/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * lustre/mgc/mgc_request.c
- * Lustre Management Client
+ * GPL HEADER START
*
- * Copyright (C) 2006 Cluster File Systems, Inc.
- * Author: Nathan Rutman <nathan@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * This file is part of Lustre, http://www.lustre.org
+ * 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 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 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 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
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
- * 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.
+ * 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 (c) 2007, 2010, Oracle and/or its affiliates. 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/mgc/mgc_request.c
+ *
+ * Author: Nathan Rutman <nathan@clusterfs.com>
+ */
+
#ifndef EXPORT_SYMTAB
# define EXPORT_SYMTAB
#endif
#include <obd_class.h>
#include <lustre_dlm.h>
+#include <lprocfs_status.h>
#include <lustre_log.h>
#include <lustre_fsfilt.h>
#include <lustre_disk.h>
+#include "mgc_internal.h"
-
-int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id)
+static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id)
{
- char *name_end;
- int len;
__u64 resname = 0;
-
- /* fsname is at most 8 chars long at the beginning of the logname
- e.g. "lustre-MDT0001" or "lustre" */
- name_end = strrchr(logname, '-');
- if (name_end)
- len = name_end - logname;
- else
- len = strlen(logname);
+
if (len > 8) {
- CERROR("fsname too long: %s\n", logname);
+ CERROR("name too long: %s\n", name);
return -EINVAL;
}
if (len <= 0) {
- CERROR("missing fsname: %s\n", logname);
+ CERROR("missing name: %s\n", name);
return -EINVAL;
}
- memcpy(&resname, logname, len);
+ memcpy(&resname, name, len);
memset(res_id, 0, sizeof(*res_id));
+
/* Always use the same endianness for the resid */
res_id->name[0] = cpu_to_le64(resname);
- CDEBUG(D_MGC, "log %s to resid "LPX64"/"LPX64" (%.8s)\n", logname,
+ CDEBUG(D_MGC, "log %s to resid "LPX64"/"LPX64" (%.8s)\n", name,
res_id->name[0], res_id->name[1], (char *)&res_id->name[0]);
return 0;
}
-EXPORT_SYMBOL(mgc_logname2resid);
+
+int mgc_fsname2resid(char *fsname, struct ldlm_res_id *res_id)
+{
+ /* fsname is at most 8 chars long, maybe contain "-".
+ * e.g. "lustre", "SUN-000" */
+ return mgc_name2resid(fsname, strlen(fsname), res_id);
+}
+EXPORT_SYMBOL(mgc_fsname2resid);
+
+int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id)
+{
+ char *name_end;
+ int len;
+
+ /* logname consists of "fsname-nodetype".
+ * e.g. "lustre-MDT0001", "SUN-000-client" */
+ name_end = strrchr(logname, '-');
+ LASSERT(name_end);
+ len = name_end - logname;
+ return mgc_name2resid(logname, len, res_id);
+}
/********************** config llog list **********************/
-static struct list_head config_llog_list = LIST_HEAD_INIT(config_llog_list);
-static spinlock_t config_list_lock = SPIN_LOCK_UNLOCKED;
+static CFS_LIST_HEAD(config_llog_list);
+static cfs_spinlock_t config_list_lock = CFS_SPIN_LOCK_UNLOCKED;
/* Take a reference to a config log */
static int config_log_get(struct config_llog_data *cld)
ENTRY;
if (cld->cld_stopping)
RETURN(1);
- atomic_inc(&cld->cld_refcount);
+ cfs_atomic_inc(&cld->cld_refcount);
CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
- atomic_read(&cld->cld_refcount));
+ cfs_atomic_read(&cld->cld_refcount));
RETURN(0);
}
-/* Drop a reference to a config log. When no longer referenced,
+/* Drop a reference to a config log. When no longer referenced,
we can free the config log data */
static void config_log_put(struct config_llog_data *cld)
{
ENTRY;
+
CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
- atomic_read(&cld->cld_refcount));
- if (atomic_dec_and_test(&cld->cld_refcount)) {
+ cfs_atomic_read(&cld->cld_refcount));
+ LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0);
+
+ /* spinlock to make sure no item with 0 refcount in the list */
+ cfs_spin_lock(&config_list_lock);
+ if (unlikely(cfs_atomic_dec_and_test(&cld->cld_refcount))) {
+ cfs_list_del(&cld->cld_list_chain);
+ cfs_spin_unlock(&config_list_lock);
+
CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname);
+
+ if (cld->cld_sptlrpc)
+ config_log_put(cld->cld_sptlrpc);
+ if (cld->cld_is_sptlrpc)
+ sptlrpc_conf_log_stop(cld->cld_logname);
+
class_export_put(cld->cld_mgcexp);
- spin_lock(&config_list_lock);
- list_del(&cld->cld_list_chain);
- spin_unlock(&config_list_lock);
OBD_FREE(cld->cld_logname, strlen(cld->cld_logname) + 1);
if (cld->cld_cfg.cfg_instance != NULL)
- OBD_FREE(cld->cld_cfg.cfg_instance,
+ OBD_FREE(cld->cld_cfg.cfg_instance,
strlen(cld->cld_cfg.cfg_instance) + 1);
OBD_FREE(cld, sizeof(*cld));
+ } else {
+ cfs_spin_unlock(&config_list_lock);
}
+
EXIT;
}
/* Find a config log by name */
-static struct config_llog_data *config_log_find(char *logname,
- struct config_llog_instance *cfg)
+static
+struct config_llog_data *config_log_find(char *logname,
+ struct config_llog_instance *cfg)
{
- struct list_head *tmp;
struct config_llog_data *cld;
char *logid = logname;
int match_instance = 0;
RETURN(ERR_PTR(-EINVAL));
}
- spin_lock(&config_list_lock);
- list_for_each(tmp, &config_llog_list) {
- cld = list_entry(tmp, struct config_llog_data, cld_list_chain);
- if (match_instance && cld->cld_cfg.cfg_instance &&
+ cfs_spin_lock(&config_list_lock);
+ cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
+ if (match_instance && cld->cld_cfg.cfg_instance &&
strcmp(logid, cld->cld_cfg.cfg_instance) == 0)
goto out_found;
- if (!match_instance &&
+ if (!match_instance &&
strcmp(logid, cld->cld_logname) == 0)
goto out_found;
}
- spin_unlock(&config_list_lock);
+ cfs_spin_unlock(&config_list_lock);
- CERROR("can't get log %s\n", logid);
+ CDEBUG(D_CONFIG, "can't get log %s\n", logid);
RETURN(ERR_PTR(-ENOENT));
out_found:
- atomic_inc(&cld->cld_refcount);
- spin_unlock(&config_list_lock);
+ cfs_atomic_inc(&cld->cld_refcount);
+ cfs_spin_unlock(&config_list_lock);
+ LASSERT(cld->cld_stopping == 0 || cld->cld_is_sptlrpc == 0);
RETURN(cld);
}
-/* Add this log to our list of active logs.
- We have one active log per "mount" - client instance or servername.
- Each instance may be at a different point in the log. */
-static int config_log_add(char *logname, struct config_llog_instance *cfg,
- struct super_block *sb)
+static
+struct config_llog_data *do_config_log_add(struct obd_device *obd,
+ char *logname,
+ unsigned int is_sptlrpc,
+ struct config_llog_instance *cfg,
+ struct super_block *sb)
{
struct config_llog_data *cld;
- struct lustre_sb_info *lsi = s2lsi(sb);
- int rc;
+ int rc;
ENTRY;
- CDEBUG(D_MGC, "adding config log %s:%s\n", logname, cfg->cfg_instance);
-
+ CDEBUG(D_MGC, "do adding config log %s:%s\n", logname,
+ cfg ? cfg->cfg_instance : "NULL");
+
OBD_ALLOC(cld, sizeof(*cld));
- if (!cld)
- RETURN(-ENOMEM);
+ if (!cld)
+ RETURN(ERR_PTR(-ENOMEM));
OBD_ALLOC(cld->cld_logname, strlen(logname) + 1);
- if (!cld->cld_logname) {
+ if (!cld->cld_logname) {
OBD_FREE(cld, sizeof(*cld));
- RETURN(-ENOMEM);
+ RETURN(ERR_PTR(-ENOMEM));
}
strcpy(cld->cld_logname, logname);
- cld->cld_cfg = *cfg;
+ if (cfg)
+ cld->cld_cfg = *cfg;
+ cfs_mutex_init(&cld->cld_lock);
cld->cld_cfg.cfg_last_idx = 0;
cld->cld_cfg.cfg_flags = 0;
cld->cld_cfg.cfg_sb = sb;
- atomic_set(&cld->cld_refcount, 1);
-
+ cld->cld_is_sptlrpc = is_sptlrpc;
+ cfs_atomic_set(&cld->cld_refcount, 1);
+
/* Keep the mgc around until we are done */
- cld->cld_mgcexp = class_export_get(lsi->lsi_mgc->obd_self_export);
-
- if (cfg->cfg_instance != NULL) {
- OBD_ALLOC(cld->cld_cfg.cfg_instance,
+ cld->cld_mgcexp = class_export_get(obd->obd_self_export);
+
+ if (cfg && cfg->cfg_instance != NULL) {
+ OBD_ALLOC(cld->cld_cfg.cfg_instance,
strlen(cfg->cfg_instance) + 1);
strcpy(cld->cld_cfg.cfg_instance, cfg->cfg_instance);
}
+
+ if (is_sptlrpc) {
+ sptlrpc_conf_log_start(logname);
+ cld->cld_cfg.cfg_obdname = obd->obd_name;
+ }
+
rc = mgc_logname2resid(logname, &cld->cld_resid);
- spin_lock(&config_list_lock);
- list_add(&cld->cld_list_chain, &config_llog_list);
- spin_unlock(&config_list_lock);
-
+
+ cfs_spin_lock(&config_list_lock);
+ cfs_list_add(&cld->cld_list_chain, &config_llog_list);
+ cfs_spin_unlock(&config_list_lock);
+
if (rc) {
config_log_put(cld);
- RETURN(rc);
+ RETURN(ERR_PTR(rc));
}
-
- RETURN(rc);
+
+ if (is_sptlrpc) {
+ rc = mgc_process_log(obd, cld);
+ if (rc)
+ CERROR("failed processing sptlrpc log: %d\n", rc);
+ }
+
+ RETURN(cld);
}
-/* Stop watching for updates on this log. */
+/** Add this log to the list of active logs watched by an MGC.
+ * Active means we're watching for updates.
+ * We have one active log per "mount" - client instance or servername.
+ * Each instance may be at a different point in the log.
+ */
+static int config_log_add(struct obd_device *obd, char *logname,
+ struct config_llog_instance *cfg,
+ struct super_block *sb)
+{
+ struct config_llog_data *cld, *sptlrpc_cld;
+ char seclogname[20];
+ char *ptr;
+ ENTRY;
+
+ CDEBUG(D_MGC, "adding config log %s:%s\n", logname, cfg->cfg_instance);
+
+ /*
+ * for each regular log, the depended sptlrpc log name is
+ * <fsname>-sptlrpc. multiple regular logs may share one sptlrpc log.
+ */
+ ptr = strrchr(logname, '-');
+ if (ptr == NULL || ptr - logname > 8) {
+ CERROR("logname %s is too long\n", logname);
+ RETURN(-EINVAL);
+ }
+
+ memcpy(seclogname, logname, ptr - logname);
+ strcpy(seclogname + (ptr - logname), "-sptlrpc");
+
+ sptlrpc_cld = config_log_find(seclogname, NULL);
+ if (IS_ERR(sptlrpc_cld)) {
+ sptlrpc_cld = do_config_log_add(obd, seclogname, 1, NULL, NULL);
+ if (IS_ERR(sptlrpc_cld)) {
+ CERROR("can't create sptlrpc log: %s\n", seclogname);
+ RETURN(PTR_ERR(sptlrpc_cld));
+ }
+ }
+
+ cld = do_config_log_add(obd, logname, 0, cfg, sb);
+ if (IS_ERR(cld)) {
+ CERROR("can't create log: %s\n", logname);
+ config_log_put(sptlrpc_cld);
+ RETURN(PTR_ERR(cld));
+ }
+
+ cld->cld_sptlrpc = sptlrpc_cld;
+
+ RETURN(0);
+}
+
+CFS_DECLARE_MUTEX(llog_process_lock);
+
+/** Stop watching for updates on this log.
+ */
static int config_log_end(char *logname, struct config_llog_instance *cfg)
-{
- struct config_llog_data *cld;
+{
+ struct config_llog_data *cld, *cld_sptlrpc = NULL;
int rc = 0;
ENTRY;
-
+
cld = config_log_find(logname, cfg);
- if (IS_ERR(cld))
+ if (IS_ERR(cld))
RETURN(PTR_ERR(cld));
- /* drop the ref from the find */
- config_log_put(cld);
+
+ cfs_mutex_lock(&cld->cld_lock);
+ /*
+ * if cld_stopping is set, it means we didn't start the log thus
+ * not owning the start ref. this can happen after previous umount:
+ * the cld still hanging there waiting for lock cancel, and we
+ * remount again but failed in the middle and call log_end without
+ * calling start_log.
+ */
+ if (unlikely(cld->cld_stopping)) {
+ cfs_mutex_unlock(&cld->cld_lock);
+ /* drop the ref from the find */
+ config_log_put(cld);
+ RETURN(rc);
+ }
cld->cld_stopping = 1;
+ cfs_mutex_unlock(&cld->cld_lock);
+
+ cfs_spin_lock(&config_list_lock);
+ cld_sptlrpc = cld->cld_sptlrpc;
+ cld->cld_sptlrpc = NULL;
+ cfs_spin_unlock(&config_list_lock);
+
+ if (cld_sptlrpc)
+ config_log_put(cld_sptlrpc);
+
+ /* drop the ref from the find */
+ config_log_put(cld);
/* drop the start ref */
config_log_put(cld);
+
CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client",
rc);
RETURN(rc);
static int rq_state = 0;
static cfs_waitq_t rq_waitq;
-static int mgc_process_log(struct obd_device *mgc,
- struct config_llog_data *cld);
+static int mgc_requeue_add(struct config_llog_data *cld, int later);
+
+static void do_requeue(struct config_llog_data *cld)
+{
+ LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0);
+
+ /* Do not run mgc_process_log on a disconnected export or an
+ export which is being disconnected. Take the client
+ semaphore to make the check non-racy. */
+ cfs_down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
+ if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) {
+ CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname);
+ mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
+ } else {
+ CDEBUG(D_MGC, "disconnecting, won't update log %s\n",
+ cld->cld_logname);
+ }
+ cfs_up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
+
+ /* Whether we enqueued again or not in mgc_process_log, we're done
+ * with the ref from the old enqueue */
+ config_log_put(cld);
+}
static int mgc_requeue_thread(void *data)
{
struct l_wait_info lwi_now, lwi_later;
- struct config_llog_data *cld, *n;
+ struct config_llog_data *cld, *cld_next, *cld_prev;
char name[] = "ll_cfg_requeue";
int rc = 0;
ENTRY;
-
- ptlrpc_daemonize(name);
-
+
+ cfs_daemonize(name);
+
CDEBUG(D_MGC, "Starting requeue thread\n");
- lwi_later = LWI_TIMEOUT(60 * HZ, NULL, NULL);
+ lwi_later = LWI_TIMEOUT(60 * CFS_HZ, NULL, NULL);
l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP), &lwi_later);
/* Keep trying failed locks periodically */
- spin_lock(&config_list_lock);
+ cfs_spin_lock(&config_list_lock);
while (rq_state & (RQ_NOW | RQ_LATER)) {
/* Any new or requeued lostlocks will change the state */
- rq_state &= ~(RQ_NOW | RQ_LATER);
- spin_unlock(&config_list_lock);
+ rq_state &= ~(RQ_NOW | RQ_LATER);
+ cfs_spin_unlock(&config_list_lock);
- /* Always wait a few seconds to allow the server who
+ /* Always wait a few seconds to allow the server who
caused the lock revocation to finish its setup, plus some
random so everyone doesn't try to reconnect at once. */
- lwi_now = LWI_TIMEOUT(3 * HZ + (ll_rand() & 0xff) * (HZ / 100),
+ lwi_now = LWI_TIMEOUT(3 * CFS_HZ + (cfs_rand() & 0xff) * \
+ (CFS_HZ / 100),
NULL, NULL);
l_wait_event(rq_waitq, rq_state & RQ_STOP, &lwi_now);
-
- spin_lock(&config_list_lock);
- list_for_each_entry_safe(cld, n, &config_llog_list,
- cld_list_chain) {
- spin_unlock(&config_list_lock);
-
+
+ /*
+ * iterate & processing through the list. for each cld, process
+ * its depending sptlrpc cld firstly (if any) and then itself.
+ *
+ * it's guaranteed any item in the list must have
+ * reference > 0; and if cld_lostlock is set, at
+ * least one reference is taken by the previous enqueue.
+ *
+ * Note: releasing a cld might lead to itself and its depended
+ * sptlrpc cld be unlinked from the list. to safely iterate
+ * we need to take a reference on next cld before processing.
+ */
+ cld_prev = NULL;
+
+ cfs_spin_lock(&config_list_lock);
+ cfs_list_for_each_entry_safe(cld, cld_next, &config_llog_list,
+ cld_list_chain) {
+ if (cld->cld_list_chain.next != &config_llog_list)
+ cfs_atomic_inc(&cld_next->cld_refcount);
+
if (cld->cld_lostlock) {
- CDEBUG(D_MGC, "updating log %s\n",
- cld->cld_logname);
+ if (cld->cld_sptlrpc &&
+ cld->cld_sptlrpc->cld_lostlock) {
+ cld->cld_sptlrpc->cld_lostlock = 0;
+
+ cfs_spin_unlock(&config_list_lock);
+ do_requeue(cld->cld_sptlrpc);
+ cfs_spin_lock(&config_list_lock);
+ LASSERT(cld->cld_lostlock);
+ }
+
cld->cld_lostlock = 0;
- rc = mgc_process_log(cld->cld_mgcexp->exp_obd,
- cld);
- /* Whether we enqueued again or not in
- mgc_process_log, we're done with the ref
- from the old enqueue */
- config_log_put(cld);
+
+ cfs_spin_unlock(&config_list_lock);
+ do_requeue(cld);
+ cfs_spin_lock(&config_list_lock);
}
- spin_lock(&config_list_lock);
+
+ if (cld_prev) {
+ cfs_spin_unlock(&config_list_lock);
+ config_log_put(cld_prev);
+ cfs_spin_lock(&config_list_lock);
+ }
+
+ cld_prev = cld_next;
}
- spin_unlock(&config_list_lock);
-
+ cfs_spin_unlock(&config_list_lock);
+
/* Wait a bit to see if anyone else needs a requeue */
l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
&lwi_later);
- spin_lock(&config_list_lock);
+ cfs_spin_lock(&config_list_lock);
}
/* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */
rq_state &= ~RQ_RUNNING;
- spin_unlock(&config_list_lock);
-
+ cfs_spin_unlock(&config_list_lock);
+
CDEBUG(D_MGC, "Ending requeue thread\n");
RETURN(rc);
}
{
int rc = 0;
- CDEBUG(D_INFO, "log %s: requeue (l=%d r=%d sp=%d st=%x)\n",
- cld->cld_logname, later, atomic_read(&cld->cld_refcount),
+ CDEBUG(D_INFO, "log %s: requeue (l=%d r=%d sp=%d st=%x)\n",
+ cld->cld_logname, later, cfs_atomic_read(&cld->cld_refcount),
cld->cld_stopping, rq_state);
+ LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0);
/* Hold lock for rq_state */
- spin_lock(&config_list_lock);
- cld->cld_lostlock = 1;
+ cfs_spin_lock(&config_list_lock);
if (cld->cld_stopping || (rq_state & RQ_STOP)) {
- spin_unlock(&config_list_lock);
+ cld->cld_lostlock = 0;
+ cfs_spin_unlock(&config_list_lock);
config_log_put(cld);
RETURN(0);
}
+ cld->cld_lostlock = 1;
+
if (!(rq_state & RQ_RUNNING)) {
LASSERT(rq_state == 0);
rq_state = RQ_RUNNING | (later ? RQ_LATER : RQ_NOW);
- spin_unlock(&config_list_lock);
