/*
* 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, version 2 only,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
#include <getopt.h>
#include <grp.h>
#include <limits.h>
+#include <paths.h>
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <sys/wait.h>
#include <urcu/uatomic.h>
#include <unistd.h>
-#include <config.h>
#include <common/common.h>
#include <common/compat/socket.h>
+#include <common/compat/getenv.h>
#include <common/defaults.h>
#include <common/kernel-consumer/kernel-consumer.h>
#include <common/futex.h>
#include <common/relayd/relayd.h>
#include <common/utils.h>
+#include <common/daemonize.h>
+#include <common/config/session-config.h>
#include "lttng-sessiond.h"
#include "buffer-registry.h"
#include "ust-consumer.h"
#include "utils.h"
#include "fd-limit.h"
-#include "health.h"
+#include "health-sessiond.h"
#include "testpoint.h"
#include "ust-thread.h"
+#include "agent-thread.h"
+#include "save.h"
+#include "load-session-thread.h"
+#include "syscall.h"
+#include "agent.h"
#define CONSUMERD_FILE "lttng-consumerd"
-/* Const values */
-const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
-
const char *progname;
-const char *opt_tracing_group;
-static const char *opt_pidfile;
+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_daemon, opt_background;
static int opt_no_kernel;
-static int is_root; /* Set to 1 if the daemon is running as root */
+static char *opt_load_session_path;
static pid_t ppid; /* Parent PID for --sig-parent option */
+static pid_t child_ppid; /* Internal parent PID use with daemonize. */
static char *rundir;
+static int lockfile_fd = -1;
+
+/* 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
.cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
- .metadata_sock.fd = -1,
.pid_mutex = PTHREAD_MUTEX_INITIALIZER,
.lock = PTHREAD_MUTEX_INITIALIZER,
.cond = PTHREAD_COND_INITIALIZER,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
- .metadata_sock.fd = -1,
.pid_mutex = PTHREAD_MUTEX_INITIALIZER,
.lock = PTHREAD_MUTEX_INITIALIZER,
.cond = PTHREAD_COND_INITIALIZER,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
- .metadata_sock.fd = -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", required_argument, 0, 'c' },
+ { "apps-sock", required_argument, 0, 'a' },
+ { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
+ { "kconsumerd-err-sock", required_argument, 0, '\0' },
+ { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
+ { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
+ { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
+ { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
+ { "consumerd32-path", required_argument, 0, '\0' },
+ { "consumerd32-libdir", required_argument, 0, '\0' },
+ { "consumerd64-path", required_argument, 0, '\0' },
+ { "consumerd64-libdir", required_argument, 0, '\0' },
+ { "daemonize", no_argument, 0, 'd' },
+ { "background", no_argument, 0, 'b' },
+ { "sig-parent", no_argument, 0, 'S' },
+ { "help", no_argument, 0, 'h' },
+ { "group", required_argument, 0, 'g' },
+ { "version", no_argument, 0, 'V' },
+ { "quiet", no_argument, 0, 'q' },
+ { "verbose", no_argument, 0, 'v' },
+ { "verbose-consumer", no_argument, 0, '\0' },
+ { "no-kernel", no_argument, 0, '\0' },
+ { "pidfile", required_argument, 0, 'p' },
+ { "agent-tcp-port", required_argument, 0, '\0' },
+ { "config", required_argument, 0, 'f' },
+ { "load", required_argument, 0, 'l' },
+ { "kmod-probes", required_argument, 0, '\0' },
+ { "extra-kmod-probes", required_argument, 0, '\0' },
+ { 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;
* for all threads when receiving an event on the pipe.
*/
static int thread_quit_pipe[2] = { -1, -1 };
+static int ht_cleanup_quit_pipe[2] = { -1, -1 };
/*
* This pipe is used to inform the thread managing application communication
static pthread_t kernel_thread;
static pthread_t dispatch_thread;
static pthread_t health_thread;
+static pthread_t ht_cleanup_thread;
+static pthread_t agent_reg_thread;
+static pthread_t load_session_thread;
/*
* UST registration command queue. This queue is tied with a futex and uses a N
* wakers / 1 waiter implemented and detailed in futex.c/.h
*
- * The thread_manage_apps and thread_dispatch_ust_registration interact with
- * this queue and the wait/wake scheme.
+ * The thread_registration_apps and thread_dispatch_ust_registration uses this
+ * queue along with the wait/wake scheme. The thread_manage_apps receives down
+ * the line new application socket and monitors it for any I/O error or clean
+ * close that triggers an unregistration of the application.
*/
static struct ust_cmd_queue ust_cmd_queue;
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";
*/
static int app_socket_timeout;
+/* Set in main() with the current page size. */
+long page_size;
+
+/* Application health monitoring */
+struct health_app *health_sessiond;
+
+/* Agent TCP port for registration. Used by the agent thread. */
+unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
+
+/* Am I root or not. */
+int is_root; /* Set to 1 if the daemon is running as root */
+
+const char * const config_section_name = "sessiond";
+
+/* Load session thread information to operate. */
+struct load_session_thread_data *load_info;
+
+/* Global hash tables */
+struct lttng_ht *agent_apps_ht_by_sock = NULL;
+
+/*
+ * Whether sessiond is ready for commands/health check requests.
+ * NR_LTTNG_SESSIOND_READY must match the number of calls to
+ * sessiond_notify_ready().
+ */
+#define NR_LTTNG_SESSIOND_READY 3
+int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
+
+/* Notify parents that we are ready for cmd and health check */
+LTTNG_HIDDEN
+void sessiond_notify_ready(void)
+{
+ if (uatomic_sub_return(<tng_sessiond_ready, 1) == 0) {
+ /*
+ * 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 || opt_background) {
+ kill(child_ppid, SIGUSR1);
+ }
+ }
+}
+
static
void setup_consumerd_path(void)
{
/*
* runtime env. var. overrides the build default.
*/
- bin = getenv("LTTNG_CONSUMERD32_BIN");
+ bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
if (bin) {
consumerd32_bin = bin;
}
- bin = getenv("LTTNG_CONSUMERD64_BIN");
+ bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
if (bin) {
consumerd64_bin = bin;
}
- libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
+ libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
if (libdir) {
consumerd32_libdir = libdir;
}
- libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
+ libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
if (libdir) {
consumerd64_libdir = libdir;
}
}
-/*
- * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
- */
-int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
+static
+int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
+ int *a_pipe)
{
int ret;
}
/* Add quit pipe */
- ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
+ ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
if (ret < 0) {
goto error;
}
}
/*
- * Check if the thread quit pipe was triggered.
- *
- * Return 1 if it was triggered else 0;
+ * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
*/
-int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
+int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
+{
+ return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
+}
+
+/*
+ * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
+ */
+int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
+ size_t size)
+{
+ return __sessiond_set_thread_pollset(events, size,
+ ht_cleanup_quit_pipe);
+}
+
+static
+int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
{
- if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
+ if (fd == a_pipe && (events & LPOLLIN)) {
return 1;
}
-
return 0;
}
/*
- * Return group ID of the tracing group or -1 if not found.
+ * Check if the thread quit pipe was triggered.
+ *
+ * Return 1 if it was triggered else 0;
*/
-static gid_t allowed_group(void)
+int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
{
- struct group *grp;
+ return __sessiond_check_thread_quit_pipe(fd, events,
+ thread_quit_pipe[0]);
+}
- if (opt_tracing_group) {
- grp = getgrnam(opt_tracing_group);
- } else {
- grp = getgrnam(default_tracing_group);
- }
- if (!grp) {
- return -1;
- } else {
- return grp->gr_gid;
- }
+/*
+ * Check if the ht_cleanup thread quit pipe was triggered.
+ *
+ * Return 1 if it was triggered else 0;
+ */
+int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
+{
+ return __sessiond_check_thread_quit_pipe(fd, events,
+ ht_cleanup_quit_pipe[0]);
}
/*
*
* Return -1 on error or 0 if all pipes are created.
*/
-static int init_thread_quit_pipe(void)
+static int __init_thread_quit_pipe(int *a_pipe)
{
int ret, i;
- ret = pipe(thread_quit_pipe);
+ ret = pipe(a_pipe);
if (ret < 0) {
PERROR("thread quit pipe");
goto error;
}
for (i = 0; i < 2; i++) {
- ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
+ ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
if (ret < 0) {
PERROR("fcntl");
goto error;
return ret;
}
+static int init_thread_quit_pipe(void)
+{
+ return __init_thread_quit_pipe(thread_quit_pipe);
+}
+
+static int init_ht_cleanup_quit_pipe(void)
+{
+ return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
+}
+
/*
* Stop all threads by closing the thread quit pipe.
*/
}
/*
- * Cleanup the daemon
+ * 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");
+ }
+ }
+}
+
+/*
+ * Generate the full lock file path using the rundir.
+ *
+ * Return the snprintf() return value thus a negative value is an error.
+ */
+static int generate_lock_file_path(char *path, size_t len)
+{
+ int ret;
+
+ assert(path);
+ assert(rundir);
+
+ /* Build lockfile path from rundir. */
+ ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf lockfile path");
+ }
+
+ return ret;
+}
+
+/*
+ * Wait on consumer process termination.
+ *
+ * Need to be called with the consumer data lock held or from a context
+ * ensuring no concurrent access to data (e.g: cleanup).
+ */
+static void wait_consumer(struct consumer_data *consumer_data)
+{
+ pid_t ret;
+ int status;
+
+ if (consumer_data->pid <= 0) {
+ return;
+ }
+
+ DBG("Waiting for complete teardown of consumerd (PID: %d)",
+ consumer_data->pid);
+ ret = waitpid(consumer_data->pid, &status, 0);
+ if (ret == -1) {
+ PERROR("consumerd waitpid pid: %d", consumer_data->pid)
+ }
+ if (!WIFEXITED(status)) {
+ ERR("consumerd termination with error: %d",
+ WEXITSTATUS(ret));
+ }
+ consumer_data->pid = 0;
+}
+
+/*
+ * Cleanup the session daemon's data structures.
*/
-static void cleanup(void)
+static void sessiond_cleanup(void)
{
int ret;
- char *cmd = NULL;
struct ltt_session *sess, *stmp;
+ char path[PATH_MAX];
- DBG("Cleaning up");
+ DBG("Cleanup sessiond");
- /* First thing first, stop all threads */
+ /*
+ * Close the thread quit pipe. It has already done its job,
+ * since we are now called.
+ */
utils_close_pipe(thread_quit_pipe);
/*
}
}
- DBG("Removing %s directory", rundir);
- ret = asprintf(&cmd, "rm -rf %s", rundir);
- if (ret < 0) {
- ERR("asprintf failed. Something is really wrong!");
- }
-
- /* Remove lttng run directory */
- ret = system(cmd);
- if (ret < 0) {
- ERR("Unable to clean %s", rundir);
- }
- free(cmd);
- free(rundir);
+ 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_AGENTPORT_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);
DBG("Cleaning up all sessions");
}
}
+ wait_consumer(&kconsumer_data);
+ wait_consumer(&ustconsumer64_data);
+ wait_consumer(&ustconsumer32_data);
+
+ DBG("Cleaning up all agent apps");
+ agent_app_ht_clean();
+
DBG("Closing all UST sockets");
ust_app_clean_list();
buffer_reg_destroy_registries();
}
DBG("Unloading kernel modules");
modprobe_remove_lttng_all();
+ free(syscall_table);
+ }
+
+ close_consumer_sockets();
+
+ if (load_info) {
+ load_session_destroy_data(load_info);
+ free(load_info);
+ }
+
+ /*
+ * Cleanup lock file by deleting it and finaly closing it which will
+ * release the file system lock.
+ */
+ if (lockfile_fd >= 0) {
+ char lockfile_path[PATH_MAX];
+
+ ret = generate_lock_file_path(lockfile_path,
+ sizeof(lockfile_path));
+ if (ret > 0) {
+ ret = remove(lockfile_path);
+ if (ret < 0) {
+ PERROR("remove lock file");
+ }
+ ret = close(lockfile_fd);
+ if (ret < 0) {
+ PERROR("close lock file");
+ }
+ }
+ }
+
+ /*
+ * 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);
+}
+
+/*
+ * Cleanup the daemon's option data structures.
+ */
+static void sessiond_cleanup_options(void)
+{
+ DBG("Cleaning up options");
+
+ /*
+ * 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.
+ */
+ if (tracing_group_name_override) {
+ free((void *) tracing_group_name);
+ }
+ if (consumerd32_bin_override) {
+ free((void *) consumerd32_bin);
+ }
+ if (consumerd64_bin_override) {
+ free((void *) consumerd64_bin);
+ }
+ if (consumerd32_libdir_override) {
+ free((void *) consumerd32_libdir);
}
+ if (consumerd64_libdir_override) {
+ free((void *) consumerd64_libdir);
+ }
+
+ free(opt_pidfile);
+ free(opt_load_session_path);
+ free(kmod_probes_list);
+ free(kmod_extra_probes_list);
+
+ run_as_destroy_worker();
/* <fun> */
DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
* right amount of memory and copying the original information from the lsm
* structure.
*
- * Return total size of the buffer pointed by buf.
+ * Return 0 on success, negative value on error.
*/
-static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
+static int setup_lttng_msg(struct command_ctx *cmd_ctx,
+ const void *payload_buf, size_t payload_len,
+ const void *cmd_header_buf, size_t cmd_header_len)
{
- int ret, buf_size;
+ int ret = 0;
+ const size_t header_len = sizeof(struct lttcomm_lttng_msg);
+ const size_t cmd_header_offset = header_len;
+ const size_t payload_offset = cmd_header_offset + cmd_header_len;
+ const size_t total_msg_size = header_len + cmd_header_len + payload_len;
- buf_size = size;
+ cmd_ctx->llm = zmalloc(total_msg_size);
- cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
if (cmd_ctx->llm == NULL) {
PERROR("zmalloc");
ret = -ENOMEM;
- goto error;
+ goto end;
}
/* Copy common data */
cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
+ cmd_ctx->llm->cmd_header_size = cmd_header_len;
+ cmd_ctx->llm->data_size = payload_len;
+ cmd_ctx->lttng_msg_size = total_msg_size;
- cmd_ctx->llm->data_size = size;
- cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
+ /* Copy command header */
+ if (cmd_header_len) {
+ memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
+ cmd_header_len);
+ }
- return buf_size;
+ /* Copy payload */
+ if (payload_len) {
+ memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
+ payload_len);
+ }
-error:
+end:
return ret;
}
+/*
+ * Version of setup_lttng_msg() without command header.
+ */
+static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
+ void *payload_buf, size_t payload_len)
+{
+ return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
+}
/*
* Update the kernel poll set of all channel fd available over all tracing
* session. Add the wakeup pipe at the end of the set.
}
ksess = session->kernel_session;
- 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;
- }
+ cds_list_for_each_entry(channel,
+ &ksess->channel_list.head, list) {
+ struct lttng_ht_iter iter;
+ struct consumer_socket *socket;
- /*
- * 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 (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) {
- /* Code flow error */
- assert(socket->fd >= 0);
-
- pthread_mutex_lock(socket->lock);
- ret = kernel_consumer_send_channel_stream(socket,
- channel, ksess);
- pthread_mutex_unlock(socket->lock);
- if (ret < 0) {
- rcu_read_unlock();
- goto error;
- }
- }
+ if (channel->fd != fd) {
+ continue;
+ }
+ 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 (ksess->consumer_fds_sent != 1
+ || ksess->consumer == NULL) {
+ ret = -1;
+ goto error;
+ }
+
+ 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;
}
- goto error;
}
+ rcu_read_unlock();
}
session_unlock(session);
}
{
struct ltt_session *sess, *stmp;
+ /* 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) {
+ struct ust_app *app;
+
session_lock(sess);
- if (sess->ust_session) {
- ust_app_global_update(sess->ust_session, app_sock);
+ if (!sess->ust_session) {
+ goto unlock_session;
}
+
+ rcu_read_lock();
+ assert(app_sock >= 0);
+ app = ust_app_find_by_sock(app_sock);
+ if (app == NULL) {
+ /*
+ * Application can be unregistered before so
+ * this is possible hence simply stopping the
+ * update.
+ */
+ DBG3("UST app update failed to find app sock %d",
+ app_sock);
+ goto unlock_rcu;
+ }
+ ust_app_global_update(sess->ust_session, app);
+ unlock_rcu:
+ rcu_read_unlock();
+ unlock_session:
session_unlock(sess);
}
}
DBG("[thread] Thread manage kernel started");
- health_register(HEALTH_TYPE_KERNEL);
+ 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 zero it before the loop.
+ * non NULL pointers so initialize it before the loop.
*/
- memset(&events, 0, sizeof(events));
+ lttng_poll_init(&events);
- if (testpoint(thread_manage_kernel)) {
+ if (testpoint(sessiond_thread_manage_kernel)) {
goto error_testpoint;
}
health_code_update();
- if (testpoint(thread_manage_kernel_before_loop)) {
+ if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
goto error_testpoint;
}
update_poll_flag = 0;
}
- DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
+ DBG("Thread kernel polling");
/* Poll infinite value of time */
restart:
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ DBG("Thread kernel return from poll on %d fds",
+ LTTNG_POLL_GETNB(&events));
health_poll_exit();
if (ret < 0) {
/*
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
}
/* Check for data on kernel pipe */
- if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
- do {
- ret = read(kernel_poll_pipe[0], &tmp, 1);
- } while (ret < 0 && errno == EINTR);
- /*
- * Ret value is useless here, if this pipe gets any actions an
- * update is required anyway.
- */
- update_poll_flag = 1;
- continue;
- } else {
- /*
- * New CPU detected by the kernel. Adding kernel stream to
- * kernel session and updating the kernel consumer
- */
- if (revents & LPOLLIN) {
+ if (revents & LPOLLIN) {
+ if (pollfd == kernel_poll_pipe[0]) {
+ (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 {
+ /*
+ * New CPU detected by the kernel. Adding kernel stream to
+ * kernel session and updating the kernel consumer
+ */
ret = update_kernel_stream(&kconsumer_data, pollfd);
if (ret < 0) {
continue;
}
break;
- /*
- * TODO: We might want to handle the LPOLLERR | LPOLLHUP
- * and unregister kernel stream at this point.
- */
}
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ update_poll_flag = 1;
+ continue;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
}
WARN("Kernel thread died unexpectedly. "
"Kernel tracing can continue but CPU hotplug is disabled.");
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Kernel thread dying");
return NULL;
}
*/
static void *thread_manage_consumer(void *data)
{
- int sock = -1, i, ret, pollfd, err = -1;
+ int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
uint32_t revents, nb_fd;
enum lttcomm_return_code code;
struct lttng_poll_event events;
DBG("[thread] Manage consumer started");
- health_register(HEALTH_TYPE_CONSUMER);
+ rcu_register_thread();
+ rcu_thread_online();
+
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
health_code_update();
restart:
health_poll_entry();
- if (testpoint(thread_manage_consumer)) {
+ if (testpoint(sessiond_thread_manage_consumer)) {
goto error;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
/* Event on the registration socket */
if (pollfd == consumer_data->err_sock) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ if (revents & LPOLLIN) {
+ continue;
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("consumer err socket poll error");
goto error;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
}
}
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);
- consumer_data->metadata_sock.fd =
+ consumer_data->metadata_fd =
lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
- if (consumer_data->cmd_sock < 0 ||
- consumer_data->metadata_sock.fd < 0) {
+ 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) {
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_sock.fd);
+ consumer_data->metadata_fd);
} else {
ERR("consumer error when waiting for SOCK_READY : %s",
lttcomm_get_readable_code(-code));
}
/* Add metadata socket that is successfully connected. */
- ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
+ ret = lttng_poll_add(&events, consumer_data->metadata_fd,
LPOLLIN | LPOLLRDHUP);
if (ret < 0) {
goto error;
/* Infinite blocking call, waiting for transmission */
restart_poll:
while (1) {
+ health_code_update();
+
+ /* Exit the thread because the thread quit pipe has been triggered. */
+ if (should_quit) {
+ /* Not a health error. */
+ err = 0;
+ goto exit;
+ }
+
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
health_poll_exit();
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 (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
}
+ /*
+ * Thread quit pipe has been triggered, flag that we should stop
+ * but continue the current loop to handle potential data from
+ * consumer.
