/* -*- 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 Lustre, http://www.lustre.org.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * 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.
+ * 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.
*
- * 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.
+ * 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
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * 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
#define DEBUG_SUBSYSTEM S_OSC
#ifdef __KERNEL__
-# include <linux/version.h>
-# include <linux/module.h>
-# include <linux/mm.h>
-# include <linux/highmem.h>
-# include <linux/ctype.h>
-# include <linux/init.h>
-# if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
-# include <linux/workqueue.h>
-# include <linux/smp_lock.h>
-# else
-# include <linux/locks.h>
-# endif
+# include <libcfs/libcfs.h>
#else /* __KERNEL__ */
# include <liblustre.h>
#endif
# include <ctype.h>
#endif
-# include <linux/lustre_dlm.h>
-#include <linux/obd_class.h>
+#include <lustre_dlm.h>
+#include <obd_class.h>
#include "osc_internal.h"
-static int osc_interpret_create(struct ptlrpc_request *req, void *data,
- int rc)
+/* XXX need AT adjust ? */
+#define osc_create_timeout (obd_timeout / 2)
+
+struct osc_create_async_args {
+ struct osc_creator *rq_oscc;
+ struct lov_stripe_md *rq_lsm;
+ struct obd_info *rq_oinfo;
+};
+
+static int oscc_internal_create(struct osc_creator *oscc);
+static int handle_async_create(struct ptlrpc_request *req, 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;
+ struct ptlrpc_request *fake_req, *pos;
ENTRY;
if (req->rq_repmsg) {
- body = lustre_swab_repbuf(req, 0, sizeof(*body),
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
lustre_swab_ost_body);
if (body == NULL && rc == 0)
rc = -EPROTO;
}
oscc = req->rq_async_args.pointer_arg[0];
+ LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
+
spin_lock(&oscc->oscc_lock);
- if (body)
- oscc->oscc_last_id = body->oa.o_id;
- if (rc == -ENOSPC) {
- DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
+ oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
+ switch (rc) {
+ case 0: {
+ if (body) {
+ int diff = body->oa.o_id - oscc->oscc_last_id;
+
+ /* 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 -ENOSPC:
+ case -EROFS:
+ case -EFBIG: {
oscc->oscc_flags |= OSCC_FLAG_NOSPC;
- } else if (rc != 0 && rc != -EIO) {
- DEBUG_REQ(D_ERROR, req,
- "unknown rc %d from async create: failing oscc",
- rc);
+ if (body && rc == -ENOSPC) {
+ oscc->oscc_grow_count = OST_MIN_PRECREATE;
+ oscc->oscc_last_id = body->oa.o_id;
+ }
+ spin_unlock(&oscc->oscc_lock);
+ DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
+ break;
+ }
+ case -EIO: {
+ /* 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;
+ spin_unlock(&oscc->oscc_lock);
+ break;
+ }
+ case -EWOULDBLOCK: {
+ /* aka EAGAIN we should not delay create if import failed -
+ * this avoid client stick in create and avoid race with
+ * delorphan */
oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
- ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
+ /* oscc->oscc_grow_count = OST_MIN_PRECREATE; */
+ spin_unlock(&oscc->oscc_lock);
+ break;
+ }
+ default: {
+ oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
+ oscc->oscc_grow_count = OST_MIN_PRECREATE;
+ spin_unlock(&oscc->oscc_lock);
+ DEBUG_REQ(D_ERROR, req,
+ "Unknown rc %d from async create: failing oscc", rc);
+ ptlrpc_fail_import(req->rq_import,
+ lustre_msg_get_conn_cnt(req->rq_reqmsg));
+ }
}
- oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
- spin_unlock(&oscc->oscc_lock);
- CDEBUG(D_HA, "preallocated through id "LPU64" (last used "LPU64")\n",
+ CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n",
oscc->oscc_last_id, oscc->oscc_next_id);
- wake_up(&oscc->oscc_waitq);
+ spin_lock(&oscc->oscc_lock);
+ list_for_each_entry_safe(fake_req, pos,
+ &oscc->oscc_wait_create_list, rq_list) {
+ if (handle_async_create(fake_req, rc) == -EAGAIN) {
+ oscc_internal_create(oscc);
+ /* sending request should be never fail because
+ * osc use preallocated requests pool */
+ GOTO(exit_wakeup, rc);
+ }
+ }
+ spin_unlock(&oscc->oscc_lock);
+
+exit_wakeup:
+ cfs_waitq_signal(&oscc->oscc_waitq);
RETURN(rc);
}
{
struct ptlrpc_request *request;
struct ost_body *body;
- int size = sizeof(*body);
+ __u32 size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
ENTRY;
- spin_lock(&oscc->oscc_lock);
- if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
- oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
+ LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
+
+ if(oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
+ spin_unlock(&oscc->oscc_lock);
+ RETURN(0);
+ }
+
+ /* we need check it before OSCC_FLAG_CREATING - because need
+ * see lower number of precreate objects */
+ 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_flags & OSCC_FLAG_CREATING) {
spin_unlock(&oscc->oscc_lock);
RETURN(0);
}
+
+ if (oscc->oscc_grow_count > oscc->oscc_max_grow_count / 2)
+ oscc->oscc_grow_count = oscc->oscc_max_grow_count / 2;
+
oscc->oscc_flags |= OSCC_FLAG_CREATING;
spin_unlock(&oscc->oscc_lock);
- request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import, OST_CREATE,
- 1, &size, NULL);
+ request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
+ LUSTRE_OST_VERSION, OST_CREATE, 2,
+ size, NULL);
if (request == NULL) {
spin_lock(&oscc->oscc_lock);
oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
RETURN(-ENOMEM);
}
- request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
- body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
+ 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;
- /* probably we should take frequence of request into account? -bzzz */
- if (oscc->oscc_grow_count < oscc->oscc_max_grow_count) {
- oscc->oscc_grow_count *= 2;
- if (oscc->oscc_grow_count > oscc->oscc_max_grow_count)
- oscc->oscc_grow_count = oscc->oscc_max_grow_count;
- }
- body->oa.o_gr = oscc->oscc_gr;
- LASSERT(body->oa.o_gr > 0);
+ body->oa.o_gr = oscc->oscc_oa.o_gr;
+ LASSERT_MDS_GROUP(body->oa.o_gr);
body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
- CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
- body->oa.o_id, oscc->oscc_next_id);
+ 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_replen = lustre_msg_size(1, &size);
+ /* we should not resend create request - anyway we will have delorphan
+ * and kill these objects */
+ request->rq_no_delay = request->rq_no_resend = 1;
+ ptlrpc_req_set_repsize(request, 2, size);
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);
}
+static int oscc_has_objects_nolock(struct osc_creator *oscc, int count)
+{
+ return ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
+}
+
+
static int oscc_has_objects(struct osc_creator *oscc, int count)
{
int have_objs;
+
spin_lock(&oscc->oscc_lock);
- have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
+ have_objs = oscc_has_objects_nolock(oscc, count);
spin_unlock(&oscc->oscc_lock);
- if (!have_objs)
- oscc_internal_create(oscc);
-
return have_objs;
}
int ost_full;
int osc_invalid;
- have_objs = oscc_has_objects(oscc, count);
+ osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
spin_lock(&oscc->oscc_lock);
ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
- spin_unlock(&oscc->oscc_lock);
+ have_objs = oscc_has_objects_nolock(oscc, count);
+ osc_invalid |= oscc->oscc_flags & OSCC_FLAG_EXITING;
- osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
+ if (!ost_full && !osc_invalid)
+ /* they release lock himself */
+ oscc_internal_create(oscc);
+ else
+ spin_unlock(&oscc->oscc_lock);
return have_objs || ost_full || osc_invalid;
}
-static int oscc_precreate(struct osc_creator *oscc, int wait)
+static int oscc_precreate(struct osc_creator *oscc)
{
- struct l_wait_info lwi = { 0 };
+ struct l_wait_info lwi;
int rc = 0;
ENTRY;
- if (oscc_has_objects(oscc, oscc->oscc_kick_barrier))
+ if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
RETURN(0);
- if (!wait)
- RETURN(0);
+ /* we should be not block forever - because client's create rpc can
+ * stick in mds for long time and forbid client reconnect */
+ lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(osc_create_timeout)),
+ NULL, NULL);
- /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
- l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
+ rc = l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
- if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
+ if (!oscc_has_objects(oscc, 1) || (oscc->oscc_flags & OSCC_FLAG_NOSPC))
rc = -ENOSPC;
if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
RETURN(rc);
}
-int oscc_recovering(struct osc_creator *oscc)
+static int oscc_recovering(struct osc_creator *oscc)
{
- int recov = 0;
+ int recov;
spin_lock(&oscc->oscc_lock);
recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
return recov;
}
+static int oscc_in_sync(struct osc_creator *oscc)
+{
+ int sync;
+
+ spin_lock(&oscc->oscc_lock);
+ sync = oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS;
+ spin_unlock(&oscc->oscc_lock);
+
+ return sync;
+}
+
+/* decide if the OST has remaining object, return value :
+ 0 : the OST has remaining object, and don't need to do precreate.
