-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
*
- * Portal-RPC reconnection and replay operations, for use in recovery.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
- * Author: Mike Shaver <shaver@clusterfs.com>
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * This file is part of Lustre, http://www.lustre.org.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
- * Lustre is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2017, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * lustre/ptlrpc/recover.c
+ *
+ * Author: Mike Shaver <shaver@clusterfs.com>
*/
#define DEBUG_SUBSYSTEM S_RPC
-#ifdef __KERNEL__
-# include <linux/config.h>
-# include <linux/module.h>
-# include <linux/kmod.h>
-#else
-# include <liblustre.h>
-#endif
-
-#include <linux/obd_support.h>
-#include <linux/lustre_ha.h>
-#include <linux/lustre_net.h>
-#include <linux/lustre_import.h>
-#include <linux/lustre_export.h>
-#include <linux/obd.h>
-#include <linux/obd_ost.h>
-#include <linux/obd_class.h>
-#include <linux/obd_lov.h> /* for IOC_LOV_SET_OSC_ACTIVE */
+#include <linux/list.h>
+#include <libcfs/libcfs.h>
+#include <obd_support.h>
+#include <lustre_ha.h>
+#include <lustre_net.h>
+#include <lustre_import.h>
+#include <lustre_export.h>
+#include <obd.h>
+#include <obd_class.h>
#include "ptlrpc_internal.h"
-static int ptlrpc_recover_import_no_retry(struct obd_import *, char *);
-
-void ptlrpc_run_recovery_over_upcall(struct obd_device *obd)
+/**
+ * Start recovery on disconnected import.
+ * This is done by just attempting a connect
+ */
+void ptlrpc_initiate_recovery(struct obd_import *imp)
{
- char *argv[4];
- char *envp[3];
- int rc;
-
ENTRY;
- argv[0] = obd_lustre_upcall;
- argv[1] = "RECOVERY_OVER";
- argv[2] = obd->obd_uuid.uuid;
- argv[3] = NULL;
-
- envp[0] = "HOME=/";
- envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
- envp[2] = NULL;
-
- rc = USERMODEHELPER(argv[0], argv, envp);
- if (rc < 0) {
- CERROR("Error invoking recovery upcall %s %s %s: %d; check "
- "/proc/sys/lustre/upcall\n",
- argv[0], argv[1], argv[2], rc);
-
- } else {
- CERROR("Invoked upcall %s %s %s",
- argv[0], argv[1], argv[2]);
- }
-}
-void ptlrpc_run_failed_import_upcall(struct obd_import* imp)
-{
-#ifdef __KERNEL__
- char *argv[7];
- char *envp[3];
- int rc;
+ CDEBUG(D_HA, "%s: starting recovery\n", obd2cli_tgt(imp->imp_obd));
+ ptlrpc_connect_import(imp);
- ENTRY;
- argv[0] = obd_lustre_upcall;
- argv[1] = "FAILED_IMPORT";
- argv[2] = imp->imp_target_uuid.uuid;
- argv[3] = imp->imp_obd->obd_name;
- argv[4] = imp->imp_connection->c_remote_uuid.uuid;
- argv[5] = imp->imp_obd->obd_uuid.uuid;
- argv[6] = NULL;
-
- envp[0] = "HOME=/";
- envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
- envp[2] = NULL;
-
- rc = USERMODEHELPER(argv[0], argv, envp);
- if (rc < 0) {
- CERROR("Error invoking recovery upcall %s %s %s %s %s: %d; "
- "check /proc/sys/lustre/lustre_upcall\n",
- argv[0], argv[1], argv[2], argv[3], argv[4],rc);
-
- } else {
- CERROR("Invoked upcall %s %s %s %s %s\n",
- argv[0], argv[1], argv[2], argv[3], argv[4]);
- }
-#else
- ptlrpc_recover_import(imp, NULL);
-#endif
+ EXIT;
}
-int ptlrpc_replay(struct obd_import *imp)
+/**
+ * Identify what request from replay list needs to be replayed next
+ * (based on what we have already replayed) and send it to server.
+ */
+int ptlrpc_replay_next(struct obd_import *imp, int *inflight)
{
- int rc = 0;
- struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
- unsigned long flags;
- ENTRY;
-
- /* It might have committed some after we last spoke, so make sure we
- * get rid of them now.
- */
- spin_lock_irqsave(&imp->imp_lock, flags);
- ptlrpc_free_committed(imp);
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- CDEBUG(D_HA, "import %p from %s has committed "LPD64"\n",
- imp, imp->imp_target_uuid.uuid, imp->imp_peer_committed_transno);
-
- list_for_each(tmp, &imp->imp_replay_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- DEBUG_REQ(D_HA, req, "RETAINED: ");
- }
-
- /* Do I need to hold a lock across this iteration? We shouldn't be
- * racing with any additions to the list, because we're in recovery
- * and are therefore not processing additional requests to add. Calls
- * to ptlrpc_free_committed might commit requests, but nothing "newer"
- * than the one we're replaying (it can't be committed until it's
- * replayed, and we're doing that here). l_f_e_safe protects against
- * problems with the current request being committed, in the unlikely
- * event of that race. So, in conclusion, I think that it's safe to
- * perform this list-walk without the imp_lock held.
- *
- * But, the {mdc,osc}_replay_open callbacks both iterate
- * request lists, and have comments saying they assume the
- * imp_lock is being held by ptlrpc_replay, but it's not. it's
- * just a little race...
- */
- list_for_each_safe(tmp, pos, &imp->imp_replay_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
-
- DEBUG_REQ(D_HA, req, "REPLAY:");
-
- rc = ptlrpc_replay_req(req);
-
- if (rc) {
- CERROR("recovery replay error %d for req "LPD64"\n",
- rc, req->rq_xid);
- RETURN(rc);
- }
- }
-
- RETURN(0);
+ int rc = 0;
+ struct ptlrpc_request *req = NULL;
+ __u64 last_transno;
+ ENTRY;
+
+ *inflight = 0;
+
+ /* It might have committed some after we last spoke, so make sure we
+ * get rid of them now.
+ */
+ spin_lock(&imp->imp_lock);
+ imp->imp_last_transno_checked = 0;
+ ptlrpc_free_committed(imp);
+ last_transno = imp->imp_last_replay_transno;
+
+ CDEBUG(D_HA, "import %p from %s committed %llu last %llu\n",
+ imp, obd2cli_tgt(imp->imp_obd),
+ imp->imp_peer_committed_transno, last_transno);
+
+ /* Replay all the committed open requests on committed_list first */
+ if (!list_empty(&imp->imp_committed_list)) {
+ req = list_last_entry(&imp->imp_committed_list,
+ struct ptlrpc_request, rq_replay_list);
+
+ /* The last request on committed_list hasn't been replayed */
+ if (req->rq_transno > last_transno) {
+ if (!imp->imp_resend_replay ||
+ imp->imp_replay_cursor == &imp->imp_committed_list)
+ imp->imp_replay_cursor =
+ imp->imp_replay_cursor->next;
+
+ while (imp->imp_replay_cursor !=
+ &imp->imp_committed_list) {
+ req = list_entry(imp->imp_replay_cursor,
+ struct ptlrpc_request,
+ rq_replay_list);
+ if (req->rq_transno > last_transno)
+ break;
+
+ req = NULL;
+ LASSERT(!list_empty(imp->imp_replay_cursor));
+ imp->imp_replay_cursor =
+ imp->imp_replay_cursor->next;
+ }
+ } else {
+ /* All requests on committed_list have been replayed */
+ imp->imp_replay_cursor = &imp->imp_committed_list;
+ req = NULL;
+ }
+ }
+
+ /* All the requests in committed list have been replayed, let's replay
+ * the imp_replay_list */
+ if (req == NULL) {
+ struct ptlrpc_request *tmp;
+
+ list_for_each_entry(tmp, &imp->imp_replay_list,
+ rq_replay_list) {
+ if (tmp->rq_transno > last_transno) {
+ req = tmp;
+ break;
+ }
+ }
+ }
+
+ /* If need to resend the last sent transno (because a reconnect
+ * has occurred), then stop on the matching req and send it again.
+ * If, however, the last sent transno has been committed then we
+ * continue replay from the next request. */
+ if (req != NULL && imp->imp_resend_replay)
+ lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
+
+ /* ptlrpc_prepare_replay() may fail to add the reqeust into unreplied
+ * list if the request hasn't been added to replay list then. Another
+ * exception is that resend replay could have been removed from the
+ * unreplied list. */
+ if (req != NULL && list_empty(&req->rq_unreplied_list)) {
+ DEBUG_REQ(D_HA, req, "resend_replay=%d, last_transno=%llu",
+ imp->imp_resend_replay, last_transno);
+ ptlrpc_add_unreplied(req);
+ imp->imp_known_replied_xid = ptlrpc_known_replied_xid(imp);
+ }
+
+ imp->imp_resend_replay = 0;
+ spin_unlock(&imp->imp_lock);
+
+ if (req != NULL) {
+ LASSERT(!list_empty(&req->rq_unreplied_list));
+
+ rc = ptlrpc_replay_req(req);
+ if (rc) {
+ CERROR("recovery replay error %d for req %llu\n",
+ rc, req->rq_xid);
+ RETURN(rc);
+ }
+ *inflight = 1;
+ }
+ RETURN(rc);
}
+/**
+ * Schedule resending of request on sending_list. This is done after
+ * we completed replaying of requests and locks.
+ */
int ptlrpc_resend(struct obd_import *imp)
{
- struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
- unsigned long flags;
+ struct ptlrpc_request *req;
ENTRY;
*/
/* Well... what if lctl recover is called twice at the same time?
*/
- spin_lock_irqsave(&imp->imp_lock, flags);
- LASSERT(imp->imp_state == LUSTRE_IMP_RECOVER);
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- list_for_each_safe(tmp, pos, &imp->imp_sending_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- ptlrpc_resend_req(req);
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state != LUSTRE_IMP_RECOVER) {
+ spin_unlock(&imp->imp_lock);
+ RETURN(-1);
}
- RETURN(0);
+ list_for_each_entry(req, &imp->imp_sending_list, rq_list) {
+ LASSERTF((long)req > PAGE_SIZE && req != LP_POISON,
+ "req %p bad\n", req);
+ LASSERTF(req->rq_type != LI_POISON, "req %p freed\n", req);
+
+ /* If the request is allowed to be sent during replay and it
+ * is not timeout yet, then it does not need to be resent. */
+ if (!ptlrpc_no_resend(req) &&
+ (req->rq_timedout || !req->rq_allow_replay))
+ ptlrpc_resend_req(req);
+ }
+ spin_unlock(&imp->imp_lock);
+
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT, 2);
+ RETURN(0);
}
+/**
+ * Go through all requests in delayed list and wake their threads
+ * for resending
+ */
void ptlrpc_wake_delayed(struct obd_import *imp)
{
- unsigned long flags;
- struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- list_for_each_safe(tmp, pos, &imp->imp_delayed_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
-
- ptlrpc_put_connection(req->rq_connection);
- req->rq_connection =
- ptlrpc_connection_addref(req->rq_import->imp_connection);
-
- if (req->rq_set) {
- DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set);
- wake_up(&req->rq_set->set_waitq);
- } else {
- DEBUG_REQ(D_HA, req, "waking:");
- wake_up(&req->rq_reply_waitq);
- }
- }
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-}
-
-inline void ptlrpc_invalidate_import_state(struct obd_import *imp)
-{
- struct obd_device *obd = imp->imp_obd;
- struct ldlm_namespace *ns = obd->obd_namespace;
-
- ptlrpc_abort_inflight(imp);
-
- obd_invalidate_import(obd, imp);
-
- ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY | LDLM_FL_CANCEL);
-}
-
-void ptlrpc_handle_failed_import(struct obd_import *imp)
-{
- ENTRY;
- if (!imp->imp_replayable) {
- CDEBUG(D_HA,
- "import %s@%s for %s not replayable, deactivating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid,
- imp->imp_obd->obd_name);
- ptlrpc_set_import_active(imp, 0);
- }
-
- ptlrpc_run_failed_import_upcall(imp);
- EXIT;
+ struct ptlrpc_request *req;
+
+ spin_lock(&imp->imp_lock);
+ list_for_each_entry(req, &imp->imp_delayed_list, rq_list) {
+ DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set);
+ ptlrpc_client_wake_req(req);
+ }
+ spin_unlock(&imp->imp_lock);
}
void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req)
{
- int rc;
- struct obd_import *imp= failed_req->rq_import;
- unsigned long flags;
- ENTRY;
-
- CDEBUG(D_HA, "import %s of %s@%s abruptly disconnected: reconnecting\n",
- imp->imp_obd->obd_name,
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
-
- rc = ptlrpc_recover_import_no_retry(imp, NULL);
-
- if (failed_req->rq_import_generation != imp->imp_generation) {
- spin_lock_irqsave (&failed_req->rq_lock, flags);
- failed_req->rq_err = 1;
- spin_unlock_irqrestore (&failed_req->rq_lock, flags);
- }
- else {
- ptlrpc_resend_req(failed_req);
- if (rc && rc != -EALREADY)
- ptlrpc_handle_failed_import(imp);
-
- }
- EXIT;
+ struct obd_import *imp = failed_req->rq_import;
+ int conn = lustre_msg_get_conn_cnt(failed_req->rq_reqmsg);
+ ENTRY;
+
+ CDEBUG(D_HA, "import %s of %s@%s abruptly disconnected: reconnecting\n",
+ imp->imp_obd->obd_name, obd2cli_tgt(imp->imp_obd),
+ imp->imp_connection->c_remote_uuid.uuid);
+
+ if (ptlrpc_set_import_discon(imp, conn, true)) {
+ /* to control recovery via lctl {disable|enable}_recovery */
+ if (imp->imp_deactive == 0)
+ ptlrpc_connect_import(imp);
+ }
+
+ /* Wait for recovery to complete and resend. If evicted, then
+ this request will be errored out later.*/
+ spin_lock(&failed_req->rq_lock);
+ if (!failed_req->rq_no_resend)
+ failed_req->rq_resend = 1;
+ spin_unlock(&failed_req->rq_lock);
+
+ EXIT;
}
+/**
+ * Administratively active/deactive a client.
+ * This should only be called by the ioctl interface, currently
+ * - the lctl deactivate and activate commands
+ * - echo 0/1 >> /proc/osc/XXX/active
+ * - client umount -f (ll_umount_begin)
+ */
int ptlrpc_set_import_active(struct obd_import *imp, int active)
{
struct obd_device *obd = imp->imp_obd;
- unsigned long flags;
+ int rc = 0;
+ ENTRY;
LASSERT(obd);
/* When deactivating, mark import invalid, and abort in-flight
* requests. */
if (!active) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- /* This is a bit of a hack, but invalidating replayable
- * imports makes a temporary reconnect failure into a much more
- * ugly -- and hard to remedy -- situation. */
- if (!imp->imp_replayable) {
- CDEBUG(D_HA, "setting import %s INVALID\n",
- imp->imp_target_uuid.uuid);
- imp->imp_invalid = 1;
- }
- imp->imp_generation++;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_invalidate_import_state(imp);
- }
+ LCONSOLE_WARN("setting import %s INACTIVE by administrator "
+ "request\n", obd2cli_tgt(imp->imp_obd));
- /* When activating, mark import valid */
- if (active) {
- CDEBUG(D_HA, "setting import %s VALID\n",
- imp->imp_target_uuid.uuid);
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_invalid = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- }
-
- if (obd->obd_observer)
- RETURN(obd_notify(obd->obd_observer, obd, active));
-
- RETURN(0);
-}
+ /* set before invalidate to avoid messages about imp_inval
+ * set without imp_deactive in ptlrpc_import_delay_req */
+ spin_lock(&imp->imp_lock);
+ imp->imp_deactive = 1;
+ spin_unlock(&imp->imp_lock);
-void ptlrpc_fail_import(struct obd_import *imp, int generation)
-{
- unsigned long flags;
- int in_recovery = 0;
- ENTRY;
+ obd_import_event(imp->imp_obd, imp, IMP_EVENT_DEACTIVATE);
- LASSERT (!imp->imp_dlm_fake);
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- if (imp->imp_state != LUSTRE_IMP_FULL) {
- in_recovery = 1;
- } else {
- CDEBUG(D_HA, "%s: new state: DISCON\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_DISCON;
- }
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (in_recovery) {
- EXIT;
- return;
+ ptlrpc_invalidate_import(imp);
}
- ptlrpc_handle_failed_import(imp);
- EXIT;
-}
-
-static int signal_completed_replay(struct obd_import *imp)
-{
- struct ptlrpc_request *req;
- int rc;
- ENTRY;
-
- req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL, NULL);
- if (!req)
- RETURN(-ENOMEM);
-
- req->rq_replen = lustre_msg_size(0, NULL);
- req->rq_send_state = LUSTRE_IMP_REPLAY;
- req->rq_reqmsg->flags |= MSG_LAST_REPLAY;
- req->rq_timeout *= 3;
-
- rc = ptlrpc_queue_wait(req);
+ /* When activating, mark import valid, and attempt recovery */
+ if (active) {
+ CDEBUG(D_HA, "setting import %s VALID\n",
+ obd2cli_tgt(imp->imp_obd));
- ptlrpc_req_finished(req);
- RETURN(rc);
-}
+ spin_lock(&imp->imp_lock);
+ imp->imp_deactive = 0;
+ spin_unlock(&imp->imp_lock);
+ obd_import_event(imp->imp_obd, imp, IMP_EVENT_ACTIVATE);
-int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
-{
- int rc;
- ENTRY;
-
- rc = ptlrpc_recover_import_no_retry(imp, new_uuid);
-
- if (rc && rc != -EALREADY) {
- unsigned long flags;
- CDEBUG(D_HA, "recovery of %s on %s failed (%d); restarting\n",
- imp->imp_target_uuid.uuid,
- new_uuid ? new_uuid :
- (char *)imp->imp_connection->c_remote_uuid.uuid, rc);
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_state = LUSTRE_IMP_FULL;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_fail_import(imp, imp->imp_generation);
+ rc = ptlrpc_recover_import(imp, NULL, 0);
}
+
RETURN(rc);
}
+EXPORT_SYMBOL(ptlrpc_set_import_active);
-static int ptlrpc_recover_import_no_retry(struct obd_import *imp,
- char *new_uuid)
+/* Attempt to reconnect an import */
+int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid, int async)
{
- int rc;
- unsigned long flags;
- int in_recovery = 0;
- int was_invalid = 0;
- ENTRY;
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- if (imp->imp_state == LUSTRE_IMP_FULL) {
- CDEBUG(D_HA, "%s: new state: DISCON\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_DISCON;
- }
-
- if (imp->imp_state != LUSTRE_IMP_DISCON) {
- in_recovery = 1;
- } else if (imp->imp_invalid) {
- imp->imp_invalid = 0;
- was_invalid = 1;
- }
-
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (in_recovery == 1)
- RETURN(-EALREADY);
-
- down(&imp->imp_recovery_sem);
- /* If recovery happened while we waited, we're done. */
- if (imp->imp_state == LUSTRE_IMP_FULL)
- GOTO(out, rc = 0);
-
- LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
-
- if (new_uuid) {
- struct ptlrpc_connection *conn;
- struct obd_uuid uuid;
- struct ptlrpc_peer peer;
- struct obd_export *dlmexp;
-
- obd_str2uuid(&uuid, new_uuid);
- if (ptlrpc_uuid_to_peer(&uuid, &peer)) {
- CERROR("no connection found for UUID %s\n", new_uuid);
- GOTO(out, rc = -EINVAL);
- }
-
- conn = ptlrpc_get_connection(&peer, &uuid);
- if (!conn)
- GOTO(out, rc = -ENOMEM);
-
- CDEBUG(D_HA, "switching import %s/%s from %s to %s\n",
- imp->imp_target_uuid.uuid, imp->imp_obd->obd_name,
- imp->imp_connection->c_remote_uuid.uuid,
- conn->c_remote_uuid.uuid);
-
- /* Switch the import's connection and the DLM export's
- * connection (which are almost certainly the same, but we
- * keep distinct refs just to make things clearer. I think. */
- if (imp->imp_connection)
- ptlrpc_put_connection(imp->imp_connection);
- /* We hand off the ref from ptlrpc_get_connection. */
- imp->imp_connection = conn;
-
- dlmexp = class_conn2export(&imp->imp_dlm_handle);
- if (dlmexp->exp_connection)
- ptlrpc_put_connection(dlmexp->exp_connection);
- dlmexp->exp_connection = ptlrpc_connection_addref(conn);
- class_export_put(dlmexp);
-
- }
-
- connect:
- rc = ptlrpc_connect_import(imp);
-
- if (rc < 0) {
- CERROR("failed to reconnect to %s@%s: %d\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid, rc);
- GOTO(out, rc);
- }
-
- if (imp->imp_state == LUSTRE_IMP_EVICTED) {
- CDEBUG(D_HA, "evicted from %s@%s; invalidating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- ptlrpc_set_import_active(imp, 0);
- CDEBUG(D_HA, "%s: new state: RECOVER\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_RECOVER;
- }
-
- if (imp->imp_state == LUSTRE_IMP_REPLAY) {
- CDEBUG(D_HA, "replay requested by %s\n",
- imp->imp_target_uuid.uuid);
- rc = ptlrpc_replay(imp);
- if (rc)
- GOTO(out, rc);
-
- rc = ldlm_replay_locks(imp);
- if (rc)
- GOTO(out, rc);
-
- rc = signal_completed_replay(imp);
- if (rc)
- GOTO(out, rc);
- CDEBUG(D_HA, "%s: new state: RECOVER\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_RECOVER;
- }
-
- if (imp->imp_state == LUSTRE_IMP_RECOVER) {
- CDEBUG(D_HA, "reconnected to %s@%s\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
-
- ptlrpc_set_import_active(imp, 1);
- ptlrpc_resend(imp);
- spin_lock_irqsave(&imp->imp_lock, flags);
- CDEBUG(D_HA, "%s: new state: FULL\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_FULL;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_wake_delayed(imp);
- }
-
-
- LASSERT(imp->imp_state == LUSTRE_IMP_FULL);
-
- out:
- if (rc != 0) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_state = LUSTRE_IMP_DISCON;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (rc == -ENOTCONN) {
- CDEBUG(D_HA, "evicted/aborted by %s@%s during recovery;"
- "invalidating and reconnecting\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- GOTO(connect, -ENOTCONN);
- } else if (was_invalid) {
- ptlrpc_set_import_active(imp, 0);
- }
- }
- up(&imp->imp_recovery_sem);
- RETURN(rc);
+ int rc = 0;
+ ENTRY;
+
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state == LUSTRE_IMP_NEW || imp->imp_deactive ||
+ atomic_read(&imp->imp_inval_count))
+ rc = -EINVAL;
+ spin_unlock(&imp->imp_lock);
+ if (rc)
+ GOTO(out, rc);
+
+ /* force import to be disconnected. */
+ ptlrpc_set_import_discon(imp, 0, false);
+
+ if (new_uuid) {
+ struct obd_uuid uuid;
+
+ /* intruct import to use new uuid */
+ obd_str2uuid(&uuid, new_uuid);
+ rc = import_set_conn_priority(imp, &uuid);
+ if (rc)
+ GOTO(out, rc);
+ }
+
+ /* Check if reconnect is already in progress */
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state != LUSTRE_IMP_DISCON) {
+ imp->imp_force_verify = 1;
+ rc = -EALREADY;
+ }
+ spin_unlock(&imp->imp_lock);
+ if (rc)
+ GOTO(out, rc);
+
+ OBD_RACE(OBD_FAIL_PTLRPC_CONNECT_RACE);
+
+ rc = ptlrpc_connect_import(imp);
+ if (rc)
+ GOTO(out, rc);
+
+ if (!async) {
+ long timeout = cfs_time_seconds(obd_timeout);
+
+ CDEBUG(D_HA, "%s: recovery started, waiting %u jiffies\n",
+ obd2cli_tgt(imp->imp_obd), obd_timeout);
+
+ rc = wait_event_idle_timeout(imp->imp_recovery_waitq,
+ !ptlrpc_import_in_recovery(imp),
+ timeout);
+ if (rc == 0)
+ rc = -ETIMEDOUT;
+ else
+ rc = 0;
+ CDEBUG(D_HA, "%s: recovery finished\n",
+ obd2cli_tgt(imp->imp_obd));
+ }
+ EXIT;
+
+out:
+ return rc;
}
+EXPORT_SYMBOL(ptlrpc_recover_import);
-void ptlrpc_fail_export(struct obd_export *exp)
+int ptlrpc_import_in_recovery(struct obd_import *imp)
{
- int rc, already_failed;
- unsigned long flags;
-
- spin_lock_irqsave(&exp->exp_lock, flags);
- already_failed = exp->exp_failed;
- exp->exp_failed = 1;
- spin_unlock_irqrestore(&exp->exp_lock, flags);
-
- if (already_failed) {
- CDEBUG(D_HA, "disconnecting dead export %p/%s; skipping\n",
- exp, exp->exp_client_uuid.uuid);
- return;
- }
+ int in_recovery = 1;
- CDEBUG(D_HA, "disconnecting export %p/%s\n",
- exp, exp->exp_client_uuid.uuid);
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state <= LUSTRE_IMP_DISCON ||
+ imp->imp_state >= LUSTRE_IMP_FULL ||
+ imp->imp_obd->obd_no_recov)
+ in_recovery = 0;
+ spin_unlock(&imp->imp_lock);
- /* Most callers into obd_disconnect are removing their own reference
- * (request, for example) in addition to the one from the hash table.
- * We don't have such a reference here, so make one. */
- class_export_get(exp);
- rc = obd_disconnect(exp, 0);
- if (rc)
- CERROR("disconnecting export %p failed: %d\n", exp, rc);
+ return in_recovery;
}