+ */
+ should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
+
if (pollfd == sock) {
/* Event on the consumerd socket */
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
+ && !(revents & LPOLLIN)) {
ERR("consumer err socket second poll error");
goto error;
}
lttcomm_get_readable_code(-code));
goto exit;
- } else if (pollfd == consumer_data->metadata_sock.fd) {
+ } else if (pollfd == consumer_data->metadata_fd) {
+ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
+ && !(revents & LPOLLIN)) {
+ ERR("consumer err metadata socket second poll error");
+ goto error;
+ }
/* UST metadata requests */
ret = ust_consumer_metadata_request(
&consumer_data->metadata_sock);
ERR("Handling metadata request");
goto error;
}
- break;
- } else {
- ERR("Unknown pollfd");
- goto error;
}
+ /* No need for an else branch all FDs are tested prior. */
}
health_code_update();
}
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);
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 >= 0) {
- ret = close(consumer_data->metadata_sock.fd);
+ if (consumer_data->metadata_sock.fd_ptr &&
+ *consumer_data->metadata_sock.fd_ptr >= 0) {
+ ret = close(*consumer_data->metadata_sock.fd_ptr);
if (ret) {
PERROR("close");
}
}
- /* Cleanup metadata socket mutex. */
- pthread_mutex_destroy(consumer_data->metadata_sock.lock);
- free(consumer_data->metadata_sock.lock);
-
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. */
+ if (consumer_data->metadata_sock.lock) {
+ pthread_mutex_destroy(consumer_data->metadata_sock.lock);
+ free(consumer_data->metadata_sock.lock);
+ }
lttng_poll_clean(&events);
error_poll:
if (err) {
health_error();
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("consumer thread cleanup completed");
+ rcu_thread_offline();
+ rcu_unregister_thread();
+
return NULL;
}
static void *thread_manage_apps(void *data)
{
int i, ret, pollfd, err = -1;
+ ssize_t size_ret;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
rcu_register_thread();
rcu_thread_online();
- health_register(HEALTH_TYPE_APP_MANAGE);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
- if (testpoint(thread_manage_apps)) {
+ if (testpoint(sessiond_thread_manage_apps)) {
goto error_testpoint;
}
goto error;
}
- if (testpoint(thread_manage_apps_before_loop)) {
+ if (testpoint(sessiond_thread_manage_apps_before_loop)) {
goto error;
}
health_code_update();
while (1) {
- DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
+ DBG("Apps thread polling");
/* Inifinite blocking call, waiting for transmission */
restart:
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
+ DBG("Apps thread return from poll on %d fds",
+ LTTNG_POLL_GETNB(&events));
health_poll_exit();
if (ret < 0) {
/*
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
/* Inspect the apps cmd pipe */
if (pollfd == apps_cmd_pipe[0]) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
- ERR("Apps command pipe error");
- goto error;
- } else if (revents & LPOLLIN) {
+ if (revents & LPOLLIN) {
int sock;
/* Empty pipe */
- do {
- ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
- } while (ret < 0 && errno == EINTR);
- if (ret < 0 || ret < sizeof(sock)) {
+ size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
+ if (size_ret < sizeof(sock)) {
PERROR("read apps cmd pipe");
goto error;
}
health_code_update();
/*
- * We only monitor the error events of the socket. This
- * thread does not handle any incoming data from UST
- * (POLLIN).
+ * Since this is a command socket (write then read),
+ * we only monitor the error events of the socket.
*/
ret = lttng_poll_add(&events, sock,
LPOLLERR | LPOLLHUP | LPOLLRDHUP);
goto error;
}
- /* Set socket timeout for both receiving and ending */
- (void) lttcomm_setsockopt_rcv_timeout(sock,
- app_socket_timeout);
- (void) lttcomm_setsockopt_snd_timeout(sock,
- app_socket_timeout);
-
DBG("Apps with sock %d added to poll set", sock);
-
- health_code_update();
-
- break;
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ ERR("Apps command pipe error");
+ goto error;
+ } else {
+ ERR("Unknown poll events %u for sock %d", revents, pollfd);
+ goto error;
}
} else {
/*
/* Socket closed on remote end. */
ust_app_unregister(pollfd);
- break;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
health_error();
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Application communication apps thread cleanup complete");
rcu_thread_offline();
rcu_unregister_thread();
* 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)
{
- int ret;
+ ssize_t ret;
- /* Sockets MUST be set or else this should not have been called. */
- assert(fd >= 0);
- assert(sock >= 0);
+ /*
+ * 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;
+ }
- do {
- ret = write(fd, &sock, sizeof(sock));
- } while (ret < 0 && errno == EINTR);
- if (ret < 0 || ret != sizeof(sock)) {
+ ret = lttng_write(fd, &sock, sizeof(sock));
+ if (ret < sizeof(sock)) {
PERROR("write apps pipe %d", fd);
if (ret < 0) {
ret = -errno;
/* All good. Don't send back the write positive ret value. */
ret = 0;
error:
- return ret;
+ 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);
+
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
+ 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);
+ /*
+ * Silence warning of use-after-free in
+ * cds_list_for_each_entry_safe which uses
+ * __typeof__(*wait_node).
+ */
+ wait_node = NULL;
+ break;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
+ }
+ }
+ }
+
+ 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;
}
/*
*/
static void *thread_dispatch_ust_registration(void *data)
{
- int ret;
- struct cds_wfq_node *node;
+ int ret, err = -1;
+ struct cds_wfcq_node *node;
struct ust_command *ust_cmd = NULL;
- struct {
- struct ust_app *app;
- struct cds_list_head head;
- } *wait_node = NULL, *tmp_wait_node;
+ 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);
+
+ if (testpoint(sessiond_thread_app_reg_dispatch)) {
+ goto error_testpoint;
+ }
+
+ health_code_update();
- CDS_LIST_HEAD(wait_queue);
+ CDS_INIT_LIST_HEAD(&wait_queue.head);
DBG("[thread] Dispatch UST command started");
while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ health_code_update();
+
/* Atomically prepare the queue futex */
futex_nto1_prepare(&ust_cmd_queue.futex);
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);
+ node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
if (node == NULL) {
DBG("Woken up but nothing in the UST command queue");
/* Continue thread execution */
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;
}
if (ret < 0) {
PERROR("close ust sock dispatch %d", ust_cmd->sock);
}
- lttng_fd_put(1, LTTNG_FD_APPS);
+ 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);
+ cds_list_add(&wait_node->head, &wait_queue.head);
+ wait_queue.count++;
free(ust_cmd);
/*
* notify socket if found.
*/
cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
- &wait_queue, head) {
+ &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 (ret < 0) {
rcu_read_unlock();
session_unlock_list();
- /* No notify thread, stop the UST tracing. */
+ /*
+ * 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;
}
* Don't care about return value. Let the manage apps threads
* handle app unregistration upon socket close.
*/
- (void) ust_app_register_done(app->sock);
+ (void) ust_app_register_done(app);
/*
* Even if the application socket has been closed, send the app
if (ret < 0) {
rcu_read_unlock();
session_unlock_list();
- /* No apps. thread, stop the UST tracing. */
+ /*
+ * 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();
- } else {
- /* Application manager threads are not available. */
- ret = close(ust_cmd->sock);
- if (ret < 0) {
- PERROR("close ust_cmd sock");
- }
- lttng_fd_put(1, LTTNG_FD_APPS);
}
} 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) {
+ &wait_queue.head, head) {
cds_list_del(&wait_node->head);
+ wait_queue.count--;
free(wait_node);
}
+ /* Empty command queue. */
+ for (;;) {
+ /* Dequeue command for registration */
+ node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
+ if (node == NULL) {
+ break;
+ }
+ ust_cmd = caa_container_of(node, struct ust_command, node);
+ ret = close(ust_cmd->sock);
+ if (ret < 0) {
+ PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
+ }
+ lttng_fd_put(LTTNG_FD_APPS, 1);
+ free(ust_cmd);
+ }
+
+error_testpoint:
DBG("Dispatch thread dying");
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_unregister(health_sessiond);
return NULL;
}
DBG("[thread] Manage application registration started");
- health_register(HEALTH_TYPE_APP_REG);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
- if (testpoint(thread_registration_apps)) {
+ if (testpoint(sessiond_thread_registration_apps)) {
goto error_testpoint;
}
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
/* Event on the registration socket */
if (pollfd == apps_sock) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
- ERR("Register apps socket poll error");
- goto error;
- } else if (revents & LPOLLIN) {
+ if (revents & LPOLLIN) {
sock = lttcomm_accept_unix_sock(apps_sock);
if (sock < 0) {
goto error;
}
+ /*
+ * 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.
ust_cmd = zmalloc(sizeof(struct ust_command));
if (ust_cmd == NULL) {
PERROR("ust command zmalloc");
+ ret = close(sock);
+ if (ret) {
+ PERROR("close");
+ }
goto error;
}
* Lock free enqueue the registration request. The red pill
* has been taken! This apps will be part of the *system*.
*/
- cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
+ cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
/*
* Wake the registration queue futex. Implicit memory
- * barrier with the exchange in cds_wfq_enqueue.
+ * barrier with the exchange in cds_wfcq_enqueue.
*/
futex_nto1_wake(&ust_cmd_queue.futex);
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ ERR("Register apps socket poll error");
+ goto error;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
}
exit:
error:
- if (err) {
- health_error();
- ERR("Health error occurred in %s", __func__);
- }
-
/* Notify that the registration thread is gone */
notify_ust_apps(0);
error_create_poll:
error_testpoint:
DBG("UST Registration thread cleanup complete");
- health_unregister();
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_unregister(health_sessiond);
return NULL;
}
int ret, clock_ret;
struct timespec timeout;
- /* Make sure we set the readiness flag to 0 because we are NOT ready */
+ /*
+ * Make sure we set the readiness flag to 0 because we are NOT ready.
+ * This access to consumer_thread_is_ready does not need to be
+ * protected by consumer_data.cond_mutex (yet) since the consumer
+ * management thread has not been started at this point.
+ */
consumer_data->consumer_thread_is_ready = 0;
/* Setup pthread condition */
ret = pthread_condattr_init(&consumer_data->condattr);
- if (ret != 0) {
+ if (ret) {
errno = ret;
PERROR("pthread_condattr_init consumer data");
goto error;
* for a more details and how we noticed it.
*/
ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
- if (ret != 0) {
+ if (ret) {
errno = ret;
PERROR("pthread_condattr_setclock consumer data");
goto error;
}
ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
- if (ret != 0) {
+ if (ret) {
errno = ret;
PERROR("pthread_cond_init consumer data");
goto error;
ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
consumer_data);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create consumer");
ret = -1;
goto error;
if (ret != 0) {
errno = ret;
if (ret == ETIMEDOUT) {
+ int pth_ret;
+
/*
* Call has timed out so we kill the kconsumerd_thread and return
* an error.
*/
ERR("Condition timed out. The consumer thread was never ready."
" Killing it");
- ret = pthread_cancel(consumer_data->thread);
- if (ret < 0) {
+ pth_ret = pthread_cancel(consumer_data->thread);
+ if (pth_ret < 0) {
PERROR("pthread_cancel consumer thread");
}
} else {
PERROR("pthread_cond_wait failed consumer thread");
}
+ /* Caller is expecting a negative value on failure. */
+ ret = -1;
goto error;
}
int ret;
ret = kill(consumer_data->pid, SIGTERM);
if (ret) {
- ERR("Error killing consumer daemon");
+ PERROR("Error killing consumer daemon");
return ret;
}
return pthread_join(consumer_data->thread, &status);
*/
if (opt_verbose_consumer) {
verbosity = "--verbose";
- } else {
+ } else if (lttng_opt_quiet) {
verbosity = "--quiet";
+ } else {
+ verbosity = "";
}
+
switch (consumer_data->type) {
case LTTNG_CONSUMER_KERNEL:
/*
consumer_to_use = consumerd32_bin;
} else {
DBG("Could not find any valid consumerd executable");
+ ret = -EINVAL;
break;
}
DBG("Using kernel consumer at: %s", consumer_to_use);
- execl(consumer_to_use,
+ ret = execl(consumer_to_use,
"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:
char *tmp;
size_t tmplen;
- tmp = getenv("LD_LIBRARY_PATH");
+ tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
if (!tmp) {
tmp = "";
}
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);
}
- if (ret) {
- goto error;
- }
break;
}
case LTTNG_CONSUMER32_UST:
char *tmp;
size_t tmplen;
- tmp = getenv("LD_LIBRARY_PATH");
+ tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
if (!tmp) {
tmp = "";
}
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);
}
- if (ret) {
- goto error;
- }
break;
}
default:
exit(EXIT_FAILURE);
}
if (errno != 0) {
- PERROR("kernel start consumer exec");
+ PERROR("Consumer execl()");
}
+ /* Reaching this point, we got a failure on our execl(). */
exit(EXIT_FAILURE);
} else if (pid > 0) {
ret = pid;
return ret;
}
-/*
- * Compute health status of each consumer. If one of them is zero (bad
- * state), we return 0.
- */
-static int check_consumer_health(void)
-{
- int ret;
-
- ret = health_check_state(HEALTH_TYPE_CONSUMER);
-
- DBG3("Health consumer check %d", ret);
-
- return ret;
-}
-
/*
* Setup necessary data for kernel tracer action.
*/
* 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 copy_session_consumer(int domain, struct ltt_session *session)
{
* domain.
*/
if (session->kernel_session->consumer) {
- consumer_destroy_output(session->kernel_session->consumer);
+ consumer_output_put(session->kernel_session->consumer);
}
session->kernel_session->consumer =
consumer_copy_output(session->consumer);
consumer = session->kernel_session->consumer;
dir_name = DEFAULT_KERNEL_TRACE_DIR;
break;
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_LOG4J:
+ case LTTNG_DOMAIN_PYTHON:
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);
+ consumer_output_put(session->ust_session->consumer);
}
session->ust_session->consumer =
consumer_copy_output(session->consumer);
/*
* 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)
assert(session->consumer);
switch (domain->type) {
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_LOG4J:
+ case LTTNG_DOMAIN_PYTHON:
case LTTNG_DOMAIN_UST:
break;
default:
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;
-
+ if (session->shm_path[0]) {
+ strncpy(lus->root_shm_path, session->shm_path,
+ sizeof(lus->root_shm_path));
+ lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
+ strncpy(lus->shm_path, session->shm_path,
+ sizeof(lus->shm_path));
+ lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
+ strncat(lus->shm_path, "/ust",
+ sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
+ }
/* Copy session output to the newly created UST session */
ret = copy_session_consumer(domain->type, session);
if (ret != LTTNG_OK) {
session->kernel_session->consumer->dst.trace_path,
S_IRWXU | S_IRWXG, session->uid, session->gid);
if (ret < 0) {
- if (ret != -EEXIST) {
+ if (errno != 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;
* 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 int process_client_msg(struct command_ctx *cmd_ctx, int sock,
int *sock_error)
DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
+ assert(!rcu_read_ongoing());
+
*sock_error = 0;
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:
+ case LTTNG_SAVE_SESSION:
+ case LTTNG_SET_SESSION_SHM_PATH:
+ case LTTNG_METADATA_REGENERATE:
need_domain = 0;
break;
default:
case LTTNG_LIST_DOMAINS:
case LTTNG_LIST_CHANNELS:
case LTTNG_LIST_EVENTS:
+ case LTTNG_LIST_SYSCALLS:
+ case LTTNG_LIST_TRACKER_PIDS:
+ case LTTNG_DATA_PENDING:
break;
default:
/* Setup lttng message with no payload */
- ret = setup_lttng_msg(cmd_ctx, 0);
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 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_CREATE_SESSION_SNAPSHOT:
+ case LTTNG_CREATE_SESSION_LIVE:
case LTTNG_CALIBRATE:
case LTTNG_LIST_SESSIONS:
case LTTNG_LIST_TRACEPOINTS:
+ case LTTNG_LIST_SYSCALLS:
case LTTNG_LIST_TRACEPOINT_FIELDS:
+ case LTTNG_SAVE_SESSION:
need_tracing_session = 0;
break;
default:
session_lock_list();
cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
if (cmd_ctx->session == NULL) {
- if (cmd_ctx->lsm->session.name != NULL) {
- ret = LTTNG_ERR_SESS_NOT_FOUND;
- } else {
- /* If no session name specified */
- ret = LTTNG_ERR_SELECT_SESS;
- }
+ ret = LTTNG_ERR_SESS_NOT_FOUND;
goto error;
} else {
/* Acquire lock for the session */
break;
}
+ /*
+ * Commands that need a valid session but should NOT create one if none
+ * exists. Instead of creating one and destroying it when the command is
+ * handled, process that right before so we save some round trip in useless
+ * code path.
+ */
+ switch (cmd_ctx->lsm->cmd_type) {
+ case LTTNG_DISABLE_CHANNEL:
+ case LTTNG_DISABLE_EVENT:
+ switch (cmd_ctx->lsm->domain.type) {
+ case LTTNG_DOMAIN_KERNEL:
+ if (!cmd_ctx->session->kernel_session) {
+ ret = LTTNG_ERR_NO_CHANNEL;
+ goto error;
+ }
+ break;
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_LOG4J:
+ case LTTNG_DOMAIN_PYTHON:
+ case LTTNG_DOMAIN_UST:
+ if (!cmd_ctx->session->ust_session) {
+ ret = LTTNG_ERR_NO_CHANNEL;
+ goto error;
+ }
+ break;
+ default:
+ ret = LTTNG_ERR_UNKNOWN_DOMAIN;
+ goto error;
+ }
+ default:
+ break;
+ }
+
if (!need_domain) {
goto skip_domain;
}
}
break;
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_LOG4J:
+ case LTTNG_DOMAIN_PYTHON:
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;
}
/* 32-bit */
+ pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
if (consumerd32_bin[0] != '\0' &&
ustconsumer32_data.pid == 0 &&
cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
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_NONE:
+ break;
+ case LTTNG_DOMAIN_JUL:
+ case LTTNG_DOMAIN_LOG4J:
+ case LTTNG_DOMAIN_PYTHON:
case LTTNG_DOMAIN_UST:
if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
ret = LTTNG_ERR_NO_USTCONSUMERD;
goto error;
}
break;
+ default:
+ ret = LTTNG_ERR_UNKNOWN_DOMAIN;
+ goto error;
}
}
switch (cmd_ctx->lsm->cmd_type) {
case LTTNG_ADD_CONTEXT:
{
- ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.context.channel_name,
- &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
- break;
- }
- case LTTNG_DISABLE_CHANNEL:
- {
- ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.disable.channel_name);
- break;
- }
- case LTTNG_DISABLE_EVENT:
- {
- ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.disable.channel_name,
- cmd_ctx->lsm->u.disable.name);
+ /*
+ * An LTTNG_ADD_CONTEXT command might have a supplementary
+ * payload if the context being added is an application context.
+ */
+ if (cmd_ctx->lsm->u.context.ctx.ctx ==
+ LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
+ char *provider_name = NULL, *context_name = NULL;
+ size_t provider_name_len =
+ cmd_ctx->lsm->u.context.provider_name_len;
+ size_t context_name_len =
+ cmd_ctx->lsm->u.context.context_name_len;
+
+ if (provider_name_len == 0 || context_name_len == 0) {
+ /*
+ * Application provider and context names MUST
+ * be provided.
+ */
+ ret = -LTTNG_ERR_INVALID;
+ goto error;
+ }
+
+ provider_name = zmalloc(provider_name_len + 1);
+ if (!provider_name) {
+ ret = -LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
+ provider_name;
+
+ context_name = zmalloc(context_name_len + 1);
+ if (!context_name) {
+ ret = -LTTNG_ERR_NOMEM;
+ goto error_add_context;
+ }
+ cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
+ context_name;
+
+ ret = lttcomm_recv_unix_sock(sock, provider_name,
+ provider_name_len);
+ if (ret < 0) {
+ goto error_add_context;
+ }
+
+ ret = lttcomm_recv_unix_sock(sock, context_name,
+ context_name_len);
+ if (ret < 0) {
+ goto error_add_context;
+ }
+ }
+
+ /*
+ * cmd_add_context assumes ownership of the provider and context
+ * names.
+ */
+ ret = cmd_add_context(cmd_ctx->session,
+ cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.context.channel_name,
+ &cmd_ctx->lsm->u.context.ctx,
+ kernel_poll_pipe[1]);
+
+ cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
+ cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
+error_add_context:
+ free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
+ free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
+ if (ret < 0) {
+ goto error;
+ }
break;
}
- case LTTNG_DISABLE_ALL_EVENT:
+ case LTTNG_DISABLE_CHANNEL:
{
- DBG("Disabling all events");
-
- ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
cmd_ctx->lsm->u.disable.channel_name);
break;
}
+ case LTTNG_DISABLE_EVENT:
+ {
+
+ /*
+ * FIXME: handle filter; for now we just receive the filter's
+ * bytecode along with the filter expression which are sent by
+ * liblttng-ctl and discard them.
+ *
+ * This fixes an issue where the client may block while sending
+ * the filter payload and encounter an error because the session
+ * daemon closes the socket without ever handling this data.
+ */
+ size_t count = cmd_ctx->lsm->u.disable.expression_len +
+ cmd_ctx->lsm->u.disable.bytecode_len;
+
+ if (count) {
+ char data[LTTNG_FILTER_MAX_LEN];
+
+ DBG("Discarding disable event command payload of size %zu", count);
+ while (count) {
+ ret = lttcomm_recv_unix_sock(sock, data,
+ count > sizeof(data) ? sizeof(data) : count);
+ if (ret < 0) {
+ goto error;
+ }
+
+ count -= (size_t) ret;
+ }
+ }
+ /* FIXME: passing packed structure to non-packed pointer */
+ ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.disable.channel_name,
+ &cmd_ctx->lsm->u.disable.event);
+ break;
+ }
case LTTNG_ENABLE_CHANNEL:
{
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:
+ case LTTNG_TRACK_PID:
{
- ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
- cmd_ctx->lsm->u.enable.channel_name,
- &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
+ ret = cmd_track_pid(cmd_ctx->session,
+ cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.pid_tracker.pid);
+ break;
+ }
+ case LTTNG_UNTRACK_PID:
+ {
+ ret = cmd_untrack_pid(cmd_ctx->session,
+ cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.pid_tracker.pid);
break;
}
- case LTTNG_ENABLE_ALL_EVENT:
+ case LTTNG_ENABLE_EVENT:
{
- DBG("Enabling all events");
+ struct lttng_event_exclusion *exclusion = NULL;
+ struct lttng_filter_bytecode *bytecode = NULL;
+ char *filter_expression = 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;
+ }
+ }
+
+ /* Get filter expression from client. */
+ if (cmd_ctx->lsm->u.enable.expression_len > 0) {
+ size_t expression_len =
+ cmd_ctx->lsm->u.enable.expression_len;
+
+ if (expression_len > LTTNG_FILTER_MAX_LEN) {
+ ret = LTTNG_ERR_FILTER_INVAL;
+ free(exclusion);
+ goto error;
+ }
+
+ filter_expression = zmalloc(expression_len);
+ if (!filter_expression) {
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_NOMEM;
+ goto error;
+ }
+
+ /* Receive var. len. data */
+ DBG("Receiving var len filter's expression from client ...");
+ ret = lttcomm_recv_unix_sock(sock, filter_expression,
+ expression_len);
+ if (ret <= 0) {
+ DBG("Nothing recv() from client car len data... continuing");
+ *sock_error = 1;
+ free(filter_expression);
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_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(filter_expression);
+ free(exclusion);
+ goto error;
+ }
+
+ bytecode = zmalloc(bytecode_len);
+ if (!bytecode) {
+ free(filter_expression);
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_NOMEM;
+ goto error;
+ }
- ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
+ /* 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(filter_expression);
+ free(bytecode);
+ free(exclusion);
+ ret = LTTNG_ERR_FILTER_INVAL;
+ goto error;
+ }
+
+ if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
+ free(filter_expression);
+ 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.type, NULL, kernel_poll_pipe[1]);
+ &cmd_ctx->lsm->u.enable.event,
+ filter_expression, bytecode, exclusion,
+ kernel_poll_pipe[1]);
break;
}
case LTTNG_LIST_TRACEPOINTS:
struct lttng_event *events;
ssize_t nb_events;
+ session_lock_list();
nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
+ session_unlock_list();
if (nb_events < 0) {
/* Return value is a negative lttng_error_code. */
ret = -nb_events;
* 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);
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
+ sizeof(struct lttng_event) * nb_events);
+ free(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;
}
struct lttng_event_field *fields;
ssize_t nb_fields;
+ session_lock_list();
nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
&fields);
+ session_unlock_list();
if (nb_fields < 0) {
/* Return value is a negative lttng_error_code. */
ret = -nb_fields;
* 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,
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
sizeof(struct lttng_event_field) * nb_fields);
+ free(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);
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_LIST_SYSCALLS:
+ {
+ struct lttng_event *events;
+ ssize_t nb_events;
- free(fields);
+ nb_events = cmd_list_syscalls(&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_no_cmd_header(cmd_ctx, events,
+ sizeof(struct lttng_event) * nb_events);
+ free(events);
+
+ if (ret < 0) {
+ goto setup_error;
+ }
+
+ ret = LTTNG_OK;
+ break;
+ }
+ case LTTNG_LIST_TRACKER_PIDS:
+ {
+ int32_t *pids = NULL;
+ ssize_t nr_pids;
+
+ nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
+ cmd_ctx->lsm->domain.type, &pids);
+ if (nr_pids < 0) {
+ /* Return value is a negative lttng_error_code. */
+ ret = -nr_pids;
+ 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_no_cmd_header(cmd_ctx, pids,
+ sizeof(int32_t) * nr_pids);
+ free(pids);
+
+ if (ret < 0) {
+ goto setup_error;
+ }
ret = LTTNG_OK;
break;
goto error;
}
- ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
- nb_uri, uris);
+ ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
+ free(uris);
if (ret != LTTNG_OK) {
- free(uris);
goto error;
}
- /*
- * XXX: 0 means that this URI should be applied on the session. Should
- * be a DOMAIN enuam.
- */
- 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);
- }
- }
-
- free(uris);
break;
}
}
ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
- &cmd_ctx->creds);
+ &cmd_ctx->creds, 0);
free(uris);
case LTTNG_LIST_DOMAINS:
{
ssize_t nb_dom;
- struct lttng_domain *domains;
+ struct lttng_domain *domains = NULL;
nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
if (nb_dom < 0) {
goto error;
}
- ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
+ nb_dom * sizeof(struct lttng_domain));
+ free(domains);
+
if (ret < 0) {
goto setup_error;
}
- /* Copy event list into message payload */
- memcpy(cmd_ctx->llm->payload, domains,
- nb_dom * sizeof(struct lttng_domain));
-
- free(domains);
-
ret = LTTNG_OK;
break;
}
case LTTNG_LIST_CHANNELS:
{
- int nb_chan;
- struct lttng_channel *channels;
+ ssize_t payload_size;
+ struct lttng_channel *channels = NULL;
- nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
+ payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
cmd_ctx->session, &channels);
- if (nb_chan < 0) {
+ if (payload_size < 0) {
/* Return value is a negative lttng_error_code. */
- ret = -nb_chan;
+ ret = -payload_size;
goto error;
}
- ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
+ payload_size);
+ free(channels);
+
if (ret < 0) {
goto setup_error;
}
- /* Copy event list into message payload */
- memcpy(cmd_ctx->llm->payload, channels,
- nb_chan * sizeof(struct lttng_channel));
-
- free(channels);
-
ret = LTTNG_OK;
break;
}
{
ssize_t nb_event;
struct lttng_event *events = NULL;
+ struct lttcomm_event_command_header cmd_header;
+ size_t total_size;
+
+ memset(&cmd_header, 0, sizeof(cmd_header));
+ /* Extended infos are included at the end of events */
+ nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
+ cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
+ &events, &total_size);
- 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));
+ cmd_header.nb_events = nb_event;
+ ret = setup_lttng_msg(cmd_ctx, events, total_size,
+ &cmd_header, sizeof(cmd_header));
+ free(events);
+
if (ret < 0) {
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;
+ void *sessions_payload;
+ size_t payload_len;
session_lock_list();
nr_sessions = lttng_sessions_count(
LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
+ payload_len = sizeof(struct lttng_session) * nr_sessions;
+ sessions_payload = zmalloc(payload_len);
- ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
- if (ret < 0) {
+ if (!sessions_payload) {
session_unlock_list();
+ ret = -ENOMEM;
goto setup_error;
}
- /* Filled the session array */
- cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
+ cmd_list_lttng_sessions(sessions_payload,
LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
-
session_unlock_list();
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
+ payload_len);
+ free(sessions_payload);
+
+ if (ret < 0) {
+ goto setup_error;
+ }
+
ret = LTTNG_OK;
break;
}
cmd_ctx->lsm->u.reg.path, cdata);
break;
}
- case LTTNG_ENABLE_EVENT_WITH_FILTER:
+ case LTTNG_DATA_PENDING:
{
- struct lttng_filter_bytecode *bytecode;
+ int pending_ret;
+ uint8_t pending_ret_byte;
- if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
- ret = LTTNG_ERR_FILTER_INVAL;
- goto error;
+ pending_ret = cmd_data_pending(cmd_ctx->session);
+
+ /*
+ * FIXME
+ *
+ * This function may returns 0 or 1 to indicate whether or not
+ * there is data pending. In case of error, it should return an
+ * LTTNG_ERR code. However, some code paths may still return
+ * a nondescript error code, which we handle by returning an
+ * "unknown" error.
+ */
+ if (pending_ret == 0 || pending_ret == 1) {
+ /*
+ * ret will be set to LTTNG_OK at the end of
+ * this function.
+ */
+ } else if (pending_ret < 0) {
+ ret = LTTNG_ERR_UNK;
+ goto setup_error;
+ } else {
+ ret = pending_ret;
+ goto setup_error;
}
- if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
- ret = LTTNG_ERR_FILTER_INVAL;
- goto error;
+
+ pending_ret_byte = (uint8_t) pending_ret;
+
+ /* 1 byte to return whether or not data is pending */
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
+ &pending_ret_byte, 1);
+
+ if (ret < 0) {
+ goto setup_error;
}
- bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
- if (!bytecode) {
- ret = LTTNG_ERR_FILTER_NOMEM;
+
+ ret = LTTNG_OK;
+ 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;
}
- /* Receive var. len. data */
- DBG("Receiving var len data from client ...");
- ret = lttcomm_recv_unix_sock(sock, bytecode,
- cmd_ctx->lsm->u.enable.bytecode_len);
- if (ret <= 0) {
- DBG("Nothing recv() from client var len data... continuing");
- *sock_error = 1;
- ret = LTTNG_ERR_FILTER_INVAL;
- goto error;
+
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
+ sizeof(reply));
+ if (ret < 0) {
+ goto setup_error;
}
- if (bytecode->len + sizeof(*bytecode)
- != cmd_ctx->lsm->u.enable.bytecode_len) {
- free(bytecode);
- ret = LTTNG_ERR_FILTER_INVAL;
+ /* Copy output list into message payload */
+ 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 = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
- cmd_ctx->lsm->u.enable.channel_name,
- &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
+ assert((nb_output > 0 && outputs) || nb_output == 0);
+ ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
+ nb_output * sizeof(struct lttng_snapshot_output));
+ free(outputs);
+
+ if (ret < 0) {
+ goto setup_error;
+ }
+
+ ret = LTTNG_OK;
break;
}
- case LTTNG_DATA_PENDING:
+ 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;
+ }
+ case LTTNG_SAVE_SESSION:
+ {
+ ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
+ &cmd_ctx->creds);
+ break;
+ }
+ case LTTNG_SET_SESSION_SHM_PATH:
{
- ret = cmd_data_pending(cmd_ctx->session);
+ ret = cmd_set_session_shm_path(cmd_ctx->session,
+ cmd_ctx->lsm->u.set_shm_path.shm_path);
+ break;
+ }
+ case LTTNG_METADATA_REGENERATE:
+ {
+ ret = cmd_metadata_regenerate(cmd_ctx->session);
break;
}
default:
error:
if (cmd_ctx->llm == NULL) {
DBG("Missing llm structure. Allocating one.");
- if (setup_lttng_msg(cmd_ctx, 0) < 0) {
+ if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
goto setup_error;
}
}
session_unlock_list();
}
init_setup_error:
+ assert(!rcu_read_ongoing());
return ret;
}
int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
- struct lttcomm_health_msg msg;
- struct lttcomm_health_data reply;
+ 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");
+ 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");
+ goto error;
+ }
+ }
+
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
goto error;
}
+ sessiond_notify_ready();
+
while (1) {
DBG("Health check ready");
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
/* Event on the registration socket */
if (pollfd == sock) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ if (revents & LPOLLIN) {
+ continue;
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("Health socket poll error");
goto error;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
}
rcu_thread_online();
- switch (msg.component) {
- case LTTNG_HEALTH_CMD:
- reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
- break;
- case LTTNG_HEALTH_APP_MANAGE:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
- break;
- case LTTNG_HEALTH_APP_REG:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
- break;
- case LTTNG_HEALTH_KERNEL:
- reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
- break;
- case LTTNG_HEALTH_CONSUMER:
- reply.ret_code = check_consumer_health();
- break;
- case LTTNG_HEALTH_ALL:
- reply.ret_code =
- health_check_state(HEALTH_TYPE_APP_MANAGE) &&
- health_check_state(HEALTH_TYPE_APP_REG) &&
- health_check_state(HEALTH_TYPE_CMD) &&
- health_check_state(HEALTH_TYPE_KERNEL) &&
- check_consumer_health();
- break;
- default:
- reply.ret_code = LTTNG_ERR_UND;
- break;
- }
-
- /*
- * Flip ret value since 0 is a success and 1 indicates a bad health for
- * the client where in the sessiond it is the opposite. Again, this is
- * just to make things easier for us poor developer which enjoy a lot
- * lazyness.
- */
- if (reply.ret_code == 0 || reply.ret_code == 1) {
- reply.ret_code = !reply.ret_code;
+ memset(&reply, 0, sizeof(reply));
+ 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;
+ }
}
- DBG2("Health check return value %d", reply.ret_code);
+ DBG2("Health check return value %" PRIx64, reply.ret_code);
ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
if (ret < 0) {
PERROR("close");
}
}
- if (new_sock >= 0) {
- ret = close(new_sock);
- if (ret) {
- PERROR("close");
- }
- }
lttng_poll_clean(&events);
rcu_register_thread();
- health_register(HEALTH_TYPE_CMD);
-
- if (testpoint(thread_manage_clients)) {
- goto error_testpoint;
- }
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
health_code_update();
goto error;
}
- /*
- * Notify parent pid that we are ready to accept command for client side.
- */
- if (opt_sig_parent) {
- kill(ppid, SIGUSR1);
+ sessiond_notify_ready();
+ ret = sem_post(&load_info->message_thread_ready);
+ if (ret) {
+ PERROR("sem_post message_thread_ready");
+ goto error;
+ }
+
+ /* This testpoint is after we signal readiness to the parent. */
+ if (testpoint(sessiond_thread_manage_clients)) {
+ goto error;
}
- if (testpoint(thread_manage_clients_before_loop)) {
+ if (testpoint(sessiond_thread_manage_clients_before_loop)) {
goto error;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
/* Event on the registration socket */
if (pollfd == client_sock) {
- if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ if (revents & LPOLLIN) {
+ continue;
+ } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("Client socket poll error");
goto error;
+ } else {
+ ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+ goto error;
}
}
}
ret = process_client_msg(cmd_ctx, sock, &sock_error);
rcu_thread_offline();
if (ret < 0) {
- if (sock_error) {
- ret = close(sock);
- if (ret) {
- PERROR("close");
- }
- sock = -1;
+ 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
health_code_update();
- DBG("Sending response (size: %d, retcode: %s)",
+ DBG("Sending response (size: %d, retcode: %s (%d))",
cmd_ctx->lttng_msg_size,
- lttng_strerror(-cmd_ctx->llm->ret_code));
+ lttng_strerror(-cmd_ctx->llm->ret_code),
+ cmd_ctx->llm->ret_code);
ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
if (ret < 0) {
ERR("Failed to send data back to client");
error_listen:
error_create_poll:
-error_testpoint:
unlink(client_unix_sock_path);
if (client_sock >= 0) {
ret = close(client_sock);
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Client thread dying");
rcu_unregister_thread();
+
+ /*
+ * Since we are creating the consumer threads, we own them, so we need
+ * to join them before our thread exits.
+ */
+ ret = join_consumer_thread(&kconsumer_data);
+ if (ret) {
+ errno = ret;
+ PERROR("join_consumer");
+ }
+
+ ret = join_consumer_thread(&ustconsumer32_data);
+ if (ret) {
+ errno = ret;
+ PERROR("join_consumer ust32");
+ }
+
+ ret = join_consumer_thread(&ustconsumer64_data);
+ if (ret) {
+ errno = ret;
+ PERROR("join_consumer ust64");
+ }
return NULL;
}
+static int string_match(const char *str1, const char *str2)
+{
+ return (str1 && str2) && !strcmp(str1, str2);
+}
+
+/*
+ * 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 set_option(int opt, const char *arg, const char *optname)
+{
+ int ret = 0;
+
+ if (string_match(optname, "client-sock") || opt == 'c') {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-c, --client-sock");
+ } else {
+ snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "apps-sock") || opt == 'a') {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-a, --apps-sock");
+ } else {
+ snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "daemonize") || opt == 'd') {
+ opt_daemon = 1;
+ } else if (string_match(optname, "background") || opt == 'b') {
+ opt_background = 1;
+ } else if (string_match(optname, "group") || opt == 'g') {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-g, --group");
+ } else {
+ /*
+ * 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.
+ */
+ if (tracing_group_name_override) {
+ free((void *) tracing_group_name);
+ }
+ tracing_group_name = strdup(arg);
+ if (!tracing_group_name) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ tracing_group_name_override = 1;
+ }
+ } else if (string_match(optname, "help") || opt == 'h') {
+ ret = utils_show_man_page(8, "lttng-sessiond");
+ if (ret) {
+ ERR("Cannot view man page lttng-sessiond(8)");
+ perror("exec");
+ }
+ exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
+ } else if (string_match(optname, "version") || opt == 'V') {
+ fprintf(stdout, "%s\n", VERSION);
+ exit(EXIT_SUCCESS);
+ } else if (string_match(optname, "sig-parent") || opt == 'S') {
+ opt_sig_parent = 1;
+ } else if (string_match(optname, "kconsumerd-err-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--kconsumerd-err-sock");
+ } else {
+ snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "kconsumerd-cmd-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--kconsumerd-cmd-sock");
+ } else {
+ snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "ustconsumerd64-err-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--ustconsumerd64-err-sock");
+ } else {
+ snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--ustconsumerd64-cmd-sock");
+ } else {
+ snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "ustconsumerd32-err-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--ustconsumerd32-err-sock");
+ } else {
+ snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--ustconsumerd32-cmd-sock");
+ } else {
+ snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
+ }
+ } else if (string_match(optname, "no-kernel")) {
+ opt_no_kernel = 1;
+ } else if (string_match(optname, "quiet") || opt == 'q') {
+ lttng_opt_quiet = 1;
+ } else if (string_match(optname, "verbose") || opt == 'v') {
+ /* Verbose level can increase using multiple -v */
+ if (arg) {
+ /* Value obtained from config file */
+ lttng_opt_verbose = config_parse_value(arg);
+ } else {
+ /* -v used on command line */
+ lttng_opt_verbose++;
+ }
+ /* Clamp value to [0, 3] */
+ lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
+ (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
+ } else if (string_match(optname, "verbose-consumer")) {
+ if (arg) {
+ opt_verbose_consumer = config_parse_value(arg);
+ } else {
+ opt_verbose_consumer++;
+ }
+ } else if (string_match(optname, "consumerd32-path")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--consumerd32-path");
+ } else {
+ if (consumerd32_bin_override) {
+ free((void *) consumerd32_bin);
+ }
+ consumerd32_bin = strdup(arg);
+ if (!consumerd32_bin) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ consumerd32_bin_override = 1;
+ }
+ } else if (string_match(optname, "consumerd32-libdir")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--consumerd32-libdir");
+ } else {
+ if (consumerd32_libdir_override) {
+ free((void *) consumerd32_libdir);
+ }
+ consumerd32_libdir = strdup(arg);
+ if (!consumerd32_libdir) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ consumerd32_libdir_override = 1;
+ }
+ } else if (string_match(optname, "consumerd64-path")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--consumerd64-path");
+ } else {
+ if (consumerd64_bin_override) {
+ free((void *) consumerd64_bin);
+ }
+ consumerd64_bin = strdup(arg);
+ if (!consumerd64_bin) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ consumerd64_bin_override = 1;
+ }
+ } else if (string_match(optname, "consumerd64-libdir")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--consumerd64-libdir");
+ } else {
+ if (consumerd64_libdir_override) {
+ free((void *) consumerd64_libdir);
+ }
+ consumerd64_libdir = strdup(arg);
+ if (!consumerd64_libdir) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ consumerd64_libdir_override = 1;
+ }
+ } else if (string_match(optname, "pidfile") || opt == 'p') {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-p, --pidfile");
+ } else {
+ free(opt_pidfile);
+ opt_pidfile = strdup(arg);
+ if (!opt_pidfile) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ }
+ } else if (string_match(optname, "agent-tcp-port")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--agent-tcp-port");
+ } else {
+ unsigned long v;
+
+ errno = 0;
+ v = strtoul(arg, NULL, 0);
+ if (errno != 0 || !isdigit(arg[0])) {
+ ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
+ return -1;
+ }
+ if (v == 0 || v >= 65535) {
+ ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
+ return -1;
+ }
+ agent_tcp_port = (uint32_t) v;
+ DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
+ }
+ } else if (string_match(optname, "load") || opt == 'l') {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-l, --load");
+ } else {
+ free(opt_load_session_path);
+ opt_load_session_path = strdup(arg);
+ if (!opt_load_session_path) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ }
+ } else if (string_match(optname, "kmod-probes")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--kmod-probes");
+ } else {
+ free(kmod_probes_list);
+ kmod_probes_list = strdup(arg);
+ if (!kmod_probes_list) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ }
+ } else if (string_match(optname, "extra-kmod-probes")) {
+ if (!arg || *arg == '\0') {
+ ret = -EINVAL;
+ goto end;
+ }
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "--extra-kmod-probes");
+ } else {
+ free(kmod_extra_probes_list);
+ kmod_extra_probes_list = strdup(arg);
+ if (!kmod_extra_probes_list) {
+ PERROR("strdup");
+ ret = -ENOMEM;
+ }
+ }
+ } else if (string_match(optname, "config") || opt == 'f') {
+ /* This is handled in set_options() thus silent skip. */
+ goto end;
+ } else {
+ /* Unknown option or other error.
+ * Error is printed by getopt, just return */
+ ret = -1;
+ }
+
+end:
+ if (ret == -EINVAL) {
+ const char *opt_name = "unknown";
+ int i;
+
+ for (i = 0; i < sizeof(long_options) / sizeof(struct option);
+ i++) {
+ if (opt == long_options[i].val) {
+ opt_name = long_options[i].name;
+ break;
+ }
+ }
+
+ WARN("Invalid argument provided for option \"%s\", using default value.",
+ opt_name);
+ }
+
+ return ret;
+}
/*
- * usage function on stderr
+ * config_entry_handler_cb used to handle options read from a config file.
+ * See config_entry_handler_cb comment in common/config/session-config.h for the
+ * return value conventions.
*/
-static void usage(void)
+static int config_entry_handler(const struct config_entry *entry, void *unused)
{
- fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
- fprintf(stderr, " -h, --help Display this usage.\n");
- fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
- fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
- fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
- fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
- fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
- fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
- fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
- fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
- fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
- fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
- fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
- fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
- 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, " -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");
+ 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;
+ }
+
+ /*
+ * 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 argument parsing
+ * daemon configuration loading and argument parsing
*/
-static int parse_args(int argc, char **argv)
+static int set_options(int argc, char **argv)
{
- 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' },
- { "pidfile", 1, 0, 'p' },
- { NULL, 0, 0, 0 }
- };
+ 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;
+ }
+
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-f, --config");
+ } else {
+ 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) {
- int option_index = 0;
- c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
- long_options, &option_index);
+ option_index = -1;
+ /*
+ * getopt_long() will not set option_index if it encounters a
+ * short option.
+ */
+ c = getopt_long(argc, argv, optstring, long_options,
+ &option_index);
if (c == -1) {
break;
}
- switch (c) {
- case 0:
- fprintf(stderr, "option %s", long_options[option_index].name);
- if (optarg) {
- fprintf(stderr, " with arg %s\n", optarg);
- }
- 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':
- lttng_opt_quiet = 1;
- break;
- case 'v':
- /* Verbose level can increase using multiple -v */
- lttng_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;
- break;
- case 'p':
- opt_pidfile = optarg;
+ /*
+ * Pass NULL as the long option name if popt left the index
+ * unset.
+ */
+ ret = set_option(c, optarg,
+ option_index < 0 ? NULL :
+ 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;
}
/*
int ret;
gid_t gid;
- ret = allowed_group();
- if (ret < 0) {
- WARN("No tracing group detected");
- ret = 0;
- goto end;
- }
-
- gid = ret;
+ 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;
}
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 = -1;
+ 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;
}
/*
* Set open files limit to unlimited. This daemon can open a large number of
- * file descriptors in order to consumer multiple kernel traces.
+ * file descriptors in order to consume multiple kernel traces.
*/
static void set_ulimit(void)
{
int ret;
struct rlimit lim;
- /* The kernel does not allowed an infinite limit for open files */
+ /* The kernel does not allow an infinite limit for open files */
lim.rlim_cur = 65535;
lim.rlim_max = 65535;
/*
* Write pidfile using the rundir and opt_pidfile.
*/
-static void write_pidfile(void)
+static int write_pidfile(void)
{
int ret;
char pidfile_path[PATH_MAX];
assert(rundir);
if (opt_pidfile) {
- strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
+ if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
+ ret = -1;
+ goto error;
+ }
} else {
/* Build pidfile path from rundir and opt_pidfile. */
ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
}
/*
- * 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.
+ * Create pid file in rundir.
*/
- (void) utils_create_pid_file(getpid(), pidfile_path);
+ ret = utils_create_pid_file(getpid(), pidfile_path);
+error:
+ return ret;
+}
+
+/*
+ * Create lockfile using the rundir and return its fd.
+ */
+static int create_lockfile(void)
+{
+ int ret;
+ char lockfile_path[PATH_MAX];
+ ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
+ if (ret < 0) {
+ goto error;
+ }
+
+ ret = utils_create_lock_file(lockfile_path);
error:
- return;
+ return ret;
+}
+
+/*
+ * Write agent TCP port using the rundir.
+ */
+static int write_agent_port(void)
+{
+ int ret;
+ char path[PATH_MAX];
+
+ assert(rundir);
+
+ ret = snprintf(path, sizeof(path), "%s/"
+ DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf agent port path");
+ goto error;
+ }
+
+ /*
+ * Create TCP agent port file in rundir.
+ */
+ ret = utils_create_pid_file(agent_tcp_port, path);
+
+error:
+ return ret;
}
/*
*/
int main(int argc, char **argv)
{
- int ret = 0;
+ int ret = 0, retval = 0;
void *status;
const char *home_path, *env_app_timeout;
rcu_register_thread();
+ if (set_signal_handler()) {
+ retval = -1;
+ goto exit_set_signal_handler;
+ }
+
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.
+ *
+ * We have an exit_options exit path to free memory reserved by
+ * set_options. This is needed because the rest of sessiond_cleanup()
+ * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
+ * depends on set_options.
+ */
progname = argv[0];
- if ((ret = parse_args(argc, argv)) < 0) {
- goto error;
+ if (set_options(argc, argv)) {
+ retval = -1;
+ goto exit_options;
}
/* Daemonize */
- if (opt_daemon) {
+ if (opt_daemon || opt_background) {
int i;
- /*
- * fork
- * child: setsid, close FD 0, 1, 2, chdir /
- * parent: exit (if fork is successful)
- */
- ret = daemon(0, 0);
+ ret = lttng_daemonize(&child_ppid, &recv_child_signal,
+ !opt_background);
if (ret < 0) {
- PERROR("daemon");
- goto error;
+ retval = -1;
+ goto exit_options;
}
+
/*
- * We are in the child. Make sure all other file
- * descriptors are closed, in case we are called with
- * more opened file descriptors than the standard ones.
+ * 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);
}
}
+ if (run_as_create_worker(argv[0]) < 0) {
+ goto exit_create_run_as_worker_cleanup;
+ }
+
+ /*
+ * Starting from here, we can create threads. This needs to be after
+ * lttng_daemonize due to RCU.
+ */
+
+ /*
+ * 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");
+ retval = -1;
+ goto exit_health_sessiond_cleanup;
+ }
+
+ if (init_ht_cleanup_quit_pipe()) {
+ retval = -1;
+ goto exit_ht_cleanup_quit_pipe;
+ }
+
+ /* Setup the thread ht_cleanup communication pipe. */
+ if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
+ retval = -1;
+ goto exit_ht_cleanup_pipe;
+ }
+
+ /* Create thread to clean up RCU hash tables */
+ ret = pthread_create(&ht_cleanup_thread, NULL,
+ thread_ht_cleanup, (void *) NULL);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_create ht_cleanup");
+ retval = -1;
+ goto exit_ht_cleanup;
+ }
+
/* Create thread quit pipe */
- if ((ret = init_thread_quit_pipe()) < 0) {
- goto error;
+ if (init_thread_quit_pipe()) {
+ retval = -1;
+ goto exit_init_data;
}
/* Check if daemon is UID = 0 */
if (is_root) {
rundir = strdup(DEFAULT_LTTNG_RUNDIR);
+ if (!rundir) {
+ retval = -1;
+ goto exit_init_data;
+ }
/* Create global run dir with root access */
- ret = create_lttng_rundir(rundir);
- if (ret < 0) {
- goto error;
+ if (create_lttng_rundir(rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
if (strlen(apps_unix_sock_path) == 0) {
- snprintf(apps_unix_sock_path, PATH_MAX,
+ ret = snprintf(apps_unix_sock_path, PATH_MAX,
DEFAULT_GLOBAL_APPS_UNIX_SOCK);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
if (strlen(client_unix_sock_path) == 0) {
- snprintf(client_unix_sock_path, PATH_MAX,
+ ret = snprintf(client_unix_sock_path, PATH_MAX,
DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
/* Set global SHM for ust */
if (strlen(wait_shm_path) == 0) {
- snprintf(wait_shm_path, PATH_MAX,
+ ret = snprintf(wait_shm_path, PATH_MAX,
DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
if (strlen(health_unix_sock_path) == 0) {
- snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
+ ret = snprintf(health_unix_sock_path,
+ sizeof(health_unix_sock_path),
DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
/* Setup kernel consumerd path */
- snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
+ ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
- snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
+ ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
DBG2("Kernel consumer err path: %s",
kconsumer_data.err_unix_sock_path);
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.");
- ret = -EPERM;
- goto error;
+ retval = -1;
+ goto exit_init_data;
}
/*
*/
ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
if (ret < 0) {
- ret = -ENOMEM;
- goto error;
+ retval = -1;
+ goto exit_init_data;
}
- ret = create_lttng_rundir(rundir);
- if (ret < 0) {
- goto error;
+ if (create_lttng_rundir(rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
if (strlen(apps_unix_sock_path) == 0) {
- snprintf(apps_unix_sock_path, PATH_MAX,
- DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
+ ret = snprintf(apps_unix_sock_path, PATH_MAX,
+ DEFAULT_HOME_APPS_UNIX_SOCK,
+ home_path);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
/* Set the cli tool unix socket path */
if (strlen(client_unix_sock_path) == 0) {
- snprintf(client_unix_sock_path, PATH_MAX,
- DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
+ ret = snprintf(client_unix_sock_path, PATH_MAX,
+ DEFAULT_HOME_CLIENT_UNIX_SOCK,
+ home_path);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
/* Set global SHM for ust */
if (strlen(wait_shm_path) == 0) {
- snprintf(wait_shm_path, PATH_MAX,
- DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
+ ret = snprintf(wait_shm_path, PATH_MAX,
+ DEFAULT_HOME_APPS_WAIT_SHM_PATH,
+ getuid());
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
/* 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);
+ ret = snprintf(health_unix_sock_path,
+ sizeof(health_unix_sock_path),
+ DEFAULT_HOME_HEALTH_UNIX_SOCK,
+ home_path);
+ if (ret < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
}
}
+ lockfile_fd = create_lockfile();
+ if (lockfile_fd < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
+
/* Set consumer initial state */
kernel_consumerd_state = CONSUMER_STOPPED;
ust_consumerd_state = CONSUMER_STOPPED;
DBG("LTTng run directory path: %s", rundir);
/* 32 bits consumerd path setup */
- snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
+ ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
- snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
+ if (ret < 0) {
+ PERROR("snprintf 32-bit consumer error socket path");
+ retval = -1;
+ goto exit_init_data;
+ }
+ ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
+ if (ret < 0) {
+ PERROR("snprintf 32-bit consumer command socket path");
+ retval = -1;
+ goto exit_init_data;
+ }
DBG2("UST consumer 32 bits err path: %s",
ustconsumer32_data.err_unix_sock_path);
ustconsumer32_data.cmd_unix_sock_path);
/* 64 bits consumerd path setup */
- snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
+ ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
- snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
+ if (ret < 0) {
+ PERROR("snprintf 64-bit consumer error socket path");
+ retval = -1;
+ goto exit_init_data;
+ }
+ ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
+ if (ret < 0) {
+ PERROR("snprintf 64-bit consumer command socket path");
+ retval = -1;
+ goto exit_init_data;
+ }
DBG2("UST consumer 64 bits err path: %s",
ustconsumer64_data.err_unix_sock_path);
/*
* See if daemon already exist.
*/
- if ((ret = check_existing_daemon()) < 0) {
+ if (check_existing_daemon()) {
ERR("Already running daemon.\n");
/*
* We do not goto exit because we must not cleanup()
* because a daemon is already running.
*/
- goto error;
+ retval = -1;
+ goto exit_init_data;
}
/*
* Init UST app hash table. Alloc hash table before this point since
* cleanup() can get called after that point.
*/
- ust_app_ht_alloc();
+ if (ust_app_ht_alloc()) {
+ ERR("Failed to allocate UST app hash table");
+ retval = -1;
+ goto exit_init_data;
+ }
- /* After this point, we can safely call cleanup() with "goto exit" */
+ /*
+ * Initialize agent app hash table. We allocate the hash table here
+ * since cleanup() can get called after this point.
+ */
+ if (agent_app_ht_alloc()) {
+ ERR("Failed to allocate Agent app hash table");
+ retval = -1;
+ goto exit_init_data;
+ }
/*
* These actions must be executed as root. We do that *after* setting up
* kernel tracer.
*/
if (is_root) {
- ret = set_consumer_sockets(&kconsumer_data, rundir);
- if (ret < 0) {
- goto exit;
+ if (set_consumer_sockets(&kconsumer_data, rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
/* Setup kernel tracer */
if (!opt_no_kernel) {
init_kernel_tracer();
+ if (kernel_tracer_fd >= 0) {
+ ret = syscall_init_table();
+ if (ret < 0) {
+ ERR("Unable to populate syscall table. "
+ "Syscall tracing won't work "
+ "for this session daemon.");
+ }
+ }
}
/* Set ulimit for open files */
/* 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;
- }
-
- ret = set_consumer_sockets(&ustconsumer32_data, rundir);
- if (ret < 0) {
- goto exit;
+ if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
- if ((ret = set_signal_handler()) < 0) {
- goto exit;
+ if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
/* Setup the needed unix socket */
- if ((ret = init_daemon_socket()) < 0) {
- goto exit;
+ if (init_daemon_socket()) {
+ retval = -1;
+ goto exit_init_data;
}
/* Set credentials to socket */
- if (is_root && ((ret = set_permissions(rundir)) < 0)) {
- goto exit;
+ if (is_root && set_permissions(rundir)) {
+ retval = -1;
+ goto exit_init_data;
}
/* Get parent pid if -S, --sig-parent is specified. */
/* Setup the kernel pipe for waking up the kernel thread */
if (is_root && !opt_no_kernel) {
- if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
- goto exit;
+ if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
+ retval = -1;
+ goto exit_init_data;
}
}
/* Setup the thread apps communication pipe. */
- if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
- goto exit;
+ if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
+ retval = -1;
+ goto exit_init_data;
}
/* Setup the thread apps notify communication pipe. */
- if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
- goto exit;
+ if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
+ retval = -1;
+ goto exit_init_data;
}
/* Initialize global buffer per UID and PID registry. */
buffer_reg_init_pid_registry();
/* Init UST command queue. */
- cds_wfq_init(&ust_cmd_queue.queue);
+ cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
/*
- * Get session list pointer. This pointer MUST NOT be free(). This list is
- * statically declared in session.c
+ * Get session list pointer. This pointer MUST NOT be free'd. This list
+ * is statically declared in session.c
*/
session_list_ptr = session_get_list();
- /* Set up max poll set size */
- lttng_poll_set_max_size();
-
cmd_init();
/* Check for the application socket timeout env variable. */
app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
}
- write_pidfile();
+ ret = write_pidfile();
+ if (ret) {
+ ERR("Error in write_pidfile");
+ retval = -1;
+ goto exit_init_data;
+ }
+ ret = write_agent_port();
+ if (ret) {
+ ERR("Error in write_agent_port");
+ retval = -1;
+ goto exit_init_data;
+ }
+
+ /* Initialize communication library */
+ lttcomm_init();
+ /* Initialize TCP timeout values */
+ lttcomm_inet_init();
- /* Create thread to manage the client socket */
+ if (load_session_init_data(&load_info) < 0) {
+ retval = -1;
+ goto exit_init_data;
+ }
+ load_info->path = opt_load_session_path;
+
+ /* Create health-check thread */
ret = pthread_create(&health_thread, NULL,
thread_manage_health, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create health");
+ retval = -1;
goto exit_health;
}
/* Create thread to manage the client socket */
ret = pthread_create(&client_thread, NULL,
thread_manage_clients, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create clients");
+ retval = -1;
goto exit_client;
}
/* Create thread to dispatch registration */
ret = pthread_create(&dispatch_thread, NULL,
thread_dispatch_ust_registration, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create dispatch");
+ retval = -1;
goto exit_dispatch;
}
/* Create thread to manage application registration. */
ret = pthread_create(®_apps_thread, NULL,
thread_registration_apps, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create registration");
+ retval = -1;
goto exit_reg_apps;
}
/* Create thread to manage application socket */
ret = pthread_create(&apps_thread, NULL,
thread_manage_apps, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create apps");
+ retval = -1;
goto exit_apps;
}
/* 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 apps");
- goto exit_apps;
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_create notify");
+ retval = -1;
+ goto exit_apps_notify;
+ }
+
+ /* Create agent registration thread. */
+ ret = pthread_create(&agent_reg_thread, NULL,
+ agent_thread_manage_registration, (void *) NULL);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_create agent");
+ retval = -1;
+ goto exit_agent_reg;
}
/* Don't start this thread if kernel tracing is not requested nor root */
/* Create kernel thread to manage kernel event */
ret = pthread_create(&kernel_thread, NULL,
thread_manage_kernel, (void *) NULL);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_create kernel");
+ retval = -1;
goto exit_kernel;
}
+ }
+
+ /* Create session loading thread. */
+ ret = pthread_create(&load_session_thread, NULL, thread_load_session,
+ load_info);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_create load_session_thread");
+ retval = -1;
+ goto exit_load_session;
+ }
+
+ /*
+ * This is where we start awaiting program completion (e.g. through
+ * signal that asks threads to teardown).
+ */
+
+ ret = pthread_join(load_session_thread, &status);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_join load_session_thread");
+ retval = -1;
+ }
+exit_load_session:
+ if (is_root && !opt_no_kernel) {
ret = pthread_join(kernel_thread, &status);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
+ retval = -1;
}
}
-
exit_kernel:
- ret = pthread_join(apps_thread, &status);
- if (ret != 0) {
- PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
+
+ ret = pthread_join(agent_reg_thread, &status);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_join agent");
+ retval = -1;
+ }
+exit_agent_reg:
+
+ ret = pthread_join(apps_notify_thread, &status);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_join apps notify");
+ retval = -1;
}
+exit_apps_notify:
+ ret = pthread_join(apps_thread, &status);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_join apps");
+ retval = -1;
+ }
exit_apps:
+
ret = pthread_join(reg_apps_thread, &status);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
+ retval = -1;
}
-
exit_reg_apps:
+
+ /*
+ * Join dispatch thread after joining reg_apps_thread to ensure
+ * we don't leak applications in the queue.
+ */
ret = pthread_join(dispatch_thread, &status);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
+ retval = -1;
}
-
exit_dispatch:
+
ret = pthread_join(client_thread, &status);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_join");
- goto error; /* join error, exit without cleanup */
- }
-
- ret = join_consumer_thread(&kconsumer_data);
- if (ret != 0) {
- PERROR("join_consumer");
- 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 */
+ retval = -1;
}
-
exit_client:
+
ret = pthread_join(health_thread, &status);
- if (ret != 0) {
+ if (ret) {
+ errno = ret;
PERROR("pthread_join health thread");
- goto error; /* join error, exit without cleanup */
+ retval = -1;
}
-
exit_health:
-exit:
+
+exit_init_data:
/*
- * cleanup() is called when no other thread is running.
+ * sessiond_cleanup() is called when no other thread is running, except
+ * the ht_cleanup thread, which is needed to destroy the hash tables.
*/
rcu_thread_online();
- cleanup();
+ sessiond_cleanup();
rcu_thread_offline();
rcu_unregister_thread();
- if (!ret) {
+
+ /*
+ * Ensure all prior call_rcu are done. call_rcu callbacks may push
+ * hash tables to the ht_cleanup thread. Therefore, we ensure that
+ * the queue is empty before shutting down the clean-up thread.
+ */
+ rcu_barrier();
+
+ ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
+ if (ret < 0) {
+ ERR("write error on ht_cleanup quit pipe");
+ retval = -1;
+ }
+
+ ret = pthread_join(ht_cleanup_thread, &status);
+ if (ret) {
+ errno = ret;
+ PERROR("pthread_join ht cleanup thread");
+ retval = -1;
+ }
+exit_ht_cleanup:
+
+ utils_close_pipe(ht_cleanup_pipe);
+exit_ht_cleanup_pipe:
+
+ /*
+ * Close the ht_cleanup quit pipe.
+ */
+ utils_close_pipe(ht_cleanup_quit_pipe);
+exit_ht_cleanup_quit_pipe:
+
+ health_app_destroy(health_sessiond);
+exit_health_sessiond_cleanup:
+exit_create_run_as_worker_cleanup:
+
+exit_options:
+ sessiond_cleanup_options();
+
+exit_set_signal_handler:
+
+ if (!retval) {
exit(EXIT_SUCCESS);
+ } else {
+ exit(EXIT_FAILURE);
}
-error:
- exit(EXIT_FAILURE);
}