/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Lustre Light Super operations
+ * GPL HEADER START
*
- * This code is issued under the GNU General Public License.
- * See the file COPYING in this distribution
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
- * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
- * Copryright (C) 1999 Seagate Technology Inc.
- * Copryright (C) 2001 Mountain View Data, Inc.
- * Copryright (C) 2002 Cluster File Systems, Inc.
+ * 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 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).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/ptlrpc/recover.c
+ *
+ * Author: Mike Shaver <shaver@clusterfs.com>
*/
-#include <linux/config.h>
-#include <linux/module.h>
-#include <linux/kmod.h>
+#define DEBUG_SUBSYSTEM S_RPC
+#ifdef __KERNEL__
+# include <libcfs/libcfs.h>
+#else
+# include <liblustre.h>
+#endif
+
+#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_ost.h>
+#include <obd_class.h>
+#include <obd_lov.h> /* for IOC_LOV_SET_OSC_ACTIVE */
+#include <libcfs/list.h>
-#define DEBUG_SUBSYSTEM S_LLITE
+#include "ptlrpc_internal.h"
-#include <linux/lustre_lite.h>
-#include <linux/lustre_ha.h>
+static int ptlrpc_recover_import_no_retry(struct obd_import *, char *);
-int ll_reconnect(struct ptlrpc_connection *conn)
+void ptlrpc_initiate_recovery(struct obd_import *imp)
{
- struct ptlrpc_request *request;
- struct list_head *tmp;
- int rc = -EINVAL;
-
- /* XXX c_lock semantics! */
- conn->c_level = LUSTRE_CONN_CON;
-
- /* XXX this code MUST be shared with class_obd_connect! */
- list_for_each(tmp, &conn->c_imports) {
- struct obd_import *imp = list_entry(tmp, struct obd_import,
- imp_chain);
- struct obd_device *obd = imp->imp_obd;
- struct client_obd *cli = &obd->u.cli;
- int rq_opc = (obd->obd_type->typ_ops->o_brw)
- ? OST_CONNECT : MDS_CONNECT;
- int size[] = { sizeof(cli->cl_target_uuid),
- sizeof(obd->obd_uuid) };
- char *tmp[] = {cli->cl_target_uuid, obd->obd_uuid };
- struct lustre_handle old_hdl;
-
- LASSERT(imp->imp_connection == conn);
- request = ptlrpc_prep_req(imp, rq_opc, 2, size, tmp);
- request->rq_level = LUSTRE_CONN_NEW;
- request->rq_replen = lustre_msg_size(0, NULL);
- /* XXX are (addr, cookie) right? */
- request->rq_reqmsg->addr = imp->imp_handle.addr;
- request->rq_reqmsg->cookie = imp->imp_handle.cookie;
- rc = ptlrpc_queue_wait(request);
- rc = ptlrpc_check_status(request, rc);
- if (rc) {
- CERROR("cannot connect to %s@%s: rc = %d\n",
- cli->cl_target_uuid, conn->c_remote_uuid, rc);
- ptlrpc_free_req(request);
- GOTO(out_disc, rc = -ENOTCONN);
+ ENTRY;
+
+ CDEBUG(D_HA, "%s: starting recovery\n", obd2cli_tgt(imp->imp_obd));
+ ptlrpc_connect_import(imp, NULL);
+
+ EXIT;
+}
+
+int ptlrpc_replay_next(struct obd_import *imp, int *inflight)
+{
+ int rc = 0;
+ struct list_head *tmp, *pos;
+ 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;
+ spin_unlock(&imp->imp_lock);
+
+ CDEBUG(D_HA, "import %p from %s committed "LPU64" last "LPU64"\n",
+ imp, obd2cli_tgt(imp->imp_obd),
+ imp->imp_peer_committed_transno, last_transno);
+
+ /* 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_replay_list);
+
+ /* 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 (imp->imp_resend_replay &&
+ req->rq_transno == last_transno) {
+ lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
+ break;
+ }
+
+ if (req->rq_transno > last_transno) {
+ imp->imp_last_replay_transno = req->rq_transno;
+ break;
}
- old_hdl = imp->imp_handle;
- imp->imp_handle.addr = request->rq_repmsg->addr;
- imp->imp_handle.cookie = request->rq_repmsg->cookie;
- CERROR("reconnected 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);
- ptlrpc_free_req(request);
+ req = NULL;
}
- conn->c_level = LUSTRE_CONN_RECOVD;
- out_disc:
- return rc;
+ spin_lock(&imp->imp_lock);
+ imp->imp_resend_replay = 0;
+ spin_unlock(&imp->imp_lock);
+
+ if (req != NULL) {
+ rc = ptlrpc_replay_req(req);
+ if (rc) {
+ CERROR("recovery replay error %d for req "
+ LPD64"\n", rc, req->rq_xid);
+ RETURN(rc);
+ }
+ *inflight = 1;
+ }
+ RETURN(rc);
}
-static int ll_recover_upcall(struct ptlrpc_connection *conn)
+int ptlrpc_resend(struct obd_import *imp)
{
- char *argv[3];
- char *envp[3];
- int rc;
+ struct ptlrpc_request *req, *next;
ENTRY;
- conn->c_level = LUSTRE_CONN_RECOVD;
- argv[0] = obd_recovery_upcall;
- argv[1] = conn->c_remote_uuid;
- argv[2] = NULL;
-
- envp[0] = "HOME=/";
- envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
- envp[2] = NULL;
+ /* 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(&imp->imp_lock);
+ if (imp->imp_state != LUSTRE_IMP_RECOVER) {
+ spin_unlock(&imp->imp_lock);
+ RETURN(-1);
+ }
- rc = call_usermodehelper(argv[0], argv, envp);
- if (rc < 0) {
- CERROR("Error invoking recovery upcall (%s): %d\n",
- obd_recovery_upcall, rc);
- CERROR("Check /proc/sys/lustre/recovery_upcall?\n");
+ list_for_each_entry_safe(req, next, &imp->imp_sending_list, rq_list) {
+ LASSERTF((long)req > CFS_PAGE_SIZE && req != LP_POISON,
+ "req %p bad\n", req);
+ LASSERTF(req->rq_type != LI_POISON, "req %p freed\n", req);
+ if (!req->rq_no_resend)
+ ptlrpc_resend_req(req);
}
- RETURN(rc);
+ spin_unlock(&imp->imp_lock);
+
+ RETURN(0);
}
-static int ll_recover_reconnect(struct ptlrpc_connection *conn)
+void ptlrpc_wake_delayed(struct obd_import *imp)
{
- int rc = 0;
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 (set %p):", req->rq_set);
+ ptlrpc_wake_client_req(req);
+ }
+ spin_unlock(&imp->imp_lock);
+}
+
+void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req)
+{
+ struct obd_import *imp = failed_req->rq_import;
ENTRY;
- /* 1. reconnect */
- rc = ll_reconnect(conn);
- if (rc)
- RETURN(rc);
-
- /* 2. walk the request list */
- spin_lock(&conn->c_lock);
+ 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);
- list_for_each_safe(tmp, pos, &conn->c_sending_head) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
-
- /* replay what needs to be replayed */
- if (req->rq_flags & PTL_RPC_FL_REPLAY) {
- CDEBUG(D_NET, "req %Ld needs replay [last rcvd %Ld]\n",
- req->rq_xid, conn->c_last_xid);
- rc = ptlrpc_replay_req(req);
-#if 0
-#error We should not hold a spinlock over such a lengthy operation.
-#error If necessary, drop spinlock, do operation, re-get spinlock, restart loop.
-#error If we need to avoid re-processint items, then delete them from the list
-#error as they are replayed and re-add at the tail of this list, so the next
-#error item to process will always be at the head of the list.
-#endif
- if (rc) {
- CERROR("recovery replay error %d for req %Ld\n",
- rc, req->rq_xid);
- GOTO(out, rc);
- }
+ if (ptlrpc_set_import_discon(imp,
+ lustre_msg_get_conn_cnt(failed_req->rq_reqmsg))) {
+ if (!imp->imp_replayable) {
+ CDEBUG(D_HA, "import %s@%s for %s not replayable, "
+ "auto-deactivating\n",
+ obd2cli_tgt(imp->imp_obd),
+ imp->imp_connection->c_remote_uuid.uuid,
+ imp->imp_obd->obd_name);
+ ptlrpc_deactivate_import(imp);
}
+ /* to control recovery via lctl {disable|enable}_recovery */
+ if (imp->imp_deactive == 0)
+ ptlrpc_connect_import(imp, NULL);
+ }
- /* server has seen req, we have reply: skip */
- if ((req->rq_flags & PTL_RPC_FL_REPLIED) &&
- req->rq_xid <= conn->c_last_xid) {
- CDEBUG(D_NET,
- "req %Ld was complete: skip [last rcvd %Ld]\n",
- req->rq_xid, conn->c_last_xid);
- continue;
- }
+ /* 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);
- /* server has lost req, we have reply: resend, ign reply */
- if ((req->rq_flags & PTL_RPC_FL_REPLIED) &&
- req->rq_xid > conn->c_last_xid) {
- CDEBUG(D_NET, "lost req %Ld have rep: replay [last "
- "rcvd %Ld]\n", req->rq_xid, conn->c_last_xid);
- rc = ptlrpc_replay_req(req);
- if (rc) {
- CERROR("request resend error %d for req %Ld\n",
- rc, req->rq_xid);
- GOTO(out, rc);
- }
- }
+ EXIT;
+}
- /* server has seen req, we have lost reply: -ERESTARTSYS */
- if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) &&
- req->rq_xid <= conn->c_last_xid) {
- CDEBUG(D_NET, "lost rep %Ld srv did req: restart "
- "[last rcvd %Ld]\n",
- req->rq_xid, conn->c_last_xid);
- ptlrpc_restart_req(req);
- }
+/*
+ * 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;
+ int rc = 0;
- /* service has not seen req, no reply: resend */
- if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) &&
- req->rq_xid > conn->c_last_xid) {
- CDEBUG(D_NET,
- "lost rep/req %Ld: resend [last rcvd %Ld]\n",
- req->rq_xid, conn->c_last_xid);
- ptlrpc_resend_req(req);
- }
+ ENTRY;
+ LASSERT(obd);
- }
+ /* When deactivating, mark import invalid, and abort in-flight
+ * requests. */
+ if (!active) {
+ LCONSOLE_WARN("setting import %s INACTIVE by administrator "
+ "request\n", obd2cli_tgt(imp->imp_obd));
+ ptlrpc_invalidate_import(imp);
- conn->c_level = LUSTRE_CONN_FULL;
- recovd_conn_fixed(conn);
+ spin_lock(&imp->imp_lock);
+ imp->imp_deactive = 1;
+ spin_unlock(&imp->imp_lock);
+ }
- CDEBUG(D_NET, "recovery complete on conn %p(%s), waking delayed reqs\n",
- conn, conn->c_remote_uuid);
- /* Finally, continue what we delayed since recovery started */
- list_for_each_safe(tmp, pos, &conn->c_delayed_head) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- ptlrpc_continue_req(req);
+ /* When activating, mark import valid, and attempt recovery */
+ if (active) {
+ spin_lock(&imp->imp_lock);
+ imp->imp_deactive = 0;
+ spin_unlock(&imp->imp_lock);
+
+ CDEBUG(D_HA, "setting import %s VALID\n",
+ obd2cli_tgt(imp->imp_obd));
+ rc = ptlrpc_recover_import(imp, NULL);
}
- EXIT;
- out:
- spin_unlock(&conn->c_lock);
- return rc;
+ RETURN(rc);
}
-static int ll_retry_recovery(struct ptlrpc_connection *conn)
+/* Attempt to reconnect an import */
+int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
{
-#if 0
- /* XXX use a timer, sideshow bob */
- recovd_conn_fail(conn);
- /* XXX this is disabled until I fix it so that we don't just keep
- * XXX retrying in the case of a missing upcall.
- */
-#endif
- return 0;
+ int rc;
+ ENTRY;
+
+ /* force import to be disconnected. */
+ ptlrpc_set_import_discon(imp, 0);
+
+ spin_lock(&imp->imp_lock);
+ imp->imp_deactive = 0;
+ spin_unlock(&imp->imp_lock);
+
+ rc = ptlrpc_recover_import_no_retry(imp, new_uuid);
+
+ RETURN(rc);
}
-int ll_recover(struct recovd_data *rd, int phase)
+int ptlrpc_import_in_recovery(struct obd_import *imp)
{
- struct ptlrpc_connection *conn = class_rd2conn(rd);
+ int in_recovery = 1;
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state == LUSTRE_IMP_FULL ||
+ imp->imp_state == LUSTRE_IMP_CLOSED ||
+ imp->imp_state == LUSTRE_IMP_DISCON)
+ in_recovery = 0;
+ spin_unlock(&imp->imp_lock);
+ return in_recovery;
+}
- LASSERT(conn);
+static int ptlrpc_recover_import_no_retry(struct obd_import *imp,
+ char *new_uuid)
+{
+ int rc;
+ int in_recovery = 0;
+ struct l_wait_info lwi;
ENTRY;
- switch (phase) {
- case PTLRPC_RECOVD_PHASE_PREPARE:
- RETURN(ll_recover_upcall(conn));
- case PTLRPC_RECOVD_PHASE_RECOVER:
- RETURN(ll_recover_reconnect(conn));
- case PTLRPC_RECOVD_PHASE_FAILURE:
- RETURN(ll_retry_recovery(conn));
+ /* Check if reconnect is already in progress */
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state != LUSTRE_IMP_DISCON) {
+ in_recovery = 1;
}
+ spin_unlock(&imp->imp_lock);
+
+ if (in_recovery == 1)
+ RETURN(-EALREADY);
- LBUG();
- RETURN(-ENOSYS);
+ rc = ptlrpc_connect_import(imp, new_uuid);
+ if (rc)
+ RETURN(rc);
+
+ CDEBUG(D_HA, "%s: recovery started, waiting\n",
+ obd2cli_tgt(imp->imp_obd));
+
+ lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(obd_timeout)),
+ NULL, NULL);
+ rc = l_wait_event(imp->imp_recovery_waitq,
+ !ptlrpc_import_in_recovery(imp), &lwi);
+ CDEBUG(D_HA, "%s: recovery finished\n",
+ obd2cli_tgt(imp->imp_obd));
+
+ RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff