*
* Portal-RPC reconnection and replay operations, for use in recovery.
*
- * This code is issued under the GNU General Public License.
- * See the file COPYING in this distribution
+ * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
+ * Author: Mike Shaver <shaver@clusterfs.com>
*
- * Copyright (C) 1996 Peter J. Braam <braam@stelias.com>
- * Copyright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
- * Copyright (C) 1999 Seagate Technology Inc.
- * Copyright (C) 2001 Mountain View Data, Inc.
- * Copyright (C) 2002 Cluster File Systems, Inc.
+ * This file is part of Lustre, http://www.lustre.org.
*
+ * 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.
+ *
+ * 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.
+ *
+ * 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.
*/
+#define DEBUG_SUBSYSTEM S_RPC
+#ifdef __KERNEL__
#include <linux/config.h>
#include <linux/module.h>
#include <linux/kmod.h>
+#else
+#include <liblustre.h>
+#endif
-#define DEBUG_SUBSYSTEM S_RPC
-
+#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_class.h>
+#include <linux/obd_lov.h> /* for IOC_LOV_SET_OSC_ACTIVE */
-int ptlrpc_reconnect_import(struct obd_import *imp, int rq_opc)
+#include "ptlrpc_internal.h"
+
+enum reconnect_result {
+ RECON_RESULT_RECOVERING = 1,
+ RECON_RESULT_RECONNECTED = 2,
+ RECON_RESULT_EVICTED = 3,
+};
+
+int ptlrpc_reconnect_import(struct obd_import *imp)
{
struct obd_device *obd = imp->imp_obd;
- struct client_obd *cli = &obd->u.cli;
- int size[] = { sizeof(cli->cl_target_uuid), sizeof(obd->obd_uuid) };
- char *tmp[] = {cli->cl_target_uuid, obd->obd_uuid };
+ int rc, size[] = {sizeof(imp->imp_target_uuid),
+ sizeof(obd->obd_uuid),
+ sizeof(imp->imp_dlm_handle)};
+ char *tmp[] = {imp->imp_target_uuid.uuid,
+ obd->obd_uuid.uuid,
+ (char *)&imp->imp_dlm_handle};
struct ptlrpc_connection *conn = imp->imp_connection;
+ struct ptlrpc_request *req;
struct lustre_handle old_hdl;
- struct ptlrpc_request *request;
- struct obd_export *ldlmexp;
- int rc;
+ __u64 committed_before_reconnect = imp->imp_peer_committed_transno;
+
+ CERROR("reconnect handle "LPX64"\n",
+ imp->imp_dlm_handle.cookie);
+
+ req = ptlrpc_prep_req(imp, imp->imp_connect_op, 3, size, tmp);
+ if (!req)
+ RETURN(-ENOMEM);
+ req->rq_level = LUSTRE_CONN_NEW;
+ req->rq_replen = lustre_msg_size(0, NULL);
+ rc = ptlrpc_queue_wait(req);
+ if (rc) {
+ /* what if rc > 0 ??*/
+ CERROR("cannot connect to %s@%s: rc = %d\n",
+ imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid, rc);
+ GOTO(out_disc, rc);
+ }
- request = ptlrpc_prep_req(imp, rq_opc, 2, size, tmp);
- request->rq_level = LUSTRE_CONN_NEW;
- request->rq_replen = lustre_msg_size(0, NULL);
- /*
- * This address is the export that represents our client-side LDLM
- * service (for ASTs). We should only have one on this list, so we
- * just grab the first one.
- *
- * XXX tear down export, call class_obd_connect?
- */
- ldlmexp = list_entry(obd->obd_exports.next, struct obd_export,
- exp_obd_chain);
- request->rq_reqmsg->addr = (__u64)(unsigned long)ldlmexp;
- request->rq_reqmsg->cookie = ldlmexp->exp_cookie;
- rc = ptlrpc_queue_wait(request);
- switch (rc) {
- case EALREADY:
- case -EALREADY:
- /* already connected! */
+ if (lustre_msg_get_op_flags(req->rq_repmsg) & MSG_CONNECT_RECONNECT) {
memset(&old_hdl, 0, sizeof(old_hdl));
- if (!memcmp(&old_hdl.addr, &request->rq_repmsg->addr,
- sizeof (old_hdl.addr)) &&
- !memcmp(&old_hdl.cookie, &request->rq_repmsg->cookie,
- sizeof (old_hdl.cookie))) {
- CERROR("%s@%s didn't like our handle %Lx/%Lx, failed\n",
- cli->cl_target_uuid, conn->c_remote_uuid,
- (__u64)(unsigned long)ldlmexp,
- ldlmexp->exp_cookie);
+ if (!memcmp(&old_hdl, &req->rq_repmsg->handle,
+ sizeof (old_hdl))) {
+ CERROR("%s@%s didn't like our handle "LPX64
+ ", failed\n", imp->imp_target_uuid.uuid,
+ conn->c_remote_uuid.uuid,
+ imp->imp_dlm_handle.cookie);
GOTO(out_disc, rc = -ENOTCONN);
}
- old_hdl.addr = request->rq_repmsg->addr;
- old_hdl.cookie = request->rq_repmsg->cookie;
- if (memcmp(&imp->imp_handle, &old_hdl, sizeof(old_hdl))) {
- CERROR("%s@%s changed handle from %Lx/%Lx to %Lx/%Lx; "
- "copying, but this may foreshadow disaster\n",
- cli->cl_target_uuid, conn->c_remote_uuid,
- old_hdl.addr, old_hdl.cookie,
- imp->imp_handle.addr, imp->imp_handle.cookie);
- imp->imp_handle.addr = request->rq_repmsg->addr;
- imp->imp_handle.cookie = request->rq_repmsg->cookie;
- GOTO(out_disc, rc = EALREADY);
+ if (memcmp(&imp->imp_remote_handle, &req->rq_repmsg->handle,
+ sizeof(imp->imp_remote_handle))) {
+ CERROR("%s@%s changed handle from "LPX64" to "LPX64
+ "; copying, but this may foreshadow disaster\n",
+ imp->imp_target_uuid.uuid,
+ conn->c_remote_uuid.uuid,
+ imp->imp_remote_handle.cookie,
+ req->rq_repmsg->handle.cookie);
+ imp->imp_remote_handle = req->rq_repmsg->handle;
+ GOTO(out_disc, rc = RECON_RESULT_RECONNECTED);
}
-
+
CERROR("reconnected to %s@%s after partition\n",
- cli->cl_target_uuid, conn->c_remote_uuid);
- GOTO(out_disc, rc = EALREADY);
- case 0:
- old_hdl = imp->imp_handle;
- imp->imp_handle.addr = request->rq_repmsg->addr;
- imp->imp_handle.cookie = request->rq_repmsg->cookie;
- CERROR("now connected to %s@%s (%Lx/%Lx, was %Lx/%Lx)!\n",
- cli->cl_target_uuid, conn->c_remote_uuid,
- imp->imp_handle.addr, imp->imp_handle.cookie,
- old_hdl.addr, old_hdl.cookie);
- GOTO(out_disc, rc = 0);
- default:
- CERROR("cannot connect to %s@%s: rc = %d\n",
- cli->cl_target_uuid, conn->c_remote_uuid, rc);
- GOTO(out_disc, rc = -ENOTCONN); /* XXX preserve rc? */
+ imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid);
+ GOTO(out_disc, rc = RECON_RESULT_RECONNECTED);
+ } else if (lustre_msg_get_op_flags(req->rq_repmsg) & MSG_CONNECT_RECOVERING) {
+ rc = RECON_RESULT_RECOVERING;
+ } else {
+ rc = RECON_RESULT_EVICTED;
}
+
+ old_hdl = imp->imp_remote_handle;
+ imp->imp_remote_handle = req->rq_repmsg->handle;
+ CERROR("reconnected to %s@%s ("LPX64", was "LPX64")!\n",
+ imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid,
+ imp->imp_remote_handle.cookie, old_hdl.cookie);
+ if (req->rq_repmsg->last_committed < committed_before_reconnect) {
+ CERROR("%s went back in time (transno "LPD64
+ " was committed, server claims "LPD64
+ ")! is shared storage not coherent?\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_peer_committed_transno,
+ req->rq_repmsg->last_committed);
+ }
+
+ GOTO(out_disc, rc);
out_disc:
- ptlrpc_req_finished(request);
+ ptlrpc_req_finished(req);
return rc;
}
-int ptlrpc_run_recovery_upcall(struct ptlrpc_connection *conn)
+void ptlrpc_run_recovery_over_upcall(struct obd_device *obd)
{
- char *argv[3];
+ char *argv[4];
char *envp[3];
int rc;
ENTRY;
- argv[0] = obd_recovery_upcall;
- argv[1] = conn->c_remote_uuid;
- argv[2] = NULL;
+ 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 = call_usermodehelper(argv[0], argv, envp);
+ rc = USERMODEHELPER(argv[0], argv, envp);
if (rc < 0) {
- CERROR("Error invoking recovery upcall %s for %s: %d\n",
- argv[0], argv[1], rc);
- CERROR("Check /proc/sys/lustre/recovery_upcall?\n");
+ 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 for connection %s\n",
- argv[0], argv[1]);
+ CERROR("Invoked upcall %s %s %s",
+ argv[0], argv[1], argv[2]);
}
+}
- /*
- * We don't want to make this a "failed" recovery, because the system
- * administrator -- or, perhaps, tester -- may well be able to rescue
- * things by running the correct upcall.
- */
- RETURN(0);
+void ptlrpc_run_failed_import_upcall(struct obd_import* imp)
+{
+ char *argv[7];
+ char *envp[3];
+ int rc;
+
+ 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]);
+ }
}
-int ptlrpc_replay(struct obd_import *imp, int send_last_flag)
+int ptlrpc_replay(struct obd_import *imp)
{
int rc = 0;
struct list_head *tmp, *pos;
struct ptlrpc_request *req;
+ unsigned long flags;
__u64 committed = imp->imp_peer_committed_transno;
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_lock(&imp->imp_lock);
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
CDEBUG(D_HA, "import %p from %s has committed "LPD64"\n",
- imp, imp->imp_obd->u.cli.cl_target_uuid, committed);
+ imp, imp->imp_target_uuid.uuid, committed);
list_for_each(tmp, &imp->imp_replay_list) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
DEBUG_REQ(D_HA, req, "RETAINED: ");
}
- list_for_each_safe(tmp, pos, &imp->imp_replay_list) {
+ /* 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);
- if (req->rq_transno == imp->imp_max_transno &&
- send_last_flag) {
- req->rq_reqmsg->flags |= MSG_LAST_REPLAY;
- DEBUG_REQ(D_HA, req, "LAST_REPLAY:");
- } else {
- DEBUG_REQ(D_HA, req, "REPLAY:");
- }
+ DEBUG_REQ(D_HA, req, "REPLAY:");
rc = ptlrpc_replay_req(req);
- req->rq_reqmsg->flags &= ~MSG_LAST_REPLAY;
-
+
if (rc) {
- CERROR("recovery replay error %d for req %Ld\n",
+ CERROR("recovery replay error %d for req "LPD64"\n",
rc, req->rq_xid);
- GOTO(out, rc);
+ RETURN(rc);
}
}
- out:
- spin_unlock(&imp->imp_lock);
- return rc;
+ RETURN(0);
}
-#define NO_RESEND 0 /* No action required. */
-#define RESEND 1 /* Resend required. */
-#define RESEND_IGNORE 2 /* Resend, ignore the reply (already saw it). */
-#define RESTART 3 /* Have to restart the call, sorry! */
+int ptlrpc_resend(struct obd_import *imp)
+{
+ struct list_head *tmp, *pos;
+ struct ptlrpc_request *req;
+ unsigned long flags;
+
+ ENTRY;
+
+ /* As long as we're in recovery, nothing should be added to the sending
+ * list, so we don't need to hold the lock during this iteration and
+ * resend process.
+ */
+ /* Well... what if lctl recover is called twice at the same time?
+ */
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ LASSERT(imp->imp_level == LUSTRE_CONN_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);
+ }
-static int resend_type(struct ptlrpc_request *req, __u64 committed)
+ RETURN(0);
+}
+
+void ptlrpc_wake_delayed(struct obd_import *imp)
{
- if (req->rq_transno < committed) {
- if (req->rq_flags & PTL_RPC_FL_REPLIED) {
- /* Saw the reply and it was committed, no biggie. */
- DEBUG_REQ(D_HA, req, "NO_RESEND");
- return NO_RESEND;
+ 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_wait_for_rep);
}
- /* Request committed, but no reply: have to restart. */
- return RESTART;
}
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+}
+
+inline void ptlrpc_invalidate_import_state(struct obd_import *imp)
+{
+ struct ldlm_namespace *ns = imp->imp_obd->obd_namespace;
+ if (ptlrpc_ldlm_namespace_cleanup == NULL)
+ CERROR("ptlrpc/ldlm hook is NULL! Please tell phil\n");
+ else
+ ptlrpc_ldlm_namespace_cleanup(ns, 1 /* no network ops */);
+ ptlrpc_abort_inflight(imp);
+}
+
+
+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;
+}
- if (req->rq_flags & PTL_RPC_FL_REPLIED) {
- /* Saw reply, so resend and ignore new reply. */
- return RESEND_IGNORE;
+void ptlrpc_request_handle_eviction(struct ptlrpc_request *failed_req)
+{
+ int rc;
+ struct obd_import *imp= failed_req->rq_import;
+ unsigned long flags;
+ struct ptlrpc_request *req;
+ ENTRY;
+
+ CDEBUG(D_HA, "import %s of %s@%s evicted: reconnecting\n",
+ imp->imp_obd->obd_name,
+ imp->imp_target_uuid.uuid,
+ imp->imp_connection->c_remote_uuid.uuid);
+ rc = ptlrpc_recover_import(imp, NULL);
+ if (rc) {
+ ptlrpc_resend_req(failed_req);
+ if (rc != -EALREADY)
+ ptlrpc_handle_failed_import(imp);
+ } else {
+ LASSERT(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);
}
+ ptlrpc_req_finished(req);
+ EXIT;
+}
- /* Didn't see reply either, so resend. */
- return RESEND;
+int ptlrpc_set_import_active(struct obd_import *imp, int active)
+{
+ struct obd_device *notify_obd;
+ unsigned long flags;
+ int rc;
+ LASSERT(imp->imp_obd);
+
+ notify_obd = imp->imp_obd->u.cli.cl_containing_lov;
+
+ /* When deactivating, mark import invalid, and
+ abort in-flight requests. */
+ if (!active) {
+ CDEBUG(D_ERROR, "setting import %s INVALID\n", imp->imp_target_uuid.uuid);
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ imp->imp_invalid = 1;
+ imp->imp_generation++;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ ptlrpc_invalidate_import_state(imp);
+// ptlrpc_abort_inflight(imp);
+ }
+
+ if (notify_obd == NULL)
+ GOTO(out, rc = 0);
+
+ /* How gross is _this_? */
+ if (!list_empty(¬ify_obd->obd_exports)) {
+ struct lustre_handle fakeconn;
+ struct obd_ioctl_data ioc_data = { 0 };
+ struct obd_export *exp =
+ list_entry(notify_obd->obd_exports.next,
+ struct obd_export, exp_obd_chain);
+
+ fakeconn.cookie = exp->exp_handle.h_cookie;
+ ioc_data.ioc_inlbuf1 = (char *)&imp->imp_target_uuid;
+ ioc_data.ioc_offset = active;
+ rc = obd_iocontrol(IOC_LOV_SET_OSC_ACTIVE, &fakeconn,
+ sizeof ioc_data, &ioc_data, NULL);
+ if (rc)
+ CERROR("error %sabling %s on LOV %p/%s: %d\n",
+ active ? "en" : "dis",
+ imp->imp_target_uuid.uuid, notify_obd,
+ notify_obd->obd_uuid.uuid, rc);
+ } else {
+ CDEBUG(D_HA, "No exports for obd %p/%s, can't notify about "
+ "%p\n", notify_obd, notify_obd->obd_uuid.uuid,
+ imp->imp_obd->obd_uuid.uuid);
+ rc = -ENOENT;
+ }
+
+out:
+ /* When activating, mark import valid */
+ if (active) {
+ CDEBUG(D_ERROR, "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);
+ }
+
+ RETURN(rc);
}
-int ptlrpc_resend(struct obd_import *imp)
+void ptlrpc_fail_import(struct obd_import *imp, int generation)
+{
+ unsigned long flags;
+ int in_recovery = 0;
+ ENTRY;
+
+ LASSERT (!imp->imp_dlm_fake);
+
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (imp->imp_level != LUSTRE_CONN_FULL)
+ in_recovery = 1;
+ else
+ imp->imp_level = LUSTRE_CONN_NOTCONN;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+
+ if (in_recovery) {
+ EXIT;
+ return;
+ }
+
+ ptlrpc_handle_failed_import(imp);
+ EXIT;
+}
+
+static int signal_completed_replay(struct obd_import *imp)
{
- int rc = 0, type;
- struct list_head *tmp, *pos;
struct ptlrpc_request *req;
- __u64 committed = imp->imp_peer_committed_transno;
+ 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_level = LUSTRE_CONN_RECOVER;
+ req->rq_reqmsg->flags |= MSG_LAST_REPLAY;
+
+ rc = ptlrpc_queue_wait(req);
+
+ ptlrpc_req_finished(req);
+ RETURN(rc);
+}
+
+int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
+{
+ int rc;
+ unsigned long flags;
+ int in_recover = 0;
+ int recon_result;
ENTRY;
- spin_lock(&imp->imp_lock);
- list_for_each(tmp, &imp->imp_sending_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- DEBUG_REQ(D_HA, req, "SENDING: ");
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (imp->imp_level == LUSTRE_CONN_FULL ||
+ imp->imp_level == LUSTRE_CONN_NOTCONN)
+ imp->imp_level = LUSTRE_CONN_RECOVER;
+ else
+ in_recover = 1;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+
+ if (in_recover == 1)
+ RETURN(-EALREADY);
+
+ 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);
+ RETURN(-EINVAL);
+ }
+
+ conn = ptlrpc_get_connection(&peer, &uuid);
+ if (!conn)
+ RETURN(-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);
+
}
- list_for_each_safe(tmp, pos, &imp->imp_sending_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
-
- switch(resend_type(req, committed)) {
- case NO_RESEND:
- break;
-
- case RESTART:
- DEBUG_REQ(D_HA, req, "RESTART:");
- ptlrpc_restart_req(req);
- break;
-
- case RESEND_IGNORE:
- DEBUG_REQ(D_HA, req, "RESEND_IGNORE:");
- rc = ptlrpc_replay_req(req);
- if (rc) {
- DEBUG_REQ(D_ERROR, req, "error %d resending:",
- rc);
- ptlrpc_restart_req(req); /* might as well */
- }
- break;
-
- case RESEND:
- DEBUG_REQ(D_HA, req, "RESEND:");
- ptlrpc_resend_req(req);
- break;
-
- default:
- LBUG();
- }
+ recon_result = ptlrpc_reconnect_import(imp);
+
+ if (recon_result < 0) {
+ CERROR("failed to reconnect to %s@%s: %d\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_connection->c_remote_uuid.uuid, recon_result);
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ imp->imp_level = LUSTRE_CONN_NOTCONN;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ RETURN(recon_result);
+ }
+
+ if (recon_result == RECON_RESULT_RECOVERING) {
+ CDEBUG(D_HA, "replay requested by %s\n",
+ imp->imp_target_uuid.uuid);
+ rc = ptlrpc_replay(imp);
+ if (rc)
+ GOTO(out, rc);
+
+ if (ptlrpc_ldlm_replay_locks == NULL)
+ CERROR("ptlrpc/ldlm hook is NULL! Please tell phil\n");
+ else
+ rc = ptlrpc_ldlm_replay_locks(imp);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = signal_completed_replay(imp);
+ if (rc)
+ GOTO(out, rc);
+ } else if (recon_result == RECON_RESULT_RECONNECTED) {
+ CDEBUG(D_HA, "reconnected to %s@%s\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_connection->c_remote_uuid.uuid);
+ } else if (recon_result == RECON_RESULT_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);
+// ptlrpc_invalidate_import_state(imp);
+ } else {
+ LBUG();
}
+
+ ptlrpc_set_import_active(imp, 1);
+
+ rc = ptlrpc_resend(imp);
+
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ imp->imp_level = LUSTRE_CONN_FULL;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+
+ ptlrpc_wake_delayed(imp);
+ EXIT;
+ out:
+ return rc;
}
-void ptlrpc_wake_delayed(struct obd_import *imp)
+void ptlrpc_fail_export(struct obd_export *exp)
{
- struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
-
- spin_lock(&imp->imp_lock);
- list_for_each_safe(tmp, pos, &imp->imp_delayed_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- DEBUG_REQ(D_HA, req, "waking:");
- wake_up(&req->rq_wait_for_rep);
+ int rc, already_failed;
+ struct lustre_handle hdl;
+ 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;
}
- spin_unlock(&imp->imp_lock);
+
+ CDEBUG(D_HA, "disconnecting export %p/%s\n",
+ exp, exp->exp_client_uuid.uuid);
+ hdl.cookie = exp->exp_handle.h_cookie;
+ rc = obd_disconnect(&hdl, 0);
+ if (rc)
+ CERROR("disconnecting export %p failed: %d\n", exp, rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff