if (create) {
imp_conn->oic_conn = ptlrpc_conn;
imp_conn->oic_uuid = *uuid;
- item->oic_last_attempt = 0;
+ imp_conn->oic_last_attempt = 0;
if (priority)
list_add(&imp_conn->oic_item, &imp->imp_conn_list);
else
RETURN(rc);
}
-static void destroy_import(struct obd_import *imp)
+void client_destroy_import(struct obd_import *imp)
{
/* drop security policy instance after all rpc finished/aborted
* to let all busy contexts be released. */
cli->cl_dirty_max = num_physpages << (CFS_PAGE_SHIFT - 3);
CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters);
CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list);
+ CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list);
CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list);
CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list);
client_obd_list_lock_init(&cli->cl_loi_list_lock);
/* ->o_connect() method for client side (OSC and MDC and MGC) */
int client_connect_import(const struct lu_env *env,
- struct lustre_handle *dlm_handle,
+ struct obd_export **exp,
struct obd_device *obd, struct obd_uuid *cluuid,
struct obd_connect_data *data, void *localdata)
{
struct client_obd *cli = &obd->u.cli;
struct obd_import *imp = cli->cl_import;
- struct obd_export *exp;
struct obd_connect_data *ocd;
struct ldlm_namespace *to_be_freed = NULL;
+ struct lustre_handle conn = { 0 };
int rc;
ENTRY;
+ *exp = NULL;
down_write(&cli->cl_sem);
- rc = class_connect(dlm_handle, obd, cluuid);
+ if (cli->cl_conn_count > 0 )
+ GOTO(out_sem, rc = -EALREADY);
+
+ rc = class_connect(&conn, obd, cluuid);
if (rc)
GOTO(out_sem, rc);
-
+
cli->cl_conn_count++;
- if (cli->cl_conn_count > 1)
- GOTO(out_sem, rc);
- exp = class_conn2export(dlm_handle);
+ *exp = class_conn2export(&conn);
if (obd->obd_namespace != NULL)
CERROR("already have namespace!\n");
+
+ /*
+ * Deadlock case - bug 18380
+ */
+ up_write(&cli->cl_sem);
obd->obd_namespace = ldlm_namespace_new(obd, obd->obd_name,
LDLM_NAMESPACE_CLIENT,
LDLM_NAMESPACE_GREEDY);
+ down_write(&cli->cl_sem);
if (obd->obd_namespace == NULL)
GOTO(out_disco, rc = -ENOMEM);
- imp->imp_dlm_handle = *dlm_handle;
+ imp->imp_dlm_handle = conn;
rc = ptlrpc_init_import(imp);
if (rc != 0)
GOTO(out_ldlm, rc);
LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
GOTO(out_ldlm, rc);
}
- LASSERT(exp->exp_connection);
+ LASSERT((*exp)->exp_connection);
if (data) {
LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
obd->obd_namespace = NULL;
out_disco:
cli->cl_conn_count--;
- class_disconnect(exp);
- } else {
- class_export_put(exp);
+ class_disconnect(*exp);
+ *exp = NULL;
}
out_sem:
up_write(&cli->cl_sem);
if (!cli->cl_conn_count) {
CERROR("disconnecting disconnected device (%s)\n",
obd->obd_name);
- GOTO(out_sem, rc = -EINVAL);
+ GOTO(out_disconnect, rc = -EINVAL);
}
cli->cl_conn_count--;
if (cli->cl_conn_count)
- GOTO(out_no_disconnect, rc = 0);
+ GOTO(out_disconnect, rc = 0);
/* Mark import deactivated now, so we don't try to reconnect if any
* of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
to_be_freed = obd->obd_namespace;
}
+ /*
+ * there's no necessary to hold sem during diconnecting an import,
+ * and actually it may cause deadlock in gss.
+ */
+ up_write(&cli->cl_sem);
rc = ptlrpc_disconnect_import(imp, 0);
+ down_write(&cli->cl_sem);
ptlrpc_invalidate_import(imp);
/* set obd_namespace to NULL only after invalidate, because we can have
ptlrpc_free_rq_pool(imp->imp_rq_pool);
imp->imp_rq_pool = NULL;
}
- destroy_import(imp);
+ client_destroy_import(imp);
cli->cl_import = NULL;
EXIT;
- out_no_disconnect:
+
+ out_disconnect:
+ /* use server style - class_disconnect should be always called for
+ * o_disconnect */
err = class_disconnect(exp);
if (!rc && err)
rc = err;
- out_sem:
+
up_write(&cli->cl_sem);
if (to_be_freed)
ldlm_namespace_free_post(to_be_freed);
* from old lib/target.c
* -------------------------------------------------------------------------- */
-int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
- struct obd_uuid *cluuid, int mds_conn)
+static int target_handle_reconnect(struct lustre_handle *conn,
+ struct obd_export *exp,
+ struct obd_uuid *cluuid)
{
ENTRY;
- if (exp->exp_connection && exp->exp_imp_reverse && !mds_conn) {
+ if (exp->exp_connection && exp->exp_imp_reverse) {
struct lustre_handle *hdl;
hdl = &exp->exp_imp_reverse->imp_remote_handle;
/* Might be a re-connect after a partition. */
if (obd_uuid_equals(&cluuid, &target->obd_uuid))
goto dont_check_exports;
- spin_lock(&target->obd_dev_lock);
export = lustre_hash_lookup(target->obd_uuid_hash, &cluuid);
+ if (!export)
+ goto no_export;
- if (export != NULL && export->exp_connecting) { /* bug 9635, et. al. */
+ /* we've found an export in the hash */
+ if (export->exp_connecting) { /* bug 9635, et. al. */
CWARN("%s: exp %p already connecting\n",
export->exp_obd->obd_name, export);
class_export_put(export);
export = NULL;
rc = -EALREADY;
- } else if (export != NULL && export->exp_connection != NULL &&
- req->rq_peer.nid != export->exp_connection->c_peer.nid &&
- !mds_conn) {
+ } else if (mds_conn && export->exp_connection) {
+ if (req->rq_peer.nid != export->exp_connection->c_peer.nid)
+ /* mds reconnected after failover */
+ CWARN("%s: received MDS connection from NID %s,"
+ " removing former export from NID %s\n",
+ target->obd_name, libcfs_nid2str(req->rq_peer.nid),
+ libcfs_nid2str(export->exp_connection->c_peer.nid));
+ else
+ /* new mds connection from the same nid */
+ CWARN("%s: received new MDS connection from NID %s,"
+ " removing former export from same NID\n",
+ target->obd_name, libcfs_nid2str(req->rq_peer.nid));
+ class_fail_export(export);
+ class_export_put(export);
+ export = NULL;
+ rc = 0;
+ } else if (export->exp_connection != NULL &&
+ req->rq_peer.nid != export->exp_connection->c_peer.nid) {
/* in mds failover we have static uuid but nid can be
* changed*/
CWARN("%s: cookie %s seen on new NID %s when "
rc = -EALREADY;
class_export_put(export);
export = NULL;
- } else if (export != NULL) {
+ } else {
spin_lock(&export->exp_lock);
export->exp_connecting = 1;
spin_unlock(&export->exp_lock);
class_export_put(export);
- spin_unlock(&target->obd_dev_lock);
LASSERT(export->exp_obd == target);
- rc = target_handle_reconnect(&conn, export, &cluuid, mds_conn);
+ rc = target_handle_reconnect(&conn, export, &cluuid);
}
/* If we found an export, we already unlocked. */
if (!export) {
- spin_unlock(&target->obd_dev_lock);
+no_export:
OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout);
} else if (req->rq_export == NULL &&
atomic_read(&export->exp_rpc_count) > 0) {
GOTO(out, rc = -EBUSY);
} else if (req->rq_export != NULL &&
(atomic_read(&export->exp_rpc_count) > 1)) {
+ /* the current connect rpc has increased exp_rpc_count */
CWARN("%s: refuse reconnection from %s@%s to 0x%p/%d\n",
target->obd_name, cluuid.uuid,
libcfs_nid2str(req->rq_peer.nid),
- export, atomic_read(&export->exp_rpc_count));
+ export, atomic_read(&export->exp_rpc_count) - 1);
+ spin_lock(&export->exp_lock);
+ if (req->rq_export->exp_conn_cnt <
+ lustre_msg_get_conn_cnt(req->rq_reqmsg))
+ /* try to abort active requests */
+ req->rq_export->exp_abort_active_req = 1;
+ spin_unlock(&export->exp_lock);
GOTO(out, rc = -EBUSY);
- } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 &&
- !mds_conn) {
+ } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) {
CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; "
"cookies not random?\n", target->obd_name,
libcfs_nid2str(req->rq_peer.nid), cluuid.uuid);
GOTO(out, rc = -EALREADY);
} else {
OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
- if (req->rq_export == NULL && mds_conn)
- export->exp_last_request_time =
- max(export->exp_last_request_time,
- (time_t)cfs_time_current_sec());
}
if (rc < 0) {
} else {
dont_check_exports:
rc = obd_connect(req->rq_svc_thread->t_env,
- &conn, target, &cluuid, data,
+ &export, target, &cluuid, data,
client_nid);
+ if (rc == 0)
+ conn.cookie = export->exp_handle.h_cookie;
}
} else {
rc = obd_reconnect(req->rq_svc_thread->t_env,
export, target, &cluuid, data, client_nid);
+ if (rc == 0)
+ /* prevous done via class_conn2export */
+ class_export_get(export);
}
if (rc)
GOTO(out, rc);
lustre_msg_set_handle(req->rq_repmsg, &conn);
- /* ownership of this export ref transfers to the request AFTER we
- * drop any previous reference the request had, but we don't want
- * that to go to zero before we get our new export reference. */
- export = class_conn2export(&conn);
- if (!export) {
- DEBUG_REQ(D_ERROR, req, "Missing export!");
- GOTO(out, rc = -ENODEV);
- }
-
/* If the client and the server are the same node, we will already
* have an export that really points to the client's DLM export,
* because we have a shared handles table.
req->rq_export = export;
spin_lock(&export->exp_lock);
- if (mds_conn) {
- lustre_msg_set_conn_cnt(req->rq_repmsg, export->exp_conn_cnt + 1);
- } else if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
+ if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
spin_unlock(&export->exp_lock);
CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n",
cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
GOTO(out, rc = -EALREADY);
}
export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
+ export->exp_abort_active_req = 0;
/* request from liblustre? Don't evict it for not pinging. */
if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
atomic_inc(&target->obd_lock_replay_clients);
if (target->obd_connected_clients ==
target->obd_max_recoverable_clients)
- wake_up(&target->obd_next_transno_waitq);
+ cfs_waitq_signal(&target->obd_next_transno_waitq);
}
spin_unlock_bh(&target->obd_processing_task_lock);
tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
*/
sptlrpc_import_inval_all_ctx(export->exp_imp_reverse);
- destroy_import(export->exp_imp_reverse);
+ client_destroy_import(export->exp_imp_reverse);
}
/* for the rest part, we return -ENOTCONN in case of errors
else
revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
- rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx,
- req->rq_flvr.sf_rpc);
+ 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);
export->exp_imp_reverse = NULL;
/* exports created from last_rcvd data, and "fake"
exports created by lctl don't have an import */
if (exp->exp_imp_reverse != NULL)
- destroy_import(exp->exp_imp_reverse);
+ client_destroy_import(exp->exp_imp_reverse);
/* We cancel locks at disconnect time, but this will catch any locks
* granted in a race with recovery-induced disconnect. */
list_empty(&obd->obd_final_req_queue)) {
obd->obd_processing_task = 0;
} else {
- CERROR("%s: Recovery queues ( %s%s%s) are empty\n",
+ CERROR("%s: Recovery queues ( %s%s%s) are not empty\n",
obd->obd_name,
list_empty(&obd->obd_req_replay_queue) ? "" : "req ",
list_empty(&obd->obd_lock_replay_queue) ? "" : "lock ",
/* when recovery finished, cleanup orphans on mds and ost */
if (OBT(obd) && OBP(obd, postrecov)) {
int rc = OBP(obd, postrecov)(obd);
- LCONSOLE_WARN("%s: recovery %s: rc %d\n", obd->obd_name,
- rc < 0 ? "failed" : "complete", rc);
+ if (rc < 0)
+ LCONSOLE_WARN("%s: Post recovery failed, rc %d\n",
+ obd->obd_name, rc);
}
obd->obd_recovery_end = cfs_time_current_sec();
spin_lock_bh(&obd->obd_processing_task_lock);
list_splice_init(&obd->obd_lock_replay_queue, &abort_list);
spin_unlock_bh(&obd->obd_processing_task_lock);
- list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){
+ list_for_each_entry_safe(req, n, &abort_list, rq_list){
DEBUG_REQ(D_ERROR, req, "aborted:");
req->rq_status = -ENOTCONN;
if (ptlrpc_error(req)) {
spin_unlock_bh(&obd->obd_processing_task_lock);
return;
}
- CWARN("%s: starting recovery timer\n", obd->obd_name);
+ CDEBUG(D_HA, "%s: starting recovery timer\n", obd->obd_name);
obd->obd_recovery_start = cfs_time_current_sec();
/* minimum */
obd->obd_recovery_timeout = OBD_RECOVERY_FACTOR * obd_timeout;
* to replay requests that demand on already committed ones
* also, we can replay first non-committed transation */
LASSERT(req_transno != 0);
- if (req_transno == obd->obd_last_committed + 1) {
+ if (obd->obd_version_recov ||
+ req_transno == obd->obd_last_committed + 1) {
obd->obd_next_recovery_transno = req_transno;
} else if (req_transno > obd->obd_last_committed) {
/* can't continue recovery: have no needed transno */
} else if (!list_empty(&obd->obd_req_replay_queue)) {
req = list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
- target_exp_dequeue_req_replay(req);
list_del_init(&req->rq_list);
obd->obd_requests_queued_for_recovery--;
} else {
req = list_entry(obd->obd_final_req_queue.next,
struct ptlrpc_request, rq_list);
list_del_init(&req->rq_list);
+ if (req->rq_export->exp_in_recovery) {
+ spin_lock(&req->rq_export->exp_lock);
+ req->rq_export->exp_in_recovery = 0;
+ spin_unlock(&req->rq_export->exp_lock);
+ }
} else {
req = NULL;
}
return req;
}
+static inline int req_vbr_done(struct obd_export *exp)
+{
+ return (exp->exp_vbr_failed == 0);
+}
+
static inline int req_replay_done(struct obd_export *exp)
{
return (exp->exp_req_replay_needed == 0);
static int check_for_clients(struct obd_device *obd)
{
- if (obd->obd_abort_recovery)
+ if (obd->obd_abort_recovery || obd->obd_version_recov)
return 1;
LASSERT(obd->obd_connected_clients <= obd->obd_max_recoverable_clients);
if (obd->obd_no_conn == 0 &&
if (!req_replay_done(req->rq_export) ||
!lock_replay_done(req->rq_export))
reset_recovery_timer(class_exp2obd(req->rq_export),
- AT_OFF ? obd_timeout :
+ AT_OFF ? obd_timeout :
at_get(&req->rq_rqbd->rqbd_service->srv_at_estimate), 1);
+
+ /**
+ * bz18031: increase next_recovery_transno before ptlrpc_free_clone()
+ * will drop exp_rpc reference
+ */
+ if (!req_replay_done(req->rq_export)) {
+ spin_lock_bh(&req->rq_export->exp_obd->obd_processing_task_lock);
+ req->rq_export->exp_obd->obd_next_recovery_transno++;
+ spin_unlock_bh(&req->rq_export->exp_obd->obd_processing_task_lock);
+ target_exp_dequeue_req_replay(req);
+ }
ptlrpc_free_clone(req);
RETURN(0);
}
-static void resume_recovery_timer(struct obd_device *obd)
-{
- /* to be safe, make it at least OBD_RECOVERY_FACTOR * obd_timeout */
- reset_recovery_timer(obd, OBD_RECOVERY_FACTOR * obd_timeout, 1);
-}
-
static int target_recovery_thread(void *arg)
{
- struct obd_device *obd = arg;
+ struct lu_target *lut = arg;
+ struct obd_device *obd = lut->lut_obd;
struct ptlrpc_request *req;
struct target_recovery_data *trd = &obd->obd_recovery_data;
struct l_wait_info lwi = { 0 };
env.le_ctx.lc_thread = thread;
CERROR("%s: started recovery thread pid %d\n", obd->obd_name,
- current->pid);
- trd->trd_processing_task = current->pid;
+ cfs_curproc_pid());
+ trd->trd_processing_task = cfs_curproc_pid();
obd->obd_recovering = 1;
complete(&trd->trd_starting);
l_wait_event(obd->obd_next_transno_waitq,
check_for_clients(obd), &lwi);
- spin_lock_bh(&obd->obd_processing_task_lock);
- target_cancel_recovery_timer(obd);
- spin_unlock_bh(&obd->obd_processing_task_lock);
-
/* If some clients haven't connected in time, evict them */
- if (obd->obd_abort_recovery) {
+ if (obd->obd_connected_clients < obd->obd_max_recoverable_clients) {
CWARN("Some clients haven't connect in time (%d/%d),"
"evict them\n", obd->obd_connected_clients,
obd->obd_max_recoverable_clients);
- obd->obd_abort_recovery = obd->obd_stopping;
- class_disconnect_stale_exports(obd, connect_done);
+ class_disconnect_stale_exports(obd, connect_done,
+ exp_flags_from_obd(obd) |
+ OBD_OPT_ABORT_RECOV);
+ /**
+ * if recovery proceeds with versions then some clients may be
+ * timed out waiting for others and trying to reconnect.
+ * Extend timer for such reconnect cases.
+ */
+ if (obd->obd_version_recov)
+ reset_recovery_timer(obd, RECONNECT_DELAY_MAX * 2, 1);
}
+
/* next stage: replay requests */
delta = jiffies;
obd->obd_req_replaying = 1;
CDEBUG(D_INFO, "1: request replay stage - %d clients from t"LPU64"\n",
- atomic_read(&obd->obd_req_replay_clients),
- obd->obd_next_recovery_transno);
- resume_recovery_timer(obd);
+ atomic_read(&obd->obd_req_replay_clients),
+ obd->obd_next_recovery_transno);
while ((req = target_next_replay_req(obd))) {
- LASSERT(trd->trd_processing_task == current->pid);
+ LASSERT(trd->trd_processing_task == cfs_curproc_pid());
DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s",
lustre_msg_get_transno(req->rq_reqmsg),
libcfs_nid2str(req->rq_peer.nid));
handle_recovery_req(thread, req,
trd->trd_recovery_handler);
obd->obd_replayed_requests++;
- spin_lock_bh(&obd->obd_processing_task_lock);
- obd->obd_next_recovery_transno++;
- spin_unlock_bh(&obd->obd_processing_task_lock);
}
- spin_lock_bh(&obd->obd_processing_task_lock);
- target_cancel_recovery_timer(obd);
- spin_unlock_bh(&obd->obd_processing_task_lock);
-
/* If some clients haven't replayed requests in time, evict them */
if (obd->obd_abort_recovery) {
- CDEBUG(D_ERROR, "req replay timed out, aborting ...\n");
- obd->obd_abort_recovery = obd->obd_stopping;
- class_disconnect_stale_exports(obd, req_replay_done);
+ CDEBUG(D_WARNING, "req replay is aborted\n");
+ class_disconnect_stale_exports(obd, req_replay_done,
+ exp_flags_from_obd(obd) |
+ OBD_OPT_ABORT_RECOV);
abort_req_replay_queue(obd);
}
+ LASSERT(list_empty(&obd->obd_req_replay_queue));
/* The second stage: replay locks */
CDEBUG(D_INFO, "2: lock replay stage - %d clients\n",
atomic_read(&obd->obd_lock_replay_clients));
- resume_recovery_timer(obd);
while ((req = target_next_replay_lock(obd))) {
- LASSERT(trd->trd_processing_task == current->pid);
- DEBUG_REQ(D_HA|D_WARNING, req, "processing lock from %s: ",
+ LASSERT(trd->trd_processing_task == cfs_curproc_pid());
+ DEBUG_REQ(D_HA, req, "processing lock from %s: ",
libcfs_nid2str(req->rq_peer.nid));
handle_recovery_req(thread, req,
trd->trd_recovery_handler);
obd->obd_replayed_locks++;
}
- spin_lock_bh(&obd->obd_processing_task_lock);
- target_cancel_recovery_timer(obd);
- spin_unlock_bh(&obd->obd_processing_task_lock);
/* If some clients haven't replayed requests in time, evict them */
if (obd->obd_abort_recovery) {
- int stale;
- CERROR("lock replay timed out, aborting ...\n");
- obd->obd_abort_recovery = obd->obd_stopping;
- stale = class_disconnect_stale_exports(obd, lock_replay_done);
+ CERROR("lock replay is aborted\n");
+ class_disconnect_stale_exports(obd, lock_replay_done,
+ exp_flags_from_obd(obd) |
+ OBD_OPT_ABORT_RECOV);
abort_lock_replay_queue(obd);
}
+ LASSERT(list_empty(&obd->obd_lock_replay_queue));
+ /* The third stage: reply on final pings */
+ CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n");
+ /** evict exports failed VBR */
+ class_disconnect_stale_exports(obd, req_vbr_done,
+ exp_flags_from_obd(obd) |
+ OBD_OPT_ABORT_RECOV);
+ /** Update server last boot epoch */
+ lut_boot_epoch_update(lut);
/* We drop recoverying flag to forward all new requests
* to regular mds_handle() since now */
spin_lock_bh(&obd->obd_processing_task_lock);
obd->obd_recovering = obd->obd_abort_recovery = 0;
+ target_cancel_recovery_timer(obd);
spin_unlock_bh(&obd->obd_processing_task_lock);
- /* The third stage: reply on final pings */
- CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n");
while ((req = target_next_final_ping(obd))) {
- LASSERT(trd->trd_processing_task == current->pid);
+ LASSERT(trd->trd_processing_task == cfs_curproc_pid());
DEBUG_REQ(D_HA, req, "processing final ping from %s: ",
libcfs_nid2str(req->rq_peer.nid));
handle_recovery_req(thread, req,
delta, obd->obd_replayed_requests, obd->obd_replayed_locks);
LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0);
LASSERT(atomic_read(&obd->obd_lock_replay_clients) == 0);
- if (delta > obd_timeout * 2) {
+ if (delta > obd_timeout * OBD_RECOVERY_FACTOR) {
CWARN("too long recovery - read logs\n");
libcfs_debug_dumplog();
}
RETURN(rc);
}
-int target_start_recovery_thread(struct obd_device *obd, svc_handler_t handler)
+static int target_start_recovery_thread(struct lu_target *lut,
+ svc_handler_t handler)
{
+ struct obd_device *obd = lut->lut_obd;
int rc = 0;
struct target_recovery_data *trd = &obd->obd_recovery_data;
init_completion(&trd->trd_finishing);
trd->trd_recovery_handler = handler;
- if (kernel_thread(target_recovery_thread, obd, 0) > 0) {
+ if (kernel_thread(target_recovery_thread, lut, 0) > 0) {
wait_for_completion(&trd->trd_starting);
LASSERT(obd->obd_recovering != 0);
} else
struct target_recovery_data *trd = &obd->obd_recovery_data;
CERROR("%s: Aborting recovery\n", obd->obd_name);
obd->obd_abort_recovery = 1;
- wake_up(&obd->obd_next_transno_waitq);
+ cfs_waitq_signal(&obd->obd_next_transno_waitq);
spin_unlock_bh(&obd->obd_processing_task_lock);
wait_for_completion(&trd->trd_finishing);
} else {
static void target_recovery_expired(unsigned long castmeharder)
{
struct obd_device *obd = (struct obd_device *)castmeharder;
- LCONSOLE_WARN("%s: recovery timed out; %d clients never reconnected "
- "after %lds (%d clients did)\n",
- obd->obd_name, obd->obd_recoverable_clients,
- cfs_time_current_sec()- obd->obd_recovery_start,
- obd->obd_connected_clients);
+ CDEBUG(D_HA, "%s: recovery timed out; %d clients never reconnected "
+ "after %lds (%d clients did)\n",
+ obd->obd_name, obd->obd_recoverable_clients,
+ cfs_time_current_sec()- obd->obd_recovery_start,
+ obd->obd_connected_clients);
+
spin_lock_bh(&obd->obd_processing_task_lock);
- if (obd->obd_recovering)
- obd->obd_abort_recovery = 1;
+ obd->obd_version_recov = 1;
+ CDEBUG(D_INFO, "VBR is used for %d clients from t"LPU64"\n",
+ atomic_read(&obd->obd_req_replay_clients),
+ obd->obd_next_recovery_transno);
cfs_waitq_signal(&obd->obd_next_transno_waitq);
spin_unlock_bh(&obd->obd_processing_task_lock);
}
-void target_recovery_init(struct obd_device *obd, svc_handler_t handler)
+void target_recovery_init(struct lu_target *lut, svc_handler_t handler)
{
- if (obd->obd_max_recoverable_clients == 0)
+ struct obd_device *obd = lut->lut_obd;
+ if (obd->obd_max_recoverable_clients == 0) {
+ /** Update server last boot epoch */
+ lut_boot_epoch_update(lut);
return;
+ }
CWARN("RECOVERY: service %s, %d recoverable clients, "
"last_transno "LPU64"\n", obd->obd_name,
obd->obd_max_recoverable_clients, obd->obd_last_committed);
+ LASSERT(obd->obd_stopping == 0);
obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
obd->obd_recovery_start = 0;
obd->obd_recovery_end = 0;
/* bz13079: this should be set to desired value for ost but not for mds */
obd->obd_recovery_max_time = OBD_RECOVERY_MAX_TIME;
cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd);
- target_start_recovery_thread(obd, handler);
+ target_start_recovery_thread(lut, handler);
}
EXPORT_SYMBOL(target_recovery_init);
#endif
-int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req)
+static int target_process_req_flags(struct obd_device *obd,
+ struct ptlrpc_request *req)
{
struct obd_export *exp = req->rq_export;
LASSERT(exp != NULL);
obd->obd_recoverable_clients--;
if (atomic_read(&obd->obd_req_replay_clients) == 0)
CDEBUG(D_HA, "all clients have replayed reqs\n");
- wake_up(&obd->obd_next_transno_waitq);
}
spin_unlock_bh(&obd->obd_processing_task_lock);
}
atomic_dec(&obd->obd_lock_replay_clients);
if (atomic_read(&obd->obd_lock_replay_clients) == 0)
CDEBUG(D_HA, "all clients have replayed locks\n");
- wake_up(&obd->obd_next_transno_waitq);
}
spin_unlock_bh(&obd->obd_processing_task_lock);
}
struct list_head *tmp;
int inserted = 0;
__u64 transno = lustre_msg_get_transno(req->rq_reqmsg);
-
ENTRY;
if (obd->obd_recovery_data.trd_processing_task == cfs_curproc_pid()) {
RETURN(-ENOMEM);
DEBUG_REQ(D_HA, req, "queue final req");
spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_waitq_signal(&obd->obd_next_transno_waitq);
if (obd->obd_recovering)
list_add_tail(&req->rq_list, &obd->obd_final_req_queue);
else {
RETURN(-ENOMEM);
DEBUG_REQ(D_HA, req, "queue lock replay req");
spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_waitq_signal(&obd->obd_next_transno_waitq);
LASSERT(obd->obd_recovering);
/* usually due to recovery abort */
if (!req->rq_export->exp_in_recovery) {
LASSERT(req->rq_export->exp_lock_replay_needed);
list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue);
spin_unlock_bh(&obd->obd_processing_task_lock);
- wake_up(&obd->obd_next_transno_waitq);
RETURN(0);
}
list_add_tail(&req->rq_list, &obd->obd_req_replay_queue);
obd->obd_requests_queued_for_recovery++;
- wake_up(&obd->obd_next_transno_waitq);
+ cfs_waitq_signal(&obd->obd_next_transno_waitq);
spin_unlock_bh(&obd->obd_processing_task_lock);
RETURN(0);
}
rs->rs_xid = req->rq_xid;
rs->rs_transno = req->rq_transno;
rs->rs_export = exp;
+ rs->rs_opc = lustre_msg_get_opc(rs->rs_msg);
- spin_lock(&obd->obd_uncommitted_replies_lock);
-
+ spin_lock(&exp->exp_uncommitted_replies_lock);
CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n",
- rs->rs_transno, obd->obd_last_committed);
- if (rs->rs_transno > obd->obd_last_committed) {
+ rs->rs_transno, exp->exp_last_committed);
+ if (rs->rs_transno > exp->exp_last_committed) {
/* not committed already */
- list_add_tail (&rs->rs_obd_list,
- &obd->obd_uncommitted_replies);
+ list_add_tail(&rs->rs_obd_list,
+ &exp->exp_uncommitted_replies);
}
+ spin_unlock (&exp->exp_uncommitted_replies_lock);
- spin_unlock (&obd->obd_uncommitted_replies_lock);
- spin_lock (&exp->exp_lock);
-
- list_add_tail (&rs->rs_exp_list, &exp->exp_outstanding_replies);
-
+ spin_lock(&exp->exp_lock);
+ list_add_tail(&rs->rs_exp_list, &exp->exp_outstanding_replies);
spin_unlock(&exp->exp_lock);
netrc = target_send_reply_msg (req, rc, fail_id);
}
spin_lock(&rs->rs_lock);
- if (rs->rs_transno <= obd->obd_last_committed ||
+ if (rs->rs_transno <= exp->exp_last_committed ||
(!rs->rs_on_net && !rs->rs_no_ack) ||
list_empty(&rs->rs_exp_list) || /* completed already */
list_empty(&rs->rs_obd_list)) {
void target_committed_to_req(struct ptlrpc_request *req)
{
- struct obd_device *obd;
-
- if (req == NULL || req->rq_export == NULL)
- return;
-
- obd = req->rq_export->exp_obd;
- if (obd == NULL)
- return;
+ struct obd_export *exp = req->rq_export;
- if (!obd->obd_no_transno && req->rq_repmsg != NULL)
+ if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL)
lustre_msg_set_last_committed(req->rq_repmsg,
- obd->obd_last_committed);
+ exp->exp_last_committed);
else
DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/"
- "%d)", obd->obd_no_transno, req->rq_repmsg == NULL);
+ "%d)", exp->exp_obd->obd_no_transno,
+ req->rq_repmsg == NULL);
CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n",
- obd->obd_last_committed, req->rq_transno, req->rq_xid);
+ exp->exp_last_committed, req->rq_transno, req->rq_xid);
}
-
EXPORT_SYMBOL(target_committed_to_req);
int target_handle_qc_callback(struct ptlrpc_request *req)
LASSERT(req->rq_export);
- OBD_ALLOC(qdata, sizeof(struct qunit_data));
- if (!qdata)
- RETURN(-ENOMEM);
- rc = quota_get_qdata(req, qdata, QUOTA_REQUEST, QUOTA_EXPORT);
- if (rc < 0) {
+ qdata = quota_get_qdata(req, QUOTA_REQUEST, QUOTA_EXPORT);
+ if (IS_ERR(qdata)) {
+ rc = PTR_ERR(qdata);
CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc);
+ req->rq_status = rc;
GOTO(out, rc);
}
if (!obd->obd_observer || !obd->obd_observer->obd_observer) {
CERROR("Can't find the observer, it is recovering\n");
req->rq_status = -EAGAIN;
- GOTO(send_reply, rc = -EAGAIN);
+ GOTO(out, rc);
}
master_obd = obd->obd_observer->obd_observer;
CDEBUG(D_QUOTA, "quota_type not processed yet, return "
"-EAGAIN\n");
req->rq_status = -EAGAIN;
- rc = ptlrpc_reply(req);
GOTO(out, rc);
}
CDEBUG(D_QUOTA, "quota_ctxt is not ready yet, return "
"-EAGAIN\n");
req->rq_status = -EAGAIN;
- rc = ptlrpc_reply(req);
GOTO(out, rc);
}
up_read(&obt->obt_rwsem);
if (rc && rc != -EDQUOT)
CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR,
- "dqacq failed! (rc:%d)\n", rc);
+ "dqacq/dqrel failed! (rc:%d)\n", rc);
req->rq_status = rc;
- /* there are three forms of qunit(historic causes), so we need to
- * adjust the same form to different forms slaves needed */
rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT);
if (rc < 0) {
- CDEBUG(D_ERROR, "Can't pack qunit_data(rc: %d)\n", rc);
+ CERROR("Can't pack qunit_data(rc: %d)\n", rc);
GOTO(out, rc);
}
/* Block the quota req. b=14840 */
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout);
-send_reply:
- rc = ptlrpc_reply(req);
+ EXIT;
+
out:
- OBD_FREE(qdata, sizeof(struct qunit_data));
- RETURN(rc);
+ rc = ptlrpc_reply(req);
+ return rc;
#else
return 0;
#endif /* !__KERNEL__ */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff