int pcaa_initial_connect;
};
+static void __import_set_state(struct obd_import *imp,
+ enum lustre_imp_state state)
+{
+ imp->imp_state = state;
+ imp->imp_state_hist[imp->imp_state_hist_idx].ish_state = state;
+ imp->imp_state_hist[imp->imp_state_hist_idx].ish_time =
+ cfs_time_current_sec();
+ imp->imp_state_hist_idx = (imp->imp_state_hist_idx + 1) %
+ IMP_STATE_HIST_LEN;
+}
+
/* A CLOSED import should remain so. */
#define IMPORT_SET_STATE_NOLOCK(imp, state) \
do { \
imp, obd2cli_tgt(imp->imp_obd), \
ptlrpc_import_state_name(imp->imp_state), \
ptlrpc_import_state_name(state)); \
- imp->imp_state = state; \
+ __import_set_state(imp, state); \
} \
} while(0)
/* Calculate max timeout for waiting on rpcs to error
* out. Use obd_timeout if calculated value is smaller
* than it. */
- timeout = ptlrpc_inflight_timeout(imp);
- timeout += timeout / 3;
+ if (!OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK)) {
+ timeout = ptlrpc_inflight_timeout(imp);
+ timeout += timeout / 3;
- if (timeout == 0)
- timeout = obd_timeout;
+ if (timeout == 0)
+ timeout = obd_timeout;
+ } else {
+ /* decrease the interval to increase race condition */
+ timeout = 1;
+ }
CDEBUG(D_RPCTRACE,"Sleeping %d sec for inflight to error out\n",
timeout);
* have been locally cancelled by ptlrpc_abort_inflight. */
lwi = LWI_TIMEOUT_INTERVAL(
cfs_timeout_cap(cfs_time_seconds(timeout)),
- cfs_time_seconds(1), NULL, NULL);
+ (timeout > 1)?cfs_time_seconds(1):cfs_time_seconds(1)/2,
+ NULL, NULL);
rc = l_wait_event(imp->imp_recovery_waitq,
(atomic_read(&imp->imp_inflight) == 0), &lwi);
if (rc) {
cli_tgt, rc, atomic_read(&imp->imp_inflight));
spin_lock(&imp->imp_lock);
- list_for_each_safe(tmp, n, &imp->imp_sending_list) {
- req = list_entry(tmp, struct ptlrpc_request,
- rq_list);
- DEBUG_REQ(D_ERROR, req,"still on sending list");
- }
- list_for_each_safe(tmp, n, &imp->imp_delayed_list) {
- req = list_entry(tmp, struct ptlrpc_request,
- rq_list);
- DEBUG_REQ(D_ERROR, req,"still on delayed list");
- }
+ if (atomic_read(&imp->imp_inflight) == 0) {
+ int count = atomic_read(&imp->imp_unregistering);
- if (atomic_read(&imp->imp_unregistering) == 0) {
- /* We know that only "unregistering" rpcs may
- * still survive in sending or delaying lists
- * (They are waiting for long reply unlink in
+ /* We know that "unregistering" rpcs only can
+ * survive in sending or delaying lists (they
+ * maybe waiting for long reply unlink in
* sluggish nets). Let's check this. If there
- * is no unregistering and inflight != 0 this
+ * is no inflight and unregistering != 0, this
* is bug. */
- LASSERT(atomic_read(&imp->imp_inflight) == 0);
+ LASSERTF(count == 0, "Some RPCs are still "
+ "unregistering: %d\n", count);
/* Let's save one loop as soon as inflight have
* dropped to zero. No new inflights possible at
* this point. */
rc = 0;
} else {
+ list_for_each_safe(tmp, n,
+ &imp->imp_sending_list) {
+ req = list_entry(tmp,
+ struct ptlrpc_request,
+ rq_list);
+ DEBUG_REQ(D_ERROR, req,
+ "still on sending list");
+ }
+ list_for_each_safe(tmp, n,
+ &imp->imp_delayed_list) {
+ req = list_entry(tmp,
+ struct ptlrpc_request,
+ rq_list);
+ DEBUG_REQ(D_ERROR, req,
+ "still on delayed list");
+ }
+
CERROR("%s: RPCs in \"%s\" phase found (%d). "
"Network is sluggish? Waiting them "
"to error out.\n", cli_tgt,
continue;
}
- /* If we have not tried this connection since the
+ /* If we have not tried this connection since
the last successful attempt, go with this one */
if ((conn->oic_last_attempt == 0) ||
cfs_time_beforeq_64(conn->oic_last_attempt,
}
/* if not found, simply choose the current one */
- if (!imp_conn) {
+ if (!imp_conn || imp->imp_force_reconnect) {
LASSERT(imp->imp_conn_current);
imp_conn = imp->imp_conn_current;
tried_all = 0;
if (imp->imp_conn_current != imp_conn) {
if (imp->imp_conn_current)
- LCONSOLE_INFO("Changing connection for %s to %s/%s\n",
- imp->imp_obd->obd_name,
- imp_conn->oic_uuid.uuid,
- libcfs_nid2str(imp_conn->oic_conn->c_peer.nid));
+ CDEBUG(D_HA, "Changing connection for %s to %s/%s\n",
+ imp->imp_obd->obd_name, imp_conn->oic_uuid.uuid,
+ libcfs_nid2str(imp_conn->oic_conn->c_peer.nid));
imp->imp_conn_current = imp_conn;
}
wake_pinger = 1;
}
#else
- /* liblustre has no pinger thead, so we wakup pinger anyway */
+ /* liblustre has no pinger thread, so we wakeup pinger anyway */
wake_pinger = 1;
#endif
EXIT;
}
+static int ptlrpc_busy_reconnect(int rc)
+{
+ return (rc == -EBUSY) || (rc == -EAGAIN);
+}
+
+
static int ptlrpc_connect_interpret(const struct lu_env *env,
struct ptlrpc_request *request,
void *data, int rc)
spin_unlock(&imp->imp_lock);
RETURN(0);
}
- spin_unlock(&imp->imp_lock);
- if (rc)
+ if (rc) {
+ /* if this reconnect to busy export - not need select new target
+ * for connecting*/
+ imp->imp_force_reconnect = ptlrpc_busy_reconnect(rc);
+ spin_unlock(&imp->imp_lock);
GOTO(out, rc);
+ }
LASSERT(imp->imp_conn_current);
msg_flags = lustre_msg_get_op_flags(request->rq_repmsg);
/* All imports are pingable */
- spin_lock(&imp->imp_lock);
imp->imp_pingable = 1;
+ imp->imp_force_reconnect = 0;
if (aa->pcaa_initial_connect) {
if (msg_flags & MSG_CONNECT_REPLAYABLE) {
/* Reset ns_connect_flags only for initial connect. It might be
* changed in while using FS and if we reset it in reconnect
- * this leads to lossing user settings done before such as
+ * this leads to losing user settings done before such as
* disable lru_resize, etc. */
if (old_connect_flags != exp->exp_connect_flags ||
aa->pcaa_initial_connect) {
req->rq_send_state = LUSTRE_IMP_REPLAY_WAIT;
lustre_msg_add_flags(req->rq_reqmsg,
MSG_LOCK_REPLAY_DONE | MSG_REQ_REPLAY_DONE);
- req->rq_timeout *= 3;
+ if (AT_OFF)
+ req->rq_timeout *= 3;
req->rq_interpret_reply = completed_replay_interpret;
ptlrpcd_add_req(req, PSCOPE_OTHER);
ENTRY;
- ptlrpc_daemonize("ll_imp_inval");
+ cfs_daemonize_ctxt("ll_imp_inval");
CDEBUG(D_HA, "thread invalidate import %s to %s@%s\n",
imp->imp_obd->obd_name, obd2cli_tgt(imp->imp_obd),
#ifdef __KERNEL__
/* bug 17802: XXX client_disconnect_export vs connect request
* race. if client will evicted at this time, we start
- * invalidate thread without referece to import and import can
+ * invalidate thread without reference to import and import can
* be freed at same time. */
class_import_get(imp);
rc = cfs_kernel_thread(ptlrpc_invalidate_import_thread, imp,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff