#include <urcu.h>
#include <dirent.h>
#include <sys/types.h>
+#include <pthread.h>
#include <common/common.h>
+#include <common/utils.h>
+#include <common/trace-chunk.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <lttng/location-internal.h>
+#include "lttng-sessiond.h"
+#include "kernel.h"
#include "session.h"
#include "utils.h"
static struct ltt_session_list ltt_session_list = {
.head = CDS_LIST_HEAD_INIT(ltt_session_list.head),
.lock = PTHREAD_MUTEX_INITIALIZER,
+ .removal_cond = PTHREAD_COND_INITIALIZER,
.next_uuid = 0,
};
/* Global hash table to keep the sessions, indexed by id. */
static struct lttng_ht *ltt_sessions_ht_by_id = NULL;
+struct consumer_create_chunk_transaction {
+ struct consumer_socket *socket;
+ struct lttng_trace_chunk *new_chunk;
+ struct lttng_trace_chunk *previous_chunk;
+ bool new_chunk_created;
+};
+
/*
* Validate the session name for forbidden characters.
*
return <t_session_list;
}
+/*
+ * Returns once the session list is empty.
+ */
+void session_list_wait_empty(void)
+{
+ pthread_mutex_lock(<t_session_list.lock);
+ while (!cds_list_empty(<t_session_list.head)) {
+ pthread_cond_wait(<t_session_list.removal_cond,
+ <t_session_list.lock);
+ }
+ pthread_mutex_unlock(<t_session_list.lock);
+}
+
/*
* Acquire session list lock
*/
pthread_mutex_unlock(&session->lock);
}
+static
+int _session_set_trace_chunk_no_lock_check(struct ltt_session *session,
+ struct lttng_trace_chunk *new_trace_chunk)
+{
+ int ret;
+ unsigned int i, refs_to_acquire = 0, refs_acquired = 0, refs_to_release = 0;
+ unsigned int consumer_count = 0;
+ /*
+ * The maximum amount of consumers to reach is 3
+ * (32/64 userspace + kernel).
+ */
+ struct consumer_create_chunk_transaction transactions[3] = {};
+ struct cds_lfht_iter iter;
+ struct consumer_socket *socket;
+ bool close_error_occured = false;
+
+ if (new_trace_chunk) {
+ uint64_t chunk_id;
+ enum lttng_trace_chunk_status chunk_status =
+ lttng_trace_chunk_get_id(new_trace_chunk,
+ &chunk_id);
+
+ assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
+ LTTNG_OPTIONAL_SET(&session->last_trace_chunk_id, chunk_id)
+ }
+
+ if (new_trace_chunk) {
+ refs_to_acquire = 1;
+ refs_to_acquire += !!session->ust_session;
+ refs_to_acquire += !!session->kernel_session;
+ }
+
+ /*
+ * Build a list of consumers to reach to announce the new trace chunk.
+ *
+ * Rolling back the annoucement in case of an error is important since
+ * not doing so would result in a leak; the chunk will not be
+ * "reclaimed" by the consumer(s) since they have no concept of the
+ * lifetime of a session.
+ */
+ if (session->ust_session) {
+ cds_lfht_for_each_entry(
+ session->ust_session->consumer->socks->ht,
+ &iter, socket, node.node) {
+ transactions[consumer_count].socket = socket;
+ transactions[consumer_count].new_chunk = new_trace_chunk;
+ transactions[consumer_count].previous_chunk =
+ session->current_trace_chunk;
+ consumer_count++;
+ assert(consumer_count <= 3);
+ }
+ }
+ if (session->kernel_session) {
+ cds_lfht_for_each_entry(
+ session->kernel_session->consumer->socks->ht,
+ &iter, socket, node.node) {
+ transactions[consumer_count].socket = socket;
+ transactions[consumer_count].new_chunk = new_trace_chunk;
+ transactions[consumer_count].previous_chunk =
+ session->current_trace_chunk;
+ consumer_count++;
+ assert(consumer_count <= 3);
+ }
+ }
+ for (refs_acquired = 0; refs_acquired < refs_to_acquire; refs_acquired++) {
+ if (new_trace_chunk && !lttng_trace_chunk_get(new_trace_chunk)) {
+ ERR("Failed to acquire reference to new current trace chunk of session \"%s\"",
+ session->name);
+ goto error;
+ }
+ }
+
+ /*
+ * Close the previous chunk on remote peers (consumers and relayd).
+ */
+ for (i = 0; i < consumer_count; i++) {
+ if (!transactions[i].previous_chunk) {
+ continue;
+ }
+ pthread_mutex_lock(transactions[i].socket->lock);
+ ret = consumer_close_trace_chunk(transactions[i].socket,
+ session->consumer->net_seq_index,
+ session->id,
+ transactions[i].previous_chunk);
+ pthread_mutex_unlock(transactions[i].socket->lock);
+ if (ret) {
+ ERR("Failed to close trace chunk on consumer");
+ close_error_occured = true;
+ }
+ }
+
+ if (close_error_occured) {
+ /*
+ * Skip the creation of the new trace chunk and report the
+ * error.
+ */
+ goto error;
+ }
+
+ /* Create the new chunk on remote peers (consumers and relayd) */
+ if (new_trace_chunk) {
+ for (i = 0; i < consumer_count; i++) {
+ pthread_mutex_lock(transactions[i].socket->lock);
+ ret = consumer_create_trace_chunk(transactions[i].socket,
+ session->consumer->net_seq_index,
+ session->id,
+ transactions[i].new_chunk);
+ pthread_mutex_unlock(transactions[i].socket->lock);
+ if (ret) {
+ ERR("Failed to create trace chunk on consumer");
+ goto error;
+ }
+ /* This will have to be rolled-back on error. */
+ transactions[i].new_chunk_created = true;
+ }
+ }
+
+ lttng_trace_chunk_put(session->current_trace_chunk);
+ session->current_trace_chunk = NULL;
+ if (session->ust_session) {
+ lttng_trace_chunk_put(
+ session->ust_session->current_trace_chunk);
+ session->ust_session->current_trace_chunk = NULL;
+ }
+ if (session->kernel_session) {
+ lttng_trace_chunk_put(
+ session->kernel_session->current_trace_chunk);
+ session->kernel_session->current_trace_chunk = NULL;
+ }
+
+ /*
+ * Update local current trace chunk state last, only if all remote
+ * annoucements succeeded.
+ */
+ session->current_trace_chunk = new_trace_chunk;
+ if (session->ust_session) {
+ session->ust_session->current_trace_chunk = new_trace_chunk;
+ }
+ if (session->kernel_session) {
+ session->kernel_session->current_trace_chunk =
+ new_trace_chunk;
+ }
+
+ return 0;
+error:
+ /*
+ * Release references taken in the case where all references could not
+ * be acquired.
+ */
+ refs_to_release = refs_to_acquire - refs_acquired;
+ for (i = 0; i < refs_to_release; i++) {
+ lttng_trace_chunk_put(new_trace_chunk);
+ }
+
+ /*
+ * Close the newly-created chunk from remote peers (consumers and
+ * relayd).
+ */
+ DBG("Rolling back the creation of the new trace chunk on consumers");
+ for (i = 0; i < consumer_count; i++) {
+ if (!transactions[i].new_chunk_created) {
+ continue;
+ }
+
+ pthread_mutex_lock(transactions[i].socket->lock);
+ ret = consumer_close_trace_chunk(transactions[i].socket,
+ session->consumer->net_seq_index,
+ session->id,
+ transactions[i].new_chunk);
+ pthread_mutex_unlock(transactions[i].socket->lock);
+ if (ret) {
+ ERR("Failed to close trace chunk on consumer");
+ close_error_occured = true;
+ }
+ }
+
+ return -1;
+}
+
+static
+bool output_supports_chunks(const struct ltt_session *session)
+{
+ if (session->consumer->type == CONSUMER_DST_LOCAL) {
+ return true;
+ } else {
+ struct consumer_output *output;
+
+ if (session->ust_session) {
+ output = session->ust_session->consumer;
+ } else if (session->kernel_session) {
+ output = session->kernel_session->consumer;
+ } else {
+ abort();
+ }
+
+ if (output->relay_major_version > 2) {
+ return true;
+ } else if (output->relay_major_version == 2 &&
+ output->relay_minor_version >= 11) {
+ return true;
+ }
+ }
+ return false;
+}
+
+enum lttng_error_code session_switch_trace_chunk(struct ltt_session *session,
+ const char *session_base_path_override,
+ const char *chunk_name_override)
+{
+ int ret;
+ enum lttng_error_code ret_code = LTTNG_OK;
+ struct lttng_trace_chunk *trace_chunk = NULL;
+ enum lttng_trace_chunk_status chunk_status;
+ const time_t timestamp_begin = time(NULL);
+ const bool is_local_trace =
+ session->consumer->type == CONSUMER_DST_LOCAL;
+ const char *base_path = session_base_path_override ? :
+ session_get_base_path(session);
+ struct lttng_directory_handle session_output_directory;
+ const struct lttng_credentials session_credentials = {
+ .uid = session->uid,
+ .gid = session->gid,
+ };
+ uint64_t next_chunk_id;
+
+ if (timestamp_begin == (time_t) -1) {
+ PERROR("Failed to sample time while changing session \"%s\" trace chunk",
+ session->name);
+ ret_code = LTTNG_ERR_FATAL;
+ goto error;
+ }
+ session->current_chunk_start_ts = timestamp_begin;
+
+ if (!output_supports_chunks(session)) {
+ goto end;
+ }
+ next_chunk_id = session->last_trace_chunk_id.is_set ?
+ session->last_trace_chunk_id.value + 1 : 0;
+
+ trace_chunk = lttng_trace_chunk_create(next_chunk_id, timestamp_begin);
+ if (!trace_chunk) {
+ ret_code = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ if (chunk_name_override) {
+ chunk_status = lttng_trace_chunk_override_name(trace_chunk,
+ chunk_name_override);
+ switch (chunk_status) {
+ case LTTNG_TRACE_CHUNK_STATUS_OK:
+ break;
+ case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
+ ret_code = LTTNG_ERR_INVALID;
+ goto error;
+ default:
+ ret_code = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ }
+
+ if (!is_local_trace) {
+ /*
+ * No need to set crendentials and output directory
+ * for remote trace chunks.
+ */
+ goto publish;
+ }
+
+ chunk_status = lttng_trace_chunk_set_credentials(trace_chunk,
+ &session_credentials);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret_code = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ if (!session->current_trace_chunk) {
+ DBG("Creating base output directory of session \"%s\" at %s",
+ session->name, base_path);
+ }
+ ret = utils_mkdir_recursive(base_path, S_IRWXU | S_IRWXG,
+ session->uid, session->gid);
+ if (ret) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+ ret = lttng_directory_handle_init(&session_output_directory,
+ base_path);
+ if (ret) {
+ ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+ chunk_status = lttng_trace_chunk_set_as_owner(trace_chunk,
+ &session_output_directory);
+ lttng_directory_handle_fini(&session_output_directory);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = LTTNG_ERR_CREATE_DIR_FAIL;
+ goto error;
+ }
+publish:
+ ret = session_set_trace_chunk(session, trace_chunk);
+ if (ret) {
+ ret_code = LTTNG_ERR_FATAL;
+ goto error;
+ }
+error:
+ lttng_trace_chunk_put(trace_chunk);
+end:
+ return ret_code;
+}
+
+/*
+ * Set a session's current trace chunk.
+ *
+ * Must be called with the session lock held.
+ */
+int session_set_trace_chunk(struct ltt_session *session,
+ struct lttng_trace_chunk *new_trace_chunk)
+{
+ ASSERT_LOCKED(session->lock);
+ return _session_set_trace_chunk_no_lock_check(session, new_trace_chunk);
+}
+
+static
+void session_release(struct urcu_ref *ref)
+{
+ int ret;
+ struct ltt_ust_session *usess;
+ struct ltt_kernel_session *ksess;
+ struct ltt_session *session = container_of(ref, typeof(*session), ref);
+
+ usess = session->ust_session;
+ ksess = session->kernel_session;
+ (void) _session_set_trace_chunk_no_lock_check(session, NULL);
+
+ /* Clean kernel session teardown */
+ kernel_destroy_session(ksess);
+ session->kernel_session = NULL;
+
+ /* UST session teardown */
+ if (usess) {
+ /* Close any relayd session */
+ consumer_output_send_destroy_relayd(usess->consumer);
+
+ /* Destroy every UST application related to this session. */
+ ret = ust_app_destroy_trace_all(usess);
+ if (ret) {
+ ERR("Error in ust_app_destroy_trace_all");
+ }
+
+ /* Clean up the rest. */
+ trace_ust_destroy_session(usess);
+ session->ust_session = NULL;
+ }
+
+ /*
+ * Must notify the kernel thread here to update it's poll set in order to
+ * remove the channel(s)' fd just destroyed.
+ */
+ ret = notify_thread_pipe(kernel_poll_pipe[1]);
+ if (ret < 0) {
+ PERROR("write kernel poll pipe");
+ }
+
+ DBG("Destroying session %s (id %" PRIu64 ")", session->name, session->id);
+
+ consumer_output_put(session->consumer);
+ snapshot_destroy(&session->snapshot);
+
+ pthread_mutex_destroy(&session->lock);
+
+ if (session->published) {
+ ASSERT_LOCKED(ltt_session_list.lock);
+ del_session_list(session);
+ del_session_ht(session);
+ pthread_cond_broadcast(<t_session_list.removal_cond);
+ }
+ free(session);
+}
+
+/*
+ * Acquire a reference to a session.
+ * This function may fail (return false); its return value must be checked.
+ */
+bool session_get(struct ltt_session *session)
+{
+ return urcu_ref_get_unless_zero(&session->ref);
+}
+
+/*
+ * Release a reference to a session.
+ */
+void session_put(struct ltt_session *session)
+{
+ if (!session) {
+ return;
+ }
+ /*
+ * The session list lock must be held as any session_put()
+ * may cause the removal of the session from the session_list.
+ */
+ ASSERT_LOCKED(ltt_session_list.lock);
+ assert(session->ref.refcount);
+ urcu_ref_put(&session->ref, session_release);
+}
+
+/*
+ * Destroy a session.
+ *
+ * This method does not immediately release/free the session as other
+ * components may still hold a reference to the session. However,
+ * the session should no longer be presented to the user.
+ *
+ * Releases the session list's reference to the session
+ * and marks it as destroyed. Iterations on the session list should be
+ * mindful of the "destroyed" flag.
+ */
+void session_destroy(struct ltt_session *session)
+{
+ assert(!session->destroyed);
+ session->destroyed = true;
+ session_put(session);
+}
+
/*
* Return a ltt_session structure ptr that matches name. If no session found,
* NULL is returned. This must be called with the session list lock held using
* session_lock_list and session_unlock_list.
+ * A reference to the session is implicitly acquired by this function.
*/
struct ltt_session *session_find_by_name(const char *name)
{
struct ltt_session *iter;
assert(name);
+ ASSERT_LOCKED(ltt_session_list.lock);
DBG2("Trying to find session by name %s", name);
cds_list_for_each_entry(iter, <t_session_list.head, list) {
- if (strncmp(iter->name, name, NAME_MAX) == 0) {
+ if (!strncmp(iter->name, name, NAME_MAX) &&
+ !iter->destroyed) {
goto found;
}
}
- iter = NULL;
-
+ return NULL;
found:
- return iter;
+ return session_get(iter) ? iter : NULL;
}
/*
struct lttng_ht_iter iter;
struct ltt_session *ls;
+ ASSERT_LOCKED(ltt_session_list.lock);
+
if (!ltt_sessions_ht_by_id) {
goto end;
}
ls = caa_container_of(node, struct ltt_session, node);
DBG3("Session %" PRIu64 " found by id.", id);
- return ls;
+ return session_get(ls) ? ls : NULL;
end:
DBG3("Session %" PRIu64 " NOT found by id", id);
}
/*
- * Delete session from the session list and free the memory.
- *
- * Return -1 if no session is found. On success, return 1;
- * Should *NOT* be called with RCU read-side lock held.
- */
-int session_destroy(struct ltt_session *session)
-{
- /* Safety check */
- assert(session);
-
- DBG("Destroying session %s (id %" PRIu64 ")", session->name, session->id);
- del_session_list(session);
- pthread_mutex_destroy(&session->lock);
- del_session_ht(session);
-
- consumer_output_put(session->consumer);
- snapshot_destroy(&session->snapshot);
- free(session);
-
- return LTTNG_OK;
-}
-
-/*
- * Create a brand new session and add it to the session list.
+ * Create a new session and add it to the session list.
+ * Session list lock must be held by the caller.
*/
-int session_create(char *name, uid_t uid, gid_t gid)
+enum lttng_error_code session_create(const char *name, uid_t uid, gid_t gid,
+ struct ltt_session **out_session)
{
int ret;
- struct ltt_session *new_session;
+ enum lttng_error_code ret_code;
+ struct ltt_session *new_session = NULL;
- /* Allocate session data structure */
+ ASSERT_LOCKED(ltt_session_list.lock);
+ if (name) {
+ struct ltt_session *clashing_session;
+
+ clashing_session = session_find_by_name(name);
+ if (clashing_session) {
+ session_put(clashing_session);
+ ret_code = LTTNG_ERR_EXIST_SESS;
+ goto error;
+ }
+ }
new_session = zmalloc(sizeof(struct ltt_session));
- if (new_session == NULL) {
- PERROR("zmalloc");
- ret = LTTNG_ERR_FATAL;
- goto error_malloc;
+ if (!new_session) {
+ PERROR("Failed to allocate an ltt_session structure");
+ ret_code = LTTNG_ERR_NOMEM;
+ goto error;
}
- /* Define session name */
- if (name != NULL) {
- if (snprintf(new_session->name, NAME_MAX, "%s", name) < 0) {
- ret = LTTNG_ERR_FATAL;
- goto error_asprintf;
- }
- } else {
- ERR("No session name given");
- ret = LTTNG_ERR_FATAL;
+ urcu_ref_init(&new_session->ref);
+ pthread_mutex_init(&new_session->lock, NULL);
+
+ new_session->creation_time = time(NULL);
+ if (new_session->creation_time == (time_t) -1) {
+ PERROR("Failed to sample session creation time");
+ ret_code = LTTNG_ERR_SESSION_FAIL;
goto error;
}
- ret = validate_name(name);
- if (ret < 0) {
- ret = LTTNG_ERR_SESSION_INVALID_CHAR;
+ /* Create default consumer output. */
+ new_session->consumer = consumer_create_output(CONSUMER_DST_LOCAL);
+ if (new_session->consumer == NULL) {
+ ret_code = LTTNG_ERR_NOMEM;
goto error;
}
+ if (name) {
+ ret = lttng_strncpy(new_session->name, name, sizeof(new_session->name));
+ if (ret) {
+ ret_code = LTTNG_ERR_SESSION_INVALID_CHAR;
+ goto error;
+ }
+ ret = validate_name(name);
+ if (ret < 0) {
+ ret_code = LTTNG_ERR_SESSION_INVALID_CHAR;
+ goto error;
+ }
+ } else {
+ int i = 0;
+ bool found_name = false;
+ char datetime[16];
+ struct tm *timeinfo;
+
+ timeinfo = localtime(&new_session->creation_time);
+ if (!timeinfo) {
+ ret_code = LTTNG_ERR_SESSION_FAIL;
+ goto error;
+ }
+ strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", timeinfo);
+ for (i = 0; i < INT_MAX; i++) {
+ struct ltt_session *clashing_session;
+
+ if (i == 0) {
+ ret = snprintf(new_session->name,
+ sizeof(new_session->name),
+ "%s-%s",
+ DEFAULT_SESSION_NAME,
+ datetime);
+ } else {
+ ret = snprintf(new_session->name,
+ sizeof(new_session->name),
+ "%s%d-%s",
+ DEFAULT_SESSION_NAME, i,
+ datetime);
+ }
+ if (ret == -1 || ret >= sizeof(new_session->name)) {
+ /*
+ * Null-terminate in case the name is used
+ * in logging statements.
+ */
+ new_session->name[sizeof(new_session->name) - 1] = '\0';
+ ret_code = LTTNG_ERR_SESSION_FAIL;
+ goto error;
+ }
+
+ clashing_session =
+ session_find_by_name(new_session->name);
+ session_put(clashing_session);
+ if (!clashing_session) {
+ found_name = true;
+ break;
+ }
+ }
+ if (found_name) {
+ DBG("Generated session name \"%s\"", new_session->name);
+ new_session->has_auto_generated_name = true;
+ } else {
+ ERR("Failed to auto-generate a session name");
+ ret_code = LTTNG_ERR_SESSION_FAIL;
+ goto error;
+ }
+ }
+
ret = gethostname(new_session->hostname, sizeof(new_session->hostname));
if (ret < 0) {
if (errno == ENAMETOOLONG) {
new_session->hostname[sizeof(new_session->hostname) - 1] = '\0';
+ ERR("Hostname exceeds the maximal permitted length and has been truncated to %s",
+ new_session->hostname);
} else {
- ret = LTTNG_ERR_FATAL;
+ ret_code = LTTNG_ERR_SESSION_FAIL;
goto error;
}
}
- /* Init kernel session */
- new_session->kernel_session = NULL;
- new_session->ust_session = NULL;
-
- /* Init lock */
- pthread_mutex_init(&new_session->lock, NULL);
-
new_session->uid = uid;
new_session->gid = gid;
ret = snapshot_init(&new_session->snapshot);
if (ret < 0) {
- ret = LTTNG_ERR_NOMEM;
+ ret_code = LTTNG_ERR_NOMEM;
goto error;
}
- new_session->rotation_pending_local = false;
- new_session->rotation_pending_relay = false;
new_session->rotation_state = LTTNG_ROTATION_STATE_NO_ROTATION;
- new_session->rotation_pending_check_timer_enabled = false;
- new_session->rotation_schedule_timer_enabled = false;
-
- /* Add new session to the session list */
- session_lock_list();
+ /* Add new session to the session list. */
new_session->id = add_session_list(new_session);
+
/*
* Add the new session to the ltt_sessions_ht_by_id.
* No ownership is taken by the hash table; it is merely
* by session id.
*/
add_session_ht(new_session);
- session_unlock_list();
+ new_session->published = true;
/*
- * Consumer is let to NULL since the create_session_uri command will set it
- * up and, if valid, assign it to the session.
+ * Consumer is left to NULL since the create_session_uri command will
+ * set it up and, if valid, assign it to the session.
*/
- DBG("Tracing session %s created with ID %" PRIu64 " by UID %d GID %d",
- name, new_session->id, new_session->uid, new_session->gid);
-
- return LTTNG_OK;
-
+ DBG("Tracing session %s created with ID %" PRIu64 " by uid = %d, gid = %d",
+ new_session->name, new_session->id, new_session->uid,
+ new_session->gid);
+ ret_code = LTTNG_OK;
+end:
+ if (new_session) {
+ (void) session_get(new_session);
+ *out_session = new_session;
+ }
+ return ret_code;
error:
-error_asprintf:
- free(new_session);
-
-error_malloc:
- return ret;
+ session_put(new_session);
+ new_session = NULL;
+ goto end;
}
/*
projects / lttng-tools.git / blobdiff