* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
+ * Copyright (c) 2011 Whamcloud, Inc.
+ */
+/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*/
struct obd_uuid server_uuid;
int rq_portal, rp_portal, connect_op;
char *name = obddev->obd_type->typ_name;
+ ldlm_ns_type_t ns_type = LDLM_NS_TYPE_UNKNOWN;
int rc;
ENTRY;
connect_op = OST_CONNECT;
cli->cl_sp_me = LUSTRE_SP_CLI;
cli->cl_sp_to = LUSTRE_SP_OST;
+ ns_type = LDLM_NS_TYPE_OSC;
+
} else if (!strcmp(name, LUSTRE_MDC_NAME)) {
rq_portal = MDS_REQUEST_PORTAL;
rp_portal = MDC_REPLY_PORTAL;
connect_op = MDS_CONNECT;
cli->cl_sp_me = LUSTRE_SP_CLI;
cli->cl_sp_to = LUSTRE_SP_MDT;
+ ns_type = LDLM_NS_TYPE_MDC;
+
} else if (!strcmp(name, LUSTRE_MGC_NAME)) {
rq_portal = MGS_REQUEST_PORTAL;
rp_portal = MGC_REPLY_PORTAL;
cli->cl_sp_me = LUSTRE_SP_MGC;
cli->cl_sp_to = LUSTRE_SP_MGS;
cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID;
+ ns_type = LDLM_NS_TYPE_MGC;
+
} else {
CERROR("unknown client OBD type \"%s\", can't setup\n",
name);
GOTO(err_ldlm, rc = -ENOENT);
imp->imp_client = &obddev->obd_ldlm_client;
imp->imp_connect_op = connect_op;
- imp->imp_initial_recov = 1;
- imp->imp_initial_recov_bk = 0;
CFS_INIT_LIST_HEAD(&imp->imp_pinger_chain);
memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
LUSTRE_CFG_BUFLEN(lcfg, 1));
obddev->obd_namespace = ldlm_namespace_new(obddev, obddev->obd_name,
LDLM_NAMESPACE_CLIENT,
- LDLM_NAMESPACE_GREEDY);
+ LDLM_NAMESPACE_GREEDY,
+ ns_type);
if (obddev->obd_namespace == NULL) {
CERROR("Unable to create client namespace - %s\n",
obddev->obd_name);
if (obd->obd_namespace != NULL) {
/* obd_force == local only */
ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
- obd->obd_force ? LDLM_FL_LOCAL_ONLY:0,
- NULL);
+ obd->obd_force ? LCF_LOCAL : 0, NULL);
ldlm_namespace_free_prior(obd->obd_namespace, imp, obd->obd_force);
}
struct ptlrpc_reply_state, rs_exp_list);
struct ptlrpc_service *svc = rs->rs_service;
- cfs_spin_lock(&svc->srv_lock);
+ cfs_spin_lock(&svc->srv_rs_lock);
cfs_list_del_init(&rs->rs_exp_list);
cfs_spin_lock(&rs->rs_lock);
ptlrpc_schedule_difficult_reply(rs);
cfs_spin_unlock(&rs->rs_lock);
- cfs_spin_unlock(&svc->srv_lock);
+ cfs_spin_unlock(&svc->srv_rs_lock);
}
cfs_spin_unlock(&exp->exp_lock);
- /* release nid stat refererence */
- lprocfs_exp_cleanup(exp);
-
RETURN(rc);
}
int rc = 0;
int mds_conn = 0;
struct obd_connect_data *data, *tmpdata;
+ int size, tmpsize;
lnet_nid_t *client_nid = NULL;
ENTRY;
conn = *tmp;
+ size = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA,
+ RCL_CLIENT);
data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
if (!data)
GOTO(out, rc = -EPROTO);
export, (long)cfs_time_current_sec(),
export ? (long)export->exp_last_request_time : 0);
- /* Tell the client if we're in recovery. */
- if (target->obd_recovering) {
- lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
- /* If this is the first time a client connects,
- reset the recovery timer */
- if (rc == 0)
- target_start_and_reset_recovery_timer(target, req,
- !export);
- }
+ /* If this is the first time a client connects,
+ * reset the recovery timer */
+ if (rc == 0 && target->obd_recovering)
+ target_start_and_reset_recovery_timer(target, req, !export);
/* We want to handle EALREADY but *not* -EALREADY from
* target_handle_reconnect(), return reconnection state in a flag */
/* Return only the parts of obd_connect_data that we understand, so the
* client knows that we don't understand the rest. */
if (data) {
- tmpdata = req_capsule_server_get(&req->rq_pill,
- &RMF_CONNECT_DATA);
- //data->ocd_connect_flags &= OBD_CONNECT_SUPPORTED;
- *tmpdata = *data;
+ tmpsize = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA,
+ RCL_SERVER);
+ tmpdata = req_capsule_server_get(&req->rq_pill,
+ &RMF_CONNECT_DATA);
+ /* Don't use struct assignment here, because the client reply
+ * buffer may be smaller/larger than the local struct
+ * obd_connect_data. */
+ memcpy(tmpdata, data, min(tmpsize, size));
}
/* If all else goes well, this is our RPC return code. */
req->rq_self,
&remote_uuid);
if (cfs_hlist_unhashed(&export->exp_nid_hash)) {
- cfs_hash_add_unique(export->exp_obd->obd_nid_hash,
- &export->exp_connection->c_peer.nid,
- &export->exp_nid_hash);
+ cfs_hash_add(export->exp_obd->obd_nid_hash,
+ &export->exp_connection->c_peer.nid,
+ &export->exp_nid_hash);
}
- cfs_spin_lock_bh(&target->obd_processing_task_lock);
+ cfs_spin_lock(&target->obd_recovery_task_lock);
if (target->obd_recovering && !export->exp_in_recovery) {
cfs_spin_lock(&export->exp_lock);
export->exp_in_recovery = 1;
target->obd_max_recoverable_clients)
cfs_waitq_signal(&target->obd_next_transno_waitq);
}
- cfs_spin_unlock_bh(&target->obd_processing_task_lock);
+ cfs_spin_unlock(&target->obd_recovery_task_lock);
+
+ /* Tell the client we're in recovery, when client is involved in it. */
+ if (target->obd_recovering)
+ lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
+
tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
conn = *tmp;
else
revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
+ if ((export->exp_connect_flags & OBD_CONNECT_FULL20) &&
+ (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1))
+ revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18;
+ else
+ revimp->imp_msghdr_flags &= ~MSGHDR_CKSUM_INCOMPAT18;
+
rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, &req->rq_flvr);
if (rc) {
CERROR("Failed to get sec for reverse import: %d\n", rc);
if (exp->exp_imp_reverse != NULL)
client_destroy_import(exp->exp_imp_reverse);
- LASSERT(cfs_atomic_read(&exp->exp_locks_count) == 0);
- LASSERT(cfs_atomic_read(&exp->exp_rpc_count) == 0);
- LASSERT(cfs_atomic_read(&exp->exp_cb_count) == 0);
- LASSERT(cfs_atomic_read(&exp->exp_replay_count) == 0);
+ LASSERT_ATOMIC_ZERO(&exp->exp_locks_count);
+ LASSERT_ATOMIC_ZERO(&exp->exp_rpc_count);
+ LASSERT_ATOMIC_ZERO(&exp->exp_cb_count);
+ LASSERT_ATOMIC_ZERO(&exp->exp_replay_count);
}
/*
class_export_rpc_get(req->rq_export);
LASSERT(cfs_list_empty(&req->rq_list));
CFS_INIT_LIST_HEAD(&req->rq_replay_list);
+
/* increase refcount to keep request in queue */
- LASSERT(cfs_atomic_read(&req->rq_refcount));
cfs_atomic_inc(&req->rq_refcount);
/** let export know it has replays to be handled */
cfs_atomic_inc(&req->rq_export->exp_replay_count);
- /* release service thread while request is queued
- * we are moving the request from active processing
- * to waiting on the replay queue */
- ptlrpc_server_active_request_dec(req);
}
static void target_request_copy_put(struct ptlrpc_request *req)
{
LASSERT(cfs_list_empty(&req->rq_replay_list));
- LASSERT(cfs_atomic_read(&req->rq_export->exp_replay_count) > 0);
+ LASSERT_ATOMIC_POS(&req->rq_export->exp_replay_count);
+
cfs_atomic_dec(&req->rq_export->exp_replay_count);
class_export_rpc_put(req->rq_export);
- /* ptlrpc_server_drop_request() assumes the request is active */
- ptlrpc_server_active_request_inc(req);
ptlrpc_server_drop_request(req);
}
obd->obd_name);
ldlm_reprocess_all_ns(obd->obd_namespace);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_req_replay_queue) ||
!cfs_list_empty(&obd->obd_lock_replay_queue) ||
!cfs_list_empty(&obd->obd_final_req_queue)) {
"" : "lock ",
cfs_list_empty(&obd->obd_final_req_queue) ? \
"" : "final ");
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
LBUG();
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
obd->obd_recovery_end = cfs_time_current_sec();
cfs_list_t abort_list;
CFS_INIT_LIST_HEAD(&abort_list);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
cfs_list_splice_init(&obd->obd_req_replay_queue, &abort_list);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
cfs_list_for_each_entry_safe(req, n, &abort_list, rq_list) {
DEBUG_REQ(D_WARNING, req, "aborted:");
req->rq_status = -ENOTCONN;
cfs_list_t abort_list;
CFS_INIT_LIST_HEAD(&abort_list);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
cfs_list_splice_init(&obd->obd_lock_replay_queue, &abort_list);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
cfs_list_for_each_entry_safe(req, n, &abort_list, rq_list){
DEBUG_REQ(D_ERROR, req, "aborted:");
req->rq_status = -ENOTCONN;
ENTRY;
CFS_INIT_LIST_HEAD(&clean_list);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_dev_lock);
if (!obd->obd_recovering) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_dev_lock);
EXIT;
return;
}
obd->obd_recovering = obd->obd_abort_recovery = 0;
- target_cancel_recovery_timer(obd);
+ cfs_spin_unlock(&obd->obd_dev_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
+ target_cancel_recovery_timer(obd);
cfs_list_splice_init(&obd->obd_req_replay_queue, &clean_list);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
cfs_list_for_each_entry_safe(req, n, &clean_list, rq_list) {
LASSERT(req->rq_reply_state == 0);
target_request_copy_put(req);
}
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
cfs_list_splice_init(&obd->obd_lock_replay_queue, &clean_list);
cfs_list_splice_init(&obd->obd_final_req_queue, &clean_list);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
cfs_list_for_each_entry_safe(req, n, &clean_list, rq_list){
LASSERT(req->rq_reply_state == 0);
EXIT;
}
-/* obd_processing_task_lock should be held */
+/* obd_recovery_task_lock should be held */
void target_cancel_recovery_timer(struct obd_device *obd)
{
CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
cfs_time_t now = cfs_time_current_sec();
cfs_duration_t left;
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!obd->obd_recovering || obd->obd_abort_recovery) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
return;
}
left = cfs_time_sub(obd->obd_recovery_end, now);
cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(left));
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n",
obd->obd_name, (unsigned)left);
}
static void check_and_start_recovery_timer(struct obd_device *obd)
{
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (cfs_timer_is_armed(&obd->obd_recovery_timer)) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
return;
}
CDEBUG(D_HA, "%s: starting recovery timer\n", obd->obd_name);
obd->obd_recovery_start = cfs_time_current_sec();
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
reset_recovery_timer(obd, obd->obd_recovery_timeout, 0);
}
int wake_up = 0, connected, completed, queue_len;
__u64 next_transno, req_transno;
ENTRY;
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_req_replay_queue)) {
req = cfs_list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
obd->obd_next_recovery_transno = req_transno;
wake_up = 1;
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
return wake_up;
}
{
int wake_up = 0;
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_lock_replay_queue)) {
CDEBUG(D_HA, "waking for next lock\n");
wake_up = 1;
CDEBUG(D_HA, "waking for expired recovery\n");
wake_up = 1;
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
return wake_up;
}
int (*check_routine)(struct obd_device *),
int (*health_check)(struct obd_export *))
{
- int abort = 0, expired = 1;
-
- do {
- cfs_wait_event(obd->obd_next_transno_waitq, check_routine(obd));
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
- abort = obd->obd_abort_recovery;
- expired = obd->obd_recovery_expired;
+repeat:
+ cfs_wait_event(obd->obd_next_transno_waitq, check_routine(obd));
+ if (obd->obd_abort_recovery) {
+ CWARN("recovery is aborted, evict exports in recovery\n");
+ /** evict exports which didn't finish recovery yet */
+ class_disconnect_stale_exports(obd, exp_finished);
+ return 1;
+ } else if (obd->obd_recovery_expired) {
obd->obd_recovery_expired = 0;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
- if (abort) {
- CWARN("recovery is aborted, evict exports in recovery\n");
- /** evict exports which didn't finish recovery yet */
- class_disconnect_stale_exports(obd, exp_finished);
- } else if (expired) {
- /** If some clients died being recovered, evict them */
- CDEBUG(D_WARNING, "recovery is timed out, evict stale exports\n");
- /** evict cexports with no replay in queue, they are stalled */
- class_disconnect_stale_exports(obd, health_check);
- /** continue with VBR */
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
- obd->obd_version_recov = 1;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
- /**
- * reset timer, recovery will proceed with versions now,
- * timeout is set just to handle reconnection delays
- */
- reset_recovery_timer(obd, RECONNECT_DELAY_MAX, 1);
- /** Wait for recovery events again, after evicting bad clients */
- }
- } while (!abort && expired);
-
- return abort;
+ /** If some clients died being recovered, evict them */
+ CDEBUG(D_WARNING,
+ "recovery is timed out, evict stale exports\n");
+ /** evict cexports with no replay in queue, they are stalled */
+ class_disconnect_stale_exports(obd, health_check);
+ /** continue with VBR */
+ cfs_spin_lock(&obd->obd_dev_lock);
+ obd->obd_version_recov = 1;
+ cfs_spin_unlock(&obd->obd_dev_lock);
+ /**
+ * reset timer, recovery will proceed with versions now,
+ * timeout is set just to handle reconnection delays
+ */
+ reset_recovery_timer(obd, RECONNECT_DELAY_MAX, 1);
+ /** Wait for recovery events again, after evicting bad clients */
+ goto repeat;
+ }
+ return 0;
}
static struct ptlrpc_request *target_next_replay_req(struct obd_device *obd)
abort_lock_replay_queue(obd);
}
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_req_replay_queue)) {
req = cfs_list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
cfs_list_del_init(&req->rq_list);
obd->obd_requests_queued_for_recovery--;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
} else {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
LASSERT(cfs_list_empty(&obd->obd_req_replay_queue));
LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0);
/** evict exports failed VBR */
exp_lock_replay_healthy))
abort_lock_replay_queue(obd);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_lock_replay_queue)) {
req = cfs_list_entry(obd->obd_lock_replay_queue.next,
struct ptlrpc_request, rq_list);
cfs_list_del_init(&req->rq_list);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
} else {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
LASSERT(cfs_list_empty(&obd->obd_lock_replay_queue));
LASSERT(cfs_atomic_read(&obd->obd_lock_replay_clients) == 0);
/** evict exports failed VBR */
{
struct ptlrpc_request *req = NULL;
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (!cfs_list_empty(&obd->obd_final_req_queue)) {
req = cfs_list_entry(obd->obd_final_req_queue.next,
struct ptlrpc_request, rq_list);
cfs_list_del_init(&req->rq_list);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
if (req->rq_export->exp_in_recovery) {
cfs_spin_lock(&req->rq_export->exp_lock);
req->rq_export->exp_in_recovery = 0;
cfs_spin_unlock(&req->rq_export->exp_lock);
}
+ } else {
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
return req;
}
int rc;
ENTRY;
- rc = lu_context_init(&req->rq_recov_session, LCT_SESSION);
- if (rc) {
- CERROR("Failure to initialize session: %d\n", rc);
- GOTO(reqcopy_put, rc);
- }
/**
* export can be evicted during recovery, no need to handle replays for
* it after that, discard such request silently
*/
if (req->rq_export->exp_disconnected)
+ GOTO(reqcopy_put, rc = 0);
+
+ rc = lu_context_init(&req->rq_recov_session, LCT_SESSION);
+ if (rc) {
+ CERROR("Failure to initialize session: %d\n", rc);
GOTO(reqcopy_put, rc);
+ }
req->rq_recov_session.lc_thread = thread;
lu_context_enter(&req->rq_recov_session);
thread->t_env = &env;
thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */
env.le_ctx.lc_thread = thread;
+ thread->t_data = NULL;
CERROR("%s: started recovery thread pid %d\n", obd->obd_name,
cfs_curproc_pid());
trd->trd_processing_task = cfs_curproc_pid();
+ cfs_spin_lock(&obd->obd_dev_lock);
obd->obd_recovering = 1;
+ cfs_spin_unlock(&obd->obd_dev_lock);
cfs_complete(&trd->trd_starting);
/* first of all, we have to know the first transno to replay */
/* next stage: replay requests */
delta = jiffies;
- obd->obd_req_replaying = 1;
CDEBUG(D_INFO, "1: request replay stage - %d clients from t"LPU64"\n",
cfs_atomic_read(&obd->obd_req_replay_clients),
obd->obd_next_recovery_transno);
* bz18031: increase next_recovery_transno before
* target_request_copy_put() will drop exp_rpc reference
*/
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
obd->obd_next_recovery_transno++;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
target_exp_dequeue_req_replay(req);
target_request_copy_put(req);
obd->obd_replayed_requests++;
lut_boot_epoch_update(lut);
/* We drop recoverying flag to forward all new requests
* to regular mds_handle() since now */
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_dev_lock);
obd->obd_recovering = obd->obd_abort_recovery = 0;
+ cfs_spin_unlock(&obd->obd_dev_lock);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
target_cancel_recovery_timer(obd);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
while ((req = target_next_final_ping(obd))) {
LASSERT(trd->trd_processing_task == cfs_curproc_pid());
DEBUG_REQ(D_HA, req, "processing final ping from %s: ",
cfs_init_completion(&trd->trd_finishing);
trd->trd_recovery_handler = handler;
- if (cfs_kernel_thread(target_recovery_thread, lut, 0) > 0) {
+ if (cfs_create_thread(target_recovery_thread, lut, 0) > 0) {
cfs_wait_for_completion(&trd->trd_starting);
LASSERT(obd->obd_recovering != 0);
} else
void target_stop_recovery_thread(struct obd_device *obd)
{
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
if (obd->obd_recovery_data.trd_processing_task > 0) {
struct target_recovery_data *trd = &obd->obd_recovery_data;
/** recovery can be done but postrecovery is not yet */
+ cfs_spin_lock(&obd->obd_dev_lock);
if (obd->obd_recovering) {
CERROR("%s: Aborting recovery\n", obd->obd_name);
obd->obd_abort_recovery = 1;
cfs_waitq_signal(&obd->obd_next_transno_waitq);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_dev_lock);
cfs_wait_for_completion(&trd->trd_finishing);
- } else {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
}
}
cfs_time_current_sec()- obd->obd_recovery_start,
obd->obd_connected_clients);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
obd->obd_recovery_expired = 1;
cfs_waitq_signal(&obd->obd_next_transno_waitq);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
}
void target_recovery_init(struct lu_target *lut, svc_handler_t handler)
LASSERT(exp != NULL);
if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
/* client declares he's ready to replay locks */
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&exp->exp_lock);
if (exp->exp_req_replay_needed) {
- LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) >
- 0);
- cfs_spin_lock(&exp->exp_lock);
exp->exp_req_replay_needed = 0;
cfs_spin_unlock(&exp->exp_lock);
+
+ LASSERT_ATOMIC_POS(&obd->obd_req_replay_clients);
cfs_atomic_dec(&obd->obd_req_replay_clients);
+ } else {
+ cfs_spin_unlock(&exp->exp_lock);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
}
if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
/* client declares he's ready to complete recovery
* so, we put the request on th final queue */
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&exp->exp_lock);
if (exp->exp_lock_replay_needed) {
- LASSERT(cfs_atomic_read(&obd->obd_lock_replay_clients) >
- 0);
- cfs_spin_lock(&exp->exp_lock);
exp->exp_lock_replay_needed = 0;
cfs_spin_unlock(&exp->exp_lock);
+
+ LASSERT_ATOMIC_POS(&obd->obd_lock_replay_clients);
cfs_atomic_dec(&obd->obd_lock_replay_clients);
+ } else {
+ cfs_spin_unlock(&exp->exp_lock);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
}
-
return 0;
}
* so, we put the request on th final queue */
target_request_copy_get(req);
DEBUG_REQ(D_HA, req, "queue final req");
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
cfs_waitq_signal(&obd->obd_next_transno_waitq);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
if (obd->obd_recovering) {
cfs_list_add_tail(&req->rq_list,
&obd->obd_final_req_queue);
} else {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
target_request_copy_put(req);
RETURN(obd->obd_stopping ? -ENOTCONN : 1);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
RETURN(0);
}
if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
/* client declares he's ready to replay locks */
target_request_copy_get(req);
DEBUG_REQ(D_HA, req, "queue lock replay req");
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
cfs_waitq_signal(&obd->obd_next_transno_waitq);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
LASSERT(obd->obd_recovering);
/* usually due to recovery abort */
if (!req->rq_export->exp_in_recovery) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
target_request_copy_put(req);
RETURN(-ENOTCONN);
}
LASSERT(req->rq_export->exp_lock_replay_needed);
cfs_list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
RETURN(0);
}
RETURN(1);
}
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
-
/* If we're processing the queue, we want don't want to queue this
* message.
*
* Also, a resent, replayed request that has already been
* handled will pass through here and be processed immediately.
*/
- CWARN("Next recovery transno: "LPU64", current: "LPU64", replaying: %i\n",
- obd->obd_next_recovery_transno, transno, obd->obd_req_replaying);
- if (transno < obd->obd_next_recovery_transno && obd->obd_req_replaying) {
+ CWARN("Next recovery transno: "LPU64", current: "LPU64", replaying\n",
+ obd->obd_next_recovery_transno, transno);
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
+ if (transno < obd->obd_next_recovery_transno) {
/* Processing the queue right now, don't re-add. */
LASSERT(cfs_list_empty(&req->rq_list));
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
RETURN(1);
}
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
if (OBD_FAIL_CHECK(OBD_FAIL_TGT_REPLAY_DROP))
RETURN(0);
target_request_copy_get(req);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
- LASSERT(obd->obd_recovering);
if (!req->rq_export->exp_in_recovery) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
target_request_copy_put(req);
RETURN(-ENOTCONN);
}
LASSERT(req->rq_export->exp_req_replay_needed);
if (target_exp_enqueue_req_replay(req)) {
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
DEBUG_REQ(D_ERROR, req, "dropping resent queued req");
target_request_copy_put(req);
RETURN(0);
}
/* XXX O(n^2) */
+ cfs_spin_lock(&obd->obd_recovery_task_lock);
+ LASSERT(obd->obd_recovering);
cfs_list_for_each(tmp, &obd->obd_req_replay_queue) {
struct ptlrpc_request *reqiter =
cfs_list_entry(tmp, struct ptlrpc_request, rq_list);
transno)) {
DEBUG_REQ(D_ERROR, req, "dropping replay: transno "
"has been claimed by another client");
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
target_exp_dequeue_req_replay(req);
target_request_copy_put(req);
RETURN(0);
cfs_list_add_tail(&req->rq_list, &obd->obd_req_replay_queue);
obd->obd_requests_queued_for_recovery++;
+ cfs_spin_unlock(&obd->obd_recovery_task_lock);
cfs_waitq_signal(&obd->obd_next_transno_waitq);
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
RETURN(0);
}
netrc = target_send_reply_msg (req, rc, fail_id);
- cfs_spin_lock(&svc->srv_lock);
+ cfs_spin_lock(&svc->srv_rs_lock);
cfs_atomic_inc(&svc->srv_n_difficult_replies);
* reply_out_callback leaves alone) */
rs->rs_on_net = 0;
ptlrpc_rs_addref(rs);
- cfs_atomic_inc (&svc->srv_outstanding_replies);
}
cfs_spin_lock(&rs->rs_lock);
rs->rs_scheduled = 0; /* allow notifier to schedule */
}
cfs_spin_unlock(&rs->rs_lock);
- cfs_spin_unlock(&svc->srv_lock);
+ cfs_spin_unlock(&svc->srv_rs_lock);
EXIT;
}
if (((int)error) < 0) /* cast to signed type */
result = error; /* as ldlm_error_t can be unsigned */
else {
- CERROR("Invalid DLM result code: %i\n", error);
+ CERROR("Invalid DLM result code: %d\n", error);
result = -EPROTO;
}
}
cfs_spin_unlock(&exp->exp_locks_list_guard);
}
#endif
+
+static int target_bulk_timeout(void *data)
+{
+ ENTRY;
+ /* We don't fail the connection here, because having the export
+ * killed makes the (vital) call to commitrw very sad.
+ */
+ RETURN(1);
+}
+
+static inline char *bulk2type(struct ptlrpc_bulk_desc *desc)
+{
+ return desc->bd_type == BULK_GET_SINK ? "GET" : "PUT";
+}
+
+int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc,
+ struct l_wait_info *lwi)
+{
+ struct ptlrpc_request *req = desc->bd_req;
+ int rc = 0;
+ ENTRY;
+
+ /* Check if there is eviction in progress, and if so, wait for
+ * it to finish */
+ if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+ *lwi = LWI_INTR(NULL, NULL);
+ rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
+ !cfs_atomic_read(&exp->exp_obd->
+ obd_evict_inprogress),
+ lwi);
+ }
+
+ /* Check if client was evicted or tried to reconnect already */
+ if (exp->exp_failed || exp->exp_abort_active_req) {
+ rc = -ENOTCONN;
+ } else {
+ if (desc->bd_type == BULK_PUT_SINK)
+ rc = sptlrpc_svc_wrap_bulk(req, desc);
+ if (rc == 0)
+ rc = ptlrpc_start_bulk_transfer(desc);
+ }
+
+ if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
+ ptlrpc_abort_bulk(desc);
+ } else if (rc == 0) {
+ time_t start = cfs_time_current_sec();
+ do {
+ long timeoutl = req->rq_deadline - cfs_time_current_sec();
+ cfs_duration_t timeout = timeoutl <= 0 ?
+ CFS_TICK : cfs_time_seconds(timeoutl);
+ *lwi = LWI_TIMEOUT_INTERVAL(timeout,
+ cfs_time_seconds(1),
+ target_bulk_timeout,
+ desc);
+ rc = l_wait_event(desc->bd_waitq,
+ !ptlrpc_server_bulk_active(desc) ||
+ exp->exp_failed ||
+ exp->exp_abort_active_req,
+ lwi);
+ LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ /* Wait again if we changed deadline */
+ } while ((rc == -ETIMEDOUT) &&
+ (req->rq_deadline > cfs_time_current_sec()));
+
+ if (rc == -ETIMEDOUT) {
+ DEBUG_REQ(D_ERROR, req,
+ "timeout on bulk %s after %ld%+lds",
+ bulk2type(desc),
+ req->rq_deadline - start,
+ cfs_time_current_sec() -
+ req->rq_deadline);
+ ptlrpc_abort_bulk(desc);
+ } else if (exp->exp_failed) {
+ DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s",
+ bulk2type(desc));
+ rc = -ENOTCONN;
+ ptlrpc_abort_bulk(desc);
+ } else if (exp->exp_abort_active_req) {
+ DEBUG_REQ(D_ERROR, req, "Reconnect on bulk %s",
+ bulk2type(desc));
+ /* we don't reply anyway */
+ rc = -ETIMEDOUT;
+ ptlrpc_abort_bulk(desc);
+ } else if (!desc->bd_success ||
+ desc->bd_nob_transferred != desc->bd_nob) {
+ DEBUG_REQ(D_ERROR, req, "%s bulk %s %d(%d)",
+ desc->bd_success ?
+ "truncated" : "network error on",
+ bulk2type(desc),
+ desc->bd_nob_transferred,
+ desc->bd_nob);
+ /* XXX should this be a different errno? */
+ rc = -ETIMEDOUT;
+ } else if (desc->bd_type == BULK_GET_SINK) {
+ rc = sptlrpc_svc_unwrap_bulk(req, desc);
+ }
+ } else {
+ DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d",
+ bulk2type(desc), rc);
+ }
+
+ RETURN(rc);
+}
+EXPORT_SYMBOL(target_bulk_io);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff