#include <inttypes.h>
#include <urcu/list.h>
#include <urcu/uatomic.h>
+#include <sys/stat.h>
#include <common/defaults.h>
#include <common/common.h>
#include <lttng/action/action.h>
#include <lttng/channel.h>
#include <lttng/channel-internal.h>
+#include <lttng/rotate-internal.h>
+#include <lttng/location-internal.h>
+#include <lttng/userspace-probe-internal.h>
#include <common/string-utils/string-utils.h>
#include "channel.h"
#include "kernel-consumer.h"
#include "lttng-sessiond.h"
#include "utils.h"
-#include "syscall.h"
+#include "lttng-syscall.h"
#include "agent.h"
#include "buffer-registry.h"
#include "notification-thread.h"
#include "notification-thread-commands.h"
+#include "rotate.h"
+#include "rotation-thread.h"
+#include "sessiond-timer.h"
+#include "agent-thread.h"
#include "cmd.h"
+/* Sleep for 100ms between each check for the shm path's deletion. */
+#define SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US 100000
+
+static enum lttng_error_code wait_on_path(void *path);
+
+/*
+ * Command completion handler that is used by the destroy command
+ * when a session that has a non-default shm_path is being destroyed.
+ *
+ * See comment in cmd_destroy_session() for the rationale.
+ */
+static struct destroy_completion_handler {
+ struct cmd_completion_handler handler;
+ char shm_path[member_sizeof(struct ltt_session, shm_path)];
+} destroy_completion_handler = {
+ .handler = {
+ .run = wait_on_path,
+ .data = destroy_completion_handler.shm_path
+ },
+ .shm_path = { 0 },
+};
+
+static struct cmd_completion_handler *current_completion_handler;
+
/*
* Used to keep a unique index for each relayd socket created where this value
* is associated with streams on the consumer so it can match the right relayd
goto end;
}
- ret = consumer_get_discarded_events(session->id, kchan->fd,
+ ret = consumer_get_discarded_events(session->id, kchan->key,
session->kernel_session->consumer,
discarded_events);
if (ret < 0) {
goto end;
}
- ret = consumer_get_lost_packets(session->id, kchan->fd,
+ ret = consumer_get_lost_packets(session->id, kchan->key,
session->kernel_session->consumer,
lost_packets);
if (ret < 0) {
}
}
-static void increment_extended_len(const char *filter_expression,
- struct lttng_event_exclusion *exclusion, size_t *extended_len)
+static int increment_extended_len(const char *filter_expression,
+ struct lttng_event_exclusion *exclusion,
+ const struct lttng_userspace_probe_location *probe_location,
+ size_t *extended_len)
{
+ int ret = 0;
+
*extended_len += sizeof(struct lttcomm_event_extended_header);
if (filter_expression) {
if (exclusion) {
*extended_len += exclusion->count * LTTNG_SYMBOL_NAME_LEN;
}
+
+ if (probe_location) {
+ ret = lttng_userspace_probe_location_serialize(probe_location,
+ NULL, NULL);
+ if (ret < 0) {
+ goto end;
+ }
+ *extended_len += ret;
+ }
+ ret = 0;
+end:
+ return ret;
}
-static void append_extended_info(const char *filter_expression,
- struct lttng_event_exclusion *exclusion, void **extended_at)
+static int append_extended_info(const char *filter_expression,
+ struct lttng_event_exclusion *exclusion,
+ struct lttng_userspace_probe_location *probe_location,
+ void **extended_at)
{
- struct lttcomm_event_extended_header extended_header;
+ int ret = 0;
size_t filter_len = 0;
size_t nb_exclusions = 0;
+ size_t userspace_probe_location_len = 0;
+ struct lttng_dynamic_buffer location_buffer;
+ struct lttcomm_event_extended_header extended_header;
if (filter_expression) {
filter_len = strlen(filter_expression) + 1;
nb_exclusions = exclusion->count;
}
+ if (probe_location) {
+ lttng_dynamic_buffer_init(&location_buffer);
+ ret = lttng_userspace_probe_location_serialize(probe_location,
+ &location_buffer, NULL);
+ if (ret < 0) {
+ ret = -1;
+ goto end;
+ }
+ userspace_probe_location_len = location_buffer.size;
+ }
+
/* Set header fields */
extended_header.filter_len = filter_len;
extended_header.nb_exclusions = nb_exclusions;
+ extended_header.userspace_probe_location_len = userspace_probe_location_len;
/* Copy header */
memcpy(*extended_at, &extended_header, sizeof(extended_header));
memcpy(*extended_at, &exclusion->names, len);
*extended_at += len;
}
+
+ if (probe_location) {
+ memcpy(*extended_at, location_buffer.data, location_buffer.size);
+ *extended_at += location_buffer.size;
+ lttng_dynamic_buffer_reset(&location_buffer);
+ }
+ ret = 0;
+end:
+ return ret;
}
/*
int i = 0, ret = 0;
unsigned int nb_event = 0;
struct agent_event *event;
- struct lttng_event *tmp_events;
+ struct lttng_event *tmp_events = NULL;
struct lttng_ht_iter iter;
size_t extended_len = 0;
void *extended_at;
*/
rcu_read_lock();
cds_lfht_for_each_entry(agt->events->ht, &iter.iter, event, node.node) {
- increment_extended_len(event->filter_expression, NULL,
+ ret = increment_extended_len(event->filter_expression, NULL, NULL,
&extended_len);
+ if (ret) {
+ DBG("Error computing the length of extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto error;
+ }
}
rcu_read_unlock();
i++;
/* Append extended info */
- append_extended_info(event->filter_expression, NULL,
+ ret = append_extended_info(event->filter_expression, NULL, NULL,
&extended_at);
+ if (ret) {
+ DBG("Error appending extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto error;
+ }
}
- rcu_read_unlock();
*events = tmp_events;
ret = nb_event;
-
-error:
assert(nb_event == i);
+
+end:
+ rcu_read_unlock();
return ret;
+error:
+ free(tmp_events);
+ goto end;
}
/*
continue;
}
- increment_extended_len(uevent->filter_expression,
- uevent->exclusion, &extended_len);
+ ret = increment_extended_len(uevent->filter_expression,
+ uevent->exclusion, NULL, &extended_len);
+ if (ret) {
+ DBG("Error computing the length of extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto end;
+ }
}
if (nb_event == 0) {
/* All events are internal, skip. */
i++;
/* Append extended info */
- append_extended_info(uevent->filter_expression,
- uevent->exclusion, &extended_at);
+ ret = append_extended_info(uevent->filter_expression,
+ uevent->exclusion, NULL, &extended_at);
+ if (ret) {
+ DBG("Error appending extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto end;
+ }
}
ret = nb_event;
/* Compute required extended infos size */
cds_list_for_each_entry(event, &kchan->events_list.head, list) {
- increment_extended_len(event->filter_expression, NULL,
+ ret = increment_extended_len(event->filter_expression, NULL,
+ event->userspace_probe_location,
&extended_len);
+ if (ret) {
+ DBG("Error computing the length of extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto error;
+ }
}
*total_size = nb_event * sizeof(struct lttng_event) + extended_len;
*events = zmalloc(*total_size);
if (*events == NULL) {
- ret = LTTNG_ERR_FATAL;
+ ret = -LTTNG_ERR_FATAL;
goto error;
}
memcpy(&(*events)[i].attr.probe, &event->event->u.kprobe,
sizeof(struct lttng_kernel_kprobe));
break;
+ case LTTNG_KERNEL_UPROBE:
+ (*events)[i].type = LTTNG_EVENT_USERSPACE_PROBE;
+ break;
case LTTNG_KERNEL_FUNCTION:
(*events)[i].type = LTTNG_EVENT_FUNCTION;
memcpy(&((*events)[i].attr.ftrace), &event->event->u.ftrace,
(*events)[i].type = LTTNG_EVENT_SYSCALL;
break;
case LTTNG_KERNEL_ALL:
+ /* fall-through. */
+ default:
assert(0);
break;
}
i++;
/* Append extended info */
- append_extended_info(event->filter_expression, NULL,
- &extended_at);
+ ret = append_extended_info(event->filter_expression, NULL,
+ event->userspace_probe_location, &extended_at);
+ if (ret) {
+ DBG("Error appending extended info message");
+ ret = -LTTNG_ERR_FATAL;
+ goto error;
+ }
}
end:
break;
case LTTNG_DST_PATH:
DBG2("Setting trace directory path from URI to %s", uri->dst.path);
- memset(consumer->dst.trace_path, 0,
- sizeof(consumer->dst.trace_path));
+ memset(consumer->dst.session_root_path, 0,
+ sizeof(consumer->dst.session_root_path));
/* Explicit length checks for strcpy and strcat. */
if (strlen(uri->dst.path) + strlen(default_trace_dir)
- >= sizeof(consumer->dst.trace_path)) {
+ >= sizeof(consumer->dst.session_root_path)) {
ret = LTTNG_ERR_FATAL;
goto error;
}
- strcpy(consumer->dst.trace_path, uri->dst.path);
+ strcpy(consumer->dst.session_root_path, uri->dst.path);
/* Append default trace dir */
- strcat(consumer->dst.trace_path, default_trace_dir);
+ strcat(consumer->dst.session_root_path, default_trace_dir);
/* Flag consumer as local. */
consumer->type = CONSUMER_DST_LOCAL;
break;
/* Check relayd version */
ret = relayd_version_check(rsock);
- if (ret < 0) {
- ret = LTTNG_ERR_RELAYD_VERSION_FAIL;
+ if (ret == LTTNG_ERR_RELAYD_VERSION_FAIL) {
+ goto close_sock;
+ } else if (ret < 0) {
+ ERR("Unable to reach lttng-relayd");
+ ret = LTTNG_ERR_RELAYD_CONNECT_FAIL;
goto close_sock;
}
consumer->relay_major_version = rsock->major;
*
* The consumer socket lock must be held by the caller.
*/
-static int send_consumer_relayd_socket(enum lttng_domain_type domain,
- unsigned int session_id, struct lttng_uri *relayd_uri,
+static int send_consumer_relayd_socket(unsigned int session_id,
+ struct lttng_uri *relayd_uri,
struct consumer_output *consumer,
struct consumer_socket *consumer_sock,
char *session_name, char *hostname, int session_live_timer)
/* Sending control relayd socket. */
if (!sock->control_sock_sent) {
- ret = send_consumer_relayd_socket(domain, session_id,
+ ret = send_consumer_relayd_socket(session_id,
&consumer->dst.net.control, consumer, sock,
session_name, hostname, session_live_timer);
if (ret != LTTNG_OK) {
/* Sending data relayd socket. */
if (!sock->data_sock_sent) {
- ret = send_consumer_relayd_socket(domain, session_id,
+ ret = send_consumer_relayd_socket(session_id,
&consumer->dst.net.data, consumer, sock,
session_name, hostname, session_live_timer);
if (ret != LTTNG_OK) {
}
/* Quiescent wait after starting trace */
- kernel_wait_quiescent(kernel_tracer_fd);
+ kernel_wait_quiescent(wpipe);
ksess->active = 1;
break;
}
case LTTNG_DOMAIN_UST:
+ break;
case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_LOG4J:
case LTTNG_DOMAIN_PYTHON:
+ if (!agent_tracing_is_enabled()) {
+ DBG("Attempted to enable a channel in an agent domain but the agent thread is not running");
+ ret = LTTNG_ERR_AGENT_TRACING_DISABLED;
+ goto error;
+ }
break;
default:
ret = LTTNG_ERR_UNKNOWN_DOMAIN;
int ret, chan_kern_created = 0, chan_ust_created = 0;
char *app_ctx_provider_name = NULL, *app_ctx_name = NULL;
+ /*
+ * Don't try to add a context if the session has been started at
+ * some point in time before. The tracer does not allow it and would
+ * result in a corrupted trace.
+ */
+ if (session->has_been_started) {
+ ret = LTTNG_ERR_TRACE_ALREADY_STARTED;
+ goto end;
+ }
+
if (ctx->ctx == LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
app_ctx_provider_name = ctx->u.app_ctx.provider_name;
app_ctx_name = ctx->u.app_ctx.ctx_name;
break;
}
case LTTNG_EVENT_PROBE:
+ case LTTNG_EVENT_USERSPACE_PROBE:
case LTTNG_EVENT_FUNCTION:
case LTTNG_EVENT_FUNCTION_ENTRY:
case LTTNG_EVENT_TRACEPOINT:
ret = validate_ust_event_name(event->name);
if (ret) {
WARN("Userspace event name %s failed validation.",
- event->name ?
- event->name : "NULL");
+ event->name);
ret = LTTNG_ERR_INVALID_EVENT_NAME;
goto error;
}
assert(usess);
+ if (!agent_tracing_is_enabled()) {
+ DBG("Attempted to enable an event in an agent domain but the agent thread is not running");
+ ret = LTTNG_ERR_AGENT_TRACING_DISABLED;
+ goto error;
+ }
+
agt = trace_ust_find_agent(usess, domain->type);
if (!agt) {
agt = agent_create(domain->type);
return -ret;
}
+static
+int domain_mkdir(const struct consumer_output *output,
+ const struct ltt_session *session,
+ uid_t uid, gid_t gid)
+{
+ struct consumer_socket *socket;
+ struct lttng_ht_iter iter;
+ int ret;
+ char *path = NULL;
+
+ if (!output || !output->socks) {
+ ERR("No consumer output found");
+ ret = -1;
+ goto end;
+ }
+
+ path = zmalloc(LTTNG_PATH_MAX * sizeof(char));
+ if (!path) {
+ ERR("Cannot allocate mkdir path");
+ ret = -1;
+ goto end;
+ }
+
+ ret = snprintf(path, LTTNG_PATH_MAX, "%s%s%s",
+ session_get_base_path(session),
+ output->chunk_path, output->subdir);
+ if (ret < 0 || ret >= LTTNG_PATH_MAX) {
+ ERR("Format path");
+ ret = -1;
+ goto end;
+ }
+
+ DBG("Domain mkdir %s for session %" PRIu64, path, session->id);
+ rcu_read_lock();
+ /*
+ * We have to iterate to find a socket, but we only need to send the
+ * rename command to one consumer, so we break after the first one.
+ */
+ cds_lfht_for_each_entry(output->socks->ht, &iter.iter, socket, node.node) {
+ pthread_mutex_lock(socket->lock);
+ ret = consumer_mkdir(socket, session->id, output, path, uid, gid);
+ pthread_mutex_unlock(socket->lock);
+ if (ret) {
+ ERR("Consumer mkdir");
+ ret = -1;
+ goto end_unlock;
+ }
+ break;
+ }
+
+ ret = 0;
+
+end_unlock:
+ rcu_read_unlock();
+end:
+ free(path);
+ return ret;
+}
+
+static
+int session_mkdir(const struct ltt_session *session)
+{
+ int ret;
+ struct consumer_output *output;
+ uid_t uid;
+ gid_t gid;
+
+ /*
+ * Unsupported feature in lttng-relayd before 2.11, not an error since it
+ * is only needed for session rotation and the user will get an error
+ * on rotate.
+ */
+ if (session->consumer->type == CONSUMER_DST_NET &&
+ session->consumer->relay_major_version == 2 &&
+ session->consumer->relay_minor_version < 11) {
+ ret = 0;
+ goto end;
+ }
+
+ if (session->kernel_session) {
+ output = session->kernel_session->consumer;
+ uid = session->kernel_session->uid;
+ gid = session->kernel_session->gid;
+ ret = domain_mkdir(output, session, uid, gid);
+ if (ret) {
+ ERR("Mkdir kernel");
+ goto end;
+ }
+ }
+
+ if (session->ust_session) {
+ output = session->ust_session->consumer;
+ uid = session->ust_session->uid;
+ gid = session->ust_session->gid;
+ ret = domain_mkdir(output, session, uid, gid);
+ if (ret) {
+ ERR("Mkdir UST");
+ goto end;
+ }
+ }
+
+ ret = 0;
+
+end:
+ return ret;
+}
+
/*
* Command LTTNG_START_TRACE processed by the client thread.
*
goto error;
}
+ /*
+ * Record the timestamp of the first time the session is started for
+ * an eventual session rotation call.
+ */
+ if (!session->has_been_started) {
+ session->current_chunk_start_ts = time(NULL);
+ if (session->current_chunk_start_ts == (time_t) -1) {
+ PERROR("Failed to retrieve the \"%s\" session's start time",
+ session->name);
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+ if (!session->snapshot_mode && session->output_traces) {
+ ret = session_mkdir(session);
+ if (ret) {
+ ERR("Failed to create the session directories");
+ ret = LTTNG_ERR_CREATE_DIR_FAIL;
+ goto error;
+ }
+ }
+ }
+
/* Kernel tracing */
if (ksession != NULL) {
+ DBG("Start kernel tracing session %s", session->name);
ret = start_kernel_session(ksession, kernel_tracer_fd);
if (ret != LTTNG_OK) {
goto error;
session->has_been_started = 1;
session->active = 1;
+ /*
+ * Clear the flag that indicates that a rotation was done while the
+ * session was stopped.
+ */
+ session->rotated_after_last_stop = false;
+
+ if (session->rotate_timer_period) {
+ ret = sessiond_rotate_timer_start(session,
+ session->rotate_timer_period);
+ if (ret < 0) {
+ ERR("Failed to enable rotate timer");
+ ret = LTTNG_ERR_UNK;
+ goto error;
+ }
+ }
+
ret = LTTNG_OK;
error:
return ret;
}
+static
+int rename_active_chunk(struct ltt_session *session)
+{
+ int ret;
+
+ session->current_archive_id++;
+
+ /*
+ * The currently active tracing path is now the folder we
+ * want to rename.
+ */
+ ret = lttng_strncpy(session->rotation_chunk.current_rotate_path,
+ session->rotation_chunk.active_tracing_path,
+ sizeof(session->rotation_chunk.current_rotate_path));
+ if (ret) {
+ ERR("Failed to copy active tracing path");
+ goto end;
+ }
+
+ ret = rename_complete_chunk(session, time(NULL));
+ if (ret < 0) {
+ ERR("Failed to rename current rotate path");
+ goto end;
+ }
+
+ /*
+ * We just renamed, the folder, we didn't do an actual rotation, so
+ * the active tracing path is now the renamed folder and we have to
+ * restore the rotate count.
+ */
+ ret = lttng_strncpy(session->rotation_chunk.active_tracing_path,
+ session->rotation_chunk.current_rotate_path,
+ sizeof(session->rotation_chunk.active_tracing_path));
+ if (ret) {
+ ERR("Failed to rename active session chunk tracing path");
+ goto end;
+ }
+end:
+ session->current_archive_id--;
+ return ret;
+}
+
/*
* Command LTTNG_STOP_TRACE processed by the client thread.
*/
struct ltt_kernel_channel *kchan;
struct ltt_kernel_session *ksession;
struct ltt_ust_session *usess;
+ bool error_occured = false;
assert(session);
+ DBG("Begin stop session %s (id %" PRIu64 ")", session->name, session->id);
/* Short cut */
ksession = session->kernel_session;
usess = session->ust_session;
goto error;
}
+ if (session->rotate_relay_pending_timer_enabled) {
+ sessiond_timer_rotate_pending_stop(session);
+ }
+
+ if (session->rotate_timer_enabled) {
+ sessiond_rotate_timer_stop(session);
+ }
+
+ if (session->current_archive_id > 0 && !session->rotate_pending) {
+ ret = rename_active_chunk(session);
+ if (ret) {
+ /*
+ * This error should not prevent the user from stopping
+ * the session. However, it will be reported at the end.
+ */
+ error_occured = true;
+ }
+ }
+
/* Kernel tracer */
if (ksession && ksession->active) {
DBG("Stop kernel tracing");
}
ksession->active = 0;
+ DBG("Kernel session stopped %s (id %" PRIu64 ")", session->name,
+ session->id);
}
if (usess && usess->active) {
/* Flag inactive after a successful stop. */
session->active = 0;
- ret = LTTNG_OK;
+ ret = !error_occured ? LTTNG_OK : LTTNG_ERR_UNK;
error:
return ret;
*
* Called with session lock held.
*/
-int cmd_destroy_session(struct ltt_session *session, int wpipe)
+int cmd_destroy_session(struct ltt_session *session, int wpipe,
+ struct notification_thread_handle *notification_thread_handle)
{
int ret;
struct ltt_ust_session *usess;
usess = session->ust_session;
ksess = session->kernel_session;
+ DBG("Begin destroy session %s (id %" PRIu64 ")", session->name, session->id);
+
+ if (session->rotate_relay_pending_timer_enabled) {
+ sessiond_timer_rotate_pending_stop(session);
+ }
+
+ if (session->rotate_timer_enabled) {
+ sessiond_rotate_timer_stop(session);
+ }
+
+ if (session->rotate_size) {
+ unsubscribe_session_consumed_size_rotation(session, notification_thread_handle);
+ session->rotate_size = 0;
+ }
+
+ /*
+ * The rename of the current chunk is performed at stop, but if we rotated
+ * the session after the previous stop command, we need to rename the
+ * new (and empty) chunk that was started in between.
+ */
+ if (session->rotated_after_last_stop) {
+ rename_active_chunk(session);
+ }
+
/* Clean kernel session teardown */
kernel_destroy_session(ksess);
PERROR("write kernel poll pipe");
}
+ if (session->shm_path[0]) {
+ /*
+ * When a session is created with an explicit shm_path,
+ * the consumer daemon will create its shared memory files
+ * at that location and will *not* unlink them. This is normal
+ * as the intention of that feature is to make it possible
+ * to retrieve the content of those files should a crash occur.
+ *
+ * To ensure the content of those files can be used, the
+ * sessiond daemon will replicate the content of the metadata
+ * cache in a metadata file.
+ *
+ * On clean-up, it is expected that the consumer daemon will
+ * unlink the shared memory files and that the session daemon
+ * will unlink the metadata file. Then, the session's directory
+ * in the shm path can be removed.
+ *
+ * Unfortunately, a flaw in the design of the sessiond's and
+ * consumerd's tear down of channels makes it impossible to
+ * determine when the sessiond _and_ the consumerd have both
+ * destroyed their representation of a channel. For one, the
+ * unlinking, close, and rmdir happen in deferred 'call_rcu'
+ * callbacks in both daemons.
+ *
+ * However, it is also impossible for the sessiond to know when
+ * the consumer daemon is done destroying its channel(s) since
+ * it occurs as a reaction to the closing of the channel's file
+ * descriptor. There is no resulting communication initiated
+ * from the consumerd to the sessiond to confirm that the
+ * operation is completed (and was successful).
+ *
+ * Until this is all fixed, the session daemon checks for the
+ * removal of the session's shm path which makes it possible
+ * to safely advertise a session as having been destroyed.
+ *
+ * Prior to this fix, it was not possible to reliably save
+ * a session making use of the --shm-path option, destroy it,
+ * and load it again. This is because the creation of the
+ * session would fail upon seeing the session's shm path
+ * already in existence.
+ *
+ * Note that none of the error paths in the check for the
+ * directory's existence return an error. This is normal
+ * as there isn't much that can be done. The session will
+ * be destroyed properly, except that we can't offer the
+ * guarantee that the same session can be re-created.
+ */
+ current_completion_handler = &destroy_completion_handler.handler;
+ ret = lttng_strncpy(destroy_completion_handler.shm_path,
+ session->shm_path,
+ sizeof(destroy_completion_handler.shm_path));
+ assert(!ret);
+ }
ret = session_destroy(session);
return ret;
sizeof(sessions[i].path), session);
} else {
ret = snprintf(sessions[i].path, sizeof(sessions[i].path), "%s",
- session->consumer->dst.trace_path);
+ session->consumer->dst.session_root_path);
}
if (ret < 0) {
PERROR("snprintf session path");
assert(session);
+ DBG("Data pending for session %s", session->name);
+
/* Session MUST be stopped to ask for data availability. */
if (session->active) {
ret = LTTNG_ERR_SESSION_STARTED;
}
}
+ /*
+ * A rotation is still pending, we have to wait.
+ */
+ if (session->rotate_pending) {
+ DBG("Rotate still pending for session %s", session->name);
+ ret = 1;
+ goto error;
+ }
+
if (ksess && ksess->consumer) {
ret = consumer_is_data_pending(ksess->id, ksess->consumer);
if (ret == 1) {
}
if (output->consumer->type == CONSUMER_DST_LOCAL) {
if (lttng_strncpy(list[idx].ctrl_url,
- output->consumer->dst.trace_path,
+ output->consumer->dst.session_root_path,
sizeof(list[idx].ctrl_url))) {
ret = -LTTNG_ERR_INVALID;
goto error;
int clear_metadata_file(int fd)
{
int ret;
+ off_t lseek_ret;
- ret = lseek(fd, 0, SEEK_SET);
- if (ret < 0) {
+ lseek_ret = lseek(fd, 0, SEEK_SET);
+ if (lseek_ret < 0) {
PERROR("lseek");
+ ret = -1;
goto end;
}
rcu_read_lock();
cds_lfht_for_each_entry(snap_output->consumer->socks->ht, &iter.iter,
socket, node.node) {
+ pthread_mutex_lock(socket->lock);
ret = send_consumer_relayd_sockets(0, session->id,
snap_output->consumer, socket,
session->name, session->hostname,
session->live_timer);
+ pthread_mutex_unlock(socket->lock);
if (ret != LTTNG_OK) {
rcu_read_unlock();
goto error;
return 0;
}
+/*
+ * Command LTTNG_ROTATE_SESSION from the lttng-ctl library.
+ *
+ * Ask the consumer to rotate the session output directory.
+ * The session lock must be held.
+ *
+ * Return LTTNG_OK on success or else a LTTNG_ERR code.
+ */
+int cmd_rotate_session(struct ltt_session *session,
+ struct lttng_rotate_session_return *rotate_return)
+{
+ int ret;
+ size_t strf_ret;
+ struct tm *timeinfo;
+ char datetime[21];
+ time_t now;
+ bool ust_active = false;
+
+ assert(session);
+
+ if (!session->has_been_started) {
+ ret = -LTTNG_ERR_START_SESSION_ONCE;
+ goto end;
+ }
+
+ if (session->live_timer || session->snapshot_mode ||
+ !session->output_traces) {
+ ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE;
+ goto end;
+ }
+
+ /*
+ * Unsupported feature in lttng-relayd before 2.11.
+ */
+ if (session->consumer->type == CONSUMER_DST_NET &&
+ (session->consumer->relay_major_version == 2 &&
+ session->consumer->relay_minor_version < 11)) {
+ ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE_RELAY;
+ goto end;
+ }
+
+ if (session->rotate_pending || session->rotate_pending_relay) {
+ ret = -LTTNG_ERR_ROTATION_PENDING;
+ DBG("Rotate already in progress");
+ goto end;
+ }
+
+ /*
+ * After a stop, we only allow one rotation to occur, the other ones are
+ * useless until a new start.
+ */
+ if (session->rotated_after_last_stop) {
+ DBG("Session \"%s\" was already rotated after stop, refusing rotation",
+ session->name);
+ ret = -LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP;
+ goto end;
+ }
+
+ /* Special case for the first rotation. */
+ if (session->current_archive_id == 0) {
+ const char *base_path = NULL;
+
+ /* Either one of the two sessions is enough to get the root path. */
+ if (session->kernel_session) {
+ base_path = session_get_base_path(session);
+ } else if (session->ust_session) {
+ base_path = session_get_base_path(session);
+ } else {
+ assert(0);
+ }
+ assert(base_path);
+ ret = lttng_strncpy(session->rotation_chunk.current_rotate_path,
+ base_path,
+ sizeof(session->rotation_chunk.current_rotate_path));
+ if (ret) {
+ ERR("Failed to copy session base path to current rotation chunk path");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ } else {
+ /*
+ * The currently active tracing path is now the folder we
+ * want to rotate.
+ */
+ ret = lttng_strncpy(session->rotation_chunk.current_rotate_path,
+ session->rotation_chunk.active_tracing_path,
+ sizeof(session->rotation_chunk.current_rotate_path));
+ if (ret) {
+ ERR("Failed to copy the active tracing path to the current rotate path");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ }
+ DBG("Current rotate path %s", session->rotation_chunk.current_rotate_path);
+
+ session->current_archive_id++;
+ session->rotate_pending = true;
+ session->rotation_state = LTTNG_ROTATION_STATE_ONGOING;
+ ret = notification_thread_command_session_rotation_ongoing(
+ notification_thread_handle,
+ session->name, session->uid, session->gid,
+ session->current_archive_id);
+ if (ret != LTTNG_OK) {
+ ERR("Failed to notify notification thread that a session rotation is ongoing for session %s",
+ session->name);
+ }
+
+ /*
+ * Create the path name for the next chunk.
+ */
+ now = time(NULL);
+ if (now == (time_t) -1) {
+ ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE;
+ goto end;
+ }
+ session->last_chunk_start_ts = session->current_chunk_start_ts;
+ session->current_chunk_start_ts = now;
+
+ timeinfo = localtime(&now);
+ if (!timeinfo) {
+ PERROR("Failed to sample local time in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ strf_ret = strftime(datetime, sizeof(datetime), "%Y%m%dT%H%M%S%z",
+ timeinfo);
+ if (!strf_ret) {
+ ERR("Failed to format local time timestamp in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ if (session->kernel_session) {
+ /*
+ * The active path for the next rotation/destroy.
+ * Ex: ~/lttng-traces/auto-20170922-111748/20170922-111754-42
+ */
+ ret = snprintf(session->rotation_chunk.active_tracing_path,
+ sizeof(session->rotation_chunk.active_tracing_path),
+ "%s/%s-%" PRIu64,
+ session_get_base_path(session),
+ datetime, session->current_archive_id + 1);
+ if (ret < 0 || ret == sizeof(session->rotation_chunk.active_tracing_path)) {
+ ERR("Failed to format active kernel tracing path in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ /*
+ * The sub-directory for the consumer
+ * Ex: /20170922-111754-42/kernel
+ */
+ ret = snprintf(session->kernel_session->consumer->chunk_path,
+ sizeof(session->kernel_session->consumer->chunk_path),
+ "/%s-%" PRIu64, datetime,
+ session->current_archive_id + 1);
+ if (ret < 0 || ret == sizeof(session->kernel_session->consumer->chunk_path)) {
+ ERR("Failed to format the kernel consumer's sub-directory in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ /*
+ * Create the new chunk folder, before the rotation begins so we don't
+ * race with the consumer/tracer activity.
+ */
+ ret = domain_mkdir(session->kernel_session->consumer, session,
+ session->kernel_session->uid,
+ session->kernel_session->gid);
+ if (ret) {
+ ERR("Failed to create kernel session tracing path at %s",
+ session->kernel_session->consumer->chunk_path);
+ ret = -LTTNG_ERR_CREATE_DIR_FAIL;
+ goto end;
+ }
+ ret = kernel_rotate_session(session);
+ if (ret != LTTNG_OK) {
+ ret = -ret;
+ goto end;
+ }
+ }
+ if (session->ust_session) {
+ ret = snprintf(session->rotation_chunk.active_tracing_path,
+ PATH_MAX, "%s/%s-%" PRIu64,
+ session_get_base_path(session),
+ datetime, session->current_archive_id + 1);
+ if (ret < 0) {
+ ERR("Failed to format active UST tracing path in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ ret = snprintf(session->ust_session->consumer->chunk_path,
+ PATH_MAX, "/%s-%" PRIu64, datetime,
+ session->current_archive_id + 1);
+ if (ret < 0) {
+ ERR("Failed to format the UST consumer's sub-directory in rotate session command");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+ /*
+ * Create the new chunk folder, before the rotation begins so we don't
+ * race with the consumer/tracer activity.
+ */
+ ret = domain_mkdir(session->ust_session->consumer, session,
+ session->ust_session->uid,
+ session->ust_session->gid);
+ if (ret) {
+ ret = -LTTNG_ERR_CREATE_DIR_FAIL;
+ goto end;
+ }
+ ret = ust_app_rotate_session(session, &ust_active);
+ if (ret != LTTNG_OK) {
+ goto end;
+ }
+ /*
+ * Handle the case where we did not start a rotation on any channel.
+ * The consumer will never wake up the rotation thread to perform the
+ * rename, so we have to do it here while we hold the session and
+ * session_list locks.
+ */
+ if (!session->kernel_session && !ust_active) {
+ struct lttng_trace_archive_location *location;
+
+ session->rotate_pending = false;
+ session->rotation_state = LTTNG_ROTATION_STATE_COMPLETED;
+ ret = rename_complete_chunk(session, now);
+ if (ret < 0) {
+ ERR("Failed to rename completed rotation chunk");
+ goto end;
+ }
+
+ /* Ownership of location is transferred. */
+ location = session_get_trace_archive_location(session);
+ ret = notification_thread_command_session_rotation_completed(
+ notification_thread_handle,
+ session->name,
+ session->uid,
+ session->gid,
+ session->current_archive_id,
+ location);
+ if (ret != LTTNG_OK) {
+ ERR("Failed to notify notification thread that rotation is complete for session %s",
+ session->name);
+ }
+ }
+ }
+
+ if (!session->active) {
+ session->rotated_after_last_stop = true;
+ }
+
+ if (rotate_return) {
+ rotate_return->rotation_id = session->current_archive_id;
+ }
+
+ DBG("Cmd rotate session %s, current_archive_id %" PRIu64 " sent",
+ session->name, session->current_archive_id);
+ ret = LTTNG_OK;
+
+end:
+ return ret;
+}
+
+/*
+ * Command LTTNG_ROTATION_GET_INFO from the lttng-ctl library.
+ *
+ * Check if the session has finished its rotation.
+ *
+ * Return 0 on success or else a LTTNG_ERR code.
+ */
+int cmd_rotate_get_info(struct ltt_session *session,
+ struct lttng_rotation_get_info_return *info_return,
+ uint64_t rotation_id)
+{
+ int ret;
+
+ assert(session);
+
+ DBG("Cmd rotate_get_info session %s, rotation id %" PRIu64, session->name,
+ session->current_archive_id);
+
+ if (session->current_archive_id != rotation_id) {
+ info_return->status = (int32_t) LTTNG_ROTATION_STATE_EXPIRED;
+ ret = LTTNG_OK;
+ goto end;
+ }
+
+ switch (session->rotation_state) {
+ case LTTNG_ROTATION_STATE_ONGOING:
+ DBG("Reporting that rotation id %" PRIu64 " of session %s is still pending",
+ rotation_id, session->name);
+ break;
+ case LTTNG_ROTATION_STATE_COMPLETED:
+ {
+ char *current_tracing_path_reply;
+ size_t current_tracing_path_reply_len;
+
+ switch (session_get_consumer_destination_type(session)) {
+ case CONSUMER_DST_LOCAL:
+ current_tracing_path_reply =
+ info_return->location.local.absolute_path;
+ current_tracing_path_reply_len =
+ sizeof(info_return->location.local.absolute_path);
+ info_return->location_type =
+ (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_LOCAL;
+ break;
+ case CONSUMER_DST_NET:
+ current_tracing_path_reply =
+ info_return->location.relay.relative_path;
+ current_tracing_path_reply_len =
+ sizeof(info_return->location.relay.relative_path);
+ /* Currently the only supported relay protocol. */
+ info_return->location.relay.protocol =
+ (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_RELAY_PROTOCOL_TYPE_TCP;
+
+ ret = lttng_strncpy(info_return->location.relay.host,
+ session_get_net_consumer_hostname(session),
+ sizeof(info_return->location.relay.host));
+ if (ret) {
+ ERR("Failed to host name to rotate_get_info reply");
+ info_return->status = LTTNG_ROTATION_STATUS_ERROR;
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+
+ session_get_net_consumer_ports(session,
+ &info_return->location.relay.ports.control,
+ &info_return->location.relay.ports.data);
+ info_return->location_type =
+ (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_RELAY;
+ break;
+ default:
+ abort();
+ }
+ ret = lttng_strncpy(current_tracing_path_reply,
+ session->rotation_chunk.current_rotate_path,
+ current_tracing_path_reply_len);
+ if (ret) {
+ ERR("Failed to copy current tracing path to rotate_get_info reply");
+ info_return->status = LTTNG_ROTATION_STATUS_ERROR;
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+
+ break;
+ }
+ case LTTNG_ROTATION_STATE_ERROR:
+ DBG("Reporting that an error occurred during rotation %" PRIu64 " of session %s",
+ rotation_id, session->name);
+ break;
+ default:
+ abort();
+ }
+
+ info_return->status = (int32_t) session->rotation_state;
+ ret = LTTNG_OK;
+end:
+ return ret;
+}
+
+/*
+ * Command LTTNG_ROTATION_SET_SCHEDULE from the lttng-ctl library.
+ *
+ * Configure the automatic rotation parameters.
+ * 'activate' to true means activate the rotation schedule type with 'new_value'.
+ * 'activate' to false means deactivate the rotation schedule and validate that
+ * 'new_value' has the same value as the currently active value.
+ *
+ * Return 0 on success or else a positive LTTNG_ERR code.
+ */
+int cmd_rotation_set_schedule(struct ltt_session *session,
+ bool activate, enum lttng_rotation_schedule_type schedule_type,
+ uint64_t new_value,
+ struct notification_thread_handle *notification_thread_handle)
+{
+ int ret;
+ uint64_t *parameter_value;
+
+ assert(session);
+
+ DBG("Cmd rotate set schedule session %s", session->name);
+
+ if (session->live_timer || session->snapshot_mode ||
+ !session->output_traces) {
+ DBG("Failing ROTATION_SET_SCHEDULE command as the rotation feature is not available for this session");
+ ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE;
+ goto end;
+ }
+
+ switch (schedule_type) {
+ case LTTNG_ROTATION_SCHEDULE_TYPE_SIZE_THRESHOLD:
+ parameter_value = &session->rotate_size;
+ break;
+ case LTTNG_ROTATION_SCHEDULE_TYPE_PERIODIC:
+ parameter_value = &session->rotate_timer_period;
+ if (new_value >= UINT_MAX) {
+ DBG("Failing ROTATION_SET_SCHEDULE command as the value requested for a periodic rotation schedule is invalid: %" PRIu64 " > %u (UINT_MAX)",
+ new_value, UINT_MAX);
+ ret = LTTNG_ERR_INVALID;
+ goto end;
+ }
+ break;
+ default:
+ WARN("Failing ROTATION_SET_SCHEDULE command on unknown schedule type");
+ ret = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ /* Improper use of the API. */
+ if (new_value == -1ULL) {
+ WARN("Failing ROTATION_SET_SCHEDULE command as the value requested is -1");
+ ret = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ /*
+ * As indicated in struct ltt_session's comments, a value of == 0 means
+ * this schedule rotation type is not in use.
+ *
+ * Reject the command if we were asked to activate a schedule that was
+ * already active.
+ */
+ if (activate && *parameter_value != 0) {
+ DBG("Failing ROTATION_SET_SCHEDULE (activate) command as the schedule is already active");
+ ret = LTTNG_ERR_ROTATION_SCHEDULE_SET;
+ goto end;
+ }
+
+ /*
+ * Reject the command if we were asked to deactivate a schedule that was
+ * not active.
+ */
+ if (!activate && *parameter_value == 0) {
+ DBG("Failing ROTATION_SET_SCHEDULE (deactivate) command as the schedule is already inactive");
+ ret = LTTNG_ERR_ROTATION_SCHEDULE_NOT_SET;
+ goto end;
+ }
+
+ /*
+ * Reject the command if we were asked to deactivate a schedule that
+ * doesn't exist.
+ */
+ if (!activate && *parameter_value != new_value) {
+ DBG("Failing ROTATION_SET_SCHEDULE (deactivate) command as an inexistant schedule was provided");
+ ret = LTTNG_ERR_ROTATION_SCHEDULE_NOT_SET;
+ goto end;
+ }
+
+ *parameter_value = activate ? new_value : 0;
+
+ switch (schedule_type) {
+ case LTTNG_ROTATION_SCHEDULE_TYPE_PERIODIC:
+ if (activate && session->active) {
+ /*
+ * Only start the timer if the session is active,
+ * otherwise it will be started when the session starts.
+ */
+ ret = sessiond_rotate_timer_start(session, new_value);
+ if (ret) {
+ ERR("Failed to enable session rotation timer in ROTATION_SET_SCHEDULE command");
+ ret = LTTNG_ERR_UNK;
+ goto end;
+ }
+ } else {
+ ret = sessiond_rotate_timer_stop(session);
+ if (ret) {
+ ERR("Failed to disable session rotation timer in ROTATION_SET_SCHEDULE command");
+ ret = LTTNG_ERR_UNK;
+ goto end;
+ }
+ }
+ break;
+ case LTTNG_ROTATION_SCHEDULE_TYPE_SIZE_THRESHOLD:
+ if (activate) {
+ ret = subscribe_session_consumed_size_rotation(session,
+ new_value, notification_thread_handle);
+ if (ret) {
+ ERR("Failed to enable consumed-size notification in ROTATION_SET_SCHEDULE command");
+ ret = LTTNG_ERR_UNK;
+ goto end;
+ }
+ } else {
+ ret = unsubscribe_session_consumed_size_rotation(session,
+ notification_thread_handle);
+ if (ret) {
+ ERR("Failed to disable consumed-size notification in ROTATION_SET_SCHEDULE command");
+ ret = LTTNG_ERR_UNK;
+ goto end;
+ }
+
+ }
+ break;
+ default:
+ /* Would have been caught before. */
+ abort();
+ }
+
+ ret = LTTNG_OK;
+
+ goto end;
+
+end:
+ return ret;
+}
+
+/*
+ * Command ROTATE_GET_CURRENT_PATH from the lttng-ctl library.
+ *
+ * Configure the automatic rotation parameters.
+ * Set to -1ULL to disable them.
+ *
+ * Return LTTNG_OK on success or else a LTTNG_ERR code.
+ */
+int cmd_session_get_current_output(struct ltt_session *session,
+ struct lttng_session_get_current_output_return *output_return)
+{
+ int ret;
+ const char *path;
+
+ if (!session->snapshot_mode) {
+ if (session->current_archive_id == 0) {
+ if (session->kernel_session) {
+ path = session_get_base_path(session);
+ } else if (session->ust_session) {
+ path = session_get_base_path(session);
+ } else {
+ abort();
+ }
+ assert(path);
+ } else {
+ path = session->rotation_chunk.active_tracing_path;
+ }
+ } else {
+ /*
+ * A snapshot session does not have a "current" trace archive
+ * location.
+ */
+ path = "";
+ }
+
+ DBG("Cmd get current output for session %s, returning %s",
+ session->name, path);
+
+ ret = lttng_strncpy(output_return->path,
+ path,
+ sizeof(output_return->path));
+ if (ret) {
+ ERR("Failed to copy trace output path to session get current output command reply");
+ ret = -LTTNG_ERR_UNK;
+ goto end;
+ }
+
+ ret = LTTNG_OK;
+end:
+ return ret;
+}
+
+/* Wait for a given path to be removed before continuing. */
+static enum lttng_error_code wait_on_path(void *path_data)
+{
+ const char *shm_path = path_data;
+
+ DBG("Waiting for the shm path at %s to be removed before completing session destruction",
+ shm_path);
+ while (true) {
+ int ret;
+ struct stat st;
+
+ ret = stat(shm_path, &st);
+ if (ret) {
+ if (errno != ENOENT) {
+ PERROR("stat() returned an error while checking for the existence of the shm path");
+ } else {
+ DBG("shm path no longer exists, completing the destruction of session");
+ }
+ break;
+ } else {
+ if (!S_ISDIR(st.st_mode)) {
+ ERR("The type of shm path %s returned by stat() is not a directory; aborting the wait for shm path removal",
+ shm_path);
+ break;
+ }
+ }
+ usleep(SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US);
+ }
+ return LTTNG_OK;
+}
+
+/*
+ * Returns a pointer to a handler to run on completion of a command.
+ * Returns NULL if no handler has to be run for the last command executed.
+ */
+const struct cmd_completion_handler *cmd_pop_completion_handler(void)
+{
+ struct cmd_completion_handler *handler = current_completion_handler;
+
+ current_completion_handler = NULL;
+ return handler;
+}
+
/*
* Init command subsystem.
*/