+ 1 : the OST has no remaining object, and will send a RPC for precreate.
+ 2 : the OST has no remaining object, and will not get any for
+ a potentially very long time
+ 1000 : unusable
+ */
+int osc_precreate(struct obd_export *exp)
+{
+ struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
+ struct obd_import *imp = exp->exp_imp_reverse;
+ ENTRY;
+
+ LASSERT(oscc != NULL);
+ if (imp != NULL && imp->imp_deactive)
+ RETURN(1000);
+
+ /* until oscc in recovery - other flags is wrong */
+ if (oscc_recovering(oscc))
+ RETURN(2);
+
+ if (oscc->oscc_flags & OSCC_FLAG_NOSPC)
+ RETURN(1000);
+
+ if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
+ RETURN(0);
+
+ spin_lock(&oscc->oscc_lock);
+ if ((oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) ||
+ (oscc->oscc_flags & OSCC_FLAG_CREATING)) {
+ spin_unlock(&oscc->oscc_lock);
+ RETURN(1);
+ }
+
+ oscc_internal_create(oscc);
+ RETURN(1);
+}
+
+static int handle_async_create(struct ptlrpc_request *req, int rc)
+{
+ struct osc_create_async_args *args = ptlrpc_req_async_args(req);
+ struct osc_creator *oscc = args->rq_oscc;
+ struct lov_stripe_md *lsm = args->rq_lsm;
+ struct obd_info *oinfo = args->rq_oinfo;
+ struct obdo *oa = oinfo->oi_oa;
+
+ LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
+
+ if(rc)
+ GOTO(out_wake, rc);
+
+ if ((oscc->oscc_flags & OSCC_FLAG_EXITING))
+ GOTO(out_wake, rc = -EIO);
+
+ if (oscc_has_objects_nolock(oscc, 1)) {
+ memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
+ oa->o_id = oscc->oscc_next_id;
+ lsm->lsm_object_id = oscc->oscc_next_id;
+ oscc->oscc_next_id++;
+
+ CDEBUG(D_RPCTRACE, " set oscc_next_id = "LPU64"\n",
+ oscc->oscc_next_id);
+ GOTO(out_wake, rc = 0);
+ }
+
+ /* should be try wait until recovery finished */
+ if(oscc->oscc_flags & OSCC_FLAG_RECOVERING)
+ RETURN(-EAGAIN);
+
+ if (oscc->oscc_flags & OSCC_FLAG_NOSPC)
+ GOTO(out_wake, rc = -ENOSPC);
+
+ /* we not have objects now - continue wait */
+ RETURN(-EAGAIN);
+
+out_wake:
+
+ rc = oinfo->oi_cb_up(oinfo, rc);
+ ptlrpc_fakereq_finished(req);
+
+ RETURN(rc);
+}
+
+static int async_create_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req, void *data, int rc)
+{
+ struct osc_create_async_args *args = ptlrpc_req_async_args(req);
+ struct osc_creator *oscc = args->rq_oscc;
+ int ret;
+
+ spin_lock(&oscc->oscc_lock);
+ ret = handle_async_create(req, rc);
+ spin_unlock(&oscc->oscc_lock);
+
+ return ret;
+}
+
+int osc_create_async(struct obd_export *exp, struct obd_info *oinfo,
+ struct lov_stripe_md **ea, struct obd_trans_info *oti)
+{
+ int rc;
+ struct ptlrpc_request *fake_req;
+ struct osc_create_async_args *args;
+ struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
+ struct obdo *oa = oinfo->oi_oa;
+ ENTRY;
+
+ if ((oa->o_valid & OBD_MD_FLGROUP) && (oa->o_gr != 0)){
+ rc = osc_real_create(exp, oinfo->oi_oa, ea, oti);
+ rc = oinfo->oi_cb_up(oinfo, rc);
+ RETURN(rc);
+ }
+
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ oa->o_flags == OBD_FL_RECREATE_OBJS) {
+ rc = osc_real_create(exp, oinfo->oi_oa, ea, oti);
+ rc = oinfo->oi_cb_up(oinfo, rc);
+ RETURN(rc);
+ }
+
+ LASSERT((*ea) != NULL);
+
+ fake_req = ptlrpc_prep_fakereq(oscc->oscc_obd->u.cli.cl_import,
+ osc_create_timeout,
+ async_create_interpret);
+ if (fake_req == NULL) {
+ rc = oinfo->oi_cb_up(oinfo, -ENOMEM);
+ RETURN(-ENOMEM);
+ }
+
+ args = ptlrpc_req_async_args(fake_req);
+ CLASSERT(sizeof(*args) <= sizeof(fake_req->rq_async_args));
+
+ args->rq_oscc = oscc;
+ args->rq_lsm = *ea;
+ args->rq_oinfo = oinfo;
+
+ spin_lock(&oscc->oscc_lock);
+ /* try fast path */
+ rc = handle_async_create(fake_req, 0);
+ if (rc == -EAGAIN) {
+ int is_add;
+ /* we not have objects - try wait */
+ is_add = ptlrpcd_add_req(fake_req, PSCOPE_OTHER);
+ if (!is_add)
+ list_add(&fake_req->rq_list,
+ &oscc->oscc_wait_create_list);
+ else
+ rc = is_add;
+ }
+ spin_unlock(&oscc->oscc_lock);
+
+ if (rc != -EAGAIN)
+ /* need free request if was error hit or
+ * objects already allocated */
+ ptlrpc_req_finished(fake_req);
+ else
+ /* EAGAIN mean - request is delayed */
+ rc = 0;
+
+ RETURN(rc);
+}
+
int osc_create(struct obd_export *exp, struct obdo *oa,
struct lov_stripe_md **ea, struct obd_trans_info *oti)
{
- struct lov_stripe_md *lsm;
struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
- int try_again = 1, rc = 0;
+ struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
+ struct lov_stripe_md *lsm;
+ int rc = 0;
ENTRY;
+
LASSERT(oa);
LASSERT(ea);
+ LASSERT_MDS_GROUP(oa->o_gr);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
- LASSERT(oa->o_gr > 0);
if ((oa->o_valid & OBD_MD_FLFLAGS) &&
oa->o_flags == OBD_FL_RECREATE_OBJS) {
- /* Exceptional case where we are trying to repair missing
- * objects for various groups. We have already validated that
- * this is a valid group for the file. Don't set oscc->oscc_gr.
- */
RETURN(osc_real_create(exp, oa, ea, oti));
}
- LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr);
- oscc->oscc_gr = oa->o_gr;
-
if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
RETURN(osc_real_create(exp, oa, ea, oti));
- lsm = *ea;
- if (lsm == NULL) {
- rc = obd_alloc_memmd(exp, &lsm);
- if (rc < 0)
- RETURN(rc);
- }
+ /* this is the special case where create removes orphans */
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ oa->o_flags == OBD_FL_DELORPHAN) {
+ spin_lock(&oscc->oscc_lock);
+ if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
+ spin_unlock(&oscc->oscc_lock);
+ RETURN(-EBUSY);
+ }
+ if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
+ spin_unlock(&oscc->oscc_lock);
+ RETURN(0);
+ }
- /* this is the special case where create removes orphans */
- if ((oa->o_valid & OBD_MD_FLFLAGS) &&
- oa->o_flags == OBD_FL_DELORPHAN) {
- CDEBUG(D_HA, "%s; oscc recovery started\n",
- exp->exp_obd->obd_name);
- LASSERT(oscc->oscc_flags & OSCC_FLAG_RECOVERING);
+ 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);
/* delete from next_id on up */
oa->o_valid |= OBD_MD_FLID;
oa->o_id = oscc->oscc_next_id - 1;
- CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
- exp->exp_obd->obd_name, oa->o_id);
-
rc = osc_real_create(exp, oa, ea, NULL);
- if (oscc->oscc_obd == NULL) {
- CWARN("the obd for oscc %p has been freed\n", oscc);
- RETURN(rc);
- }
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: %d\n",
- exp->exp_obd->obd_name, rc);
- wake_up(&oscc->oscc_waitq);
-
+ CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
+ LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
+ oscc->oscc_last_id, rc);
} else {
- CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
- exp->exp_obd->obd_name, rc);
+ CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
+ oscc->oscc_obd->obd_name, rc);
}
+
+ cfs_waitq_signal(&oscc->oscc_waitq);
spin_unlock(&oscc->oscc_lock);
- RETURN(rc);
- }
-
- while (try_again) {
- /* If orphans are being recovered, then we must wait until
- it is finished before we can continue with create. */
- if (oscc_recovering(oscc)) {
- struct l_wait_info lwi;
-
- CDEBUG(D_HA,"%s: oscc sync in progress, waiting\n",
- exp->exp_obd->obd_name);
-
- lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
- rc = l_wait_event(oscc->oscc_waitq,
- !oscc_recovering(oscc), &lwi);
- LASSERT(rc == 0 || rc == -ETIMEDOUT);
- if (rc == -ETIMEDOUT) {
- CDEBUG(D_HA, "%s: timed out waiting for sync\n",
- exp->exp_obd->obd_name);
- RETURN(rc);
- }
- CDEBUG(D_HA, "%s: oscc sync over, waking up\n",
- exp->exp_obd->obd_name);
+ if (rc < 0)
+ RETURN(rc);
+ }
+
+ lsm = *ea;
+ if (lsm == NULL) {
+ rc = obd_alloc_memmd(exp, &lsm);
+ if (rc < 0)
+ RETURN(rc);
+ }
+
+ while (1) {
+ if (oscc_in_sync(oscc))
+ CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n",
+ oscc->oscc_obd->obd_name);
+
+ rc = oscc_precreate(oscc);
+ if (rc) {
+ CDEBUG(D_HA,"%s: error create %d\n",
+ oscc->oscc_obd->obd_name, rc);
+ break;
}
-
+
spin_lock(&oscc->oscc_lock);
- if (oscc->oscc_last_id >= oscc->oscc_next_id) {
+ if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
+ spin_unlock(&oscc->oscc_lock);
+ break;
+ }
+ /* wakeup but recovery not finished */
+ if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
+ rc = -EIO;
+ spin_unlock(&oscc->oscc_lock);
+ break;
+ }
+
+ if (oscc_has_objects_nolock(oscc, 1)) {
memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
oa->o_id = oscc->oscc_next_id;
- oa->o_gr = oscc->oscc_gr;
lsm->lsm_object_id = oscc->oscc_next_id;
- lsm->lsm_object_gr = oscc->oscc_gr;
*ea = lsm;
oscc->oscc_next_id++;
- try_again = 0;
+ spin_unlock(&oscc->oscc_lock);
+
+ CDEBUG(D_RPCTRACE, "%s: set oscc_next_id = "LPU64"\n",
+ exp->exp_obd->obd_name, oscc->oscc_next_id);
+ break;
} else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
rc = -ENOSPC;
spin_unlock(&oscc->oscc_lock);
break;
}
+
spin_unlock(&oscc->oscc_lock);
- rc = oscc_precreate(oscc, try_again);
- if (rc == -EIO)
- break;
}
if (rc == 0)
- CDEBUG(D_INFO, "returning objid "LPU64"/"LPU64"\n",
- lsm->lsm_object_id, lsm->lsm_object_gr);
+ CDEBUG(D_INFO, "%s: returning objid "LPU64"\n",
+ obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id);
else if (*ea == NULL)
obd_free_memmd(exp, &lsm);
RETURN(rc);
oscc = &obd->u.cli.cl_oscc;
memset(oscc, 0, sizeof(*oscc));
- INIT_LIST_HEAD(&oscc->oscc_list);
- init_waitqueue_head(&oscc->oscc_waitq);
+
+ cfs_waitq_init(&oscc->oscc_waitq);
spin_lock_init(&oscc->oscc_lock);
oscc->oscc_obd = obd;
- oscc->oscc_kick_barrier = 100;
- oscc->oscc_grow_count = 36;
- oscc->oscc_max_grow_count = 2000;
+ 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;
oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
+
+ CFS_INIT_LIST_HEAD(&oscc->oscc_wait_create_list);
+
/* XXX the export handle should give the oscc the last object */
/* oed->oed_oscc.oscc_last_id = exph->....; */
}
+
+void oscc_fini(struct obd_device *obd)
+{
+ struct osc_creator *oscc = &obd->u.cli.cl_oscc;
+ ENTRY;
+
+
+ spin_lock(&oscc->oscc_lock);
+ oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
+ oscc->oscc_flags |= OSCC_FLAG_EXITING;
+ spin_unlock(&oscc->oscc_lock);
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff