock for the session */
- session_lock(cmd_ctx->session);
- }
- break;
- }
-
- if (!need_domain) {
- goto skip_domain;
- }
/*
* Check domain type for specific "pre-action".
*/
switch (cmd_ctx->lsm->domain.type) {
case LTTNG_DOMAIN_KERNEL:
if (!is_root) {
- ret = LTTCOMM_KERN_NA;
+ ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
goto error;
}
}
}
+ /* Consumer is in an ERROR state. Report back to client */
+ if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
+ ret = LTTNG_ERR_NO_KERNCONSUMERD;
+ goto error;
+ }
+
/* Need a session for kernel command */
if (need_tracing_session) {
if (cmd_ctx->session->kernel_session == NULL) {
ret = create_kernel_session(cmd_ctx->session);
if (ret < 0) {
- ret = LTTCOMM_KERN_SESS_FAIL;
+ ret = LTTNG_ERR_KERN_SESS_FAIL;
goto error;
}
}
pthread_mutex_unlock(&kconsumer_data.pid_mutex);
ret = start_consumerd(&kconsumer_data);
if (ret < 0) {
- ret = LTTCOMM_KERN_CONSUMER_FAIL;
+ ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
goto error;
}
+ uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
} else {
pthread_mutex_unlock(&kconsumer_data.pid_mutex);
}
+
+ /*
+ * The consumer was just spawned so we need to add the socket to
+ * the consumer output of the session if exist.
+ */
+ ret = consumer_create_socket(&kconsumer_data,
+ cmd_ctx->session->kernel_session->consumer);
+ if (ret < 0) {
+ goto error;
+ }
}
+
break;
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
{
+ if (!ust_app_supported()) {
+ ret = LTTNG_ERR_NO_UST;
+ goto error;
+ }
+ /* Consumer is in an ERROR state. Report back to client */
+ if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
+ ret = LTTNG_ERR_NO_USTCONSUMERD;
+ goto error;
+ }
+
if (need_tracing_session) {
+ /* Create UST session if none exist. */
if (cmd_ctx->session->ust_session == NULL) {
ret = create_ust_session(cmd_ctx->session,
&cmd_ctx->lsm->domain);
- if (ret != LTTCOMM_OK) {
+ if (ret != LTTNG_OK) {
goto error;
}
}
+
/* Start the UST consumer daemons */
/* 64-bit */
pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
ret = start_consumerd(&ustconsumer64_data);
if (ret < 0) {
- ret = LTTCOMM_UST_CONSUMER64_FAIL;
- ust_consumerd64_fd = -EINVAL;
+ ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
+ uatomic_set(&ust_consumerd64_fd, -EINVAL);
goto error;
}
- ust_consumerd64_fd = ustconsumer64_data.cmd_sock;
+ uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
+ uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
} else {
pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
}
+
+ /*
+ * Setup socket for consumer 64 bit. No need for atomic access
+ * since it was set above and can ONLY be set in this thread.
+ */
+ ret = consumer_create_socket(&ustconsumer64_data,
+ cmd_ctx->session->ust_session->consumer);
+ if (ret < 0) {
+ goto error;
+ }
+
/* 32-bit */
if (consumerd32_bin[0] != '\0' &&
ustconsumer32_data.pid == 0 &&
pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
ret = start_consumerd(&ustconsumer32_data);
if (ret < 0) {
- ret = LTTCOMM_UST_CONSUMER32_FAIL;
- ust_consumerd32_fd = -EINVAL;
+ ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
+ uatomic_set(&ust_consumerd32_fd, -EINVAL);
goto error;
}
- ust_consumerd32_fd = ustconsumer32_data.cmd_sock;
+
+ uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
+ uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
} else {
pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
}
+
+ /*
+ * Setup socket for consumer 64 bit. No need for atomic access
+ * since it was set above and can ONLY be set in this thread.
+ */
+ ret = consumer_create_socket(&ustconsumer32_data,
+ cmd_ctx->session->ust_session->consumer);
+ if (ret < 0) {
+ goto error;
+ }
}
break;
}
}
skip_domain:
+ /* Validate consumer daemon state when start/stop trace command */
+ if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
+ cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
+ switch (cmd_ctx->lsm->domain.type) {
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_UST:
+ if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
+ ret = LTTNG_ERR_NO_USTCONSUMERD;
+ goto error;
+ }
+ break;
+ case LTTNG_DOMAIN_KERNEL:
+ if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
+ ret = LTTNG_ERR_NO_KERNCONSUMERD;
+ goto error;
+ }
+ break;
+ }
+ }
+
/*
* Check that the UID or GID match that of the tracing session.
* The root user can interact with all sessions.
if (!session_access_ok(cmd_ctx->session,
LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
- ret = LTTCOMM_EPERM;
+ ret = LTTNG_ERR_EPERM;
+ goto error;
+ }
+ }
+
+ /*
+ * Send relayd information to consumer as soon as we have a domain and a
+ * session defined.
+ */
+ if (cmd_ctx->session && need_domain) {
+ /*
+ * Setup relayd if not done yet. If the relayd information was already
+ * sent to the consumer, this call will gracefully return.
+ */
+ ret = cmd_setup_relayd(cmd_ctx->session);
+ if (ret != LTTNG_OK) {
goto error;
}
}
{
ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
cmd_ctx->lsm->u.context.channel_name,
- cmd_ctx->lsm->u.context.event_name,
- &cmd_ctx->lsm->u.context.ctx);
+ &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
break;
}
case LTTNG_DISABLE_CHANNEL:
}
case LTTNG_ENABLE_CHANNEL:
{
- ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
- &cmd_ctx->lsm->u.channel.chan);
+ ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
+ &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
break;
}
case LTTNG_ENABLE_EVENT:
{
- ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ struct lttng_event_exclusion *exclusion = NULL;
+ struct lttng_filter_bytecode *bytecode = NULL;
+
+ /* Handle exclusion events and receive it from the client. */
+ if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
+ size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
+
+ exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
+ (count * LTTNG_SYMBOL_NAME_LEN));
+ if (!exclusion) {
+ ret = LTTNG_ERR_EXCLUSION_NOMEM;
+ goto error;
+ }
+
+ DBG("Receiving var len exclusion event list from client ...");
+ exclusion->count = count;
+ ret = lttcomm_recv_unix_sock(sock, exclusion->names,
+ count * LTTNG_SYMBOL_NAME_LEN);
+ if (ret <= 0) {
+ DBG("Nothing recv() from client var len data... continuing");
+ *sock_error = 1;
+ free(exclusion);
+ ret = LTTNG_ERR_EXCLUSION_INVAL;
+ goto error;
+ }
+ }
+
+ /* Handle filter and get bytecode from client. */
+ if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
+ size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
+
+ if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
+ ret = LTTNG_ERR_FILTER_INVAL;
+ free(exclusion);
+ goto error;
+ }
+
+ bytecode = zmalloc(bytecode_len);
+ if (!bytecode) {
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_NOMEM;
+ goto error;
+ }
+
+ /* Receive var. len. data */
+ DBG("Receiving var len filter's bytecode from client ...");
+ ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
+ if (ret <= 0) {
+ DBG("Nothing recv() from client car len data... continuing");
+ *sock_error = 1;
+ free(bytecode);
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_INVAL;
+ goto error;
+ }
+
+ if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
+ free(bytecode);
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_INVAL;
+ goto error;
+ }
+ }
+
+ ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
cmd_ctx->lsm->u.enable.channel_name,
- &cmd_ctx->lsm->u.enable.event);
+ &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
+ kernel_poll_pipe[1]);
break;
}
case LTTNG_ENABLE_ALL_EVENT:
{
DBG("Enabling all events");
- ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
cmd_ctx->lsm->u.enable.channel_name,
- cmd_ctx->lsm->u.enable.event.type);
+ cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
break;
}
case LTTNG_LIST_TRACEPOINTS:
struct lttng_event *events;
ssize_t nb_events;
- nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
- if (nb_events < 0) {
- ret = -nb_events;
+ nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
+ if (nb_events < 0) {
+ /* Return value is a negative lttng_error_code. */
+ ret = -nb_events;
+ goto error;
+ }
+
+ /*
+ * Setup lttng message with payload size set to the event list size in
+ * bytes and then copy list into the llm payload.
+ */
+ ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
+ if (ret < 0) {
+ free(events);
+ goto setup_error;
+ }
+
+ /* Copy event list into message payload */
+ memcpy(cmd_ctx->llm->payload, events,
+ sizeof(struct lttng_event) * nb_events);
+
+ free(events);
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_LIST_TRACEPOINT_FIELDS:
+ {
+ struct lttng_event_field *fields;
+ ssize_t nb_fields;
+
+ nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
+ &fields);
+ if (nb_fields < 0) {
+ /* Return value is a negative lttng_error_code. */
+ ret = -nb_fields;
+ goto error;
+ }
+
+ /*
+ * Setup lttng message with payload size set to the event list size in
+ * bytes and then copy list into the llm payload.
+ */
+ ret = setup_lttng_msg(cmd_ctx,
+ sizeof(struct lttng_event_field) * nb_fields);
+ if (ret < 0) {
+ free(fields);
+ goto setup_error;
+ }
+
+ /* Copy event list into message payload */
+ memcpy(cmd_ctx->llm->payload, fields,
+ sizeof(struct lttng_event_field) * nb_fields);
+
+ free(fields);
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_SET_CONSUMER_URI:
+ {
+ size_t nb_uri, len;
+ struct lttng_uri *uris;
+
+ nb_uri = cmd_ctx->lsm->u.uri.size;
+ len = nb_uri * sizeof(struct lttng_uri);
+
+ if (nb_uri == 0) {
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+
+ uris = zmalloc(len);
+ if (uris == NULL) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ /* Receive variable len data */
+ DBG("Receiving %zu URI(s) from client ...", nb_uri);
+ ret = lttcomm_recv_unix_sock(sock, uris, len);
+ if (ret <= 0) {
+ DBG("No URIs received from client... continuing");
+ *sock_error = 1;
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+
+ ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
+ nb_uri, uris);
+ if (ret != LTTNG_OK) {
+ free(uris);
goto error;
}
/*
- * Setup lttng message with payload size set to the event list size in
- * bytes and then copy list into the llm payload.
+ * XXX: 0 means that this URI should be applied on the session. Should
+ * be a DOMAIN enuam.
*/
- ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
- if (ret < 0) {
- free(events);
- goto setup_error;
+ if (cmd_ctx->lsm->domain.type == 0) {
+ /* Add the URI for the UST session if a consumer is present. */
+ if (cmd_ctx->session->ust_session &&
+ cmd_ctx->session->ust_session->consumer) {
+ ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
+ nb_uri, uris);
+ } else if (cmd_ctx->session->kernel_session &&
+ cmd_ctx->session->kernel_session->consumer) {
+ ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
+ cmd_ctx->session, nb_uri, uris);
+ }
}
- /* Copy event list into message payload */
- memcpy(cmd_ctx->llm->payload, events,
- sizeof(struct lttng_event) * nb_events);
-
- free(events);
+ free(uris);
- ret = LTTCOMM_OK;
break;
}
case LTTNG_START_TRACE:
}
case LTTNG_CREATE_SESSION:
{
- ret = cmd_create_session(cmd_ctx->lsm->session.name,
- cmd_ctx->lsm->session.path, &cmd_ctx->creds);
+ size_t nb_uri, len;
+ struct lttng_uri *uris = NULL;
+
+ nb_uri = cmd_ctx->lsm->u.uri.size;
+ len = nb_uri * sizeof(struct lttng_uri);
+
+ if (nb_uri > 0) {
+ uris = zmalloc(len);
+ if (uris == NULL) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ /* Receive variable len data */
+ DBG("Waiting for %zu URIs from client ...", nb_uri);
+ ret = lttcomm_recv_unix_sock(sock, uris, len);
+ if (ret <= 0) {
+ DBG("No URIs received from client... continuing");
+ *sock_error = 1;
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+
+ if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
+ DBG("Creating session with ONE network URI is a bad call");
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+ }
+
+ ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
+ &cmd_ctx->creds, 0);
+
+ free(uris);
+
break;
}
case LTTNG_DESTROY_SESSION:
{
- ret = cmd_destroy_session(cmd_ctx->session,
- cmd_ctx->lsm->session.name);
+ ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
+
+ /* Set session to NULL so we do not unlock it after free. */
+ cmd_ctx->session = NULL;
break;
}
case LTTNG_LIST_DOMAINS:
nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
if (nb_dom < 0) {
+ /* Return value is a negative lttng_error_code. */
ret = -nb_dom;
goto error;
}
ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
if (ret < 0) {
+ free(domains);
goto setup_error;
}
free(domains);
- ret = LTTCOMM_OK;
+ ret = LTTNG_OK;
break;
}
case LTTNG_LIST_CHANNELS:
{
- size_t nb_chan;
+ int nb_chan;
struct lttng_channel *channels;
nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
cmd_ctx->session, &channels);
if (nb_chan < 0) {
+ /* Return value is a negative lttng_error_code. */
ret = -nb_chan;
goto error;
}
ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
if (ret < 0) {
+ free(channels);
goto setup_error;
}
free(channels);
- ret = LTTCOMM_OK;
+ ret = LTTNG_OK;
break;
}
case LTTNG_LIST_EVENTS:
nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
cmd_ctx->lsm->u.list.channel_name, &events);
if (nb_event < 0) {
+ /* Return value is a negative lttng_error_code. */
ret = -nb_event;
goto error;
}
- ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
- if (ret < 0) {
- goto setup_error;
+ ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
+ if (ret < 0) {
+ free(events);
+ goto setup_error;
+ }
+
+ /* Copy event list into message payload */
+ memcpy(cmd_ctx->llm->payload, events,
+ nb_event * sizeof(struct lttng_event));
+
+ free(events);
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_LIST_SESSIONS:
+ {
+ unsigned int nr_sessions;
+
+ session_lock_list();
+ nr_sessions = lttng_sessions_count(
+ LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
+ LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
+
+ ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
+ if (ret < 0) {
+ session_unlock_list();
+ goto setup_error;
+ }
+
+ /* Filled the session array */
+ cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
+ LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
+ LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
+
+ session_unlock_list();
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_CALIBRATE:
+ {
+ ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
+ &cmd_ctx->lsm->u.calibrate);
+ break;
+ }
+ case LTTNG_REGISTER_CONSUMER:
+ {
+ struct consumer_data *cdata;
+
+ switch (cmd_ctx->lsm->domain.type) {
+ case LTTNG_DOMAIN_KERNEL:
+ cdata = &kconsumer_data;
+ break;
+ default:
+ ret = LTTNG_ERR_UND;
+ goto error;
+ }
+
+ ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.reg.path, cdata);
+ break;
+ }
+ case LTTNG_DATA_PENDING:
+ {
+ ret = cmd_data_pending(cmd_ctx->session);
+ break;
+ }
+ case LTTNG_SNAPSHOT_ADD_OUTPUT:
+ {
+ struct lttcomm_lttng_output_id reply;
+
+ ret = cmd_snapshot_add_output(cmd_ctx->session,
+ &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
+ if (ret != LTTNG_OK) {
+ goto error;
+ }
+
+ ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
+ if (ret < 0) {
+ goto setup_error;
+ }
+
+ /* Copy output list into message payload */
+ memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_SNAPSHOT_DEL_OUTPUT:
+ {
+ ret = cmd_snapshot_del_output(cmd_ctx->session,
+ &cmd_ctx->lsm->u.snapshot_output.output);
+ break;
+ }
+ case LTTNG_SNAPSHOT_LIST_OUTPUT:
+ {
+ ssize_t nb_output;
+ struct lttng_snapshot_output *outputs = NULL;
+
+ nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
+ if (nb_output < 0) {
+ ret = -nb_output;
+ goto error;
+ }
+
+ ret = setup_lttng_msg(cmd_ctx,
+ nb_output * sizeof(struct lttng_snapshot_output));
+ if (ret < 0) {
+ free(outputs);
+ goto setup_error;
+ }
+
+ if (outputs) {
+ /* Copy output list into message payload */
+ memcpy(cmd_ctx->llm->payload, outputs,
+ nb_output * sizeof(struct lttng_snapshot_output));
+ free(outputs);
+ }
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_SNAPSHOT_RECORD:
+ {
+ ret = cmd_snapshot_record(cmd_ctx->session,
+ &cmd_ctx->lsm->u.snapshot_record.output,
+ cmd_ctx->lsm->u.snapshot_record.wait);
+ break;
+ }
+ case LTTNG_CREATE_SESSION_SNAPSHOT:
+ {
+ size_t nb_uri, len;
+ struct lttng_uri *uris = NULL;
+
+ nb_uri = cmd_ctx->lsm->u.uri.size;
+ len = nb_uri * sizeof(struct lttng_uri);
+
+ if (nb_uri > 0) {
+ uris = zmalloc(len);
+ if (uris == NULL) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ /* Receive variable len data */
+ DBG("Waiting for %zu URIs from client ...", nb_uri);
+ ret = lttcomm_recv_unix_sock(sock, uris, len);
+ if (ret <= 0) {
+ DBG("No URIs received from client... continuing");
+ *sock_error = 1;
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+
+ if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
+ DBG("Creating session with ONE network URI is a bad call");
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+ }
+
+ ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
+ nb_uri, &cmd_ctx->creds);
+ free(uris);
+ break;
+ }
+ case LTTNG_CREATE_SESSION_LIVE:
+ {
+ size_t nb_uri, len;
+ struct lttng_uri *uris = NULL;
+
+ nb_uri = cmd_ctx->lsm->u.uri.size;
+ len = nb_uri * sizeof(struct lttng_uri);
+
+ if (nb_uri > 0) {
+ uris = zmalloc(len);
+ if (uris == NULL) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ /* Receive variable len data */
+ DBG("Waiting for %zu URIs from client ...", nb_uri);
+ ret = lttcomm_recv_unix_sock(sock, uris, len);
+ if (ret <= 0) {
+ DBG("No URIs received from client... continuing");
+ *sock_error = 1;
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+
+ if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
+ DBG("Creating session with ONE network URI is a bad call");
+ ret = LTTNG_ERR_SESSION_FAIL;
+ free(uris);
+ goto error;
+ }
+ }
+
+ ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
+ nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
+ free(uris);
+ break;
+ }
+ default:
+ ret = LTTNG_ERR_UND;
+ break;
+ }
+
+error:
+ if (cmd_ctx->llm == NULL) {
+ DBG("Missing llm structure. Allocating one.");
+ if (setup_lttng_msg(cmd_ctx, 0) < 0) {
+ goto setup_error;
+ }
+ }
+ /* Set return code */
+ cmd_ctx->llm->ret_code = ret;
+setup_error:
+ if (cmd_ctx->session) {
+ session_unlock(cmd_ctx->session);
+ }
+ if (need_tracing_session) {
+ session_unlock_list();
+ }
+init_setup_error:
+ return ret;
+}
+
+/*
+ * Thread managing health check socket.
+ */
+static void *thread_manage_health(void *data)
+{
+ int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
+ uint32_t revents, nb_fd;
+ struct lttng_poll_event events;
+ struct health_comm_msg msg;
+ struct health_comm_reply reply;
+
+ DBG("[thread] Manage health check started");
+
+ rcu_register_thread();
+
+ /* We might hit an error path before this is created. */
+ lttng_poll_init(&events);
+
+ /* Create unix socket */
+ sock = lttcomm_create_unix_sock(health_unix_sock_path);
+ if (sock < 0) {
+ ERR("Unable to create health check Unix socket");
+ ret = -1;
+ goto error;
+ }
+
+ if (is_root) {
+ /* lttng health client socket path permissions */
+ ret = chown(health_unix_sock_path, 0,
+ utils_get_group_id(tracing_group_name));
+ if (ret < 0) {
+ ERR("Unable to set group on %s", health_unix_sock_path);
+ PERROR("chown");
+ ret = -1;
+ goto error;
+ }
+
+ ret = chmod(health_unix_sock_path,
+ S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
+ if (ret < 0) {
+ ERR("Unable to set permissions on %s", health_unix_sock_path);
+ PERROR("chmod");
+ ret = -1;
+ goto error;
+ }
+ }
+
+ /*
+ * Set the CLOEXEC flag. Return code is useless because either way, the
+ * show must go on.
+ */
+ (void) utils_set_fd_cloexec(sock);
+
+ ret = lttcomm_listen_unix_sock(sock);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /*
+ * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
+ * more will be added to this poll set.
+ */
+ ret = sessiond_set_thread_pollset(&events, 2);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Add the application registration socket */
+ ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
+ if (ret < 0) {
+ goto error;
+ }
+
+ while (1) {
+ DBG("Health check ready");
+
+ /* Inifinite blocking call, waiting for transmission */
+restart:
+ ret = lttng_poll_wait(&events, -1);
+ if (ret < 0) {
+ /*
+ * Restart interrupted system call.
+ */
+ if (errno == EINTR) {
+ goto restart;
+ }
+ goto error;
+ }
+
+ nb_fd = ret;
+
+ for (i = 0; i < nb_fd; i++) {
+ /* Fetch once the poll data */
+ revents = LTTNG_POLL_GETEV(&events, i);
+ pollfd = LTTNG_POLL_GETFD(&events, i);
+
+ /* Thread quit pipe has been closed. Killing thread. */
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
+ if (ret) {
+ err = 0;
+ goto exit;
+ }
+
+ /* Event on the registration socket */
+ if (pollfd == sock) {
+ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ ERR("Health socket poll error");
+ goto error;
+ }
+ }
+ }
+
+ new_sock = lttcomm_accept_unix_sock(sock);
+ if (new_sock < 0) {
+ goto error;
}
- /* Copy event list into message payload */
- memcpy(cmd_ctx->llm->payload, events,
- nb_event * sizeof(struct lttng_event));
-
- free(events);
+ /*
+ * Set the CLOEXEC flag. Return code is useless because either way, the
+ * show must go on.
+ */
+ (void) utils_set_fd_cloexec(new_sock);
- ret = LTTCOMM_OK;
- break;
- }
- case LTTNG_LIST_SESSIONS:
- {
- unsigned int nr_sessions;
+ DBG("Receiving data from client for health...");
+ ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
+ if (ret <= 0) {
+ DBG("Nothing recv() from client... continuing");
+ ret = close(new_sock);
+ if (ret) {
+ PERROR("close");
+ }
+ new_sock = -1;
+ continue;
+ }
- session_lock_list();
- nr_sessions = lttng_sessions_count(
- LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
- LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
+ rcu_thread_online();
- ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
- if (ret < 0) {
- session_unlock_list();
- goto setup_error;
+ reply.ret_code = 0;
+ for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
+ /*
+ * health_check_state returns 0 if health is
+ * bad.
+ */
+ if (!health_check_state(health_sessiond, i)) {
+ reply.ret_code |= 1ULL << i;
+ }
}
- /* Filled the session array */
- list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
- LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
- LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
+ DBG2("Health check return value %" PRIx64, reply.ret_code);
- session_unlock_list();
+ ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
+ if (ret < 0) {
+ ERR("Failed to send health data back to client");
+ }
- ret = LTTCOMM_OK;
- break;
- }
- case LTTNG_CALIBRATE:
- {
- ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
- &cmd_ctx->lsm->u.calibrate);
- break;
- }
- case LTTNG_REGISTER_CONSUMER:
- {
- ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.reg.path);
- break;
- }
- default:
- ret = LTTCOMM_UND;
- break;
+ /* End of transmission */
+ ret = close(new_sock);
+ if (ret) {
+ PERROR("close");
+ }
+ new_sock = -1;
}
+exit:
error:
- if (cmd_ctx->llm == NULL) {
- DBG("Missing llm structure. Allocating one.");
- if (setup_lttng_msg(cmd_ctx, 0) < 0) {
- goto setup_error;
- }
+ if (err) {
+ ERR("Health error occurred in %s", __func__);
}
- /* Set return code */
- cmd_ctx->llm->ret_code = ret;
-setup_error:
- if (cmd_ctx->session) {
- session_unlock(cmd_ctx->session);
+ DBG("Health check thread dying");
+ unlink(health_unix_sock_path);
+ if (sock >= 0) {
+ ret = close(sock);
+ if (ret) {
+ PERROR("close");
+ }
}
-init_setup_error:
- return ret;
+
+ lttng_poll_clean(&events);
+
+ rcu_unregister_thread();
+ return NULL;
}
/*
*/
static void *thread_manage_clients(void *data)
{
- int sock = -1, ret, i, pollfd;
+ int sock = -1, ret, i, pollfd, err = -1;
+ int sock_error;
uint32_t revents, nb_fd;
struct command_ctx *cmd_ctx = NULL;
struct lttng_poll_event events;
rcu_register_thread();
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
+
+ if (testpoint(thread_manage_clients)) {
+ goto error_testpoint;
+ }
+
+ health_code_update();
+
ret = lttcomm_listen_unix_sock(client_sock);
if (ret < 0) {
- goto error;
+ goto error_listen;
}
/*
* Pass 2 as size here for the thread quit pipe and client_sock. Nothing
* more will be added to this poll set.
*/
- ret = create_thread_poll_set(&events, 2);
+ ret = sessiond_set_thread_pollset(&events, 2);
if (ret < 0) {
- goto error;
+ goto error_create_poll;
}
/* Add the application registration socket */
/*
* Notify parent pid that we are ready to accept command for client side.
+ * This ppid is the one from the external process that spawned us.
*/
if (opt_sig_parent) {
kill(ppid, SIGUSR1);
}
+ /* Notify the parent of the fork() process that we are ready. */
+ if (opt_daemon) {
+ kill(child_ppid, SIGUSR1);
+ }
+
+ if (testpoint(thread_manage_clients_before_loop)) {
+ goto error;
+ }
+
+ health_code_update();
+
while (1) {
DBG("Accepting client command ...");
- nb_fd = LTTNG_POLL_GETNB(&events);
-
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
goto error;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update();
+
/* Thread quit pipe has been closed. Killing thread. */
- ret = check_thread_quit_pipe(pollfd, revents);
+ ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
DBG("Wait for client response");
+ health_code_update();
+
sock = lttcomm_accept_unix_sock(client_sock);
if (sock < 0) {
goto error;
}
+ /*
+ * Set the CLOEXEC flag. Return code is useless because either way, the
+ * show must go on.
+ */
+ (void) utils_set_fd_cloexec(sock);
+
/* Set socket option for credentials retrieval */
ret = lttcomm_setsockopt_creds_unix_sock(sock);
if (ret < 0) {
cmd_ctx->llm = NULL;
cmd_ctx->session = NULL;
+ health_code_update();
+
/*
* Data is received from the lttng client. The struct
* lttcomm_session_msg (lsm) contains the command and data request of
PERROR("close");
}
sock = -1;
- free(cmd_ctx);
+ clean_command_ctx(&cmd_ctx);
continue;
}
+ health_code_update();
+
// TODO: Validate cmd_ctx including sanity check for
// security purpose.
* informations for the client. The command context struct contains
* everything this function may needs.
*/
- ret = process_client_msg(cmd_ctx);
+ ret = process_client_msg(cmd_ctx, sock, &sock_error);
rcu_thread_offline();
if (ret < 0) {
+ ret = close(sock);
+ if (ret) {
+ PERROR("close");
+ }
+ sock = -1;
/*
* TODO: Inform client somehow of the fatal error. At
* this point, ret < 0 means that a zmalloc failed
- * (ENOMEM). Error detected but still accept command.
+ * (ENOMEM). Error detected but still accept
+ * command, unless a socket error has been
+ * detected.
*/
clean_command_ctx(&cmd_ctx);
continue;
}
+ health_code_update();
+
DBG("Sending response (size: %d, retcode: %s)",
cmd_ctx->lttng_msg_size,
lttng_strerror(-cmd_ctx->llm->ret_code));
sock = -1;
clean_command_ctx(&cmd_ctx);
+
+ health_code_update();
}
+exit:
error:
- DBG("Client thread dying");
- unlink(client_unix_sock_path);
- if (client_sock >= 0) {
- ret = close(client_sock);
- if (ret) {
- PERROR("close");
- }
- }
if (sock >= 0) {
ret = close(sock);
if (ret) {
lttng_poll_clean(&events);
clean_command_ctx(&cmd_ctx);
+error_listen:
+error_create_poll:
+error_testpoint:
+ unlink(client_unix_sock_path);
+ if (client_sock >= 0) {
+ ret = close(client_sock);
+ if (ret) {
+ PERROR("close");
+ }
+ }
+
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+
+ health_unregister(health_sessiond);
+
+ DBG("Client thread dying");
+
rcu_unregister_thread();
return NULL;
}
fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
fprintf(stderr, " -V, --version Show version number.\n");
- fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
+ fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
fprintf(stderr, " -q, --quiet No output at all.\n");
fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
+ fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
fprintf(stderr, " --no-kernel Disable kernel tracer\n");
+ fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
+ fprintf(stderr, " -f --config Load daemon configuration file\n");
}
/*
- * daemon argument parsing
+ * Take an option from the getopt output and set it in the right variable to be
+ * used later.
+ *
+ * Return 0 on success else a negative value.
*/
-static int parse_args(int argc, char **argv)
+static int set_option(int opt, const char *arg, const char *optname)
{
- int c;
-
- static struct option long_options[] = {
- { "client-sock", 1, 0, 'c' },
- { "apps-sock", 1, 0, 'a' },
- { "kconsumerd-cmd-sock", 1, 0, 'C' },
- { "kconsumerd-err-sock", 1, 0, 'E' },
- { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
- { "ustconsumerd32-err-sock", 1, 0, 'H' },
- { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
- { "ustconsumerd64-err-sock", 1, 0, 'F' },
- { "consumerd32-path", 1, 0, 'u' },
- { "consumerd32-libdir", 1, 0, 'U' },
- { "consumerd64-path", 1, 0, 't' },
- { "consumerd64-libdir", 1, 0, 'T' },
- { "daemonize", 0, 0, 'd' },
- { "sig-parent", 0, 0, 'S' },
- { "help", 0, 0, 'h' },
- { "group", 1, 0, 'g' },
- { "version", 0, 0, 'V' },
- { "quiet", 0, 0, 'q' },
- { "verbose", 0, 0, 'v' },
- { "verbose-consumer", 0, 0, 'Z' },
- { "no-kernel", 0, 0, 'N' },
- { NULL, 0, 0, 0 }
- };
+ int ret = 0;
- while (1) {
- int option_index = 0;
- c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
- long_options, &option_index);
- if (c == -1) {
- break;
+ switch (opt) {
+ case 0:
+ fprintf(stderr, "option %s", optname);
+ if (arg) {
+ fprintf(stderr, " with arg %s\n", arg);
+ }
+ break;
+ case 'c':
+ snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'a':
+ snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'd':
+ opt_daemon = 1;
+ break;
+ case 'g':
+ tracing_group_name = strdup(arg);
+ break;
+ case 'h':
+ usage();
+ exit(EXIT_FAILURE);
+ case 'V':
+ fprintf(stdout, "%s\n", VERSION);
+ exit(EXIT_SUCCESS);
+ case 'S':
+ opt_sig_parent = 1;
+ break;
+ case 'E':
+ snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'C':
+ snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'F':
+ snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'D':
+ snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'H':
+ snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'G':
+ snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ break;
+ case 'N':
+ opt_no_kernel = 1;
+ break;
+ case 'q':
+ lttng_opt_quiet = 1;
+ break;
+ case 'v':
+ /* Verbose level can increase using multiple -v */
+ if (arg) {
+ lttng_opt_verbose = config_parse_value(arg);
+ } else {
+ lttng_opt_verbose += 1;
+ }
+ break;
+ case 'Z':
+ if (arg) {
+ opt_verbose_consumer = config_parse_value(arg);
+ } else {
+ opt_verbose_consumer += 1;
+ }
+ break;
+ case 'u':
+ consumerd32_bin = strdup(arg);
+ consumerd32_bin_override = 1;
+ break;
+ case 'U':
+ consumerd32_libdir = strdup(arg);
+ consumerd32_libdir_override = 1;
+ break;
+ case 't':
+ consumerd64_bin = strdup(arg);
+ consumerd64_bin_override = 1;
+ break;
+ case 'T':
+ consumerd64_libdir = strdup(arg);
+ consumerd64_libdir_override = 1;
+ break;
+ case 'p':
+ opt_pidfile = strdup(arg);
+ break;
+ case 'J': /* JUL TCP port. */
+ {
+ unsigned long v;
+
+ errno = 0;
+ v = strtoul(arg, NULL, 0);
+ if (errno != 0 || !isdigit(arg[0])) {
+ ERR("Wrong value in --jul-tcp-port parameter: %s", arg);
+ return -1;
+ }
+ if (v == 0 || v >= 65535) {
+ ERR("Port overflow in --jul-tcp-port parameter: %s", arg);
+ return -1;
+ }
+ jul_tcp_port = (uint32_t) v;
+ DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
+ break;
+ }
+ default:
+ /* Unknown option or other error.
+ * Error is printed by getopt, just return */
+ ret = -1;
+ }
+
+ return ret;
+}
+
+/*
+ * config_entry_handler_cb used to handle options read from a config file.
+ * See config_entry_handler_cb comment in common/config/config.h for the
+ * return value conventions.
+ */
+static int config_entry_handler(const struct config_entry *entry, void *unused)
+{
+ int ret = 0, i;
+
+ if (!entry || !entry->name || !entry->value) {
+ ret = -EINVAL;
+ goto end;
+ }
+
+ /* Check if the option is to be ignored */
+ for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
+ if (!strcmp(entry->name, config_ignore_options[i])) {
+ goto end;
+ }
+ }
+
+ for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
+ i++) {
+
+ /* Ignore if not fully matched. */
+ if (strcmp(entry->name, long_options[i].name)) {
+ continue;
}
- switch (c) {
- case 0:
- fprintf(stderr, "option %s", long_options[option_index].name);
- if (optarg) {
- fprintf(stderr, " with arg %s\n", optarg);
+ /*
+ * If the option takes no argument on the command line, we have to
+ * check if the value is "true". We support non-zero numeric values,
+ * true, on and yes.
+ */
+ if (!long_options[i].has_arg) {
+ ret = config_parse_value(entry->value);
+ if (ret <= 0) {
+ if (ret) {
+ WARN("Invalid configuration value \"%s\" for option %s",
+ entry->value, entry->name);
+ }
+ /* False, skip boolean config option. */
+ goto end;
}
+ }
+
+ ret = set_option(long_options[i].val, entry->value, entry->name);
+ goto end;
+ }
+
+ WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
+
+end:
+ return ret;
+}
+
+/*
+ * daemon configuration loading and argument parsing
+ */
+static int set_options(int argc, char **argv)
+{
+ int ret = 0, c = 0, option_index = 0;
+ int orig_optopt = optopt, orig_optind = optind;
+ char *optstring;
+ const char *config_path = NULL;
+
+ optstring = utils_generate_optstring(long_options,
+ sizeof(long_options) / sizeof(struct option));
+ if (!optstring) {
+ ret = -ENOMEM;
+ goto end;
+ }
+
+ /* Check for the --config option */
+ while ((c = getopt_long(argc, argv, optstring, long_options,
+ &option_index)) != -1) {
+ if (c == '?') {
+ ret = -EINVAL;
+ goto end;
+ } else if (c != 'f') {
+ /* if not equal to --config option. */
+ continue;
+ }
+
+ config_path = utils_expand_path(optarg);
+ if (!config_path) {
+ ERR("Failed to resolve path: %s", optarg);
+ }
+ }
+
+ ret = config_get_section_entries(config_path, config_section_name,
+ config_entry_handler, NULL);
+ if (ret) {
+ if (ret > 0) {
+ ERR("Invalid configuration option at line %i", ret);
+ ret = -1;
+ }
+ goto end;
+ }
+
+ /* Reset getopt's global state */
+ optopt = orig_optopt;
+ optind = orig_optind;
+ while (1) {
+ c = getopt_long(argc, argv, optstring, long_options, &option_index);
+ if (c == -1) {
break;
- case 'c':
- snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'a':
- snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'd':
- opt_daemon = 1;
- break;
- case 'g':
- opt_tracing_group = optarg;
- break;
- case 'h':
- usage();
- exit(EXIT_FAILURE);
- case 'V':
- fprintf(stdout, "%s\n", VERSION);
- exit(EXIT_SUCCESS);
- case 'S':
- opt_sig_parent = 1;
- break;
- case 'E':
- snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'C':
- snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'F':
- snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'D':
- snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'H':
- snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'G':
- snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
- break;
- case 'N':
- opt_no_kernel = 1;
- break;
- case 'q':
- opt_quiet = 1;
- break;
- case 'v':
- /* Verbose level can increase using multiple -v */
- opt_verbose += 1;
- break;
- case 'Z':
- opt_verbose_consumer += 1;
- break;
- case 'u':
- consumerd32_bin= optarg;
- break;
- case 'U':
- consumerd32_libdir = optarg;
- break;
- case 't':
- consumerd64_bin = optarg;
- break;
- case 'T':
- consumerd64_libdir = optarg;
+ }
+
+ ret = set_option(c, optarg, long_options[option_index].name);
+ if (ret < 0) {
break;
- default:
- /* Unknown option or other error.
- * Error is printed by getopt, just return */
- return -1;
}
}
- return 0;
+end:
+ free(optstring);
+ return ret;
}
/*
goto end;
}
+ /* Set the cloexec flag */
+ ret = utils_set_fd_cloexec(client_sock);
+ if (ret < 0) {
+ ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
+ "Continuing but note that the consumer daemon will have a "
+ "reference to this socket on exec()", client_sock);
+ }
+
/* File permission MUST be 660 */
ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (ret < 0) {
goto end;
}
+ /* Set the cloexec flag */
+ ret = utils_set_fd_cloexec(apps_sock);
+ if (ret < 0) {
+ ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
+ "Continuing but note that the consumer daemon will have a "
+ "reference to this socket on exec()", apps_sock);
+ }
+
/* File permission MUST be 666 */
ret = chmod(apps_unix_sock_path,
S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
goto end;
}
+ DBG3("Session daemon client socket %d and application socket %d created",
+ client_sock, apps_sock);
+
end:
umask(old_umask);
return ret;
int ret;
gid_t gid;
- gid = allowed_group();
- if (gid < 0) {
- WARN("No tracing group detected");
- ret = 0;
- goto end;
- }
+ gid = utils_get_group_id(tracing_group_name);
/* Set lttng run dir */
ret = chown(rundir, 0, gid);
PERROR("chown");
}
- /* Ensure tracing group can search the run dir */
- ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
+ /*
+ * Ensure all applications and tracing group can search the run
+ * dir. Allow everyone to read the directory, since it does not
+ * buy us anything to hide its content.
+ */
+ ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
if (ret < 0) {
ERR("Unable to set permissions on %s", rundir);
PERROR("chmod");
}
/* kconsumer error socket path */
- ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
+ ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
PERROR("chown");
}
/* 64-bit ustconsumer error socket path */
- ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
+ ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
PERROR("chown");
}
/* 32-bit ustconsumer compat32 error socket path */
- ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
+ ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
PERROR("chown");
DBG("All permissions are set");
-end:
- return ret;
-}
-
-/*
- * Create the pipe used to wake up the kernel thread.
- * Closed in cleanup().
- */
-static int create_kernel_poll_pipe(void)
-{
- int ret, i;
-
- ret = pipe(kernel_poll_pipe);
- if (ret < 0) {
- PERROR("kernel poll pipe");
- goto error;
- }
-
- for (i = 0; i < 2; i++) {
- ret = fcntl(kernel_poll_pipe[i], F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
- PERROR("fcntl kernel_poll_pipe");
- goto error;
- }
- }
-
-error:
- return ret;
-}
-
-/*
- * Create the application command pipe to wake thread_manage_apps.
- * Closed in cleanup().
- */
-static int create_apps_cmd_pipe(void)
-{
- int ret, i;
-
- ret = pipe(apps_cmd_pipe);
- if (ret < 0) {
- PERROR("apps cmd pipe");
- goto error;
- }
-
- for (i = 0; i < 2; i++) {
- ret = fcntl(apps_cmd_pipe[i], F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
- PERROR("fcntl apps_cmd_pipe");
- goto error;
- }
- }
-
-error:
return ret;
}
int ret;
char path[PATH_MAX];
- switch (consumer_data->type) {
+ switch (consumer_data->type) {
case LTTNG_CONSUMER_KERNEL:
snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
break;
DBG2("Creating consumer directory: %s", path);
- ret = mkdir(path, S_IRWXU);
+ ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
if (ret < 0) {
if (errno != EEXIST) {
+ PERROR("mkdir");
ERR("Failed to create %s", path);
goto error;
}
- ret = 0;
+ ret = -1;
+ }
+ if (is_root) {
+ ret = chown(path, 0, utils_get_group_id(tracing_group_name));
+ if (ret < 0) {
+ ERR("Unable to set group on %s", path);
+ PERROR("chown");
+ goto error;
+ }
}
/* Create the kconsumerd error unix socket */
goto error;
}
+ /*
+ * Set the CLOEXEC flag. Return code is useless because either way, the
+ * show must go on.
+ */
+ ret = utils_set_fd_cloexec(consumer_data->err_sock);
+ if (ret < 0) {
+ PERROR("utils_set_fd_cloexec");
+ /* continue anyway */
+ }
+
/* File permission MUST be 660 */
ret = chmod(consumer_data->err_unix_sock_path,
S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
DBG("SIGTERM caught");
stop_threads();
break;
+ case SIGUSR1:
+ CMM_STORE_SHARED(recv_child_signal, 1);
+ break;
default:
break;
}
return ret;
}
- DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
+ if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
+ PERROR("sigaction");
+ return ret;
+ }
+
+ DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
return ret;
}
}
}
+/*
+ * Write pidfile using the rundir and opt_pidfile.
+ */
+static void write_pidfile(void)
+{
+ int ret;
+ char pidfile_path[PATH_MAX];
+
+ assert(rundir);
+
+ if (opt_pidfile) {
+ strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
+ } else {
+ /* Build pidfile path from rundir and opt_pidfile. */
+ ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
+ DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf pidfile path");
+ goto error;
+ }
+ }
+
+ /*
+ * Create pid file in rundir. Return value is of no importance. The
+ * execution will continue even though we are not able to write the file.
+ */
+ (void) utils_create_pid_file(getpid(), pidfile_path);
+
+error:
+ return;
+}
+
+/*
+ * Write JUL TCP port using the rundir.
+ */
+static void write_julport(void)
+{
+ int ret;
+ char path[PATH_MAX];
+
+ assert(rundir);
+
+ ret = snprintf(path, sizeof(path), "%s/"
+ DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf julport path");
+ goto error;
+ }
+
+ /*
+ * Create TCP JUL port file in rundir. Return value is of no importance.
+ * The execution will continue even though we are not able to write the
+ * file.
+ */
+ (void) utils_create_pid_file(jul_tcp_port, path);
+
+error:
+ return;
+}
+
+/*
+ * Daemonize this process by forking and making the parent wait for the child
+ * to signal it indicating readiness. Once received, the parent successfully
+ * quits.
+ *
+ * The child process undergoes the same action that daemon(3) does meaning
+ * setsid, chdir, and dup /dev/null into 0, 1 and 2.
+ *
+ * Return 0 on success else -1 on error.
+ */
+static int daemonize(void)
+{
+ int ret;
+ pid_t pid;
+
+ /* Get parent pid of this process. */
+ child_ppid = getppid();
+
+ pid = fork();
+ if (pid < 0) {
+ PERROR("fork");
+ goto error;
+ } else if (pid == 0) {
+ int fd;
+ pid_t sid;
+
+ /* Child */
+
+ /*
+ * Get the newly created parent pid so we can signal that process when
+ * we are ready to operate.
+ */
+ child_ppid = getppid();
+
+ sid = setsid();
+ if (sid < 0) {
+ PERROR("setsid");
+ goto error;
+ }
+
+ /* Try to change directory to /. If we can't well at least notify. */
+ ret = chdir("/");
+ if (ret < 0) {
+ PERROR("chdir");
+ }
+
+ fd = open(_PATH_DEVNULL, O_RDWR, 0);
+ if (fd < 0) {
+ PERROR("open %s", _PATH_DEVNULL);
+ /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
+ } else {
+ (void) dup2(fd, STDIN_FILENO);
+ (void) dup2(fd, STDOUT_FILENO);
+ (void) dup2(fd, STDERR_FILENO);
+ if (fd > 2) {
+ ret = close(fd);
+ if (ret < 0) {
+ PERROR("close");
+ }
+ }
+ }
+ goto end;
+ } else {
+ /* Parent */
+
+ /*
+ * Waiting for child to notify this parent that it can exit. Note that
+ * sleep() is interrupted before the 1 second delay as soon as the
+ * signal is received, so it will not cause visible delay for the
+ * user.
+ */
+ while (!CMM_LOAD_SHARED(recv_child_signal)) {
+ int status;
+ pid_t ret;
+
+ /*
+ * Check if child exists without blocking. If so, we have to stop
+ * this parent process and return an error.
+ */
+ ret = waitpid(pid, &status, WNOHANG);
+ if (ret < 0 || (ret != 0 && WIFEXITED(status))) {
+ /* The child exited somehow or was not valid. */
+ goto error;
+ }
+ sleep(1);
+ }
+
+ /*
+ * From this point on, the parent can exit and the child is now an
+ * operationnal session daemon ready to serve clients and applications.
+ */
+ exit(EXIT_SUCCESS);
+ }
+
+end:
+ return 0;
+
+error:
+ return -1;
+}
+
/*
* main
*/
{
int ret = 0;
void *status;
- const char *home_path;
+ const char *home_path, *env_app_timeout;
init_kernel_workarounds();
rcu_register_thread();
- /* Create thread quit pipe */
- if ((ret = init_thread_quit_pipe()) < 0) {
+ if ((ret = set_signal_handler()) < 0) {
goto error;
}
setup_consumerd_path();
- /* Parse arguments */
+ page_size = sysconf(_SC_PAGESIZE);
+ if (page_size < 0) {
+ PERROR("sysconf _SC_PAGESIZE");
+ page_size = LONG_MAX;
+ WARN("Fallback page size to %ld", page_size);
+ }
+
+ /* Parse arguments and load the daemon configuration file */
progname = argv[0];
- if ((ret = parse_args(argc, argv) < 0)) {
+ if ((ret = set_options(argc, argv)) < 0) {
goto error;
}
/* Daemonize */
if (opt_daemon) {
- ret = daemon(0, 0);
+ int i;
+
+ ret = daemonize();
if (ret < 0) {
- PERROR("daemon");
goto error;
}
+
+ /*
+ * 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);
+ }
+ }
+
+ /* Create thread quit pipe */
+ if ((ret = init_thread_quit_pipe()) < 0) {
+ goto error;
}
/* Check if daemon is UID = 0 */
DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
}
+ if (strlen(health_unix_sock_path) == 0) {
+ snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
+ DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
+ }
+
/* Setup kernel consumerd path */
snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
DBG2("Kernel consumer cmd path: %s",
kconsumer_data.cmd_unix_sock_path);
} else {
- home_path = get_home_dir();
+ home_path = utils_get_home_dir();
if (home_path == NULL) {
/* TODO: Add --socket PATH option */
ERR("Can't get HOME directory for sockets creation.");
/* Set global SHM for ust */
if (strlen(wait_shm_path) == 0) {
snprintf(wait_shm_path, PATH_MAX,
- DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
+ DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
+ }
+
+ /* Set health check Unix path */
+ if (strlen(health_unix_sock_path) == 0) {
+ snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
+ DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
}
}
+ /* Set consumer initial state */
+ kernel_consumerd_state = CONSUMER_STOPPED;
+ ust_consumerd_state = CONSUMER_STOPPED;
+
DBG("Client socket path %s", client_unix_sock_path);
DBG("Application socket path %s", apps_unix_sock_path);
+ DBG("Application wait path %s", wait_shm_path);
DBG("LTTng run directory path: %s", rundir);
/* 32 bits consumerd path setup */
goto error;
}
+ /*
+ * Init UST app hash table. Alloc hash table before this point since
+ * cleanup() can get called after that point.
+ */
+ ust_app_ht_alloc();
+
+ /* Initialize JUL domain subsystem. */
+ if ((ret = jul_init()) < 0) {
+ /* ENOMEM at this point. */
+ goto error;
+ }
+
/* After this point, we can safely call cleanup() with "goto exit" */
/*
/* Set ulimit for open files */
set_ulimit();
}
+ /* init lttng_fd tracking must be done after set_ulimit. */
+ lttng_fd_init();
ret = set_consumer_sockets(&ustconsumer64_data, rundir);
if (ret < 0) {
goto exit;
}
- if ((ret = set_signal_handler()) < 0) {
- goto exit;
- }
-
/* Setup the needed unix socket */
if ((ret = init_daemon_socket()) < 0) {
goto exit;
}
/* Setup the kernel pipe for waking up the kernel thread */
- if ((ret = create_kernel_poll_pipe()) < 0) {
+ if (is_root && !opt_no_kernel) {
+ if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
+ goto exit;
+ }
+ }
+
+ /* Setup the thread ht_cleanup communication pipe. */
+ if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
goto exit;
}
/* Setup the thread apps communication pipe. */
- if ((ret = create_apps_cmd_pipe()) < 0) {
+ if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
+ goto exit;
+ }
+
+ /* Setup the thread apps notify communication pipe. */
+ if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
goto exit;
}
+ /* Initialize global buffer per UID and PID registry. */
+ buffer_reg_init_uid_registry();
+ buffer_reg_init_pid_registry();
+
/* Init UST command queue. */
cds_wfq_init(&ust_cmd_queue.queue);
- /* Init UST app hash table */
- ust_app_ht_alloc();
-
/*
* Get session list pointer. This pointer MUST NOT be free(). This list is
* statically declared in session.c
/* Set up max poll set size */
lttng_poll_set_max_size();
+ cmd_init();
+
+ /* Check for the application socket timeout env variable. */
+ env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
+ if (env_app_timeout) {
+ app_socket_timeout = atoi(env_app_timeout);
+ } else {
+ app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
+ }
+
+ write_pidfile();
+ write_julport();
+
+ /* Initialize communication library */
+ lttcomm_init();
+ /* This is to get the TCP timeout value. */
+ lttcomm_inet_init();
+
+ /*
+ * Initialize the health check subsystem. This call should set the
+ * appropriate time values.
+ */
+ health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
+ if (!health_sessiond) {
+ PERROR("health_app_create error");
+ goto exit_health_sessiond_cleanup;
+ }
+
+ /* Create thread to clean up RCU hash tables */
+ ret = pthread_create(&ht_cleanup_thread, NULL,
+ thread_ht_cleanup, (void *) NULL);
+ if (ret != 0) {
+ PERROR("pthread_create ht_cleanup");
+ goto exit_ht_cleanup;
+ }
+
+ /* Create health-check thread */
+ ret = pthread_create(&health_thread, NULL,
+ thread_manage_health, (void *) NULL);
+ if (ret != 0) {
+ PERROR("pthread_create health");
+ goto exit_health;
+ }
+
/* Create thread to manage the client socket */
ret = pthread_create(&client_thread, NULL,
thread_manage_clients, (void *) NULL);
goto exit_apps;
}
- /* Create kernel thread to manage kernel event */
- ret = pthread_create(&kernel_thread, NULL,
- thread_manage_kernel, (void *) NULL);
+ /* Create thread to manage application notify socket */
+ ret = pthread_create(&apps_notify_thread, NULL,
+ ust_thread_manage_notify, (void *) NULL);
if (ret != 0) {
- PERROR("pthread_create kernel");
- goto exit_kernel;
+ PERROR("pthread_create apps");
+ goto exit_apps_notify;
}
- ret = pthread_join(kernel_thread, &status);
+ /* Create JUL registration thread. */
+ ret = pthread_create(&jul_reg_thread, NULL,
+ jul_thread_manage_registration, (void *) NULL);
if (ret != 0) {
- PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
+ PERROR("pthread_create apps");
+ goto exit_jul_reg;
+ }
+
+ /* Don't start this thread if kernel tracing is not requested nor root */
+ if (is_root && !opt_no_kernel) {
+ /* Create kernel thread to manage kernel event */
+ ret = pthread_create(&kernel_thread, NULL,
+ thread_manage_kernel, (void *) NULL);
+ if (ret != 0) {
+ PERROR("pthread_create kernel");
+ goto exit_kernel;
+ }
+
+ ret = pthread_join(kernel_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join");
+ goto error; /* join error, exit without cleanup */
+ }
}
exit_kernel:
+ ret = pthread_join(jul_reg_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join JUL");
+ goto error; /* join error, exit without cleanup */
+ }
+
+exit_jul_reg:
+ ret = pthread_join(apps_notify_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join apps notify");
+ goto error; /* join error, exit without cleanup */
+ }
+
+exit_apps_notify:
ret = pthread_join(apps_thread, &status);
if (ret != 0) {
- PERROR("pthread_join");
+ PERROR("pthread_join apps");
goto error; /* join error, exit without cleanup */
}
+
exit_apps:
ret = pthread_join(reg_apps_thread, &status);
if (ret != 0) {
goto error; /* join error, exit without cleanup */
}
+ ret = join_consumer_thread(&ustconsumer32_data);
+ if (ret != 0) {
+ PERROR("join_consumer ust32");
+ goto error; /* join error, exit without cleanup */
+ }
+
+ ret = join_consumer_thread(&ustconsumer64_data);
+ if (ret != 0) {
+ PERROR("join_consumer ust64");
+ goto error; /* join error, exit without cleanup */
+ }
+
exit_client:
+ ret = pthread_join(health_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join health thread");
+ goto error; /* join error, exit without cleanup */
+ }
+
+exit_health:
+ ret = pthread_join(ht_cleanup_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join ht cleanup thread");
+ goto error; /* join error, exit without cleanup */
+ }
+exit_ht_cleanup:
+ health_app_destroy(health_sessiond);
+exit_health_sessiond_cleanup:
exit:
/*
* cleanup() is called when no other thread is running.