/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2001-2003 Cluster File Systems, Inc.
- * Author Peter Braam <braam@clusterfs.com>
+ * GPL HEADER START
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * You may have signed or agreed to another license before downloading
- * this software. If so, you are bound by the terms and conditions
- * of that agreement, and the following does not apply to you. See the
- * LICENSE file included with this distribution for more information.
+ * 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.
*
- * If you did not agree to a different license, then this copy of Lustre
- * is open source 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.
+ * 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).
*
- * In either case, 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
- * license text for more details.
+ * 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
*
- * For testing and management it is treated as an obd_device,
- * although * it does not export a full OBD method table (the
- * requests are coming * in over the wire, so object target modules
- * do not have a full * method table.)
+ * 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/osc/osc_create.c
+ * For testing and management it is treated as an obd_device,
+ * although * it does not export a full OBD method table (the
+ * requests are coming * in over the wire, so object target modules
+ * do not have a full * method table.)
+ *
+ * Author: Peter Braam <braam@clusterfs.com>
*/
#ifndef EXPORT_SYMTAB
#include <obd_class.h>
#include "osc_internal.h"
-static int osc_interpret_create(struct ptlrpc_request *req, void *data, int rc)
+static int osc_interpret_create(const struct lu_env *env,
+ struct ptlrpc_request *req, void *data, int rc)
{
struct osc_creator *oscc;
struct ost_body *body = NULL;
oscc = req->rq_async_args.pointer_arg[0];
LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
-
+
spin_lock(&oscc->oscc_lock);
oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
switch (rc) {
if (body) {
int diff = body->oa.o_id - oscc->oscc_last_id;
- if (diff < oscc->oscc_grow_count)
- oscc->oscc_grow_count =
- max(diff/3, OST_MIN_PRECREATE);
- else
+ /* oscc_internal_create() stores the original value of
+ * grow_count in rq_async_args.space[0].
+ * We can't compare against oscc_grow_count directly,
+ * because it may have been increased while the RPC
+ * is in flight, so we would always find ourselves
+ * having created fewer objects and decreasing the
+ * precreate request size. b=18577 */
+ if (diff < (int) req->rq_async_args.space[0]) {
+ /* the OST has not managed to create all the
+ * objects we asked for */
+ oscc->oscc_grow_count = max(diff,
+ OST_MIN_PRECREATE);
+ /* don't bump grow_count next time */
+ oscc->oscc_flags |= OSCC_FLAG_LOW;
+ } else {
+ /* the OST is able to keep up with the work,
+ * we could consider increasing grow_count
+ * next time if needed */
oscc->oscc_flags &= ~OSCC_FLAG_LOW;
+ }
oscc->oscc_last_id = body->oa.o_id;
}
spin_unlock(&oscc->oscc_lock);
break;
}
+ case -EAGAIN:
+ /* valid race delorphan vs create, or somthing after resend */
+ spin_unlock(&oscc->oscc_lock);
+ DEBUG_REQ(D_INODE, req, "Got EAGAIN - resend \n");
+ break;
case -ENOSPC:
- case -EROFS: {
+ case -EROFS:
+ case -EFBIG: {
oscc->oscc_flags |= OSCC_FLAG_NOSPC;
if (body && rc == -ENOSPC) {
oscc->oscc_grow_count = OST_MIN_PRECREATE;
break;
}
case -EIO: {
- /* filter always set body->oa.o_id as the last_id
+ /* filter always set body->oa.o_id as the last_id
* of filter (see filter_handle_precreate for detail)*/
if (body && body->oa.o_id > oscc->oscc_last_id)
oscc->oscc_last_id = body->oa.o_id;
{
struct ptlrpc_request *request;
struct ost_body *body;
- int size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ __u32 size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
ENTRY;
LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
- if (oscc->oscc_grow_count < OST_MAX_PRECREATE &&
- !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) &&
+ if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
+ oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
+ spin_unlock(&oscc->oscc_lock);
+ RETURN(0);
+ }
+
+ if (oscc->oscc_grow_count < oscc->oscc_max_grow_count &&
+ ((oscc->oscc_flags & OSCC_FLAG_LOW) == 0) &&
(__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
(oscc->oscc_grow_count / 4 + 1)) {
oscc->oscc_flags |= OSCC_FLAG_LOW;
oscc->oscc_grow_count *= 2;
}
- if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2)
- oscc->oscc_grow_count = OST_MAX_PRECREATE / 2;
+ if (oscc->oscc_grow_count > oscc->oscc_max_grow_count / 2)
+ oscc->oscc_grow_count = oscc->oscc_max_grow_count / 2;
- if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
- oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
- spin_unlock(&oscc->oscc_lock);
- RETURN(0);
- }
oscc->oscc_flags |= OSCC_FLAG_CREATING;
spin_unlock(&oscc->oscc_lock);
RETURN(-ENOMEM);
}
- request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
+ request->rq_request_portal = OST_CREATE_PORTAL;
+ ptlrpc_at_set_req_timeout(request);
body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
spin_lock(&oscc->oscc_lock);
body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
body->oa.o_gr = oscc->oscc_oa.o_gr;
- LASSERT(body->oa.o_gr > 0);
+ LASSERT_MDS_GROUP(body->oa.o_gr);
body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
+ request->rq_async_args.space[0] = oscc->oscc_grow_count;
spin_unlock(&oscc->oscc_lock);
CDEBUG(D_RPCTRACE, "prealloc through id "LPU64" (last seen "LPU64")\n",
body->oa.o_id, oscc->oscc_last_id);
request->rq_async_args.pointer_arg[0] = oscc;
request->rq_interpret_reply = osc_interpret_create;
- ptlrpcd_add_req(request);
+ ptlrpcd_add_req(request, PSCOPE_OTHER);
RETURN(0);
}
if (imp != NULL && imp->imp_deactive)
RETURN(1000);
+ if (oscc_recovering(oscc))
+ RETURN(2);
+
+ if (oscc->oscc_flags & OSCC_FLAG_NOSPC)
+ RETURN(1000);
+
if (oscc->oscc_last_id < oscc->oscc_next_id) {
- spin_lock(&oscc->oscc_lock);
- if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
- spin_unlock(&oscc->oscc_lock);
- RETURN(1000);
- }
- if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
- spin_unlock(&oscc->oscc_lock);
+ if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS)
RETURN(1);
- }
- if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
- spin_unlock(&oscc->oscc_lock);
- RETURN(2);
- }
+ spin_lock(&oscc->oscc_lock);
if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
spin_unlock(&oscc->oscc_lock);
RETURN(1);
struct lov_stripe_md **ea, struct obd_trans_info *oti)
{
struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
+ struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
struct lov_stripe_md *lsm;
int try_again = 1, rc = 0;
ENTRY;
LASSERT(oa);
LASSERT(ea);
- LASSERT(oa->o_gr > 0);
+ LASSERT_MDS_GROUP(oa->o_gr);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
if ((oa->o_valid & OBD_MD_FLFLAGS) &&
RETURN(0);
}
oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
+ /* seting flag LOW we prevent extra grow precreate size
+ * and enforce use last assigned size */
+ oscc->oscc_flags |= OSCC_FLAG_LOW;
spin_unlock(&oscc->oscc_lock);
CDEBUG(D_HA, "%s: oscc recovery started - delete to "LPU64"\n",
oscc->oscc_obd->obd_name, oscc->oscc_next_id - 1);
spin_lock(&oscc->oscc_lock);
oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
if (rc == 0 || rc == -ENOSPC) {
+ struct obd_connect_data *ocd;
+
if (rc == -ENOSPC)
oscc->oscc_flags |= OSCC_FLAG_NOSPC;
oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
+
oscc->oscc_last_id = oa->o_id;
+ ocd = &imp->imp_connect_data;
+ if (ocd->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN) {
+ CDEBUG(D_HA, "%s: Skip orphan set, reset last "
+ "objid\n", oscc->oscc_obd->obd_name);
+ oscc->oscc_next_id = oa->o_id + 1;
+ }
+
+ /* sanity check for next objid. see bug 17025 */
+ LASSERT(oscc->oscc_next_id == oa->o_id + 1);
+
CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
oscc->oscc_last_id, rc);
CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n",
oscc->oscc_obd->obd_name);
- lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(obd_timeout/4)),
- NULL, NULL);
+ lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(
+ obd_timeout / 4)), NULL, NULL);
rc = l_wait_event(oscc->oscc_waitq,
!oscc_recovering(oscc), &lwi);
LASSERT(rc == 0 || rc == -ETIMEDOUT);
spin_lock_init(&oscc->oscc_lock);
oscc->oscc_obd = obd;
oscc->oscc_grow_count = OST_MIN_PRECREATE;
+ oscc->oscc_max_grow_count = OST_MAX_PRECREATE;
oscc->oscc_next_id = 2;
oscc->oscc_last_id = 1;