+ cfs_spin_unlock(&config_list_lock);
rc = cfs_kernel_thread(mgc_requeue_thread, 0,
CLONE_VM | CLONE_FILES);
if (rc < 0) {
}
} else {
rq_state |= later ? RQ_LATER : RQ_NOW;
- spin_unlock(&config_list_lock);
+ cfs_spin_unlock(&config_list_lock);
cfs_waitq_signal(&rq_waitq);
}
/********************** class fns **********************/
-static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb,
+static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb,
struct vfsmount *mnt)
{
struct lvfs_run_ctxt saved;
LASSERT(lsi->lsi_srv_mnt == mnt);
/* The mgc fs exclusion sem. Only one fs can be setup at a time. */
- down(&cli->cl_mgc_sem);
+ cfs_down(&cli->cl_mgc_sem);
- cleanup_group_info();
+ cfs_cleanup_group_info();
obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
if (IS_ERR(obd->obd_fsops)) {
- up(&cli->cl_mgc_sem);
- CERROR("No fstype %s rc=%ld\n", MT_STR(lsi->lsi_ldd),
+ cfs_up(&cli->cl_mgc_sem);
+ CERROR("No fstype %s rc=%ld\n", MT_STR(lsi->lsi_ldd),
PTR_ERR(obd->obd_fsops));
RETURN(PTR_ERR(obd->obd_fsops));
}
obd->obd_lvfs_ctxt.fs = get_ds();
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- dentry = lookup_one_len(MOUNT_CONFIGS_DIR, current->fs->pwd,
+ dentry = lookup_one_len(MOUNT_CONFIGS_DIR, cfs_fs_pwd(current->fs),
strlen(MOUNT_CONFIGS_DIR));
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
if (IS_ERR(dentry)) {
err = PTR_ERR(dentry);
- CERROR("cannot lookup %s directory: rc = %d\n",
+ CERROR("cannot lookup %s directory: rc = %d\n",
MOUNT_CONFIGS_DIR, err);
GOTO(err_ops, err);
}
/* We take an obd ref to insure that we can't get to mgc_cleanup
without calling mgc_fs_cleanup first. */
- class_incref(obd);
+ class_incref(obd, "mgc_fs", obd);
label = fsfilt_get_label(obd, mnt->mnt_sb);
if (label)
/* We keep the cl_mgc_sem until mgc_fs_cleanup */
RETURN(0);
-err_ops:
+err_ops:
fsfilt_put_ops(obd->obd_fsops);
obd->obd_fsops = NULL;
cli->cl_mgc_vfsmnt = NULL;
- up(&cli->cl_mgc_sem);
+ cfs_up(&cli->cl_mgc_sem);
RETURN(err);
}
struct lvfs_run_ctxt saved;
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
l_dput(cli->cl_mgc_configs_dir);
- cli->cl_mgc_configs_dir = NULL;
+ cli->cl_mgc_configs_dir = NULL;
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- class_decref(obd);
+ class_decref(obd, "mgc_fs", obd);
}
cli->cl_mgc_vfsmnt = NULL;
- if (obd->obd_fsops)
+ if (obd->obd_fsops)
fsfilt_put_ops(obd->obd_fsops);
-
- up(&cli->cl_mgc_sem);
+
+ cfs_up(&cli->cl_mgc_sem);
RETURN(rc);
}
-static atomic_t mgc_count = ATOMIC_INIT(0);
+static cfs_atomic_t mgc_count = CFS_ATOMIC_INIT(0);
static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
{
int rc = 0;
ENTRY;
switch (stage) {
- case OBD_CLEANUP_EARLY:
+ case OBD_CLEANUP_EARLY:
break;
case OBD_CLEANUP_EXPORTS:
- if (atomic_dec_and_test(&mgc_count)) {
- /* Kick the requeue waitq - cld's should all be
+ if (cfs_atomic_dec_and_test(&mgc_count)) {
+ /* Kick the requeue waitq - cld's should all be
stopping */
- spin_lock(&config_list_lock);
+ cfs_spin_lock(&config_list_lock);
rq_state |= RQ_STOP;
- spin_unlock(&config_list_lock);
+ cfs_spin_unlock(&config_list_lock);
cfs_waitq_signal(&rq_waitq);
}
- break;
- case OBD_CLEANUP_SELF_EXP:
rc = obd_llog_finish(obd, 0);
if (rc != 0)
CERROR("failed to cleanup llogging subsystems\n");
break;
- case OBD_CLEANUP_OBD:
- break;
}
RETURN(rc);
}
ENTRY;
LASSERT(cli->cl_mgc_vfsmnt == NULL);
-
- /* COMPAT_146 - old config logs may have added profiles we don't
+
+ /* COMPAT_146 - old config logs may have added profiles we don't
know about */
- if (obd->obd_type->typ_refcnt <= 1)
+ if (obd->obd_type->typ_refcnt <= 1)
/* Only for the last mgc */
class_del_profiles();
+ lprocfs_obd_cleanup(obd);
ptlrpcd_decref();
rc = client_obd_cleanup(obd);
RETURN(rc);
}
-static int mgc_setup(struct obd_device *obd, obd_count len, void *buf)
+static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
{
+ struct lprocfs_static_vars lvars;
int rc;
ENTRY;
ptlrpcd_addref();
- rc = client_obd_setup(obd, len, buf);
+ rc = client_obd_setup(obd, lcfg);
if (rc)
GOTO(err_decref, rc);
- rc = obd_llog_init(obd, obd, 0, NULL, NULL);
+ rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
if (rc) {
CERROR("failed to setup llogging subsystems\n");
GOTO(err_cleanup, rc);
}
- spin_lock(&config_list_lock);
- atomic_inc(&mgc_count);
- if (atomic_read(&mgc_count) == 1) {
+ lprocfs_mgc_init_vars(&lvars);
+ lprocfs_obd_setup(obd, lvars.obd_vars);
+ sptlrpc_lprocfs_cliobd_attach(obd);
+
+ cfs_spin_lock(&config_list_lock);
+ cfs_atomic_inc(&mgc_count);
+ if (cfs_atomic_read(&mgc_count) == 1) {
rq_state &= ~RQ_STOP;
cfs_waitq_init(&rq_waitq);
}
- spin_unlock(&config_list_lock);
-
+ cfs_spin_unlock(&config_list_lock);
+
RETURN(rc);
err_cleanup:
case LDLM_CB_CANCELING: {
/* We've given up the lock, prepare ourselves to update. */
LDLM_DEBUG(lock, "MGC cancel CB");
-
+
CDEBUG(D_MGC, "Lock res "LPX64" (%.8s)\n",
- lock->l_resource->lr_name.name[0],
+ lock->l_resource->lr_name.name[0],
(char *)&lock->l_resource->lr_name.name[0]);
-
+
if (!cld) {
CERROR("missing data, won't requeue\n");
break;
}
/* Are we done with this log? */
if (cld->cld_stopping) {
- CDEBUG(D_MGC, "log %s: stopping, won't requeue\n",
+ CDEBUG(D_MGC, "log %s: stopping, won't requeue\n",
cld->cld_logname);
- config_log_put(cld);
+ config_log_put(cld);
break;
}
/* Make sure not to re-enqueue when the mgc is stopping
!lock->l_conn_export->exp_obd->u.cli.cl_conn_count) {
CDEBUG(D_MGC, "log %s: disconnecting, won't requeue\n",
cld->cld_logname);
- config_log_put(cld);
+ config_log_put(cld);
break;
}
/* Did we fail to get the lock? */
- if (lock->l_req_mode != lock->l_granted_mode) {
+ if (lock->l_req_mode != lock->l_granted_mode &&
+ !cld->cld_is_sptlrpc) {
CDEBUG(D_MGC, "log %s: original grant failed, will "
"requeue later\n", cld->cld_logname);
/* Try to re-enqueue later */
if (rc) {
- CERROR("%s CB failed %d:\n", flag == LDLM_CB_BLOCKING ?
+ CERROR("%s CB failed %d:\n", flag == LDLM_CB_BLOCKING ?
"blocking" : "cancel", rc);
LDLM_ERROR(lock, "MGC ast");
}
RETURN(rc);
}
+/* Not sure where this should go... */
+#define MGC_ENQUEUE_LIMIT 50
+#define MGC_TARGET_REG_LIMIT 10
+#define MGC_SEND_PARAM_LIMIT 10
+
+/* Send parameter to MGS*/
+static int mgc_set_mgs_param(struct obd_export *exp,
+ struct mgs_send_param *msp)
+{
+ struct ptlrpc_request *req;
+ struct mgs_send_param *req_msp, *rep_msp;
+ int rc;
+ ENTRY;
+
+ req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+ &RQF_MGS_SET_INFO, LUSTRE_MGS_VERSION,
+ MGS_SET_INFO);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ req_msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
+ if (!req_msp) {
+ ptlrpc_req_finished(req);
+ RETURN(-ENOMEM);
+ }
+
+ memcpy(req_msp, msp, sizeof(*req_msp));
+ ptlrpc_request_set_replen(req);
+
+ /* Limit how long we will wait for the enqueue to complete */
+ req->rq_delay_limit = MGC_SEND_PARAM_LIMIT;
+ rc = ptlrpc_queue_wait(req);
+ if (!rc) {
+ rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
+ memcpy(msp, rep_msp, sizeof(*rep_msp));
+ }
+
+ ptlrpc_req_finished(req);
+
+ RETURN(rc);
+}
+
/* Take a config lock so we can get cancel notifications */
static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
__u32 type, ldlm_policy_data_t *policy, __u32 mode,
int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
void *data, __u32 lvb_len, void *lvb_swabber,
struct lustre_handle *lockh)
-{
+{
struct config_llog_data *cld = (struct config_llog_data *)data;
+ struct ldlm_enqueue_info einfo = { type, mode, mgc_blocking_ast,
+ ldlm_completion_ast, NULL, NULL, data};
+ struct ptlrpc_request *req;
+ int short_limit = cld->cld_is_sptlrpc;
int rc;
ENTRY;
CDEBUG(D_MGC, "Enqueue for %s (res "LPX64")\n", cld->cld_logname,
cld->cld_resid.name[0]);
-
+
/* We can only drop this config log ref when we drop the lock */
if (config_log_get(cld))
RETURN(ELDLM_LOCK_ABORTED);
/* We need a callback for every lockholder, so don't try to
ldlm_lock_match (see rev 1.1.2.11.2.47) */
-
- rc = ldlm_cli_enqueue(exp, NULL, cld->cld_resid,
- type, NULL, mode, flags,
- mgc_blocking_ast, ldlm_completion_ast, NULL,
- data, NULL, 0, NULL, lockh, 0);
- /* A failed enqueue should still call the mgc_blocking_ast,
- where it will be requeued if needed ("grant failed"). */
-
+ req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+ &RQF_LDLM_ENQUEUE, LUSTRE_DLM_VERSION,
+ LDLM_ENQUEUE);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+ ptlrpc_request_set_replen(req);
+ /* check if this is server or client */
+ if (cld->cld_cfg.cfg_sb) {
+ struct lustre_sb_info *lsi = s2lsi(cld->cld_cfg.cfg_sb);
+ if (lsi && (lsi->lsi_flags & LSI_SERVER))
+ short_limit = 1;
+ }
+ /* Limit how long we will wait for the enqueue to complete */
+ req->rq_delay_limit = short_limit ? 5 : MGC_ENQUEUE_LIMIT;
+ rc = ldlm_cli_enqueue(exp, &req, &einfo, &cld->cld_resid, NULL, flags,
+ NULL, 0, lockh, 0);
+ /* A failed enqueue should still call the mgc_blocking_ast,
+ where it will be requeued if needed ("grant failed"). */
+ ptlrpc_req_finished(req);
RETURN(rc);
}
RETURN(0);
}
-#if 0
-static int mgc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg)
-{
- struct obd_device *obd = exp->exp_obd;
- struct obd_ioctl_data *data = karg;
- struct llog_ctxt *ctxt;
- struct lvfs_run_ctxt saved;
- int rc;
- ENTRY;
-
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- MOD_INC_USE_COUNT;
-#else
- if (!try_module_get(THIS_MODULE)) {
- CERROR("Can't get module. Is it alive?");
- return -EINVAL;
- }
-#endif
- switch (cmd) {
- /* REPLicator context */
- case OBD_IOC_PARSE: {
- CERROR("MGC parsing llog %s\n", data->ioc_inlbuf1);
- ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
- rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
- GOTO(out, rc);
- }
-#ifdef __KERNEL__
- case OBD_IOC_LLOG_INFO:
- case OBD_IOC_LLOG_PRINT: {
- ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
- rc = llog_ioctl(ctxt, cmd, data);
-
- GOTO(out, rc);
- }
-#endif
- /* ORIGinator context */
- case OBD_IOC_DUMP_LOG: {
- ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- if (rc)
- RETURN(rc);
-
- GOTO(out, rc);
- }
- default:
- CERROR("mgc_ioctl(): unrecognised ioctl %#x\n", cmd);
- GOTO(out, rc = -ENOTTY);
- }
-out:
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- MOD_DEC_USE_COUNT;
-#else
- module_put(THIS_MODULE);
-#endif
-
- return rc;
-}
-#endif
-
/* Send target_reg message to MGS */
static int mgc_target_register(struct obd_export *exp,
struct mgs_target_info *mti)
{
- struct ptlrpc_request *req;
+ struct ptlrpc_request *req;
struct mgs_target_info *req_mti, *rep_mti;
- int size[] = { sizeof(struct ptlrpc_body), sizeof(*req_mti) };
- int rep_size[] = { sizeof(struct ptlrpc_body), sizeof(*mti) };
- int rc;
+ int rc;
ENTRY;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MGS_VERSION,
- MGS_TARGET_REG, 2, size, NULL);
- if (!req)
+ req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+ &RQF_MGS_TARGET_REG, LUSTRE_MGS_VERSION,
+ MGS_TARGET_REG);
+ if (req == NULL)
RETURN(-ENOMEM);
- req_mti = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*req_mti));
- if (!req_mti)
+ req_mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
+ if (!req_mti) {
+ ptlrpc_req_finished(req);
RETURN(-ENOMEM);
- memcpy(req_mti, mti, sizeof(*req_mti));
-
- ptlrpc_req_set_repsize(req, 2, rep_size);
+ }
+ memcpy(req_mti, mti, sizeof(*req_mti));
+ ptlrpc_request_set_replen(req);
CDEBUG(D_MGC, "register %s\n", mti->mti_svname);
-
+ /* Limit how long we will wait for the enqueue to complete */
+ req->rq_delay_limit = MGC_TARGET_REG_LIMIT;
+
rc = ptlrpc_queue_wait(req);
if (!rc) {
- rep_mti = lustre_swab_repbuf(req, REPLY_REC_OFF,
- sizeof(*rep_mti),
- lustre_swab_mgs_target_info);
+ rep_mti = req_capsule_server_get(&req->rq_pill,
+ &RMF_MGS_TARGET_INFO);
memcpy(mti, rep_mti, sizeof(*rep_mti));
CDEBUG(D_MGC, "register %s got index = %d\n",
mti->mti_svname, mti->mti_stripe_index);
}
int mgc_set_info_async(struct obd_export *exp, obd_count keylen,
- void *key, obd_count vallen, void *val,
+ void *key, obd_count vallen, void *val,
struct ptlrpc_request_set *set)
{
- struct obd_import *imp = class_exp2cliimp(exp);
int rc = -EINVAL;
ENTRY;
- /* Try to "recover" the initial connection; i.e. retry */
- if (KEY_IS(KEY_INIT_RECOV)) {
- if (vallen != sizeof(int))
- RETURN(-EINVAL);
- imp->imp_initial_recov = *(int *)val;
- CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
- exp->exp_obd->obd_name, imp->imp_initial_recov);
- RETURN(0);
- }
/* Turn off initial_recov after we try all backup servers once */
if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
+ struct obd_import *imp = class_exp2cliimp(exp);
int value;
if (vallen != sizeof(int))
RETURN(-EINVAL);
value = *(int *)val;
- imp->imp_initial_recov_bk = value > 0;
- /* Even after the initial connection, give up all comms if
- nobody answers the first time. */
- imp->imp_recon_bk = 1;
- CDEBUG(D_MGC, "InitRecov %s %d/%d:d%d:i%d:r%d:or%d:%s\n",
- imp->imp_obd->obd_name, value, imp->imp_initial_recov,
- imp->imp_deactive, imp->imp_invalid,
+ CDEBUG(D_MGC, "InitRecov %s %d/d%d:i%d:r%d:or%d:%s\n",
+ imp->imp_obd->obd_name, value,
+ imp->imp_deactive, imp->imp_invalid,
imp->imp_replayable, imp->imp_obd->obd_replayable,
ptlrpc_import_state_name(imp->imp_state));
/* Resurrect if we previously died */
- if (imp->imp_invalid || value > 1) {
- /* See client_disconnect_export */
- /* Allow reconnect attempts */
- imp->imp_obd->obd_no_recov = 0;
- /* Force a new connect attempt */
- /* (can't put these in obdclass, module loop) */
- ptlrpc_invalidate_import(imp);
- /* Do a fresh connect next time by zeroing the handle */
- ptlrpc_disconnect_import(imp, 1);
- /* Remove 'invalid' flag */
- ptlrpc_activate_import(imp);
- /* Attempt a new connect */
- ptlrpc_recover_import(imp, NULL);
- }
+ if ((imp->imp_state != LUSTRE_IMP_FULL &&
+ imp->imp_state != LUSTRE_IMP_NEW) || value > 1)
+ ptlrpc_reconnect_import(imp);
RETURN(0);
}
/* FIXME move this to mgc_process_config */
- if (KEY_IS("register_target")) {
+ if (KEY_IS(KEY_REGISTER_TARGET)) {
struct mgs_target_info *mti;
if (vallen != sizeof(struct mgs_target_info))
RETURN(-EINVAL);
rc = mgc_target_register(exp, mti);
RETURN(rc);
}
- if (KEY_IS("set_fs")) {
+ if (KEY_IS(KEY_SET_FS)) {
struct super_block *sb = (struct super_block *)val;
struct lustre_sb_info *lsi;
if (vallen != sizeof(struct super_block))
}
RETURN(rc);
}
- if (KEY_IS("clear_fs")) {
+ if (KEY_IS(KEY_CLEAR_FS)) {
if (vallen != 0)
RETURN(-EINVAL);
rc = mgc_fs_cleanup(exp->exp_obd);
}
RETURN(rc);
}
+ if (KEY_IS(KEY_SET_INFO)) {
+ struct mgs_send_param *msp;
+
+ msp = (struct mgs_send_param *)val;
+ rc = mgc_set_mgs_param(exp, msp);
+ RETURN(rc);
+ }
+ if (KEY_IS(KEY_MGSSEC)) {
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+ struct sptlrpc_flavor flvr;
+
+ /*
+ * empty string means using current flavor, if which haven't
+ * been set yet, set it as null.
+ *
+ * if flavor has been set previously, check the asking flavor
+ * must match the existing one.
+ */
+ if (vallen == 0) {
+ if (cli->cl_flvr_mgc.sf_rpc != SPTLRPC_FLVR_INVALID)
+ RETURN(0);
+ val = "null";
+ vallen = 4;
+ }
+
+ rc = sptlrpc_parse_flavor(val, &flvr);
+ if (rc) {
+ CERROR("invalid sptlrpc flavor %s to MGS\n",
+ (char *) val);
+ RETURN(rc);
+ }
+
+ /*
+ * caller already hold a mutex
+ */
+ if (cli->cl_flvr_mgc.sf_rpc == SPTLRPC_FLVR_INVALID) {
+ cli->cl_flvr_mgc = flvr;
+ } else if (memcmp(&cli->cl_flvr_mgc, &flvr,
+ sizeof(flvr)) != 0) {
+ char str[20];
+
+ sptlrpc_flavor2name(&cli->cl_flvr_mgc,
+ str, sizeof(str));
+ LCONSOLE_ERROR("asking sptlrpc flavor %s to MGS but "
+ "currently %s is in use\n",
+ (char *) val, str);
+ rc = -EPERM;
+ }
+ RETURN(rc);
+ }
RETURN(rc);
-}
+}
static int mgc_import_event(struct obd_device *obd,
struct obd_import *imp,
CDEBUG(D_MGC, "import event %#x\n", event);
switch (event) {
- case IMP_EVENT_DISCON:
+ case IMP_EVENT_DISCON:
/* MGC imports should not wait for recovery */
- ptlrpc_invalidate_import(imp);
break;
- case IMP_EVENT_INACTIVE:
+ case IMP_EVENT_INACTIVE:
break;
case IMP_EVENT_INVALIDATE: {
struct ldlm_namespace *ns = obd->obd_namespace;
ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
break;
}
- case IMP_EVENT_ACTIVE:
+ case IMP_EVENT_ACTIVE:
LCONSOLE_WARN("%s: Reactivating import\n", obd->obd_name);
+ /* Clearing obd_no_recov allows us to continue pinging */
+ obd->obd_no_recov = 0;
break;
case IMP_EVENT_OCD:
break;
RETURN(rc);
}
-static int mgc_llog_init(struct obd_device *obd, struct obd_device *tgt,
- int count, struct llog_catid *logid,
- struct obd_uuid *uuid)
+static int mgc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
+ struct obd_device *tgt, int *index)
{
struct llog_ctxt *ctxt;
int rc;
ENTRY;
- rc = llog_setup(obd, LLOG_CONFIG_ORIG_CTXT, tgt, 0, NULL,
+ LASSERT(olg == &obd->obd_olg);
+
+ rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, tgt, 0, NULL,
&llog_lvfs_ops);
if (rc)
RETURN(rc);
- rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
+ rc = llog_setup(obd, olg, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
&llog_client_ops);
if (rc == 0) {
ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
- ctxt->loc_imp = obd->u.cli.cl_import;
+ if (!ctxt) {
+ ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+ if (ctxt)
+ llog_cleanup(ctxt);
+ RETURN(-ENODEV);
+ }
+ llog_initiator_connect(ctxt);
+ llog_ctxt_put(ctxt);
+ } else {
+ ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+ if (ctxt)
+ llog_cleanup(ctxt);
}
RETURN(rc);
static int mgc_llog_finish(struct obd_device *obd, int count)
{
- int rc;
+ struct llog_ctxt *ctxt;
+ int rc = 0, rc2 = 0;
ENTRY;
- rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
- rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
+ ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
+ if (ctxt)
+ rc = llog_cleanup(ctxt);
+
+ ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+ if (ctxt)
+ rc2 = llog_cleanup(ctxt);
+
+ if (!rc)
+ rc = rc2;
RETURN(rc);
}
return(rc <= 1);
}
-static int mgc_copy_handler(struct llog_handle *llh, struct llog_rec_hdr *rec,
+static int mgc_copy_handler(struct llog_handle *llh, struct llog_rec_hdr *rec,
void *data)
{
struct llog_rec_hdr local_rec = *rec;
int rc = 0;
ENTRY;
- lcfg = (struct lustre_cfg *)cfg_buf;
-
- /* append new records */
- if (rec->lrh_index >= llog_get_size(local_llh)) {
- rc = llog_write_rec(local_llh, &local_rec, NULL, 0,
- (void *)cfg_buf, -1);
+ /* Append all records */
+ local_rec.lrh_len -= sizeof(*rec) + sizeof(struct llog_rec_tail);
+ rc = llog_write_rec(local_llh, &local_rec, NULL, 0,
+ (void *)cfg_buf, -1);
- CDEBUG(D_INFO, "idx=%d, rc=%d, len=%d, cmd %x %s %s\n",
- rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command,
- lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1));
- } else {
- CDEBUG(D_INFO, "skip idx=%d\n", rec->lrh_index);
- }
+ lcfg = (struct lustre_cfg *)cfg_buf;
+ CDEBUG(D_INFO, "idx=%d, rc=%d, len=%d, cmd %x %s %s\n",
+ rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command,
+ lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1));
RETURN(rc);
}
+/* Copy a remote log locally */
static int mgc_copy_llog(struct obd_device *obd, struct llog_ctxt *rctxt,
struct llog_ctxt *lctxt, char *logname)
{
struct llog_handle *local_llh, *remote_llh;
struct obd_uuid *uuid;
+ char *temp_log;
int rc, rc2;
ENTRY;
+ /* Write new log to a temp name, then vfs_rename over logname
+ upon successful completion. */
+
+ OBD_ALLOC(temp_log, strlen(logname) + 1);
+ if (!temp_log)
+ RETURN(-ENOMEM);
+ sprintf(temp_log, "%sT", logname);
+
+ /* Make sure there's no old temp log */
+ rc = llog_create(lctxt, &local_llh, NULL, temp_log);
+ if (rc)
+ GOTO(out, rc);
+ rc = llog_init_handle(local_llh, LLOG_F_IS_PLAIN, NULL);
+ if (rc)
+ GOTO(out, rc);
+ rc = llog_destroy(local_llh);
+ llog_free_handle(local_llh);
+ if (rc)
+ GOTO(out, rc);
+
/* open local log */
- rc = llog_create(lctxt, &local_llh, NULL, logname);
+ rc = llog_create(lctxt, &local_llh, NULL, temp_log);
if (rc)
- RETURN(rc);
+ GOTO(out, rc);
+
/* set the log header uuid for fun */
OBD_ALLOC_PTR(uuid);
obd_str2uuid(uuid, logname);
if (rc)
GOTO(out_closel, rc);
- /* FIXME write new log to a temp name, then vfs_rename over logname
- upon successful completion. */
-
/* open remote log */
rc = llog_create(rctxt, &remote_llh, NULL, logname);
if (rc)
if (rc)
GOTO(out_closer, rc);
+ /* Copy remote log */
rc = llog_process(remote_llh, mgc_copy_handler,(void *)local_llh, NULL);
out_closer:
if (!rc)
rc = rc2;
+ /* We've copied the remote log to the local temp log, now
+ replace the old local log with the temp log. */
+ if (!rc) {
+ struct client_obd *cli = &obd->u.cli;
+ LASSERT(cli);
+ LASSERT(cli->cl_mgc_configs_dir);
+ rc = lustre_rename(cli->cl_mgc_configs_dir, cli->cl_mgc_vfsmnt,
+ temp_log, logname);
+ }
CDEBUG(D_MGC, "Copied remote log %s (%d)\n", logname, rc);
+out:
+ if (rc)
+ CERROR("Failed to copy remote log %s (%d)\n", logname, rc);
+ OBD_FREE(temp_log, strlen(logname) + 1);
RETURN(rc);
}
-DECLARE_MUTEX(llog_process_lock);
-
-/* Get a config log from the MGS and process it.
- This func is called for both clients and servers. */
-static int mgc_process_log(struct obd_device *mgc,
- struct config_llog_data *cld)
+/** Get a config log from the MGS and process it.
+ * This func is called for both clients and servers.
+ * Copy the log locally before parsing it if appropriate (non-MGS server)
+ */
+int mgc_process_log(struct obd_device *mgc,
+ struct config_llog_data *cld)
{
struct llog_ctxt *ctxt, *lctxt;
struct lustre_handle lockh;
struct client_obd *cli = &mgc->u.cli;
struct lvfs_run_ctxt saved;
- struct lustre_sb_info *lsi;
+ struct lustre_sb_info *lsi = NULL;
int rc = 0, rcl, flags = 0, must_pop = 0;
ENTRY;
- if (!cld || !cld->cld_cfg.cfg_sb) {
- /* This should never happen */
- CERROR("Missing cld, aborting log update\n");
- RETURN(-EINVAL);
- }
- if (cld->cld_stopping)
+ LASSERT(cld);
+
+ /* I don't want multiple processes running process_log at once --
+ sounds like badness. It actually might be fine, as long as
+ we're not trying to update from the same log
+ simultaneously (in which case we should use a per-log sem.) */
+ cfs_mutex_lock(&cld->cld_lock);
+
+ if (cld->cld_stopping) {
+ cfs_mutex_unlock(&cld->cld_lock);
RETURN(0);
+ }
- OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PROCESS_LOG, 20);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PAUSE_PROCESS_LOG, 20);
- lsi = s2lsi(cld->cld_cfg.cfg_sb);
+ if (cld->cld_cfg.cfg_sb)
+ lsi = s2lsi(cld->cld_cfg.cfg_sb);
- CDEBUG(D_MGC, "Process log %s:%s from %d\n", cld->cld_logname,
+ CDEBUG(D_MGC, "Process log %s:%s from %d\n", cld->cld_logname,
cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1);
ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
if (!ctxt) {
CERROR("missing llog context\n");
+ cfs_mutex_unlock(&cld->cld_lock);
RETURN(-EINVAL);
}
- /* I don't want mutliple processes running process_log at once --
- sounds like badness. It actually might be fine, as long as
- we're not trying to update from the same log
- simultaneously (in which case we should use a per-log sem.) */
- down(&llog_process_lock);
-
/* Get the cfg lock on the llog */
- rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL,
- LCK_CR, &flags, NULL, NULL, NULL,
+ rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL,
+ LCK_CR, &flags, NULL, NULL, NULL,
cld, 0, NULL, &lockh);
- if (rcl)
+ if (rcl)
CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl);
-
+
lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT);
- /* Copy the setup log locally if we can. Don't mess around if we're
+ /*
+ * local copy of sptlrpc log is controlled elsewhere, don't try to
+ * read it up here.
+ */
+ if (rcl && cld->cld_is_sptlrpc)
+ GOTO(out_pop, rc);
+
+ /* Copy the setup log locally if we can. Don't mess around if we're
running an MGS though (logs are already local). */
- if (lctxt && lsi && (lsi->lsi_flags & LSI_SERVER) &&
+ if (lctxt && lsi && (lsi->lsi_flags & LSI_SERVER) &&
(lsi->lsi_srv_mnt == cli->cl_mgc_vfsmnt) &&
!IS_MGS(lsi->lsi_ldd)) {
push_ctxt(&saved, &mgc->obd_lvfs_ctxt, NULL);
must_pop++;
- if (rcl == 0)
+ if (rcl == 0)
/* Only try to copy log if we have the lock. */
rc = mgc_copy_llog(mgc, ctxt, lctxt, cld->cld_logname);
if (rcl || rc) {
if (mgc_llog_is_empty(mgc, lctxt, cld->cld_logname)) {
- LCONSOLE_ERROR("Failed to get MGS log %s "
- "and no local copy.\n",
- cld->cld_logname);
+ LCONSOLE_ERROR_MSG(0x13a, "Failed to get MGS "
+ "log %s and no local copy."
+ "\n", cld->cld_logname);
GOTO(out_pop, rc = -ENOTCONN);
}
- LCONSOLE_WARN("Failed to get MGS log %s, using "
- "local copy.\n", cld->cld_logname);
+ CDEBUG(D_MGC, "Failed to get MGS log %s, using local "
+ "copy for now, will try to update later.\n",
+ cld->cld_logname);
}
/* Now, whether we copied or not, start using the local llog.
- If we failed to copy, we'll start using whatever the old
+ If we failed to copy, we'll start using whatever the old
log has. */
+ llog_ctxt_put(ctxt);
ctxt = lctxt;
}
- /* logname and instance info should be the same, so use our
+ if (cld->cld_is_sptlrpc)
+ sptlrpc_conf_log_update_begin(cld->cld_logname);
+
+ /* logname and instance info should be the same, so use our
copy of the instance for the update. The cfg_last_idx will
be updated here. */
- rc = class_config_parse_llog(ctxt, cld->cld_logname, &cld->cld_cfg);
-
- out_pop:
- if (must_pop)
+ if (rcl == 0 || lctxt == ctxt)
+ rc = class_config_parse_llog(ctxt, cld->cld_logname, &cld->cld_cfg);
+out_pop:
+ llog_ctxt_put(ctxt);
+ if (ctxt != lctxt)
+ llog_ctxt_put(lctxt);
+ if (must_pop)
pop_ctxt(&saved, &mgc->obd_lvfs_ctxt, NULL);
- /* Now drop the lock so MGS can revoke it */
+ /*
+ * update settings on existing OBDs. doing it inside
+ * of llog_process_lock so no device is attaching/detaching
+ * in parallel.
+ * the logname must be <fsname>-sptlrpc
+ */
+ if (cld->cld_is_sptlrpc && rcl == 0) {
+ sptlrpc_conf_log_update_end(cld->cld_logname);
+ class_notify_sptlrpc_conf(cld->cld_logname,
+ strlen(cld->cld_logname) -
+ strlen("-sptlrpc"));
+ }
+
+ /* Now drop the lock so MGS can revoke it */
if (!rcl) {
- rcl = mgc_cancel(mgc->u.cli.cl_mgc_mgsexp, NULL,
+ rcl = mgc_cancel(mgc->u.cli.cl_mgc_mgsexp, NULL,
LCK_CR, &lockh);
- if (rcl)
+ if (rcl)
CERROR("Can't drop cfg lock: %d\n", rcl);
}
-
+
CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n",
mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc);
- up(&llog_process_lock);
-
+ cfs_mutex_unlock(&cld->cld_lock);
+
RETURN(rc);
}
+/** Called from lustre_process_log.
+ * LCFG_LOG_START gets the config log from the MGS, processes it to start
+ * any services, and adds it to the list logs to watch (follow).
+ */
static int mgc_process_config(struct obd_device *obd, obd_count len, void *buf)
{
struct lustre_cfg *lcfg = buf;
switch(cmd = lcfg->lcfg_command) {
case LCFG_LOV_ADD_OBD: {
- /* Add any new target, not just osts */
+ /* Overloading this cfg command: register a new target */
struct mgs_target_info *mti;
- if (LUSTRE_CFG_BUFLEN(lcfg, 1) !=
+ if (LUSTRE_CFG_BUFLEN(lcfg, 1) !=
sizeof(struct mgs_target_info))
GOTO(out, rc = -EINVAL);
mti = (struct mgs_target_info *)lustre_cfg_buf(lcfg, 1);
- CDEBUG(D_MGC, "add_target %s %#x\n",
+ CDEBUG(D_MGC, "add_target %s %#x\n",
mti->mti_svname, mti->mti_flags);
rc = mgc_target_register(obd->u.cli.cl_mgc_mgsexp, mti);
break;
}
case LCFG_LOV_DEL_OBD:
- /* Remove target from the fs? */
- /* FIXME */
+ /* Unregister has no meaning at the moment. */
CERROR("lov_del_obd unimplemented\n");
rc = -ENOSYS;
break;
+ case LCFG_SPTLRPC_CONF: {
+ rc = sptlrpc_process_config(lcfg);
+ break;
+ }
case LCFG_LOG_START: {
struct config_llog_data *cld;
struct config_llog_instance *cfg;
char *logname = lustre_cfg_string(lcfg, 1);
cfg = (struct config_llog_instance *)lustre_cfg_buf(lcfg, 2);
sb = *(struct super_block **)lustre_cfg_buf(lcfg, 3);
-
- CDEBUG(D_MGC, "parse_log %s from %d\n", logname,
+
+ CDEBUG(D_MGC, "parse_log %s from %d\n", logname,
cfg->cfg_last_idx);
/* We're only called through here on the initial mount */
- rc = config_log_add(logname, cfg, sb);
- if (rc)
+ rc = config_log_add(obd, logname, cfg, sb);
+ if (rc)
break;
cld = config_log_find(logname, cfg);
if (IS_ERR(cld)) {
rc = PTR_ERR(cld);
break;
}
-
+
/* COMPAT_146 */
/* FIXME only set this for old logs! Right now this forces
us to always skip the "inside markers" check */
cld->cld_cfg.cfg_flags |= CFG_F_COMPAT146;
-
+
rc = mgc_process_log(obd, cld);
config_log_put(cld);
-
- break;
+
+ break;
}
case LCFG_LOG_END: {
struct config_llog_instance *cfg = NULL;
int __init mgc_init(void)
{
- return class_register_type(&mgc_obd_ops, NULL, LUSTRE_MGC_NAME);
+ return class_register_type(&mgc_obd_ops, NULL, NULL,
+ LUSTRE_MGC_NAME, NULL);
}
#ifdef __KERNEL__
class_unregister_type(LUSTRE_MGC_NAME);
}
-MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
+MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("Lustre Management Client");
MODULE_LICENSE("GPL");